PHM Society Asia-Pacific Conference

A Comparative Study of K-Means Clustering and a Novel Ranking Algorithm for Risk Priority Number Analysis in FMECA

Jiaxiang Cheng — 2023-09-04

Failure mode, effects, and criticality analysis (FMECA) has become a fundamental tool for identifying critical failure modes and prioritizing maintenance activities. As part of the analysis, the risk priority number (RPN), a numeric assessment of the risk, has received much attention as it is computed using severity (S), occurrence (O), and detectability (D), which serve as the main criteria for criticality analysis in many practical FMECA cases. In this paper, we assemble and present a data set containing RPN evaluations from 20 real-world cases. We then apply K-Means clustering to identify the most critical failure modes and propose a novel ranking algorithm that prioritizes mitigation actions based on specific criteria for each failure mode. Our experimental results suggest that both clustering and ranking methods can provide effective prioritization for critical failure modes under given assumptions, while our novel ranking algorithm can adapt to general scenarios and provide more accurate prioritization that can help develop effective maintenance strategies to minimize equipment failure risk and optimize maintenance costs.

A Concept of Condition Monitoring for AC-DC Converter Output Capacitors via Discriminative Features

Akeem Bayo Kareem — 2023-09-04

This paper discusses recent research on the condition mon- itoring (CM) approach for aluminium electrolytic capacitors (AEC) used in power electronics equipment such as switched- mode power supplies (SMPS). Capacitors are identified as the most critical component with the highest percentage of failure in AEC. CM offers a better paradigm for AEC due to its long- lasting ability (endurance). This study proposes accelerated life testing through electrical stress and long-term frequency testing for the AEC component. An experiment test bench was set up to monitor the critical electrical parameters such as dissipation factor (D), equivalent series resistance (ESR), capacitance (Cp), and impedance (Z), which serve as health indicators (HI) for the evaluation of the AECs. Time-domain features were extracted from the measured data, and the best features were selected using the correlation-based technique. This research contributes to developing a cost-effective CM approach for AECs used in power electronics equipment, which can reduce downtime and maintenance costs.

A monitoring method for detecting and localizing overheat, smoke and fire faults in wind turbine nacelle

Minsoo Lee — 2023-09-04

This study presents a monitoring method that utilizes 3D object classification to accurately detect mechanical and electrical components of a wind turbine by combining a geometric and statistic feature extractor (GSFE) with a multiview approach. The proposed monitoring method also detect outlier after executing object detection to localize overheat faults in these components with fused Optical or Infrared/LiDAR measurements. The proposed method has
three key characteristics. First, the proposed outlier detection allocates two extremes of normal and faulty clusters by using 2D object classification/detection model or measuring the
standard deviation of temperature with sensor fusing measurements. Specifically, the outlier detection with sensor fusing measurements extracts the position coordinates and
temperature data to localize overheat faults, effectively detecting an overheat component. Second, the GSFE utilizes a group sampling approach to extract the local geometric feature information from neighboring point clouds, aggregating normal vectors and standard deviation. This method ensures the high accuracy of object classification. Third, a multi-view approach focuses on updating local geometric and statistic features through a graph convolution network, improving the accuracy and robustness of object classification. The proposed outlier detection is verified through overheat/fire field tests. The effectiveness of the proposed 3D object classification method is also validated by using a virtual wind turbine nacelle CAD dataset and a public CAD dataset named ModelNet40. Consequently, the proposed method is practical and effective for monitoring a fire and overheat component because it can accurately detect critical components with only a few virtual datasets because gathering bigdata for training a neural network is extremely difficult.

A Multi-periodicity and Multi-scale Network for Motor Fault Diagnosis

Pengcheng Xia — 2023-09-04

Intelligent fault diagnosis of motor is of tremendous significance to ensuring reliable industrial production, and deep learning methods have gained notable achievements recently. Most researches automatically extracted fault information from raw monitoring signals with deep models, whereas the strong periodic temporal information containing in the signals were ignored. To tackle this limitation, a multi-periodicity and multi-scale network is proposed in this paper. 1D monitoring signals are transformed into 2D space with multiple various periods, allowing for the straightforward reflection and modeling of variations both within and between different periods. Multi-scale learning is introduced to extract temporal information from the multi-periodicity representations with multiple scales in a parameter-efficient way. Experiments carried out on a motor fault dataset verified the effectiveness of the proposed method. The results demonstrate that over 99% diagnosis accuracy can be achieved with onechannel vibration signals, and superior performance is obtained under diverse noise conditions compared with other methods.

A Physics Informed Machine Learning Approach for Performance Degradation Monitoring of Gas Turbine

Yiyang Liu — 2023-09-04

The heavy-duty gas turbine is playing an increasingly significant role on power generation due to its lower-emission, higher flexibility and thermo-efficiency. Main subsystems of the gas turbine like compressor, combustor and turbine degrade over the operating time under the harsh environmental conditions, which largely impacts the efficiency and productivity of the system. Therefore, it is critical to develop effective approaches to monitor performance degradation of a heavy-duty gas turbine for system predictive maintenance thus improving the efficiency and productivity of the machine. This paper presents a new physics informed machine learning methodology to predict the degradation of gas turbine by seamlessly integrating thermodynamic heat balancing mechanism, component characteristics, multi-source data and artificial neural network model. The mechanism-based thermodynamic model is established for multiple subsystems considering the balance of flow, mass and energy, and then integrated to a system level for performance simulation of the gas turbine under different conditions. The system model is able to effectively simulate values for those parameters that are not measurable (e.g. GT exhaust flow) or inaccurately measured (e.g. fuel flow). Machine learning based data cleaning approach is employed to preprocess the multivariate raw data of the gas turbine. The difference between design performance data and corrected value obtained from the physics-informed model under ISO conditions is utilized to assess the performance degradation. A Long Short-Term Memory (LSTM) model is established from the fusion of the actual and simulation data to predict the performance degradation of the gas turbine. A comparison study with the classical Nonlinear Autoregressive Network with External Input (NARX) neural network is conducted to demonstrate the advantage of the proposed method. Key Word: Gas Turbine, Thermodynamic Balance, Performance Degradation Predict, Machine Learning, LSTM

A Reliability-Centered Maintenance Framework for Distribution Grids Based on Fault-Tree Analysis

Ting Wang — 2023-09-04

Reliability is the key issue in the supply of electrical energy in modern society, which is jeopardized by the failures occurring in different sections of distribution grids. To address this challenge, this paper presents a reliability-centered maintenance framework for transformers, switchgear panels and power cables in medium-voltage distribution grids. First, fault tree models for the different equipment are established in this paper, with which the impacts of different failures and the effects of maintenance actions are analyzed in a quantitative manner. Using the fault tree models, the influences of different maintenance strategies on the reliability indexes of equipments in long-term operations can be estimated, which provides references for the selection and prioritization of preventive maintenance actions. This research work provides a generalized and practical framework for designing reliability-centered maintenance plans for distribution grids.

A Signal Pre-processing Method for Condition Monitoring based on Vibration Signals from On-Site Manipulators

Hea-Ryeon Seo — 2023-09-04

Handling irregular and noisy field data is challenging in condition monitoring. In contrast to refined lab data, where external influences are kept to a minimum, acquired signal from accelerometer attached to mechanical devices involves a great deal of uncontrollable variables. Especially, irregular operation cycles of the process make difficult to specify significant vibration signals for monitoring without mechanical expertise and information of the manipulators' motion. In this study, we distinguish motion signals from noisy raw signal using Shannon Energy Envelope (SEE). The extracted individual motion signals are algorithmically clustered through the signal graph characteristics for each robot motion. Clusters are evaluated for the effectiveness of monitoring, and it enables users to obtain a reference whose signal can perform the same accuracy for condition monitoring with expert knowledge.

A Simple Remaining Useful Life Algorithm Using the Quadratic

Eric Bechhoefer — 2023-09-04

The goal of predictive maintenance (PdM) is to facilitate on-condition maintenance or reduce/eliminate unscheduled maintenance events. For critical systems such as aircraft,
PdM improves safety while increasing operational readiness. Aircraft operators can order the parts and ensure the correct skills and tools are available to avoid unplanned downtime.
An enabler for PdM is the need to estimate the remaining useful life (RUL). For RUL to be accurate, there needs to be an assessment of the current component health, a threshold
for when it is appropriate to do maintenance, and a degradation model. This model could be based on some physical processes, such as high-cycle fatigue failure.
However, often the exact fatigue process is unknown. In this paper, a quadratic RUL model is used to calculate RUL using a state estimator. The proposed process allows for model
validation of the RUL state estimator itself. This is demonstrated using a bearing fault, a gear fault, and oil debris example.

A Simplified Framework for Fault Prediction in Radar Transmitter based on Vector Autoregression Model

Sheriff Murtala — 2023-09-04

The prediction of faults in radar subsystems remains a challenge. It is common practice to use multiple sensors to monitor the performance of electronic components in radar. The complexity of processing the measurements increases with the number of monitored quantities. In this paper, we presented a simple method to predict the fault degradation of radar transmitter. Using historical data of monitored quantities leading to two different faults, the vector autoregression model is applied to predict future values of monitored quantities resulting in fault degradation in marine radar. The results showed that the proposed method can be useful for cases where failure in subsystem needs to be promptly detected and corrected to avoid overall system failure. We also demonstrated the performance of the proposed method on interpolated data generated from radar transmitter fault data.

A Survey of Prognostics and Health Management for Transformers: From Statistical Analysis to Condition-Based Diagnostics

Jiaxiang Cheng — 2023-09-04

Power transformers are one of the key network components for reliable and efficient operation of power grids. Over the past few decades, there have been growing research efforts in improving the prognostics and health management (PHM) for transformers, including failure analysis using time-to-event data and condition-based diagnostics for both single and multiple components. In this paper, we survey recent literature and relevant works, focusing on widely used statistical models and advanced diagnostic techniques that leverage on condition data and maintenance history. Additionally, we examine the role of artificial intelligence (AI) applications in PHM for power transformers. Finally, we summarize the current limitations and future opportunities to support new research efforts for improving the monitoring of power transformers.

Abnormal Detection Using Two-Stage Method in Combined Power Plant

Yun Hee Kim — 2023-09-04

Complex systems, such as power plants, demand precise and reliable anomaly detection mechanisms. Traditional supervised learning approaches often fall short due to the challenges of imbalanced data and the scarcity of labeled abnormal instances. This paper introduces a two-stage methodology to address these challenges. The first stage emphasizes feature engineering, mitigating redundant sensor effects, and reducing dimensionality through Kmeans clustering and PCA. The second stage employs an LSTM-Autoencoder for abnormal event detection. Validated using data from a combined power plant, our approach demonstrates superior performance over existing techniques in terms of accuracy and computational efficiency. This research not only advances the field of anomaly detection in power plants but also offers insights for other complex systems.

Advanced condition-based monitoring of CFRP under multiple impacts using monte carlo based prognostics and real-time self-sensing data

Yong Lee — 2023-09-04

Studies on self-sensing system under multiple impacts are limited. Furthermore, real-time prognostics research using electromechanical behavior for impact-damage growth is rare and the impact-damaged area analysis has limited in self-sensing. In this paper, the health state of the carbon-fiber-reinforced plastic samples were monitored in real time utilizing self-sensing data. In-depth damage analysis using C-scan and cross-sectional analysis were conducted to investigate the correlation between the electromechanical behavior analysis. Moreover, the relationship between electromechanical behavior and the impact-damaged area was investigated. The damage propagation during multiple impacts was identified in real time. Furthermore, the electromechanical behavior was predicted to prognosticate the damage propagation in the samples under multiple impacts using a particle filter. The impact damage area was determined based on the predicted electromechanical behavior. Moreover, the prediction accuracy according to data acquired was investigated. An advanced condition-based monitoring methodology can monitor current and future health states and damage propagation under multiple impacts that overcomes the previous self-sensing research. Therefore, this study showed high applicability and guidelines for future self-sensing research fields.

Advanced Weibull Modelling with Outliers

Yipeng Pang — 2023-09-04

This paper presents a comprehensive process for the advanced Weibull modelling with potential outlier inclusions. In this process, an algorithm is designed to identify if there exist any outliers (i.e., failures with different failure modes from the majority) in the failure data of the equipment of interest. Depending on the conditions of the identified outliers, a suitable statistical model is developed. To validate the model, it is compared with the estimated empirical distribution function with the inclusion of new failure data. It is shown that the proposed advanced Weibull model outperforms the two-parameter Weibull model in terms of fitting, and hence a better accuracy is achieved in the failure statistical analysis. Case study in the application of power systems is conducted to illustrate its effectiveness.

AI image analysis technologies for efficient water pipeline inspection

Ying Piao — 2024-01-12

Inspection of water pipelines with cameras under pressure is attracting attention. The inspection can be performed without digging the ground and water interruption by inserting a camera into aging water pipes while the water is flowing. However, the inspection has two problems: (1) a long-time visual check by expert engineers is required and (2) variations in the evaluation standards. To solve these problems, we have developed an AI image analysis system for automatically judging the state of degradation of water pipelines by using images captured from the in-pipe endoscope cameras. This report describes the developed technology and software to support the inspection work.

An Enhanced Model-Based Algorithm for Early Internal Short Circuit Detection in Lithium-Ion Batteries

Yiqi Jia — 2023-09-04

Electric vehicles (EVs) are becoming more popular due to concerns about fuel shortages and environmental pollution. Lithium-ion batteries are the preferred power source for EVs because they have high energy and power densities. Ensuring the efficient, safe, and reliable operation of these batteries has been a significant focus of research in recent decades. One major concern that can affect Li-ion battery performance is thermal runaway, which can cause dangerous battery fires. Internal short circuits (ISCs) are believed to be the root cause of thermal runaway incidents in batteries, making early detection of spontaneous ISCs a critical diagnostic task. This study presents a new and simple early ISC detection method for a Li-ion cell based on the augmentation of the state space of an Extended Kalman Filter (EKF) that includes voltage and surface temperature observations. The framework allows for an estimation of the cell's internal ISC state while remaining computationally efficient. The proposed approach is demonstrated in a simulated environment using dynamic stress tests that reflect a practical battery working cycle. The results demonstrate that the method can promptly detect ISC occurrences.

An improved OAKR approach to condition monitoring of rotating machinery

Kexin Zhang — 2023-09-04

Faults in main subsystems or components of a rotating machine often causes unscheduled shutdown, which may lead to not only huge economic losses, but also safety accidents. As an important part of intelligent maintenance, condition monitoring becomes a powerful tool in reducing maintenance costs through automatic fault alarming, thereby reducing potential downtime while improving system safety and reliability. An optimized auto-associative kernel
regression (OAKR) model has been proposed recently and demonstrated as a promising tool for condition monitoring of various turbomachines, which is independent of fault mode and machine type. However, the fault identification accuracy of this approach largely relies on data quality in practical applications. Data incompleteness, parameter variation and system complexity often result in the inaccuracy of fault alarming for complicated rotating machinery. This paper proposes an improved OAKR method to address these issues, including utilizing wavelet packet Bayesian thresholding method (WPB) to reduce noise in the raw multivariate data, developing the Manhattan distance to calculate the sample similarity, and constructing a multivariate health index based on Multivariate Permutation Entropy to identify potential faults in equipment condition monitoring. Parametric analysis and a comparison study with original AAKR and OAKR methods by using the actual data of a gas turbine are conducted to illustrate the effectiveness and feasibility of the proposed methodology.

Analysis of Diagnostic Capabilities for Degradation of Brushless Direct Current Motors Depending on Varying Simulation Data

Max Weigert — 2023-09-04

As the use of unmanned aerial vehicles (UAVs) becomes more widespread and their missions more complex, the need for safety measures for their technical components is also increasing. Among the components that are critical for the operation of UAVs, Brushless Direct Current (BLDC) motors are particularly important. This is due to their compact design and low number of wear parts, which make them well-suited for use in UAVs. In this work, test rig and simulation data of BLDC motors degradation are utilized to investigate the capabilities and limitations of different machine learning algorithms. For this purpose, suitable features representing the motor behavior are discussed. Classification and regression tasks are applied to analyze both the fault type and the degradation progress. The simulated data allows for an assessment of the influence of noise and degradation progress on the diagnosis performance. Furthermore, characteristics of various fault types and the representation of their degradation process in the simulation are discussed. The database and the derived features are shared publicly.

Analysis of Statistical Data Heterogeneity in Federated Fault Identification

Zahra Taghiyarrenani — 2023-09-04

Federated Learning (FL) is a setting where different clients collaboratively train a Machine Learning model in a privacy-preserving manner, i.e., without the requirement to share data. Given the importance of security and privacy in real-world applications, FL is gaining popularity in many areas, including predictive maintenance. For example, it allows independent companies to construct a model collaboratively. However, since different companies operate in different environments, their working conditions may differ, resulting in heterogeneity among their data distributions. This paper considers the fault identification problem and simulates different scenarios of data heterogeneity. Such a setting remains challenging for popular FL algorithms, and thus we demonstrate the considerations to be taken into account when designing federated predictive maintenance solutions.

Anomaly data synthesis and detection via domain randomization

Joonha Jun — 2023-09-04

The demand for a large amount of data necessary for learning is increasing with the great development of artificial intelligence. The synthesis of engineering data is challenging in that it is not only to combine data, but also to proceed with data synthesis while keeping the engineering characteristics intact. To address this problem, this work proposes a synthesis and detection model of anomalous data utilizing domain randomization. This model learns data from existing systems to identify patterns and synthesizes new data by itself with domain randomization. The learned model can accurately detect anomaly data in the system in various environments.

Application of Model-based Deep Reinforcement Learning Framework to Thermal Power Plant Operation Considering Performance Change

Yutaka Watanabe — 2023-09-04

In recent years, there have been increasing expectations for the development of advanced plant operational support systems that can automate complex tasks and autonomously
optimize operational procedures in thermal power plants. The performance of the equipment changes during operation and maintenance; hence, it is necessary to adjust the operating process to satisfy the operational constraints. In this study, we investigated a framework based on model-based deep reinforcement learning for acquiring control methods that are robust to changes in equipment performance using a digital twin model. A case study of the operational planning of a thermal power plant was presented and it was demonstrated that a stable control system can be constructed even when plant characteristics are changing.

Architecting a Digital Twin-Based Predictive Maintenance System for Modelling Cable Joint Degradation

Raymon van Dinter — 2023-09-04

The large scale adoption of wind turbines and solar panels in the Netherlands places new demands on the medium voltage power grid. For example, highly varying loads can cause failures in certain cables. Cable joints are natural weak spots prone to faults due to varying currents, creating downtime challenges for public utility companies. Predictive maintenance (PdM) practices are necessary to minimize downtime for users. We present a Model-Based System Engineering approach using formal models and UML views to provide a scalable PdM design ontology for modeling cable joint degradation. We aim to monitor cable joint degradation from different manufacturers under varying conditions throughout the Netherlands in real-time using a Digital Twin (DT) approach. Our design provides high-resolution, real-time synchronization between the DT-based PdM system and the cable joints. The proposed architecture is scalable, robust, and flexible, and the software implementation is publicly available in an open-source repository.

Assessing the Performance of Transformer for Time Series Anomaly Detection

Takuto Nakashima — 2023-09-04

This study aims to assess the effectiveness of the Transformer-based reconstruction approach for detecting anomalies in time series data. The reconstruction error-based anomaly detection method was applied to both multivariate time series from NASA SMAP/MSL and univariate time series from UCR. Four deep learning models, including Transformer, Dilated CNN, LSTM, and MLP, were compared in terms of their ability to reconstruct input data. Dilated CNN outperformed the other models in almost all experimental results, achieving a 25% higher score than Transformer on the UCR dataset when trained with random masking, and a 60% higher score when trained with middle masking. These results suggest that the Transformer did not perform as well as expected for anomaly detection based on time series reconstruction errors, and its inferiority to Dilated CNN may be attributed to the characteristics of the time series and the limited training data. Future research should focus on developing Transformer models that can better capture the properties of time series data and investigate the relationship between the model’s performance, data volume, and model complexity.

Automatic detection of hardware failures in an air quality measuring station with low cost sensors

Sylvain Poupry — 2023-09-04

Monitoring air quality to protect the population is a challenge for cities with modest budgets. With this in mind, a measuring station has been developed using low-cost sensors (LCS) arranged in Triple Modular Redundancy (TMR). However LCS technology has limitations which lead to incomplete or inaccurate air quality measurements. To improve the availability of the measuring station, and also to make the data gathered more reliable, a fault detection method is proposed in this paper. By comparing measurements collected by the LCS in TMR configuration, the proposed method synthesizes measurements for each monitored parameter and assesses the health state of the measuring station in real-time. This information can be used to promptly alert maintenance teams, facilitating timely interventions and ensuring the continuous monitoring of air quality.

Automatic Generation of Seven-Segment Display Image for Machine-Learning-Based Digital Meter Reading

Kota Gushima — 2023-09-04

This paper presents a novel approach for automating the reading of seven-segment displays using machine learning, specifically addressing the concern of acquiring training data. By developing an algorithm that can automatically generate training images, the need for seven-segment display digit image acquisition was significantly reduced, making the process more efficient and cost-effective. In addition, by automatically generating images, a large amount of training data can be acquired. The training images include noise, such as sunlight reflections, shadows, and blurring due to camera shaking. The proposed method employs a machine-learning model trained on a diverse dataset of synthetic images generated by an algorithm. This dataset includes various fonts and styles, enabling the model to predict the meter values displayed on various fonts of the seven-segment liquid crystal display. By leveraging this auto-generated image set, the model effectively eliminates the labor-intensive process of manually capturing and annotating real-world meter images. The experimental results demonstrated the effectiveness of the proposed approach, with a reading accuracy of 96.8%.

Automating daily inspection for Expressways using anomaly detection model

Yuta Shirakawa — 2023-09-04

Because of the high speed at which vehicles travel on highways, even small irregularities on the road surface can lead to serious accidents. It is important to conduct daily visual inspections to detect these abnormalities at an early stage and to repair them quickly. We are considering replacing a part of visual inspection with automatic classification using image recognition. Automating inspections will make it possible to increase inspection frequency and expected to reduce the variation in quality due to the skill of inspectors. In this paper, we report on an AI-based inspection system and evaluation results using actual highway driving video captured by an in-vehicle camera.

Bayesian-based Component Lifetime Prediction Model Using Workshop and Telematics Data

Seungyoung Park — 2023-09-04

This paper presents a Bayesian approach to predicting brake pad and battery life based on field service data from a fleet management system(FMS). The data includes vehicle driving data collected via telematics and maintenance record data managed by the workshop. The proposed approach consists of three modules: component health diagnosis, workshop data analysis and driving pattern analysis. The health diagnosis module detects domain-based transformed feature, from the driving data, changes using KL divergence. The maintenance record data from workshop analysis module estimates the prior probability of maintenance cycles. The censored nature of workshop data is validated by updating the posterior probability using driving patterns from driving data. The driving pattern analysis module classifies driving patterns for lifetime prediction. This study develops a predictive maintenance model for brakes and batteries without additional sensors using the data required for fleet operation. The mileagebased cycle maintenance approach commonly used for fleet management is improved by this model. Future FMS systems are expected to make extensive use of this concept.

Bridging the Gap: A Comparative Analysis of Regressive Remaining Useful Life Prediction and Survival Analysis Methods for Predictive Maintenance

Mahmoud Rahat — 2023-09-04

Regressive Remaining Useful Life Prediction and Survival Analysis are two lines of research with similar goals but different origins; one from engineering and the other from survival study in clinical research. Although the two research paths share a common objective of predicting the time to an event, researchers from each path typically do not compare their methods with methods from the other direction. Given the mentioned gap, we propose a framework to compare methods from the two lines of research using run-to-failure datasets. Then by utilizing the proposed framework, we compare six models incorporating three widely recognized degradation models along with two learning algorithms. The first dataset used in this study is C-MAPSS which includes simulation data from aircraft turbofan engines. The second dataset is real-world data from streamed condition monitoring of turbocharger devices installed on a fleet of Volvo trucks.

Change Point-based Spatio-temporal Process Modeling of Image Degradation for Manufacturing Process

Munwon Lim — 2023-09-04

As an advent of smart factory technology, data-driven condition-based maintenance (CBM) is developed to automatically control the production process in engineering field. CBM usually focuses on diagnosing the production status based on real-time data from the sensors. In general manufacturing field, the performance of production equipment gradually decreases due to the wear or deterioration of equipment. To determine if the process is in-control, degradation modeling of observed data from the equipment and its statistical inference is conducted. In this paper, we propose image-based degradation modeling and change-point detection using spatio-temporal process (STP). To describe the deteriorating patterns of image observation, degradation based on spatial and temporal relationship is conducted. At the same time, change-point is estimated to distinguish the degradation under normal and abnormal production status. Through the application to the image stream in real industry, the proposed monitoring scheme `vely conduct the bi-phase representation providing the change-point of manufacturing processes.

Comprehensive Failure Diagnosis Model with Degradation Indicators of Multiple Sensors

Jun Tominaga — 2023-09-04

Although monitoring system can detect abnormality of sensor reading in air-conditioning equipment, the root cause of the abnormality may not be sensor failure but other failures such as gas shortage. We propose new method that estimates the cause by the following steps. Firstly, regression model predicts the normal readings of multiple sensors (e.g., thermistor) for a given operational condition. Secondly, the gap between measured and predicted values is calculated for each parameter as a degradation indicator. Finally, our failure diagnosis model estimates the cause by considering degradation indicators of multiple sensors. Our evaluation verifies the effectiveness of our method.

Configuration and Comparative Study of Prediction Models for Indoor Air Quality

Geonhwi Lee — 2023-09-04

Since COVID-19, cultural life and working conditions have changed to be done indoors. Various harmful substances are produced indoors, and when they enter the human body through the air, they can cause serious diseases. Indoor air pollution is not visible to the naked eye, and it is not easy for people to perceive it. Human damage due to harmful indoor gases is increasing. In this study, we predict indoor air pollution occurring in daily life in advance. We collected indoor air quality data every 10 seconds from the different types of residential spaces in Seoul. For accurate prediction, we compared the prediction performances of various models, such as the ARIMA model, and the recurrent neural network (RNN) based models. In addition, the prediction performances were compared according to the size of the historical window. The comparison results revealed that for short-term interval predictions, shorter historical window sizes and simpler models were more effective. This study provides a baseline for selecting a predictive model and configuring training datasets.

Deep Learning-Enabled Statistical Model Estimation for Power Transformers with Censoring and Truncation Problems

Jiaxiang Cheng — 2023-09-04

Traditional statistical models, e.g., Weibull distributions, are popular solutions for failure modeling and degradation anal- ysis in a variety of industries. To estimate the parameters of these statistical models, maximum likelihood estimation (MLE) is often engaged through various optimization algo- rithms. However, when dealing with highly reliable or new equipment, it is challenging to fit limited or unbalanced data to obtain an accurate model. In this paper, we propose a deep learning (DL)-based model for estimating the Weibull param- eters with both censoring and truncation problems. Instead of using the conventional matrices such as concordance index, we propose a novel validation framework to examine the pre- diction accuracy of different models. We examine the perfor- mance of the proposed approach on real-world power trans- former data, and the results show that our approach can im- prove prediction accuracy and is less susceptible to the trun- cation problem. Our results also suggest that deep learning techniques can help enhance traditional statistical modeling for reliability analysis.

Demonstration of Sensor Monitoring of Lubricants

Kyoko Kojima — 2023-09-04

Due to the spread of carbon neutral, the effective use of lubricant has been drawing attention. Since the main component of lubricant is petroleum-derived hydrocarbon oil, reducing the amount used by 1 kg will reduce CO2 by approximately 3 kg. The value of CO2 reduction is very important. In order to reduce the amount of lubricant used, there is a movement to reduce the frequency of lubricant exchange or continue to use lubricant without exchanging it. However, it is known that lubricant-induced mechanical failures occur. For this reason, equipment condition monitoring using oil sensors has been spread. The color of the lubricant, also called machine blood, indicates the condition of the machine. The oil sensor measures contamination, which has a fatal effect on machine failure, and oxidation degradation, which is related to the performance of lubricant and the machine failure. Contamination includes water and wear debris, and oxidative degradation includes consumption of additives and oxidation of base oil. By digitizing the hue of wind turbine gear oil through color diagnosis using an oil sensor, the oil contamination and degradation is identified. Additives in the gear oil were quantified by liquid chromatography-mass spectrometry, and it was found that the color change of gear oil was highly correlated with the depletion of the extreme pressure additive. It is known that the depletion of the extreme pressure additive is correlated with the useful life of the gear oil. Using the technique, the remaining life diagnosis of the gear oil was shown. Demonstration in the gearbox with the oil sensor was succeeded by avoidance of the effects of air bubbles in the gear oil.

Detection and Diagnostic with Random Forest Classifier (RFC) to Improve the Maintenance Management System in Steam Boiler of Power Plant

Ghiffari Awliya Muhammad Ashfania — 2023-09-04

Industrial internet of things (IIoT), digital twin, and connected devices can continue to use smart equipment and improve access to data. While the data collected by sensors has been an invaluable asset to companies, the ability to understand and use this data to drive new insights. The development of Condition Monitoring (CM) technology and Computerized Maintenance Management System (CMMS) in power generation systems provides a validated set of operation and maintenance data with abundant event data. Maintenance decision-making is primarily based on equipment reliability and performance-based features for diagnosing equipment failure. The most critical asset and often reduces the reliability and availability of a Coal Fired Steam Power Plant (CFSPP) with the most frequency of disturbances is the steam boiler. As a departure from the idea of creating a digital twin, this article will focus on analyzing equipment health conditions and finding causes of failure of the tools, utilizing data for diagnostic purposes. Real-case used in this research are steam boilers, which are important assets in power plant generation. The online and Failure Mode and Effect Analysis (FMEA) module data will be combined to realize the concept of anomaly diagnosis which is driven by hybrid data. Hoping that accurate diagnosis result with the Random Forest Classifier (RFC) Algorithm can be obtained and be used to analyze the causes of failure and decrease in equipment performance resulting by a decrease of energy efficiency performance. The analytical approaches are carried out to have the goal of generating detection models and diagnostic insights of event data based on operational data and FMEA.

Development of an Operational Digital Twin of a Locomotive Braking System Solenoid Valve for Fault Classification

Gabriel Davidyan — 2023-09-04

In recent years, a growing role in digital technologies has been filled by model-based digital twinning. A digital twin produces a mapping of a physical structure, operating in the digital domain. Combined with sensor technology and analytics, a digital twin can provide enhanced monitoring, diagnostic, and optimization capabilities. This research harnesses the significant capabilities of digital twining for the unmitigated challenge of fault type classification of a locomotive braking system solenoid valve. We develop a digital twin of the solenoid valve and suggest a method for fault type classification based on the digital twin. The diagnostic ability of the approach is demonstrated on a large experimental dataset.

Differential Diagnosis with Active Testing

Emile van Gerwen — 2023-09-04

The diagnosis of complex systems benefits greatly from a differential, multistep approach that narrows down the list of possible conditions or failures that share the same observable effects to a single root cause. We provide a suitable and practically applicable methodology for this. In extension to existing work, it covers all types of diagnostic actions, i.e., the observation of system properties, active testing and system interventions like providing a dedicated diagnostic input or forcing the system into discriminating states, but also the replacement of components. Combining all these possible steps into one probabilistic and causal reasoning framework, we I) stepwise generate the diagnostic model systematically to correctly cover the interplay of observations and diagnostic interventions, and II) provide decision support based on counterfactuals for the selection of the next diagnostic step, countering the vast number of possible actions that arise in machine diagnostic processes. We developed and successfully tried our methodology for diagnosing cyber-physical systems in the high-tech industry, but we found that it supports more processes, such as computing intervention actions for autonomous robots.

Discovering Premature Replacements in Predictive Maintenance Time-to-Event Data

Abdallah Alabdallah — 2023-09-04

Time-To-Event (TTE) modeling using survival analysis in industrial settings faces the challenge of premature replacements of machine components, which leads to bias and errors in survival prediction. Typically, TTE survival data contains information about components and if they had failed or not up to a certain time. For failed components, the time is noted, and a failure is referred to as an event. A component that has not failed is denoted as censored. In industrial settings, in contrast to medical settings, there can be considerable uncertainty in an event; a component can be replaced before it fails to prevent operation stops or because maintenance staff believe that the component is faulty. This shows up as “no fault found” in warranty studies, where a significant proportion of replaced components may appear fault-free when tested or inspected after replacement.

In this work, we propose an expectation-maximization-like method for discovering such premature replacements in survival data. The method is a two-phase iterative algorithm employing a genetic algorithm in the maximization phase to learn better event assignments on a validation set. The learned labels through iterations are accumulated and averaged to be used to initialize the following expectation phase. The assumption is that the more often the event is selected, the more likely it is to be an actual failure and not a “no fault found”.

Experiments on synthesized and simulated data show that the proposed method can correctly detect a significant percentage of premature replacement cases.

Elastic Wave Field Neural Networks for Structural Health Monitoring: An Analytical and Numerical Study of Multiple Neurons

Arata Masuda — 2023-09-04

The purpose of this study is to develop a novel concept of smart structural systems recognizing their own structural integrity by an embodied high density sensor network. In our concept, a number of sensor nodes are embedded in the host structure, each of which reacts point-wise to the structural vibration with a simple rule. This allows the whole nodes to be mutually coupled through the elastic field, forming a neural network that incorporates the dynamic characteristics of the host structure as the coupling weights. In the previous study, we presented that a single-neuron network as its minimum configuration can exhibit a bifurcation of its dynamics behavior, which can be used to detect the change of the network due to damages. In this study, the formulation of networks with multiple neurons deployed in a structure with single-mode approximation is presented particularly focusing on the bi- furcation analysis to reveal how the behavior of the network is drastically altered depending of the network and structural parameters. Numerical analysis is conducted to examine the validity of the bifurcation analysis.

Energy Saving Structural Health Monitoring Using Semi-Active Identification

Yushin Hara — 2023-09-04

This paper presents a novel approach to achieving system identification of a structure while minimizing energy consumption. The identified structural model can be used for structural health monitoring. In the aerospace environment, energy consumption is strictly regulated. To address this issue, we propose an energy-saving identification method that utilizes piezoelectric semi-active control as an input generation technique. This approach generates control force through electric switch activation, resulting in a smaller amount of energy consumption for input generation than conventional active control. We achieved semi-active input generation suitable for identification by incorporating a novel control strategy. The semi-active control has the disadvantage of limiting the free control of inputs. The identification performance may degrade if the properties of the semi-active input deviate from the desired ones. To address this issue, we also propose a data processing method that extracts a certain input with appropriate properties for identification from the acquired input. We validated the proposed method through numerical simulations and experiments. The results confirmed the feasibility of the semi-active identification method for structural health monitoring. Additionally, we found that the total energy consumption during the 20-second experiment was only 68 mJ to identify the 50 kg structure.

Ensemble Neural Networks for Remaining Useful Life (RUL) Prediction

Abhishek Srinivasan — 2023-09-04

A core part of maintenance planning is a monitoring system that provides a good prognosis on health and degradation, often expressed as remaining useful life (RUL). Most of the current data-driven approaches for RUL prediction focus on single-point prediction. These point prediction approaches do not include the probabilistic nature of the failure. The few
probabilistic approaches to date either include the aleatoric uncertainty (which originates from the system), or the epistemic uncertainty (which originates from the model param-
eters), or both simultaneously as a total uncertainty. Here, we propose ensemble neural networks for probabilistic RUL predictions which considers both uncertainties and decouples these two uncertainties. These decoupled uncertainties are vital in knowing and interpreting the confidence of the predictions. This method is tested on NASA’s turbofan jet engine CMAPSS data-set. Our results show how these uncertainties can be modeled and how to disentangle the contribution of aleatoric and epistemic uncertainty. Additionally, our approach is evaluated on different metrics and compared against the current state-of-the-art methods.

Evaluation of Multi-Modal Learning for Predicting Coolant Pump Failures in Heavy Duty Vehicles

Yuantao Fan — 2023-09-04

Coolant Pump failures in heavy-duty vehicles can cause severe collateral damage if they are not detected and resolved in time; the engine will overheat quickly, rendering the vehicle inoperable. Nowadays, a vast amount of heterogeneous sensor data from different sources is being collected in the automotive industry. Such multi-modal data include onboard signals reflecting the overall usage of the vehicle, multi-dimensional histograms that capture the relation between physical quantities, and categorical variables that encode the physical configuration of the vehicle. This work evaluates several multi-modal learning approaches leveraging this diverse data to build a prognosis and health management system for coolant pumps in commercial heavy-duty vehicles. Four auto-encoder architectures are examined to extract features from 2D histograms. These trained models are anticipated to capture key characteristics of the healthy system operation and yield large reconstruction errors when applied on faulty, or near end-of-life samples. Such learned representations are then combined with expert-engineered features. Both early and intermediate fusion are evaluated on a real-world coolant pump replacement dataset. Results indicate that the combination of diverse features was the most effective approach, thereby motivating further research on multimodal methods.

Event detection in a noisy time series data using static smoothening and gradient variation analytics

Saravanaram T — 2023-09-04

Data-driven prognostic industrial asset health management is essential to improve reliability and availability of industrial machineries. Industrial sensors capture multiple phenomena of underlying assets like environment impact, system degradation, process variation, instrument noise, control system response or user induced actions. Some of these phenomena have distinct signature and have impact on component life and remaining useful life. Capturing these events’ signature help to apply advanced AI algorithm to categorize various failure modes and early detection. For Ex: Some of transient events can contribute to thermal cycling of component and in turn reduces life of high temperature and low thermal mass components. Due to random and nonstandard nature of these events, it is extremely challenging to detect and extract these events. Existing change point detection algorithms have limitations to detect sudden variations, which is common due to process or control actions. The noisy signal adds additional challenge to differentiate between important event and noise. In this work of time series analysis, we propose a new approach for consistent estimation of numbers and locations of the change-points. With this tunable algorithm combined with event labelling and pattern search, we can detect events of our functional needs and use them as a feature for our prediction models. This methodology has opened exciting opportunity to further analyse these events with development of classification system and time to fail prediction models and also apply large language models for time series data.

Explainable multimodal learning for predictive maintenance of steam generators

Duc An Nguyen — 2023-09-04

Prognostics and Health Management (PHM) is identified as an important lever for enhancing the development of predictive maintenance to ensure the reliability, availability, and safety of industrial systems. However, the efficiency of data- driven PHM approaches is dependent on the quality and quantity of data. Therefore, exploiting multiple data sources can provide additional, useful information than single-modal data. For instance, by incorporating multiple data sources, including condition monitoring data, images from cameras, and texts from maintenance technicians’ reports, multi-modal learning can provide a more comprehensive and accurate understanding of the system’s health. However, multi-modal deep learning is complex to understand. To address this complexity, it is crucial to incorporate explainable artificial intelligent techniques to provide clear and interpretable insights into how the model makes decisions. In this light, this paper proposes the application of the model-agnostic-explanation approach, i.e., SHAP, to explain the working mechanism of multimodal learning for the prediction of industrial steam generator degradation. Particularly, we determine the important features of each data modality and investigate how multimodal learning can overcome the issues of low-quality data from a single modality due to the additional information from other data modalities.

Failure Prediction of Hard Disk Drives in Redundant Arrays Using Disk-Level Performance Metrics

Masanao Natsumeda — 2023-09-04

Although many hard drive failure prediction methods utilize Self-Monitoring Analysis and Reporting Technology (SMART) features, they are not collected in IT systems with demanding performance requirements to meet their specification. We present a novel data-driven method for the prediction utilizing disk-level performance metrics collected by Redundant Array of Independent Disk (RAID) controllers instead of SMART features. The proposed method computes relational anomaly scores leveraging logical relationships of Hard Disk Drives (HDDs) based on RAID configuration for better failure prediction. In addition, it further utilizes error codes from HDDs to filter out false positives. We evaluate the proposed method on a real-world dataset collected for this study from 881 disks used in disk arrays of RAID-6 and 1660 disks used in disk arrays of RAID-10 in a data center. The results show consistent performance improvement by the logical relationships and error-code-based filtering. In addition, seven out of nine failures are predicted one day before the failure at the latest. This result suggests that the proposed method provides plenty of time for HDD replacement before a failure occurs.

Fully Unsupervised Defect Clustering using Adversarial Autoencoder

Taewan Kim — 2023-09-04

Identifying defect types and developing proper maintenance strategies is a major concern in modern industry. Most conventional studies have been conducted primarily based on a supervised learning scheme. However, supervised learning has a critical limitation in that it requires labeled data, which is difficult and expensive to obtain in real-world industry. Considering that there are many industries that do not perform post investigations on the defects, fully unsupervised learning methods, which do not exploit any information such as label data or the number of types, need to be developed. Accordingly, in this study, we propose a fully unsupervised defect clustering method that does not exploit any information other than the data itself. The proposed method consists of two major components. The first is dimensionality reduction into latent space via adversarial autoencoder, and the second is a Bayesian mixture model for distribution estimation in latent space. The experiments on a rolling-element-bearing dataset validate the effectiveness of our method. Specifically, our method performs defect clustering without any information other than the data itself, which is promising for real industrial applications.

Fusion with Joint Distribution and Adversarial Networks: A New Transfer Learning Approach for Intelligent Fault Diagnosis

Xueyi Li — 2023-09-04

Bearings and gears are important components in rotating machinery, and the diagnosis of faults in bearings and gears has always been an important topic. Currently, data-driven fault diagnosis is a better method. However, under actual working conditions, domain shift can easily occur due to different operating conditions, leading to difficulties in transfer learning and significantly reducing the diagnostic performance of the model. Re-labeling the fault types of the model is time-consuming and costly. To overcome these difficulties, a new unsupervised transfer learning framework based on the fusion of joint distribution and adversarial networks has been introduced for the fault diagnosis of bearings and gears in rotating machinery. The joint adaptation network learns the transfer network by aligning the joint distribution of multiple specific domain layers across domains, based on Joint Maximum Mean Discrepancy (JMMD) to achieve domain alignment. At the same time, the domain classifier in the adversarial network is used to minimize the domain classification loss as domain distribution difference to minimize domain shift. The fusion of these two methods achieves domain alignment, reduces model training time, and improves the accuracy and stability of the model. The experimental results demonstrate that the proposed model framework exhibits excellent performance in detecting and classifying different types of faults. The new model framework also demonstrates outstanding performance across various fault detection and classification tasks.

Fuzzy-membership-based labeling: a new labeling method for both classification task and regression task

Diwang Ruan — 2023-09-04

In the machine learning and deep learning field, there are two main kinds of tasks: classification and regression. The label for the former is discrete, while for the latter is continuous. Due to the big gaps in labels, these two tasks are generally re solved separately, bringing low training efficiency and waste of computing resources. To this end, this paper proposes a new labeling method based on fuzzy membership. The main idea is to build an intermediate variable, which behaves between continuous and discrete variables. Then, the relation between the intermediate variable and the discrete label can be identified with fuzzy membership. Finally, the fuzzy membership is adopted for building labels to train the source model. After training, the source model can be transferred to achieve both classification and regression tasks. To validate the new labeling method, two typical tasks in the PHM field, aging stage classification and RUL prediction, are selected as the representative for classification and regression tasks, respectively. Furthermore, LSTM with two dense layers is chosen as the benchmark source model. With the C-MAPSS dataset, the superiority of the proposed fuzzy-membership based labeling to improve the network’s task transfer learning performance has been verified.

Health Monitoring of Power Semiconductor Module Using Temperature Sensitive Electrical Parameter

Guesuk Lee — 2023-09-04

Power semiconductor modules (PSMs) are critical components in power electronics applications such as motor drives, renewable energy systems, and electric vehicles. The reliable operation of these modules is crucial for the safe and efficient operation of these systems. One of the most common failure mechanisms in PSMs is due to overheating or repeated heating and cooling, which can result in thermal stress and component degradation. Therefore, monitoring the temperature of PSMs is essential for ensuring their health and preventing catastrophic failures.

By monitoring temperature-sensitive electrical parameters (TSEPs), such as the on-state voltage drop, it is possible to detect changes in the temperature of the PSM in real-time. The change in voltage drop can be used as an early warning sign of a potential failure or degradation of the PSM. The advantage of this method is that it provides a non-invasive and real-time monitoring solution that can detect changes in the PSM's temperature distribution.

Improving Anomalous Sound Detection by Distance Matrix-Based Visualization of Measurement Flaws

Nobuaki Tanaka — 2023-09-04

Although recent DNN-based methods have improved the performance of anomalous sound detection systems, it is still difficult to deploy a system in a real environment without performance degradation. This is often due to measurement flaws such as sensor variability, poor setup, or environmental noise. Since such adverse effects are difficult to model by machine learning, a practical approach to this issue is for humans to identify such flaws and correct them. To this end, we propose a method to visualize measurement flaws as a heatmap based on the distance matrix of the samples in the dataset. This method is designed to find unexpected flaws in the measurement process. Using this method, we were able to identify measurement flaws of anomalous sound detection systems in real production lines. The robustness of anomalous sound detection can be improved by correcting the flaws found by our method.

Long-Term Preventive Failure Mitigation Strategy For Transformers Based on Markov Method

Jianzheng Wang — 2023-09-04

In this paper, we propose a preventive failure mitigation strategy in the power system based on Markov method. Specifically, we consider multiple units in the system, which are of different types and are managed by a single utility company. To characterize the operation, failure mitigation, and deterioration processes of the equipment, a continuous-time Markov model is formulated. By modelling the failure rate of equipment and the reinstallation rate after failures, the steady state of the proposed Markov model is analytically derived. Then to optimize the long-term net revenue of the utility company, the optimal failure mitigation rate is determined by considering the failure mitigation capacity for each equipment type as well as the overall failure mitigation capacity of the company. The performance of the proposed algorithms is demonstrated with three types of transformers in the simulation.

Multi-Label Fault Diagnosis of Rotary Machine via Domain Adversarial Neural Network-Based Domain Generalization Targeting High Range Rotating Speed

Sukeun Hong — 2023-09-04

Research on deep learning has been increasing in recent years in the field of fault diagnosis in rotary machine. However, compared to training data, real world data is collected from different system conditions and environments. Therefore, real world data has different data
distribution and various noise with the training data, leading to domain shift between data. Due to the problem mentioned above, deep learning often fails to apply on industrial data.
Domain generalization is an emerging deep learning technique to generalize domain discrepancy. In this study, domain adversarial neural network (DANN)-based domain
generalization is proposed for multi-label fault diagnosis of rotary machine. Frequency domain image data were generated via implementing short time fourier transform
(STFT) to the sensor data collected from the test rig. Then, the features are utilized as training data to diagnosis multi-label fault via DANN-based domain generalization. Moreover, the upper boundary of rotating speed domain of the rotary machine where domain generalization can effectively diagnosis multi-label fault is suggested.

Multiphysics-informed DeepONet of a lithium-ion battery to predict thermal runaway

Jinho Jeong — 2023-09-04

This study proposes a multiphysics-informed deep operator network (MPI-DeepONet) to predict the thermal runaway of lithium-ion batteries (LIBs) under a variety of thermal operational and abuse conditions. Specifically, this study aims to address the functional mapping from a heating curve to predict the evolution of the temperature of a LIB and dimensionless concentration of dominant components of the LIB including an anode, cathode, electrolyte, and solid electrolyte interphase. The proposed method has two key characteristics. First, the MPI-DeepONet is supervised by using ordinary and partial differential equations, which govern highly complex and nonlinear phenomena of thermal runaway of a LIB, including the chemical reaction degradation of the dominant four components and thermodynamics. This feature enables to train of the proposed neural network with a small amount of data available, suggesting that the proposed neural network is accurate and robust even though the proposed method is trained even with limited data. Second, the proposed neural network is trained with the data that is generated from high-fidelity finite element analysis under a variety of thermal operational and abuse conditions because measurements for the thermal runaway of a LIB are limitedly available. Hence, the MPIDeepONet does not require actual measurements, which is extremely difficult in field experiments. Finally, the accuracy and robustness of the proposed architecture are verified through actual measurements and other scenarios, which are different from the data trained. The analysis of results reveals that the MPI-DeepONet secures higher accuracy and robustness than purely data-driven DeepONet. The proposed surrogate model, which is faster than existing surrogate models, suggesting that this model contributes to developing a digital twin model of a LIB, which can be deployed on a battery thermal management system and provides sufficient information for effective power and energy management.

Novel Ensemble Domain Adaptation Methodology for Enhanced Multi-class Fault Diagnosis of Highly-Connected Fleet of Assets

Takanobu Minami — 2023-09-04

This paper proposes a novel methodology for enhancing multi-class classification accuracy in fault diagnosis problems among domains with highly-connected fleets of assets using time series data. The approach involves appending specially tailored models to an initial model and incorporating domain adaptation techniques to account for domain variations. The methodology is demonstrated through a case study on fault diagnosis of a fleet of hydraulic rock drills, which presents challenges due to variations in sensor data between different fault classes and individual machines. Results show significant improvements in classification accuracy, both in validation and testing, upon employing ensemble models and applying domain adaptation. While the study is limited to one case study, it lays the groundwork for exploring the applicability of the proposed methodology in other contexts.

Observation and Prediction of Instability due to RD Fluid Force in Rotating Machinery by Operational Modal Analysis

Daiki Goto — 2023-09-04

In rotating machinery, rotordynamic (RD) fluid force is generated in fluid elements such as journal bearings, impellers, and seals. This RD fluid force is generated by the interaction of shaft vibration and fluid force and is difficult to predict because of nonlinearity. The RD fluid force has a significant effect on the stability of rotating machinery and is known to cause instability. In this study, we predicted the instability caused by RD fluid force by using operational modal analysis, which is a method to identify vibration characteristics based only on operating data.

Online fault detection for industrial processes through Kalman filter

Wenyi Liu — 2023-09-04

Industrial processes suffer from a wide range of damages including normal wear, environmental changes, physical structural defects and so on. This paper describes the possibility of system health management based on a prediction model, i.e., state space model realized by Kalman filter. The categorical target was mapped to numerical values in advance for this purpose. To deal with the time-varying and streaming characteristics of the industrial process, the model is applied in an online fashion. Comparing with conventional fault detection techniques, this model has the advantages of monitoring not only the production process of interests through observation equation, but also the structural anomalies described via unseen states estimation. In addition, the process and measurement noises provide valuable information about the unstructured uncertainties caused by other reasons. Experiments have been conducted to valid the effectiveness of the proposed method.

Physical Reservoir-Based Health Monitoring of a Structure with Nonlinear Attachments

Arata Masuda — 2023-09-04

The purpose of this work is to discuss the possibility of the concept of physical reservoir computing (PRC) in the field of structural health monitoring (SHM) by regarding the target structure of SHM as the physical reservoir. To this end, the dynamics of the structure, which is assumed extrinsically linear, is tailored to be strongly nonlinear by installing nonlinear attachments. Our purpose is then to detect the change occurred in this augmented physical reservoir. As one possible methodology to achieve this, we propose in this study to train the output layer to learn a specific nonlinear mapping of the input so that the increase of the error may indicate the change of the reservoir. Numerical experiments are presented to examine the validity of the proposed concept.

Prediction of Impact Information of Composites Laminates Considering the Practicality

Saki Hasebe — 2023-09-04

Recently, carbon fiber reinforced plastics (CFRP) have been used in various applications, including aircraft. Because they are vulnerable to out-of-plane loads, internal and external
damage occurs when foreign objects impact them. Internal damage that can affect residual properties is difficult to find and judge from the outside without special devices, which are
highly costed and are sometimes difficult to conduct in some locations. In this study, surface contour information was obtained from impact tests on CFRP laminates, and the
predictability of compression after impact (CAI)strength was investigated using a conventional single-task random forest model, and a decision tree-based multi-task learning model with other objective variables related to impact tests. The models estimated CAI strength with around 75% R2, and the conventional single-task learning model showed the highest value. The importance of each model indicated that factors that contribute to impact-related objective variables (impactor shape, delamination area, and delamination length) and those to CAI strength do not have a strong relationship.

Prognosis using Bayesian Method by Incorporating Physical Constraints

Hyung Jun Park — 2023-09-04

Accurately predicting the remaining useful life (RUL) of industrial machinery is crucial for ensuring their reliability and safety. Prognostic methods that rely on Bayesian inference, such as Bayesian method (BM), Kalman and Particle filter (KF, PF), have been extensively studied for the RUL prognosis. However, these algorithms can be affected by noise when training data is limited, and the uncertainty associated with empirical models that are used in place of expensive physics models. As a result, this can lead to significant prediction errors or even infeasible RUL prediction in some cases. To overcome this challenge, three different approaches are proposed to guide the Bayesian framework by incorporating low-fidelity physical information. The proposed approaches embed inequality constraints to reduce sensitivity to local observations and achieve robust prediction. To determine an appropriate approach and its advantageous features, performance is evaluated by both numerical example and real case study for drone motor degradation.

Prognostics in Highly Accelerated Limit Testing Using Deep Learning Data Analysis

Tadahiro Shibutani — 2023-09-04

In this study, an anomaly detection analysis of electronic components was conducted using deep learning algorithms on time-series data of voltage monitored during highly accelerated limit testing (HALT) on inverters used in automobiles and other vehicles. We demonstrated that the anomaly detection technology of time-series data using deep learning could detect equipment anomalies/failures to achieve effective data representation, improving the reliability assurance technology with HALT.

Quantum Optimization for Location Assignment Problem in ASSR

Kuniaki Satori — 2023-09-04

In an Automated Storage and Retrieval System (AS/RS), a location assignment of products is important to improve the picking efficiency. In this paper, the optimization of shelf location assignment with a quantum annealing is investigated. Product pairs are considered in order of picking frequency and are assigned to empty shelves in order of distance from an outlet. Then swapping the position of product in the pair is considered as the decision variable. This reduces the number of required qubits and guarantees the feasibility of solution. The efficiency of quantum algorithm is evaluated by comparing with mixed integer programming (MIP).

Research on the method of digital twin operation and maintenance platform for intelligent early warning of wind turbine tower

Yu Jia — 2023-09-04

Wind power generators have a complex structure and operate in harsh environments, where working conditions are highly variable. As a result, the operation and maintenance of wind turbines face numerous challenges. In response to the need for the development of wind power operation and maintenance informatization, it is necessary to satisfy the requirements for multi-party collaborative monitoring to ensure the long-term safe and reliable operation of wind turbines. In this paper,we proposed a method for building an intelligent early-warning digital twin platform focused on the simulation of wind turbines and tower components. The platform construction method proposed in this article is based on the Web and from the perspective of intelligent operation and maintenance of wind turbines. It establishes a warning model for tower agent simulation and vibration signal time series prediction. The tower mechanism model is established based on the operating data set of a 4MW wind turbine at Shanghai Electric. Different physical responses of the tower under different wind speeds are simulated, and an agent model using LSTM and decision tree models is established for predictive analysis. To account for uncertainty, a Bayesian-LSTM model is established to warn against predictive errors. Finally, a data-driven digital twin wind turbine platform is achieved on the Web.

Sequential Domain Adaptation for Fault Diagnosis in Rotating Machinery

Yong Chae Kim — 2023-09-04

Fault diagnosis of the machinery system is essential to minimize the damage to the industrial field. Recently, with the development of computer and IoT technology, deep learning-based fault diagnosis has been widely researched. However, due to the domain shift, which changes the distribution of data under different operating conditions of the machinery system, the performance of the deep learning-based fault diagnosis algorithm decreases. This paper proposes a sequential domain adaptation to alleviate different distributions between different operating conditions. The proposed method has been validated in open-source datasets and shows a high performance compared to the other models.

Sound-Dr: Reliable Sound Dataset and Baseline Artificial Intelligence System for Respiratory Illnesses

Van Truong Hoang — 2023-09-04

As the burden of respiratory diseases continues to fall on society worldwide, this paper proposes a high-quality and reliable dataset of human sounds for studying respiratory illnesses, including pneumonia and COVID-19. It consists of coughing, mouth breathing, and nose breathing sounds together with metadata on related clinical characteristics. We also develop a proof-of-concept system for establishing baselines and benchmarking against multiple datasets, such as Coswara and COUGHVID. Our comprehensive experiments
show that the Sound-Dr dataset has richer features, better performance, and is more robust to dataset shifts in various machine learning tasks. It is promising for a wide range of real-time applications on mobile devices. The proposed dataset and system will serve as practical tools to support healthcare professionals in diagnosing respiratory disorders. The dataset and code are publicly available here: https://github.com/ReML-AI/Sound-Dr/.

Statistical Analysis and Runtime Monitoring for an AI-based Autonomous Centerline Tracking System

Yuning He — 2023-09-04

Autonomous Centerline Tracking (ACT) enables an unmanned aircraft to be guided down the center of the runway, using a camera-based Deep Neural Network (DNN). ACT is safety-critical. The EASA Guidelines for machine-learning based systems list numerous assurance objectives that must be met toward certification and V&V. We extend our analysis framework SYSAI to provide feedback on performance of system and AI component to the designer and describe a combination with a runtime monitoring architecture that also supports advanced risk mitigation to support safety assurance of a complex AI-based aerospace system.

Study on the intermittent fault mechanism of electromagnetic relay under complex environmental stress

Ye Jingzhong — 2023-09-04

When the relay experiences intermittent failures, its faults are difficult to reproduce, making it difficult to repair, and bringing daunting challenges to equipment reliability and mission success. Domestic and international studies have shown that environmental stresses, especially vibration stress and temperature stress, are important causes of intermittent failures of relays, but their mechanisms are not yet clear. Therefore, an in-depth analysis of the mechanism of relay intermittent failure under complex environmental stresses has become a critical challenge that needs to be accomplished. To address this issue, this paper takes a model of electromagnetic relay as the research object and conducts an in-depth study of the intermittent failure mechanism of the relay under vibration stress and temperature stress after salt spray corrosion using the technical approach of theoretical analysis, simulation analysis and experimental verification.

System-Level Simulation of 120 kW Interior Permanent Magnet Synchronous Motor Drive for Electric Vehicle Usage Under Various Types of Faults for Fault Diagnosis

Woyeong Kwon — 2023-09-04

Owing to the recent trend toward the zero-carbon emission, drive motors are gaining attention for substitute for internal combustion engine. Among them, interior permanent magnet synchronous motor(IPMSM) is being extensively used for their high-power density. Consequently, verification of safety and reliability for IPMSM is becoming an important subject. In this study, system-level simulation for IPMSM drive system under various failure modes were carried out. Fault characteristics were extracted from the phase current profile and classified by the support vector machine. It is shown that with the proposed model, faults can be detected with less cost and time-consumption.

Simulation-driven Bearing Fault Diagnosis for Condition Monitoring without Faulty Data

Iljeok Kim — 2023-09-04

The failure of rolling element bearings in complex mechanical systems is a significant cause of mechanical failures, leading to decreased productivity and safety risks. Deep learning has shown promising results in bearing fault diagnosis, but the predictive performance depends on highquality data. Domain adaptation has been studied to solve this problem, but it still has limitations when applied to real-world industrial applications. In this study, we propose a deep learning-based domain generalization framework for bearing fault diagnosis using the bearing simulation model and adversarial data augmentation method. The proposed framework was validated on a real bearing fault dataset and showed promising results in improving diagnostic performance in cases where fault data cannot be obtained or when dealing with unlearned target domains. This approach has the potential to improve industrial maintenance systems by obtaining improved generalization performance in the absence of fault datasets.

Tacholess Instantaneous Speed Estimation Considering the Characteristics of Vibration Harmonics

Jinoh Yoo — 2023-09-04

Knowledge of instantaneous shaft speed is vital for non- stationary condition monitoring of rotating machinery in real applications. To avoid installing expensive and inconvenient encoders, many researchers have developed instantaneous speed estimation methods by extracting the shaft speed from vibration signals. However, previous methods show limitations due to challenges in vibration signals. Therefore, we propose a novel instantaneous speed estimation method considering the characteristics of vibration harmonics to overcome the difficulties. Multiple harmonic components and their characteristics are utilized to obtain an accurate ridge in the time-frequency representation (TFR). The proposed method is validated and compared with the previous methods using a gear vibration simulated signal and civil aircraft engine dataset. The results show the accuracy and robustness of the proposed method.

To Trust or Not: Towards Efficient Uncertainty Quantification for Stochastic Shapley Explanations

Joseph Cohen — 2023-09-04

Recently, explainable AI (XAI) techniques have gained traction in the field of prognostics and health management (PHM) to enhance the credibility and trustworthiness of data-driven nonlinear models. Post-hoc model explanations have been popularized via algorithms such as SHapley Additive exPlanations (SHAP), but remain impractical for real-time prognostics applications due to the curse of dimensionality. As an alternative to deterministic approaches, stochastically sampled Shapley-based approximations have computational benefits for explaining model predictions. This paper will introduce and examine a new concept of explanation uncertainty through the lens of uncertainty quantification of stochastic Shapley attribution estimates. The proposed algorithm for estimating Shapley explanation uncertainty is efficiently applied for the 2021 PHM Data Challenge problem. The uncertainty in the derived explanation for a single prediction is also illustrated through personalized prediction recipe plots, improving post-hoc model visualization. Finally, important practical considerations for the implementation of Shapley-based XAI for industrial prognostics are provided.

Trip Reduction in Turbo Machinery

Pranay Mathur — 2023-09-04

Industrial plant uptime is top most importance for reliable, profitable & sustainable operation. Trip and failed start have major impact on plant reliability and all plant operators focussed on efforts required to minimise the trips & failed starts. The performance of these Critical to Quality (CTQs) are measured with 2 metrics, MTBT (Mean time between trips) and SR (Starting reliability). These metrics helps to identify top failure modes and identify units need more effort to improve plant reliability.
Baker Hughes Trip reduction program structured to reduce these unwanted trip
1. Real time machine operational parameters remotely available and capturing the signature of malfunction including related boundary condition.
2. Real time alerting system based on analytics available remotely.
3. Remote access to trip logs and alarms from control system to identify the cause of events.
4. Continuous support to field engineers by remotely connecting with subject matter expert.
5. Live tracking of key Critical to Quality (CTQs)
6. Benchmark against fleet
7. Break down to the cause of failure to component level
8. Investigate top contributor, identify design and operational root cause
9. Implement corrective and preventive action
10. Assessing effectiveness of implemented solution.
11. Develop analytics for predictive maintenance

With this approach, Baker Hughes team is able to support customer in achieving their Reliability Key performance Indicators for monitored units, huge cost savings for plant operators. This paper explains this approach while providing successful case studies, in particular where 12nos. of Liquified Natural Gas (LNG) and Pipeline operators with about 140 gas compressing line-ups has adopted these techniques and significantly reduce the number of trips and improved MTBT.

Unsupervised Anomaly Detection in Marine Diesel Engines using Transformer Neural Networks and Residual Analysis

Qin Liang — 2023-09-04

This paper presents a novel unsupervised approach for detecting anomalies in marine diesel engines using a Transformer Neural Network based autoencoder (TAE) and residual analysis with Sequential Probability Ratio Test (SPRT) and Sum of Squares of Normalized Residuals (SSNR). The proposed method can capture temporal dependencies in normal timeseries data without the need for labeled failure data. To assess the effectiveness of the proposed approach, a dataset of faulty data is generated under the same operational profile as the normal training data. The model is trained using normal data, and the faulty data is reconstructed using the trained model. SPRT and SSNR are then used to analyze the residuals from the observed and reconstructed faulty data, with significant deviations exceeding a predefined threshold being identified as anomalous behavior. The experimental results demonstrate that the proposed approach can accurately and efficiently detect anomalies in marine diesel engines. Therefore, this approach can be considered as a promising solution for early anomaly detection, leading to timely maintenance and repair, and preventing costly downtime.

Unsupervised Retrieval Based Multivariate Time Series Anomaly Detection and Diagnosis with Deep Binary Coding Models

Takehiko Mizoguchi — 2023-09-04

Retrieval based multivariate time series anomaly detection and diagnosis refer to identifying abnormal status in certain time steps and pinpointing the root cause input variables, i.e., sensors, by comparing a current time series segment and its relevant ones that are retrieved from huge amount of historical data. Binary coding with a deep neural network can be applied to reduce the computational cost of the retrieval tasks. However, it is hard to pinpoint the root cause sensors that are responsible for the anomaly, once multivariate time series segments are transformed into binary codes. In this paper, we present an unsupervised retrieval based multivariate time series anomaly detection and diagnosis method with deep binary coding model, to secure both efficiency and explainability. Specifically, we first transform input multivariate time series segments into low dimensional features with a temporal encoder. Subsequently, two hash functions predict two binary codes with different lengths from each feature. The binary codes with two different lengths can contribute to accelerate both anomaly detection and anomaly diagnosis. Experiments performed on datasets from various domains including real optical network, demonstrate the effectiveness and efficiency of the proposed method.

V-belt Tension Reduction Diagnostic Method By Using Multi-Frequency Current Power Spectrum Density

Hiroshi Inoue — 2023-09-04

Motors are used in many factories and need stable operation. A V-belt is a device that transmits motor power to a load, and its tension decreases with prolonged use, which degrades the motor efficiency and causes such abnormalities as wear and cracks. The timing of belt replacement when the belt tension decreases is presently determined by regular inspections. An automatic diagnosis of belt-tension decrease is required to lower labor costs. One automatic diagnosis method applies FFT analysis to the phase current that is applied to the motor and focuses on the signal intensity at a specific frequency. Since this method can only detect belt tension after it has progressed, early detection is necessary. Our new method using motor phase current enabled early detection of belt-a tension decrease.

Anomaly Detection in Spacecraft Propulsion System using Time Series Classification based on K-NN

Yoshiki Kato — 2023-09-04

In this paper, we propose an anomaly detection method developed by the team called “Team Tsubasa” in the PHMAP2023 Data Challenge. This is an anomaly detection competition for spacecraft propulsion systems (PHM Society, 2023). We joined the Data Challenge with the aim of deepening our knowledge of anomaly detection technology through the competition. In spacecraft propulsion systems, solenoid valve faults and bubble anomalies can occur, and it is considered important to detect them. Also, when other unknown anomalies occur, it is necessary to detect them without confusing them with known anomalies. In this paper, we propose time series classification by k-NN algorithm (Cover & Hart, 1967) as one of the methods to detect these anomalies. In this data challenge, we tried to classify anomalies by k-NN and to identify the location of the anomalies. For those classified as solenoid valve faults, we estimated the opening ratio of the solenoid valve from the similarity of the time series waveforms. As a result, the proposed method achieved a score of 99.05% based on the scoring rules given by the PHMAP 2023 Secretariat and our team won third place.

Dataset Generation Based on 1D-CAE Modeling for Fault Diagnostics in a Spacecraft Propulsion System

Kohji Tominaga — 2023-09-04

The objective of this study is to generate a classified dataset of valve faults and bubble contamination anomalies in the propellant supply pipe of spacecraft propulsion systems. The dataset is available in PHMAP23, and the paper intends to describe its characteristics. The dataset includes time and pressure information and has been generated through numerical simulations using SimlationX, a 1D-CAE software. The condition of the propulsion system is reflected by the characteristics of the pressure dynamic response generated by the water hammer in the supply pipe caused by the rapid opening and closing of the downstream solenoid valve. Therefore, accurate classification of anomalies and faults can be achieved by extracting characteristics from the pressure dynamic response waveform.

Expert-Informed Hierarchical Diagnostics of Multiple Fault Modes of a Spacecraft Propulsion System

Osarenren Kennedy Aimiyekagbon — 2023-09-04

This paper presents a comprehensive study on diagnosing a spacecraft propulsion system utilizing data provided by the Prognostics and Health Management (PHM) society, specifically obtained as part of the Asia-Pacific PHM conference’s data challenge 2023. The objective of the challenge is to identify and diagnose known faults as well as unknown anomalies in the spacecraft’s propulsion system, which is critical for ensuring the spacecraft’s proper functionality and safety. To address this challenge, the proposed method follows a systematic approach of feature extraction, feature selection, and model development. The models employed in this study are kMeans clustering and decision trees combined to ensembles, enriched with expert knowledge. With the method presented, our team was capable of reaching high accuracy in identifying anomalies as well as diagnosing faults, resulting in attaining the seventh place with a score of 93.08 %.

Hybrid Approach of XGBoost and Rule-based Model for Fault Detection and Severity Estimation in Spacecraft Propulsion System

Sang Kyung Lee — 2023-09-04

This study presents a method for fault detection and severity estimation of the spacecraft propulsion system. The spacecraft propulsion system is complicated, consisting of many valves, and operates in a harsh environment. Therefore, faults due to external factors such as bubbles or valve breakage can occur within the complex system at any time. To diagnose faults in this system, we propose a hybrid method of XGBoost-based method and rule-based method. In the XGBoost-based method, the overall fault classification, including unknown fault filtering was performed. In addition, the rule-based model was performed to estimate the fault severity. The results show that the proposed method reached a 99.94% score, which is calculated by the score matrix considering fault classification accuracy and severity estimation.

PHM for Spacecraft Propulsion Systems: Similarity-Based Model and Physics-Inspired Features

Takanobu Minami — 2023-09-04

This paper presents a methodology designed for the Prognostics and Health Management (PHM) Asia-Pacific 2023 Conference Data Challenge. In particular, this study targets the health assessment of spacecraft propulsion systems. The challenge involved analyzing and categorizing a simulation-generated dataset that included four unique spacecraft and multiple health conditions, such as normal operation, bubble anomalies, and solenoid valve faults in various system locations. The proposed approach uses a two- step process. First, a model based on similarity measures is employed to classify the data into one of four health states. Then, a model incorporating physics-inspired features is utilized in solenoid valve faults to identify the fault location and estimate the valve opening ratio. The validity of the model is confirmed through cross-validation with the training dataset, which achieved a flawless total score across all permutations. Our method effectively categorizes the test data, including cases from a spacecraft not covered in the training, thereby securing a top position in the competition. The findings highlight the strength of our proposed model, which uses physics-inspired features to predict valve opening ratios, proving useful in managing and interpreting complex, unfamiliar spacecraft health data.

A Feature Selection Method for Machine Tool Wear Diagnosis

Yuji Homma — 2023-09-04

We propose an algorithm for estimating the wear condition of tools. We have previously developed a method for predicting machining dimensions by learning features of waveform shapes such as torque during machining as explanatory variables and measured machining dimensions as objective variables. In this method, the features do not fully explain the machining dimensions because including data other than the machining operation such as tool change. In this paper, we propose a method to improve explanatory power and prediction accuracy by selecting subsequences from the machining waveform that are highly related to machining dimensions as explanatory variables. The effectiveness of the proposed method was confirmed through an evaluation using data of machining product part.

A Health Index for Satellite System Based on Characteristics of Telemetry Data

Shun Katsube — 2023-09-04

Since satellites are non-repairable systems, it is important to detect anomalies early to prevent failures. Data-driven approaches to anomaly detection, which have been actively
studied in recent years, have problems such as low explainability and insufficient training data in initial operation. Thus, we propose a health index that is commonly available for all satellites. It is possible to share training data and examples of anomalies by monitoring
health status with the same index across different satellites. In this study, we extended our health index defined for the satellite power system to the entire satellite system. Then,
we applied the health index to the operational data of the Suzaku satellite and confirmed that the index is useful for anomaly detection.

A Method for Executing Digital System Models and Digital Twins at Scale to Enrich Fleet Health Management

Daniel Newman — 2023-09-04

The digitization of the aviation industry brings the promise of increased transparency into the engineering design across aircraft lifecycles. With an increased focus on Model-Based Engineering, design engineers are constructing interoperable, standardized simulation models which can be leveraged across the Digital Thread. This advancement presents an opportunity to improve health management and prognostics by directly comparing these models against in-service data. This paper proposes a method for leveraging engineering Digital System Models and Digital Twins at-scale to facilitate and enrich health management activities. Supporting considerations include specifying model standards and ingesting fleet-level data using cloud computing.

A new approach to multivariate statistical process control and its application to wastewater treatment process monitoring

Osamu Yamanaka — 2023-09-04

This paper presents a new process monitoring and fault diagnosis approach based on a modified Multivariate Statistical Process Control (MSPC) and evaluates its applicability to municipal wastewater treatment process monitoring. A conventional MSPC based on Principal Component Analysis (PCA) is firstly adjusted to have an easy-to-understand user interface and then a new yet simplified decomposable diagnostic model is introduced. The developed user interface is designed to seamlessly connect MSPC to existing process monitoring system adopting the so-called trend graphs. The proposed diagnostic model is derived in a constructive way by aggregating small size models with one input or two inputs to improve tractability of the diagnostic model. The effectiveness of the modified MSPC is illustrated through some offline and on-line experiments by using a set of real multivariate process data at a municipal wastewater treatment plant.

A New Challenge in Predictive Maintenance Analysis for Aircraft

Takashi Takemura — 2023-09-04

JAL Engineering Co., Ltd. (JALEC) has been challenging to develop the predictive maintenance by big data analysis using the flight sensor data and the maintenance data as one of our initiatives to improve our aircraft reliability. Our analysis method is to verify hypotheses based on mechanics and engineer expertise with the data. Furthermore, in order to enhance our predictive maintenance, we introduced a new analysis method to search for hypotheses based on the information obtained from the data with AI technology. This article describes how JALEC has succeeded in developing a new analysis method with AI technology in the predictive maintenance and what we want to realize in the near future as a total engineering company.

A Quantitative Analysis of Domain Discrepancies Between General- Speed and Very Low-Speed Bearings and Its Applications to Fault Diagnosis of Very Low-Speed Bearings Using Transfer Learning

Seungyun Lee — 2023-09-04

This paper proposes a fault diagnosis framework for very low-speed bearings using transfer learning. To handle large domain discrepancies between general-speed and very lowspeed bearings, the quantitative analysis is performed using operating information and geometries of bearings. From this analysis, the domain discrepancy can be quantitively compared under various speed conditions. Furthermore, a transfer learning technique is proposed to reduce the analyzed discrepancies. The domain discrepancy can be significantly aligned by integrating the operating information and geometries of bearings into transfer learning. Also, the proposed framework is not tailored for the specific algorithm; this means that the framework can be applied to any transfer learning technique regardless of the architecture. The proposed method is validated using two bearing datasets under general-speed and very low-speed conditions. The results show that the domain discrepancy can be quantitatively measured for transfer learning. Additionally, the proposed fault diagnosis framework outperforms the existing methods by aligning domain discrepancies of bearing datasets under large different speeds.

A Simple Hybrid Model for Estimating Remaining Useful Life of SiC MOSFETs in Power Cycling Experiments

Mattias P. Eng — 2023-09-04

Recording and prediction of the accumulated damage, which will eventually lead to the failure of power electronic modules, is an aspect of high importance for power electronic systems design and, in particular, for development of Prognostic and Health Management (PHM) schemes for in-field applications. To this end, this paper presents a simple and cost-effective prognostic method for predicting the remaining useful life (RUL) of TO-247 packaged silicon carbide (SiC) metal-oxide semiconductor field-effect transistors (MOSFETs) subjected to power cycling experiments. The model assumes that the major failure mode is bond-wire lift-off and uses a damage accumulation scheme based on Paris’ crack law. The only inputs to the model are historical data on the average junction temperature swing and the temperature-compensated drain-source ON-state resistance at the peak temperature of the current cycle. Using only these two input values, the model is shown to predict RUL with surprising accuracy for the range of constant current loads determining cycling conditions under which the test data series have been acquired. This work is a first step in an ongoing project towards building more elaborate prognostic schemes for RUL-determination of SiC power MOSFETs in actual working conditions, using physics-informed neural networks (PINNs).

A study on self-diagnosis/prediction technology for LIDAR sensor of autonomous vehicles

Jaewook Lee — 2023-09-04

Along with the development of autonomous driving technology, the need for self-failure diagnosis and Remaining Useful Life (RUL) prediction technology for core parts for autonomous driving is increasing. In particular, the characteristics of the light detection and ranging (LIDAR) sensor exposed to the outside further increase the need to apply fault diagnosis and RUL prediction technology considering various environmental variables. In this study, based on the accelerated degradation test of LIDAR, the failure mode was analyzed. Through this, LIDAR failure due to thermal runaway, which is the first failure type in high temperature conditions, was identified, and whether there were major environmental data that could identify thermal runaway was identified. In the case of LIDAR's thermal runaway phenomenon, a study on an algorithm to identify the precursor symptoms of failure in an accidental failure situation is conducted. Afterwards, through the actual vehicle test process, various environmental variable information is analyzed for correlation with LIDAR internal sensor data, and the abnormal data for the temperature of the internal parts of the LIDAR is predicted through the external environmental sensor.

A Study on the Intent-based System Design Automation Method for Fault Detection and Fault Tolerance

Takayuki Kuroda — 2023-09-04

ICT systems are becoming an important infrastructure indispensable for business operations in various industries. While the requirements are becoming more diverse and complex, stable service provision is required. As a result, the burden on operations is increasing, and there are growing expectations for automation technology to solve this problem. This study investigates techniques to significantly reduce the burden on operators by automating the construction and maintenance of systems that satisfy Intent, a concise description of users' requirements. However, even using the results of previous research, it has been difficult to automate operations with consideration for the stability of service provision. Therefore, this paper describes an automated method based on intent for detecting failures that occur in the target system and an automated design method for fault-tolerant systems. First, we describe the basic concept, present the results of thought experiments, and discuss its effectiveness.

A two-step detection method for actuator and sensor failures based on self-repairing PID control

Masayoshi Hara — 2023-09-04

Our previous works have proposed a self-repairing PID (SRPID) control for plants with sensor failures. This method uses an I-controller (integrator) with a well-designed auxiliary signal to detect failures. However, the SRPID has not addressed actuator failures. This paper presents a two-step detection method for actuator and sensor failures based on the SRPID and confirms the effectiveness of the proposed detection method through experiments on a ball-and-beam system.

Acoustic Signal based Non-Contact Ball Bearing Fault Diagnosis Using Adaptive Wavelet De-Noising

Wonho Jung — 2023-09-04

This paper presents a non-contact fault diagnostic method for ball bearing using adaptive wavelet denoising, statistical-spectral acoustic features, and one-dimensional (1D) convolutional neural networks (CNN). The health conditions of the ball bearing are monitored by microphone under noisy condition. To eliminate noise, adaptive wavelet denoising method based on kurtosis-entropy (KE) index is proposed. Multiple acoustic features are extracted base on expert knowledge. The 1D ResNet is used to classify the health conditions of the bearings. Case study is presented to examine the proposed method’s capability to monitor the condition of ball bearings. The fault diagnosis results were compared with and without the adaptive wavelet denoising. The results show its effectiveness of the proposed fault diagnostic method using acoustic signals.

Advancing Predictive Maintenance: A Study of Domain Adaptation for Fault Identification in Gearbox Components

Shinya Tsuruta — 2023-09-04

This paper explores the use of machine learning in predictive maintenance, which has been increasingly demanded in recent years to reduce downtime and maintenance burden. The challenge of different data distributions between training and test data in machine learning is common in predictive maintenance where equipment operation patterns can change, leading to reduced operational efficiency. The authors validate a domain-adaptive anomaly detection method combining CNN and MMD, which achieves similar accuracy with PCA, SVD, and other dimensionality reduction methods. The study also shows that the method maintains accuracy even when the number of normal data in the target domain is 1/10 of the source domain.

Algorithmic Study for Power Restoration in Electrical Distribution Networks

Jun Kawahara — 2023-09-04

We study the automated power restoration in electrical distribution networks, from the algorithmic viewpoint. During power outages, blackout sections without faults may be able to be recovered early using the capacity margins of surrounding supply sources. However, remote supply sources must be utilized in cases where the capacity margins of the neighboring supply sources are insufficient for the scale of the power outage, which is called multi-stage power restoration. In multi-stage power restoration, the distribution network subject to control becomes broader, and in addition, even healthy sections are subjected to control. In this study, we give an efficient algorithm which determines whether multi-stage power restoration is needed for power restoration, and in either case, the algorithm calculates the switching procedure which recovers power with the minimum number of switch operations. Our algorithm employs a novel algorithmic technique of combinatorial reconfiguration, which enables to maintain power supply to the healthy sections.

An Easy-to-Use and Customizable Data Science Tool for Predictive Maintenance in Manufacturing

Naoki Sugawara — 2023-09-04

Even if many manufacturing companies have attempted to utilize data analytics for predictive maintenance but have not been able to do so. This is because predictive with data requires both knowledge of equipment in production site and knowledge of data science, and such workers are scarce. Another problem is that it takes time and effort to apply the created machine learning models on site and to develop applications for monitoring the diagnosis results.

Therefore, we developed “MELSOFT MaiLab”. The application has an intuitive interface and functions to automatically create machine learning models, making it possible for workers with no knowledge of data science to easily create machine learning models for predictive maintenance. The application also has functions for deploying the created machine learning models to the production site and monitoring the diagnostic results.

An Experimental Study of the Effect of Patrol Inspection Strategy for Improving Detection Rate of Abnormality of Industrial Plants

Akio Gofuku — 2023-09-04

Safe and stable operation is very important in large-scale plants. Patrol inspections are comprehensive and frequent on-site inspections and are indispensable works because early detection of a malfunction or failure of a component is desirable to achieve safe and stable operations. This study supposed that the detection rate of anomalies by patrol inspections is related with not only the amount of knowledge about plants but also the modalities such as vision and listening that are conscious during inspections. This study experimentally examined the difference in the detection rate of anomalies depending on patrol inspection strategies. In the experiment, a simulated small plant using the devices that are similar to plant components was constructed as a site for patrol inspections, and two types of patrol inspection strategies were examined: visual strategy that focuses on watching component conditions and auditory strategy that focuses on listening abnormal sounds. Fifteen volunteers participated in the experiment. The relationship between patrol inspection strategies and the approach and performance of patrol inspections were investigated by observing patrol inspection behaviors of participants and measuring their anomaly detection rates. As the results, although there was a small difference in the detection rate for visually detectable anomalies, the effect of the auditory strategy was suggested to improve the detection rate for not only auditory detectable anomalies but also visually detectable anomalies.

Anomaly detection for yield improvement in glass production

Haruo Yonemori — 2023-09-04

Predictive maintenance using manufacturing sensor data has attracted attention for reducing defects and selecting appropriate actions. This paper proposes an anomaly detection method using lasso regression and group-wise variable selection based on FTA (Fault Tree Analysis) domain knowledge. We evaluated our approach using real factory data and found that its precision and false positive rate are 66% and 30%, respectively. Moreover, we validate that the visualization of the contribution rate for anomaly detection is helpful for factory maintenance.

Anomaly Detection in Air Handling Units using Motor Current Signal Imaging for Belt Looseness Detection

Sung Hyun Yun — 2023-09-04

An air handling unit (AHU) is a critical component of heating, ventilation, and air conditioning (HVAC) systems. Slip of AHU is an intuitive key feature for monitoring a belt looseness fault of an AHU. However, fluctuating rotation speed of the motor and fan makes slip hard to monitor. Since the role of the belt is to deliver torque between the motor and fan, this leads to change of the motor current signal. This paper suggests a normal data-based anomaly detection that utilizes motor current signal imaging to identify belt looseness in AHUs. The overall process proceeds as followings: (1) converting 1-dimensional motor current signal into 2-dimentional image in the amplitude domain, (2) extracting features of normal data by applying convolutional neural networks, (3) calculating health index to detect the belt looseness fault. The technique to transform time-series current data to an image is based on its histogram. The image is obtained by the inner product of the histogram obtained from a current signal and its transpose. The effect of torque load on a motor induces an amplitude modulation of the current signal. Current signal imaging based on histogram provides the fault features in a robust way. To validate the proposed method, a case study using an AHU testbed is conducted. The results demonstrate that the proposed method can detect belt looseness faults in AHU using only normal data, providing an approach for early fault detection in HVAC systems.

Anomaly Detection in Airliner Centrifugal Compressor Using Sensor Data during the Climb Phase

Sadanari Shigetomi — 2023-09-04

In recent years, aircraft have been able to quickly acquire vast and diverse sensor data, leading to the expansion of predictive maintenance applications. Centrifugal compressors are crucial to aircraft air conditioning systems, which has a high need for anomaly detection due to the impact of failures. However, due to non-stationary behavior, there is a challenge in anomaly detection of the air conditioning system. In this study, we propose a method of detecting anomalies by comparison of actual and predicted behavior of centrifugal compressors for non-stationary time-series data. Multiple failure cases confirmed that bearing deterioration results in changes in behavior.

Anomaly detection of propulsion system of spacecrafts with Light GBM

Shota Iino — 2023-09-04

Future spacecrafts require robust operations for long-term missions to the Moon or Mars. Automatic anomaly detection with machine learning, in this context, plays a significant role because it enables early symptom detection and proactive redundant switching which preserves components in the long mission. In this research, we adopted Light GBM, one of the machine learning models, to investigate such anomaly. We especially focused on the telemetry data of propulsion system of H-II Transfer Vehicle (HTV) to resolve typical problems of deep-space mission spacecrafts, a thruster failure. The data was collected from multiple types of thruster maneuvers performed at simulator training. The results showed the effectiveness of the proposed method.

Anomaly Sign Detection for Automatic Ticket Gates by the Histogram Limitation Method

Ken Ueno — 2023-09-04

It is crucial for automatic ticket gates (ATGs) on railways, also known as fare collection systems, to detect anomalies at an early stage, especially in the automatic separation module for multiple tickets. It is also required for efficient and low-cost monitoring without any additional sensors especially for old-type ATGs that need to be maintained frequently. However, the failure rate is basically very low, and monitoring data contain various kinds of normal status indicators depending on complicated mechatronics controls. In addition, it is hard to collect high quality learning data because ATGs are affected by various ticket conditions or timing when releasing tickets by users, which makes detecting anomaly signs difficult. For these reasons, conventional machine learning or deep learning methods are not suitable for anomaly detection for ATGs. In this paper, we propose a simple anomaly detection method with new anomaly sign index, called the histogram limitation method (HLM), for effective monitoring to realize preventive maintenance of ATGs based only on system log data. Despite being a quite simple and compact method, HLM provides anomaly sign scores that agree adequately with assessments by maintenance service engineers in our evaluation with real field ATGs in operation.

Applications of Active Learning in Predictive Maintenance

Navid Zaman — 2023-09-04

Nowadays, the common choice in maintenance strategies is predictive maintenance (PdM), deprecating the corrective and preventive kinds. Even with various machine learning techniques to get advanced predictive models to achieve PdM, difficulties remain in the data acquisition process. While there is a plethora of unlabeled data from sensors, most of those available techniques can only process labeled data, i.e, supervised learning. To combat the fact that the availability of the labeled
data is limited, this paper proposes the use of Active Learning to label and annotate the informative instances while minimizing overall processing time. This approach maintains high performance and decreases the number of labeled instances, with support from experimental results and a discussion of the applicability of this method.

Automatic Detection of Concrete Surface Defects Using Deep Learning and Laser Ultrasonic Visualization Testing

Takahiro Saitoh — 2023-09-04

In recent years, nondestructive testing of civil engineering structures has become increasingly important. Non-destructive inspection methods for civil engineering structures generally use ultrasonic waves. However, the inspection of civil engineering structures takes much time because of the extensive scope of the inspection. Moreover, in the field of nondestructive testing, there are also concerns about a future shortage of inspectors. Therefore, this study proposes a laser ultrasonic visualization testing using AI. The proposed method will be applied to a concrete structure with surface defects to confirm the effectiveness of the proposed method.

AVIATAR – Deep dive into prediction with AVIATAR with in-service examples from airlines

Sebastian Lang — 2023-09-04

This paper demonstrates how AVIATAR prediction successfully improves operational stability.It will explain a specific use case example, presented with customers to complete the holistic view on big data models and predictive recommendations through the eyes of an airline. A deep dive int he ATAchapter36 (EBASS-suite) will illustrate how Lufthansa Technik’s predictive maintenance knowledge achieves numerous cost savings and provides increased operational stability.

AVIATAR – Optimizing airline flight operations with Predictive Health Analytics and building on digital platform technology

Mia Witzig — 2023-09-04

AVIATAR uses data from aircraft systems, airline operations, aircraft, maintenance systems and other external sources. Subsequently, it turns that data into actionable insights that empower airline technical operations teams to optimize maintenance and to reduce unscheduled interruptions. AVIATAR includes different modules that can be combined to address different operational needs and requirements.

Building The Health Monitoring and Fault Diagnosis Models For Stamping Press

Yuan-Jen Chang — 2023-09-04

A stamping press is widely used for the metal forming process. To achieve continuous automation and high precision forming, monitoring the press's health and diagnosing faults during stamping is necessary. The three primary types of faults that may occur in the stamping press are lack of lubrication oil, quality variation of lubrication oil, and clearance variation, which can lead to a decline in workpiece quality and reduced lifespan of the dies and presses. This study adopted the Prognostics and Health Management (PHM) technique to implement a predictive maintenance system for the stamping press. To extract relevant data, the National Instrument (NI) DAQ was used to acquire the three-phase currents and X, Y, and Z vibration signals. Six signals provided a total of seventy-two features, and the top three key features were selected for building a health assessment model using the Logistic regression and PCA algorithms. An early warning is triggered when the health indicator drops below the threshold, alerting the operators. Additionally, fault diagnosis was achieved using classification algorithms such as Support Vector Machine (SVM), K Nearest Neighbors (K-NN), and eXtreme Gradient Boosting (XGBoost). The fault diagnosis model achieved high accuracies of up to 99%.

Condition-based Maintenance of Brake Pads and Tires in Shared Vehicles using Cloud-based Health monitoring and prognostics.

Jeong Hae Lee — 2023-09-04

Cloud-based prognostics and health management is a centralized method for monitoring the condition of individual shared vehicles and determining their maintenance schedules.
In this study, we focused on monitoring the condition of brake pads and tires, as these crucial components require frequent and regular maintenance for safety. We developed a data acquisition system to transmit data from acoustic and vibration sensors to the cloud server. Useful and efficient features were extracted and selected from time and frequency
domains to assess the degradation of brake pads and tires. Moreover, based on feature extraction using the KruskalWallis method, we confirmed that diagnosing brake pad conditions with support vector machines (SVM) provides consistent result for classification of sevierities.. Our preliminary results suggest that cloud-based condition
monitoring can be an effective approach to managing shared vehicles.

Construction and evaluation of an anomaly detection system using System Invariant Analysis Technology(SIAT) for sound data

Tomoya Soma — 2023-09-04

Here is a summary of a paper that presents case studies on the application of SIAT, a machine learning technique, for anomaly detection in plants and industrial machinery, with a focus on sound-based anomaly detection as a new application of SIAT:

This paper explains case studies on anomaly detection using SIAT as a machine learning technique. SIAT is specialized in analyzing time-series data and is widely used for anomaly prediction in plants and industrial machinery. In recent years, the application of SIAT has been extended to sound-based anomaly detection, and this paper presents some case studies on this topic.

Specifically, the paper provides several examples of sound-based anomaly detection, such as detecting abnormal sounds or predicting machinery failures. In these cases, SIAT was used to analyze sound data collected from multiple sensors, and anomalies were detected successfully. The results of these anomaly detection methods were then used to take preventive measures such as maintenance or repairs, leading to improvements in productivity and safety.

This paper demonstrates the usefulness of SIAT for sound-based anomaly detection and suggests the potential for expanding the scope of SIAT’s applications.

Damage-Based Lifetime Modeling for Power Electronic Devices

Chao Guo — 2023-09-04

Lifetime modeling is an essential tool for ensuring the reliability of systems. The purpose is to estimate the time before the power electronic device failure so that downtime can be reduced and costly failures can be avoided in industry. This paper will first quantify the cumulative damage in the power cycling test using Junction Temperature Swing and Maximum Junction Temperature, and then formulate the cumulative damage-based lifetime model of power electronic devices. This model assumes that the lifetime is linear to the inverse of the cumulated damage, and shows superior performance in experiments compared with the well-known LESIT model.

Data-Driven Prognostics and Diagnostics of Industrial Machinery — A Turbofan Engine Case Study

Russell Graves — 2023-09-04

A machine’s Remaining Useful Life (RUL) is the expected life or usage time remaining before the machine requires repair or replacement. In data-driven methods, typical RUL estimation is performed using models trained with health condition indicator values derived from measured system data. A significant challenge in developing an RUL estimation model is transforming large, multivariate, noisy sensor datasets into useful format(s) that make the data analysis and processing pipeline efficient and extract valuable condition indicators from the data. This work uses the N-CMAPSS dataset to explore options and implications for efficiently organizing and storing large time-series datasets to support prognostics and diagnostics applications. We extend the work to demonstrate a predictive maintenance workflow and solution to (1) detect and classify faults in a turbofan engine and (2) estimate the RUL once we detect performance degradation.

Under data engineering, we investigate the impact of various file formats and file types on memory and execution time when dealing with large datasets like N-CMAPSS. We analyze, pre-process, and extract/engineer critical features from the transformed dataset by leveraging our understanding of gas turbines' operation (e.g., Brayton Cycle). We also analyze the performance of various engine submodules for different flight phases (climb, cruise, and descent). This work also explains an approach to down-sample the time series data without losing information relevant to our goals. Using the health condition indicators derived and synthesized in the data engineering stage, we train machine learning models for diagnostics (differentiate between healthy operation and seven different types of faults in the turbofan engine) and prognostics (RUL estimation).

Data-driven Pump Performance Analysis Using Online Monitoring Data Accumulated with Supervisory Control and Data Acquisition System

Ryo Namba — 2023-09-04

Pump degradation has significant impacts on the operation of water treatment plants since pumps are widely used as basic facilities to supply and distribute water. The authors develop a method for quantifying pump performance to monitor pump degradation by using online monitoring data accumulated with a supervisory control and data acquisition (SCADA) system. The developed method estimates the performance of each pump from the measured values at the confluence of pipes with the least-squares method and respective on-off signals. From the viewpoint of practicality, the usefulness of the developed method was evaluated.

Data-driven satellite monitoring method applicable to various telemetry

Noriyasu Omata — 2023-09-04

It is difficult to detect signs of faults for the rule-based health monitoring systems currently installed on artificial satellites in principle, and manual monitoring of satellite telemetry is conducted. However, due to lack of human resources, much of the data is left not yet well reviewed. In this study, a systematic telemetry monitoring method that screens anomalous ones applicable to various time-series telemetry is proposed. The proposed method estimates the normal range of future telemetry values by focusing on quantile statistics of each telemetry. The demonstrative application results to the real telemetry data are also reported.

Deep Learning Approach for Operational Transfer Path Analysis: Case Study of Electric Vehicles

Jeongmin Oh — 2023-09-12

This paper presents a new approach to fault diagnosis of the drivetrains of the electric vehicle. Most commercially available electric vehicles do not have accelerometers on electric drivetrains making it difficult to detect fault characteristics of the drivetrains of the electric vehicle, whereas accelerometers exist on the driver's seat. The proposed approach’s key idea is based on the operational transfer path analysis that determines the transfer function between the source and receiver. The transfer function is derived by training a deep learning model. The deep learning model converts the driver's seat vibration signals into drivetrains vibration signals. The validity of the proposed approach is evaluated using data from the durability test of real electric vehicles. It is anticipated that the proposed approach is effective to diagnose electric vehicle drivetrains subjected to external noise conditions.

Deep Learning/Machine Learning Techniques for Vibration Condition Monitoring of Major Facilities in Automobile Assembly/Painting Plants

Gun Sik Kim — 2023-09-04

We have been expanding the vibration monitoring system to prevent malfunctions of rotating equipment in Hyundai/Kia Motors' global factories. In this paper, a secondary analysis model was explored using an existing legacy program containing vibration trend and spectrum data. In the existing case, it goes through the steps of setting the alarm level - raising the vibration - reaching the alarm - alarm - recognizing - analyzing the vibration - drawing the result. An automation program was applied to reduce the steps to vibration increase - derivation of abnormal equipment - result analysis. In addition, we will also cover essential system components for the operation of additional development programs.

Deep Metric Learning for Abnormal Rotation Detection of Satellites from Irregularly Sampled Light Curve

Jun Yoshida — 2023-09-04

In recent years, satellites have become an indispensable infrastructure in our lives. The number of satellites is increasing yearly and becoming increasingly active. To use satellites safely, it is crucial to manage them and detect the anomaly of satellites as much as possible. However, it currently takes skilled operators to detect an anomaly, and it is difficult for even skilled operators to detect the anomaly early without the telemetry data in cases such as an abnormal rotation. To address these challenges, we tested the feasibility of using deep metric learning for early anomaly detection from the irregularly sampled light curve. One of the characteristics of a light curve is unequally spaced because the optical sensor on the ground can only observe the subject at night and not when the weather is terrible. Given an irregularly sampled light curve, our model employs a long short-term memory (LSTM) unit of encoding the temporal dynamics and learns the embedding on the feature space using triplet loss. Then, an anomaly score is calculated based on pairwise distances between segments from the learned embedding in the feature space. With actual data from the satellite being operated, we showed the effectiveness of our model and the feasibility of early anomaly detection. Also, by exploring learned embedding in the feature space, we show that our model could capture the continuous state of the satellite.

Deep Neural Network Anomaly Detection and Statistical Estimation of High Pressure Liquefied Natural Gas Pipe

Dabin Yang — 2023-09-04

Anomaly detection method using neural network is performed for diagnosis. Liquefied natural gas pipeline is designed using finite element method. To consider abnormal condition, a damage was applied to the model. Then failure mode and effect analysis are performed to determine if the location of damage is acceptable. The designed system was validated through literatures and showed that the model is suitable to replace the actual model. Data collection was done by changing each design variables in certain range from the designed model. Designable generative adversarial network was used for data augmentation and anomaly detection with adversarial network was used for anomaly detection. The performance of anomaly detection of the proposed model showed 95% of accuracy before data augmentation and 99% of accuracy after data augmentation. The result provides statistical estimation of diagnosis range for each design variables, which clearly showed the difference of performing data augmentation. By diagnosis result, the variables are used back to the designed model for validation of the result and showed accuracy of 85%.

Design of a Framework for Demand Forecasting Using Time Series Decomposition-Based Approach

Hazuki Shibayama — 2023-09-04

In recent years, artificial intelligence (AI) has made highly accurate demand forecasting possible. However, improving forecast accuracy does not necessarily mean reducing inventory costs or improving service levels in supply chain and inventory management, which are closely related to demand forecasting. Workers require not only high accuracy but also a basis for making decisions a based on forecasts. Autoregressive integrated moving average (ARIMA) and seasonal autoregressive integrated moving average (SARIMA) are demand forecasting methods with high accuracy and interpretability. However, these methods cannot provide evidence for demand fluctuations such as trends and seasonality, although they exhibit an autoregressive time-series structure. In this study, a framework for demand forecasting with high accuracy and interpretability was designed using time series decomposition and ARIMA to support decision makers in demand forecasting. The Seasonal-trend decomposition using locally estimated scatterplot smoothing (STL) is used to decompose a time series into three components trend, seasonality, and residual to provide decision makers with an easily understandable basis for demand changes. In addition, the ARIMA model is used for trends and residuals to achieve highly accurate forecasts. Comparing the prediction accuracies of the proposed STL-ARIMA and SARIMA models shows that STL-ARIMA has higher interpretability than the SARIMA model and the same accuracy.

Demonstration of model-based real-time anomaly detection in a JAXA 6.5m×5.5m low-speed wind tunnel.

Shotaro Hamato — 2023-09-04

In this study, real-time anomaly detection in a wind tunnel was conducted using a threshold based on uncertainty quantification of a numerical model. A model-based numerical model of a wind tunnel was developed, and the uncertainty consisting of input uncertainty, model form uncertainty, and numerical approximation was quantitatively evaluated. The threshold of anomaly obtained here was demonstrated in a 6.5m×5.5m wind tunnel of Japan Aerospace Exploration Agency (JAXA). Synthetic anomaly injected into the measurement system was successfully detected.

Development of a space exploration rover digital twin for damage detection

Lucio Pinello — 2023-09-04

This study focuses on the creation of a digital twin of a space exploration rover to perform damage detection. The digital twin incorporates various subsystems of real rovers to accurately simulate the rover’s behaviour. Damage detection is performed by introducing damages into the digital twin and comparing signals obtained in healthy and damaged conditions. By using the multiphysics model created by integrating different subsystems, the effect of damages can be observed in other subsystems of the rover. The study aims to demonstrate the potentiality of a digital twin for damage detection, reducing the risk of mission failure and data loss.

Development of an Amine Antioxidant Depletion Diagnosis Method Using Colorimetric Analysis of Membrane Patches

Tomomi Honda — 2023-09-04

Lubricating oils degrade by various factors under operation. Degradation causes of lubricating oils are roughly classified into two types, one is caused by solid particles and the other by oil oxidation products. Amine antioxidants are generally added in turbine oils to prevent formation of sludge. Degradation of base oil progresses rapidly, when antioxidant is used up. Thus, an on-site degradation diagnosis method of antioxidant is significant to maintain fluid property of lubricating oil. There are several methods such as a chromatography and ASTM D2272 in order to measure the residual ratio of antioxidant, but these methods cannot be used as an on-site degradation diagnosis method and need the particular experience and long time. On the other hand, ASTM D6810 can measure it in a short time. In our laboratory, we have reported that there is a good relationship between the degradation of lubricating oil and color of the membrane patch. Furthermore, we have developed degradation diagnosis method of the lubricating oils focusing on the coloration of the membrane patches with contamination. In this study, we investigated relations between oxidized products of amine antioxidant captured with membrane filter and color of membrane patch and we discussed the possibility of new depletion diagnosis method of amine antioxidants.

Development of demonstration system for fault diagnosis of rotating equipments using RK4 test rig

Hyun Joon Lee — 2023-09-04

Rotating machinery is an essential equipment in the manufacturing industry, of which the failure can lead to the interruption of whole production line. To prevent such failure, there have been a large number of studies for Prognostics and Health Management (PHM) technology. However, most studies have been focused on a specific algorithm or component, making it difficult to apply them in the field. In this study, a demonstration system for integrated fault diagnosis is developed for critical failure modes in the rotating equipment such as the mass imbalance, shaft misalignment, and the bearing fault using the vibration signals. To this end, Bently Nevada's RK4 rotor kit is revised to impose the failure modes easily. The solution is developed to detect anomalies, identify failure modes, and diagnose them in real time. Coupling ISO-based anomaly detection, rotational failure diagnosis and bearing failure diagnosis, the system is configured to enable comprehensive condition monitoring. The results are demonstrated by real-time simulation of each fault on the test-rig.

Digital Twin for condition based maintenance within a railway infrastructure testing lab

Antonio J. Guillén López — 2023-09-04

This article presents a digital twin development in a railway case, in order to improve operations and maintenance decisions, aligned with an asset management strategy. A digital framework for the sustainable management of these assets is defined with the purpose of facilitating the implementation on a cloud platform, searching the generation and sharing of the produced models and the evaluation from different perspectives. The developed digital twin allows for the digital representation of railway lines and vehicles, the connection between different entities based on an ontology, the management of data ingestion and storage, and the administration of models for the detection, diagnosis, and prognosis, as well as the representation and control of the level of risk of the assets. When emulation of railway line degradation is searched, different types of data are combined, from on-board sensors in railway vehicles, and physical behaviours, up to machine learning algorithms for estimation. In this way, the degradation behaviour model for the railway line is shown and validated through intelligent models easily replicated in several areas of the railway network, showing risk levels for each one.

Digital Twin for Diagnosis of Belt Looseness in HVAC Systems using multi-body dynamics simulation

Daeguen Lim — 2023-09-04

Most heating, ventilation, and air conditioning (HVAC) systems operate using a belt pulley system that provides high efficiency at a low cost. To monitor the health of the HVAC system, it is essential to detect looseness, which is one of the primary failure modes of belt-pulley systems. The main feature used to identify looseness is belt slip, which can be computed using the ratio of angular velocity and diameter between the drive pulley and driven pulley. However, accurately diagnosing the condition of the HVAC system based on the slip distribution calculated by measuring the rotational speed can be based on empirical criteria. To overcome this problem, this paper proposes constructing a digital twin model for accurate diagnosis of belt loosening in HVAC systems. The proposed approach involves performing multi-body dynamics (MBD) based time-domain analysis considering uncertainty. To validate the proposed model, belt looseness is applied to the HVAC system, and the slip distribution is calculated and compared with the results computed by the digital twin model. Through this comparison, it was demonstrated that the proposed model can be utilized to diagnose belt looseness in the HVAC system.

Digital Twin of Built Structures assisted by Computer Vision Techniques: Overview and Preliminary Results

Yasutaka Narazaki — 2023-09-04

Digital Twin is an effective platform for analyzing, visualizing, and interpreting the condition of as-built structures based on sensor measurement data. Based on the data inflow from the as-built structures, the associated models (typically finite element models) are updated, and/or the full behaviors of the structures are replicated in the virtual space. Despite its potential, field implementation of digital twin concepts often faces challenge, because the specific forms of digital twin, such as sensor types/locations, structural model development, and data fusion algorithms, depend strongly on the case-specific objectives. Focusing on the digital twin concepts assisted by computer vision techniques, this research aims at facilitating the implementation of those concepts by presenting the definition, setting, and preliminary results of digital twin in different application contexts, including post-earthquake building assessment and structural health monitoring based on multiple types of measurements. This research is expected to contribute to the broader impact of digital twin concepts in the structural engineering community.

Effect of Aircraft Health Management on Aircraft Maintenance Program Development by Aircraft Manufacturer

Takuro Koizumi — 2023-09-04

The objective of the research is to propose a system modeling method for aircraft maintenance program development applying the condition-based maintenance using AHM (Aircraft Health Management) from the viewpoint of aircraft manufacturer. The proposed model based on the MSG-3 (Maintenance Steering Group - 3) considers the uncertainty of aircraft maintenance environment related to the airline operation and assumed system degradation levels. Then, the proposed model is formulated by using the concept of a robust optimization method.

Efficient Inspection of Civil Engineering Structures for Railways and Roads Using Images and GNSS

Tomoya Abe — 2023-09-04

As structures deteriorate and engineers age, efficient maintenance and management of civil engineering structures becomes increasingly important. Alternative inspection of civil engineering structures using images is being recognized as an efficient inspection method. However, acquiring images of large and extensive civil engineering structures is time and equipment-intensive, which hinders mobility and is consequently not efficient. In addition, the management of existing civil engineering structures is highly specialized and lacks scalability, which means that significant effort is required to associate images with civil engineering structures. Therefore, we decided to acquire images of large and extensive civil engineering structures with video cameras to improve mobility and efficiency. In addition, as a key to linking images and civil engineering structures, we developed a technology to record latitude and longitude information as audio data in a video using GNSS (Global Navigation Satellite System) + dual-frequency RTK (Real Time Kinematic) This facilitates the rapid comparison of images and management data for existing structures with latitude and longitude information, as well as image data management in chronological order. This results in improved productivity in image-based inspection. In addition, AI processing of image data has made it possible to sample deformed areas and analyze geographical characteristics. These combined technologies have raised expectations for preventive maintenance of structures using images.

Expert Knowledge Transfer from CAE Models to CNN Models Using Enhanced Adversarial Domain Adaptation

Minseok Choi — 2023-09-04

It is a common belief that convolutional neural networks (CNN) are incapable of acquiring knowledge from domain experts for fault detection and diagnosis. To address the challenge, this paper proposes a knowledge-transfer scheme from computer-aided engineering (CAE) models to CNN models. Domain experts build the CAE models that emulate the faulty behavior of rotating machines by incorporating fault symptom and controlling the degree of fault severity. Fault data are hardly acquired from rotating machines in the field, while a sufficient number of fault data can be generated using the CAE models. Then, a domain adaption model is trained using synthetic data (i.e., normal and fault data) from the CAE models and real data (i.e., normal data only) from rotating machines. To evaluate the validity of the proposed method, a small-scale testbed is regarded as the target system that does not have any fault data. This study contributes to resolve the dearth of fault data from most safety-related engineering assets such as power plant steam turbines, wind turbines, and urban air mobility.

Extracting Broken-Rotor-Bar Fault Signature of Varying-Speed Induction Motors

Dehong Liu — 2023-09-04

An effective way to detect broken-bar faults of squirrel-cage induction motors is to extract the characteristic frequency component in the stator current as a fault signature, or so-called motor current signature analysis (MCSA). However, for inverter-fed motor drive systems, the motor is typically operating under varying-speed, varying-load, and noisy environments, which makes the fault signature extraction a very challenging problem. In this paper, we propose a sparsity-driven and graph-based method to extract the fault signature effectively, where the fault signature is modeled as a sparse component in the frequency domain for each short-time window measurement while gradually changing from window to window in the time-domain. Compared to the conventional short-time Fourier transform-based method, our method is more robust to noise and varying speed operations. Experiments are carried out to demonstrate the effectiveness of the proposed method.

Fault Detection of Rotor Bars in Inverter-Fed Induction Motors Based on Current Signature Analysis Technique

Tomoyuki Iwawaki — 2023-09-04

Induction motors, which are one key piece of equipment for power plants, waterworks facilities, and factories, must be maintained appropriately for reliable operation. A motor current signature analysis (MCSA) technique, which monitors and detects problems in motors and diagnoses devices, has already been marketed by some companies. Recently, applications of inverter-fed motors have increased for greater energy conservation. However, a fault-detection method in inverter-fed motors has been inadequately studied despite the risk of misdetection due to inverter noise. This paper shows our results that detected a broken rotor bar in an inverter-fed motor based on a MCSA technique. An abnormal motor with a broken rotor bar and a normal motor are driven by an inverter. The current supplied to both motors is measured and the frequency spectra results are compared. In the measurements, the inverter’s drive frequency is varied from 120 to 10 Hz in 10 Hz increments. In each drive frequency, the slip is varied in a range of 0-5% by adjusting the load connected to the motor. The results of comparing the current spectra show significant reinforcement of the signal intensities in abnormal motors. Some signals are reinforced by both inverter noise and a component that originated in the broken bar. The superposition that may lead to a misdiagnosis in inverter-fed motors is avoided by identifying the normal spectrum shapes of the target equipment in advance.

Few-Shot Learning for Full Ceramic Bearing Fault Diagnosis with Acoustic Emission Signals

David He — 2023-09-04

Full ceramic bearings are critical components in many full ceramic and oil-free food processing and medical equipment. Developing effective full ceramic fault diagnostic methods is important. Supervised deep learning approaches have been considered promising for fault diagnosis in the era of big data where abundantly labelled datasets are available. However, in many industrial applications, datasets with fault labels are rare. This challenge has motivated the task for developing deep learning approaches for fault diagnosis with few training examples. To meet the challenge, one attractive direction is to use available pre-trained deep learning architectures to do fault diagnosis with only few examples. Specifically, this paper investigates the effectiveness of using pre-trained deep learning architectures successfully used in natural language processing to achieve few-shot learning for full ceramic bearing fault diagnosis using acoustic emission signals.

Field study toward anomaly road damage detection with drive recorder

Masato Tsuchiya — 2023-09-04

As one of the ways to reduce road maintenance costs, road damage detection with a mobile camera is gaining attention. Most of conventional damage detection use supervised learning, nevertheless three practical drawbacks exist. Firstly, supervised learning requires a high manual cost to collect annotated data for training. Secondly, some damages are rarely observed, resulting in imbalanced data and difficulty in training an efficient model for all damage categories. Additionally, annotators may not identify such rare damages correctly. Thirdly, supervised learning cannot detect unknown categories of damages, though unknown categories are often found in a practical scene. To overcome these three drawbacks, we propose an ensemble model that combines anomaly detection and supervised damage detection. Anomaly detection can detect previously unknown and rare types of damage, while supervised damage detection ensures damages frequently observed on roads. Two different models cover wider categories of road damages. Our ensemble model is expected to achieve higher accuracy and lower manual cost.

Imbalanced Anomaly Detection Using Augmented Deeper FCDDs for Wooden Sleeper Deterioration Prognostics

Takato Yasuno — 2023-09-04

Maintaining high standards for user safety during daily railway operations is crucial for railway managers. To aid in this endeavor, top or side-view cameras and GPS positioning systems have facilitated progress toward automating periodic inspections of defective features and assessing the deteriorating status of railway components. However, collecting data on deteriorated status can be time-consuming and requires repeated data acquisition because of the extreme temporal occurrence imbalance. In supervised learning, thousands of paired data sets containing defective raw images and annotated labels are required. Concretely, the one-class classification approach offers the advantage of requiring quite a few anomalous images to optimize parameters for training large normal images. The deeper fully-convolutional data descriptions (FCDDs) were applicable to several damage data sets of concrete/steel components in structures, and fallen tree, and wooden building collapse in disasters. However, it is not yet known to feasible to railway components. In this study, we devised a prognostic discriminator pipeline to automate one class classification using the augmented deeper FCDDs for defective railway components. We also performed sensitivity analysis of the mixture and erasing augmentations, and the deeper backbone rather than the shallow baseline of convolutional neural network (CNN) with 27 layers. Furthermore, we visualized defective railway features by using transposed Gaussian upsampling. We demonstrated our application to railway inspection using a video acquisition dataset that contains wooden sleeper deterioration. Finally, we examined the usability of our approach for prognostic monitoring and fu ture work on railway component inspection.

Improvement in Identification Accuracy of a Failure Diagnostic System for a Reusable Rocket Engine

Fumihisa Nagashima — 2023-09-04

As a technology for safe and efficient operation of reusable rockets, we are developing failure diagnosis technology for reusable rocket engines. In order to follow the changes in rocket engine operating conditions, a failure diagnostic method which monitors an error vector: the difference between the predicted and measured values of the sensors was developed. The method contains anomaly detection by Mahalanobis distance and failure identification by support vector machines (SVMs). In this report, combinations of monitoring sensors of SVMs for each failure mode were optimized by using design of experiments. By using optimal sensor combinations, the F-score of SVMs were improved in all failure modes. From the results of the orthogonal table experiments, it was supposed that sensors which show the difference in failure modes are important to distinguish failure modes. In addition, a failure classifier combined with the optimized SVMs for each failure mode was developed and demonstrated. The performance of the combined failure classifier with the optimal sensor combination was mostly greater than that with all sensors. However, degradation of the classification performance was also obtained. It is necessary to consider how integrate the results of SVMs which are optimized individually.

Industrial automation findings: Smart Manufacturing Kaizen Level (SMKL) based maturity evaluation access to digital transformation and decarbonization

Nicole S. Otoki — 2023-09-04

Digital transformation is fostering the evolution within the manufacturing industry, entering a new era of smart manufacturing. Manufacturers are facing the challenge in organizing various of disruptive events, while supporting their business sustainability in a coordinated manner across manufacturing. Especially, in response to Paris Agreement, decarbonization has arisen as one of the essential keys to smart manufacturing for manufacturers to achieve their business sustainability. In this paper, we present a maturity model called Smart Manufacturing Kaizen Level (SMKL), which evaluate smart manufacturing implementation maturity by applying manufacturing data and supporting on implementation with reasonable investment. With SMKL, manufacturers can have a better understanding of their efforts in the smart manufacturing journey and how to commit to their decarbonization targets in scaling up implementation continuously. We also provide real life case studies utilizing SMKL to lower the barriers in understanding and assure the effectiveness of smart manufacturing assessment in practice.

Integration of Health Monitoring of Cutting Tools and Production Scheduling in Smart Factory

Hitoshi Komoto — 2023-09-04

Smart factory evolves by adding new functions or upgrading existing functions to meet the needs of manufacturers in the use stage of the life cycle. To reduce complexity of smart factory, these functions must be carefully designed considering interactions with other functions. This study analyzes integration of functions situated in the different layers in the functional hierarchy of smart factory. In this study, a health monitoring system for cutting tools in a shop floor, which offers a function to manage the lifetime of cutting tools, is presented. The system is integrated with a production scheduling system, which offers a function to schedule machining processes considering efficient usage of machines as well as cutting tools, while maintaining the quality machining processes. Primary evaluation of the functional integration shows that activities in the shop floor regarding selection and replacement of cutting tools are considered in defining production schedules. It also shows that such functional integration results in increase in complexity regarding the behavioral model of humans in smart factory.

Lessons Learned from Aircraft Component Failure Prediction using Full Flight Sensor Data

Changzhou Wang — 2023-09-04

Successful aircraft predictive maintenance relies on the accurate prediction of major aircraft component failures for operators to schedule and carry out maintenance operations before failure actually happens. In this paper, we share important lessons learned from our development of prognostics alerts using full flight sensor data, including various challenges of using big data, data quality issues, failure identification for data labeling, engineering-driven vs. data-driven methods, and aggregating alerts into actionable alerts. We also provide recommendations based on our experience with prognostic alerts developed and deployed for many airline operators.

Machine learning model for detecting hydrogen leakage from hydrogen pipeline using physical modeling

Yuki Suzuki — 2023-09-04

Hydrogen pipelines (HPL) are one of the hydrogen transportation systems for realizing a hydrogen society. Hydrogen leakage from HPL is a challenge because hydrogen has a wide flammable range and low minimum ignition energy. Thus, hydrogen leakage from the HPL must be rapidly detected, and appropriate actions should be taken. Leakage detection is important for safe operation of HPL. The basic leakage detection method for HPL involves monitoring the pressure and flow rate values of the sensors. However, in some cases, it is difficult to distinguish between non-leakage and leakage conditions using this method. In this study, we focus on a leakage detection method using machine learning (ML) based on the relationship between pressure and flow rate data. There are two challenges in applying the ML- based leak detection method to an HPL. First, there are insufficient operational data for ML during the process- design stage. Secondly, it is difficult to obtain the pressure and flow rate behaviors during hydrogen leakage because leakage does not occur frequently. Consequently, this study employed an unsupervised ML method based on data simulated using a physical model of the HPL. First, a physical model of the HPL (HPL model) was constructed, and an ML model for leak detection was constructed based on the data simulated by the HPL model. The leak detection capability of the ML model was verified by comparing the anomaly scores of the non-leakage and leakage conditions. From the results, the ML model can distinguish between non-leakage and leakage behaviors and identify leakage points under certain conditions. This method can contribute to the optimization of the sensors required for leak detection during the process design stage.

MLOps for PHM Systems

Mikael Yemane — 2023-09-04

Advances in machine learning (ML) techniques allow practitioners to generate substantial predictive value from historical data. Modern sensors generate vast amounts of data which inform prognostic health management (PHM) systems. As ML techniques continue to grow in importance for PHM, the system that manages and deploys ML models becomes critical for successful production software. Machine Learning Operations (MLOps) is centered around implementing continuous integration and deployment (CI/CD) practices in the context of ML applications. We will present MLOps designs for deploying machine learning based PHM software and discuss ML pipelines that automate data ingestion, model training, testing, deployment, and monitoring. The principles we will examine ensure model quality, performance, and software stability. We will call attention to important design considerations and demonstrate solutions for the full model lifecycle when building MLOps pipelines for PHM systems.

Model Based and Big Data Enabled Predictive Maintenance Capability Development Experience

Mark Mazarek — 2023-09-04

Airplane health management and predictive maintenance have been in place for many years and has been implemented across the industry on various platforms with some success and some challenges. Predictive maintenance leaders have a unique challenge of developing talent pipelines, technology focus areas, balanced with delivery of prognostic insights to customers. This paper will discuss strategies, lessons learned for how to build, grow and sustain a team of engineers, data scientists, software developers and others focused on delivering aerospace prognostic insights. It will also consider how to implement metrics for measuring performance such that the team is persistent, and motivated in their endeavors.

Modeling of Journal Bearings for Wear Diagnosis and Its Verification Using SVM

Masaki Goto — 2023-09-04

Recently, deterioration of infrastructure facilities has become problem. To reduce inspection costs, wear diagnosis of journal bearings in operating conditions using machine learning has been studied. However, constructing a highly accurate machine learning model requires training data, and it may be difficult to conduct numerous tests depending on the application. In this study, a mathematical model of a rotor system for wear diagnosis of journal bearings was developed, and the vibration data was obtained when the clearance changed due to wear. Then, changes in the features used for condition monitoring were examined. Furthermore, the important features for wear diagnosis were selected and SVM models were constructed to verify the mathematical model.

Non-destructive Prognostics for Rolling Bearings by Eddy Current Testing

Daisuke Kobayashi — 2023-09-04

Rolling bearings, which are assembled into various industrial machinery, are regularly replaced after being used for a period of time, even if they have not failed. It is important to predict failure of rolling bearings not only for safe operation of machineries but also for resource conservation. The X-ray diffraction (XRD) is known as an effective method for estimating the remaining useful life (RUL) of rolling bearings. However, it is a destructive approach sometimes requiring cutting bearing rings. Therefore, non-destructive and simple diagnostic method for estimating RUL of rolling bearings using Eddy Current Testing (ECT) has been developed by focusing on the experimental evidence that changes in microstructure of the steel cause changes in the magnetic property. Rolling contact fatigue tests were conducted using several types of rolling bearings, and it was found that the ECT measurement results on raceway surface show a determined behavior with fatigue progress. The tendency of changes did not depend on bearing type, material or heat treatment. Additionally, measurement results by ECT were related to those by XRD. Above experimental results suggest that ECT can be applied to estimate RUL of rolling bearings as a non-destructive and simple method.

Nonlinear Model Predictive Control using Neural ODE Replicas of Dynamic Simulators

Shumpei Kubosawa — 2023-09-04

We propose simulation-based nonlinear model predictive control as a first step towards autonomous decision-making for stable operation of large complex dynamical systems such as chemical plants. The effect of abrupt external disturbances should be quickly eliminated, taking into account such complex dynamic responses, to maintain stable production. In this paper, we propose a control system to eliminate these effects. The system uses engineering models, including dynamic simulators, based on chemical engineering knowledge. Dynamic simulators are generally not differentiable with respect to actions; however, differentiable models are advantageous for fast nonlinear optimization. To take advantage of both reliable dynamic simulators and differentiable models, we introduce neural ordinary differentiable equation models and clone the behaviour of simulators on them. The cloned differentiable neural replica model is then incorporated into a gradient-based nonlinear model predictive control. Evaluation of this method in a real methanol distillation plant confirms that it can significantly remove abrupt heavy rain disturbances compared to existing methods.

Observation Maintenance for Bridges Using Early Detection of Deterioration Progress

Hitoshi Ito — 2023-09-04

In Japan, the results of inspections show that many bridges require repairs. However, Japanese municipalities do not have sufficient time and budget, and the progress of bridge repair is insufficient. Almost all the bridges managed by municipalities are short-span bridges. Therefore, a management system that selects and focuses on bridge repair based on performance evaluation would be effective. This paper presents an overview of a bridge management system that applies observational maintenance based on the early detection of deterioration, and its effectiveness is demonstrated.

Operational Prognostic Model Evaluation

Shashvat Prakash — 2023-09-06

Prognostic analytic models have become a viable way to reduce operational interruptions when sufficient timely data is available. This work describes a set of evaluation metrics which can characterize model performance as a degradation estimate and as a decision enabler. The model accuracy over time is assessed against a correlation with the remaining useful life. This yields both a prediction accuracy and confidence interval. The decision can be based on the level of confidence around the prediction, which is based on both how far into the future the event is predicted and how well the current health and its deterioration is estimated. With an effective means of evaluating prognostic models, better benchmarks can be established to communicate model effectiveness and appropriately schedule routine service.

Outlier Analysis of Bridge Deflections Using Satellite SAR and Structural Simulation: A Case Study on a Collapse Accident in a Water Pipe Bridge in Japan

Kosuke Kinoshita — 2023-09-04

This paper proposes a method for outlier analysis of bridge deflections using satellite Synthetic Aperture Radar (SAR) and structural simulation based on the finite element (FE) analysis. The proposed method is unique in combining the following two points: (1) detecting anomalous displacement of bridges before a severe accident occurs using satellite SAR displacement analysis, and (2) attempting to estimate the factors anomalous responses on the bridge, such as damages on structural members, using FE analysis. This paper describes the results of the practical application of the proposed method to MUSOTA water pipe bridge in Japan, which collapsed on October 3, 2021.

Pipe Corrosion Inspection System based on Human-in-the-Loop Machine Learning

Toshihiro Shimbo — 2023-09-04

The aim of this study was to improve the efficiency of external corrosion inspection of pipes in chemical plants. Currently, the preferred method involves manual examination of images of corroded pipes; however, this places significant workload on human experts owing to the very high number of such images. To address this issue, we developed an artificial intelligence (AI)-based corrosion diagnosis system and implemented it in a factory.

Initially, interviews were conducted to understand the decision-making processes of human experts. Subsequently, we converted their tacit knowledge into explicit knowledge, which was used to define the training data for the machine learning (ML) model. The predictions of the ML model were compared with the manually obtained results, exhibiting an accuracy of 70 %.

The proposed architecture was based on human-in-the-loop ML. It included a process to retrain the ML model using manual results gathered during operation. It was operated using a collaborative approach, in which human experts supported the ML model under development.

The proposed model enhanced the efficiency of the inspection process successfully.

Power Consumption Optimization for Electric Arc Furnace with Time Series Prediction

Jaehyuk Lee — 2023-09-04

Optimizing power consumption for electric arc furnace (EAF) has a critical impact for maximizing productivity. To achieve the goal, we propose an AI based algorithm that determines optimal timing for recharging scrap to EAF. More specifically, we predict power consumption and time duration required for melting scrap considering scrap types and amounts of each type of scrap. Furthermore, with the advance in explainable AI, we offer guidance for the optimal timing of recharging scrap. We evaluate the performance on a real site and successfully reduce scrap charging time of 3% and power consumption of 7.1%, 53,802 Japanese Yen.

Prediction of Remaining Life of Turbo Pump Inducer for Spacecraft Using Cumulative Damage Model

Hatsuo Mori — 2023-09-04

In the near future, to operate reusable spacecraft safely and efficiently, it is necessary to have failure prediction technology for reusable rocket engines. Among the components that constitute a rocket engine, the failure of the turbo pump inducer has a significant impact on missions. Therefore, we have been researching to monitor the remaining life of the inducer.

This method estimates the fluctuating stress field excited in the inducer section based on the time-series signals from pressure sensors installed upstream. It predicts the remaining life using a Rain flow model to account for high- cycle fatigue (HCF) phenomena. To validate the effectiveness of the proposed method, we attempted to acquire experimental data, particularly for the challenging pressure-stress transfer section where concrete specifications are difficult to define. Through experiments using an elemental model, we demonstrated that a certain degree of remaining life can be predicted based on measurement from the upstream sensor.

This indicates that the proposed method is a promising option for solving the remaining life prediction problem of reusable spacecraft.

Predictive Maintenance for station equipment and Applications for the Space field

Ryoma Yokoyama — 2023-09-04

Our company have a lot of stations fare equipment such as
ticket gate machine. Maintenance and inspection of these
require a lot of labor and cost. In this paper, we aimed to solve
this problem by applying failure detection a form of machine
learning. Currently, the system has been installed in all of our
station equipment, and has reduced the number of failures by
20% and inspections by 30%, helping to optimize our
operations. In the future, we plan to apply this method and
our knowledge of maintenance and operation to evaluate the
health and management of satellites in the space field.

Predictive modelling for airline technical operations

Dimitri Reiswich — 2023-09-04

The digital platform AVIATAR leverages aircraft and maintenance data as well as advanced algorithms to optimize technical operations. Optimization is achieved by utilizing predictive maintenance algorithms which lead to a reduction of costs and operational incidences. The rise of AI algorithms combined with an increase of the available data to feed those algorithms provides both an opportunity as well as challenge to the traditional approaches which were in use in the aviation industry for decades. We demonstrate how AVIATAR’s data science team develops state of the art predictive maintenance models by combining advanced statistical models and AI with the unique engineering know how of Lufthansa Technik. We will show the immense value of full flight data in order to achieve this and provide an outlook into the potential future role of algorithms in the aviation industry.

Proposal of a Time Series Anomaly Detection Method Using Image Encoding Techniques

Ryo Sakurai — 2023-09-04

Time series anomaly detection is considered to play a major role in many areas of society. Models using RNN, which are well suited for time series data, have been studied. However, models using RNN have the problem of high cost; image en- coding approaches combining Gramian Angular Fields(GAF) and Autoencoder are less expensive than RNN. However, the accuracy in existing studies is not as good as RNN. In this paper, we propose a time-series anomaly detection frame- work that first focuses on the structural issues of GAF and the reconstruction accuracy of Autoencoder. Experiments were conducted to verify the effectiveness of the framework. The results showed that the approach focusing on the structural is- sues of GAF achieved a significant improvement in accuracy, while the approach focusing on improving the reconstruction accuracy of the Autoencoder network decreased the anomaly detection accuracy. The reason for the lower accuracy was found to be that the networks with higher reconstruction accuracy accurately reconstructed even the anomaly images, making anomaly detection based on L1 errors impossible. These results indicate that an approach that focuses on the structural problems of GAF is effective, while an approach that improves the reconstruction accuracy of Autoencoder is not necessarily effective.

Quality Management for Machine Learning Systems

Yutaka Oiwa — 2023-09-04

We have been developing a methodology for process-based quality management of machine learning-based AI systems. Our fruit is compiled as a guideline document named “Machine Learning Quality Management Guideline”, published as our technical report. We will describe our background motivation, surrounding situation and our proposal for quality management.

Resilient Operation Planning for CubeSat Using Reinforcement Learning

Shuntaro Kuroiwa — 2023-09-04

This study proposes an autonomous operation procedure for a CubeSat by applying reinforcement learning based on resilient engineering. The CubeSat requires rapid judgment in every visible window based on a sufficient understanding of the health conditions of the satellite from limited telemetry data due to the limited communication performance and poor protection functions from the harsh environment. This study first performs a risk analysis by using System Theoretic Process Analysis (STPA) to evaluate the risk scenario of the Cube-Sat. In order to successfully operate the missions while avoiding the risk scenarios, reinforcement learning is applied to learn adequate behaviors according to the satellite situations such as the temperature and voltage of the installed battery, the sunlight and eclipse phase and the mission progress and plan. Through numerical examples, the validity of the proposed method is illustrated.

RUL Estimation for Package Failure of Power Electronic Devices Using Integral Mean of Precursor Signal

Zhonghai Lu — 2023-09-04

Package failure like bond-wire lift-off is one common cause of failure for discrete power electronic devices such as Schottky diodes. To estimate their Remaining Useful Lifetime (RUL), forward voltage drop is often used as the precursor signal. Prior researches use the direct forward voltage feature and its derived features to construct neural networks for RUL prediction. These features can reflect the instant health condition of the device in the current time or time window, but miss to represent the accumulated effect of gradually decreasing health conditions. In the paper, we formulate the integral mean feature of forward voltage drop and propose to use it to conduct RUL estimation. By the integral mean feature, we are able to capture the device's health condition in an accumulated fashion. Our experiments show that our approach is superior in generalization performance when compared to the forward voltage feature and its statistical features based neural networks for RUL estimation.

Semi-supervised machine learning for motor eccentricity fault diagnosis

Bingnan Wang — 2023-09-04

Eccentricity is one major indicator of mechanical faults in electric machines and needs to be detected early to avoid machine failures. Data-driven techniques based on machine learning and deep learning algorithms have been proposed in recent years for motor fault detection. However, majority of these methods use supervised learning algorithms and require large, labelled datasets, which can be challenging to obtain. In this paper, we propose a semi-supervised learning method based on a deep generative model using variational auto-encoder for eccentricity fault quantification. Good prediction accuracy can be achieved when only a small subset of training data has labels.

Size Estimation of Flaking in Rolling Bearings Using Deep Learning with Explainability

Osamu Yoshimatsu — 2023-09-04

To improve the availability of rotating machines such as wind turbines, where rolling bearing replacement is costly and time-consuming, it is effective to estimate the damage progression of the rolling bearings. As one of the damage progressions, the size of flaking in rolling bearings is estimated by vibration analysis using rule-based methods. However, these rule-based methods require expert knowledge of rolling bearings. Therefore, an estimation model using deep learning was proposed and its performance was evaluated. Furthermore, it was verified that the proposed model had extracted the features of physical phenomena using Grad-CAM.

SKYWISE - Big data platform as a foundation of airlines predictive and health monitoring

William Bernard — 2023-09-04

By abstracting the complexity of data collection, combining engineering and operational data, and enabling collaboration across the industry in a rich analytical environment, airlines overcome intrinsic challenges of the predictive and health monitoring such as data silo’s, complex data management and data-driven decision enablers. Make the data visible and understandable through strong and flexible data integration. Make data actionable through digital twin and collaboration. Make data proactive by implementing predictive models and bringing machine learning into operations. Airbus Skywise, powered by Palantir Technologies and fueled by Airbus expertise, provides Airlines with the leading aviation data platform to address aircraft operation challenges.

Study of Localization of Partial Discharges in Oil-filled Transformers using Acoustic Emission Signals

Yasutomo Otake — 2023-09-04

The measurement of partial discharges (PD) in power transformers is crucial for fault detection and maintenance scheduling. In this paper, the relationship between detection intensity, type of PD source and propagation distance is investigated using an acoustic emission (AE) sensor. The AE wave intensity by the corona discharges are relatively strong. Creeping discharges were next, followed by PD in bubbles. Furthermore, two methods for calculating time difference of arrival (TDOA) in locate calculations, energy reference and generalized cross-correlation (GCC), were experimentally compared. The results showed that the energy reference method is suitable when sensors can be placed around the tank, while the GCC method is suitable when sensors are concentrated in specific parts of the tank. This finding may contribute to improving the accuracy of maintenance diagnostics.

Systemic symptom detection in telemetry of ISS with explainability using FRAM and SpecTRM

Shota Iino — 2023-09-04

Explainability is important for machine learning-based anomaly detection of safety critical systems. In this respect, we propose a new systemic symptom detection method by combining two methodologies: the Functional Resonance Analysis Method (FRAM) and the Specification Tools and Requirement Methodology-Requirement Language (SpecTRM-RL) with machine learning-based normal behavior prediction model. The method was verified with data of thermal control system of Japanese Experimental Module of the International Space Station, and the result found that the proposed method enables flight controllers and specialists to obtain additional information for identifying causes of anomaly with the method.

Study on Fault Diagnosis in a Spacecraft Propulsion System

Kazushi Adachi — 2023-09-04

The propulsion system in a spacecraft is an important subsystem for orbit transfer and attitude control. A fast and accurate fault diagnosis system contributes to the safety of the entire system. As the system becomes more complex, identifying faults, their locations, and root causes becomes increasingly difficult. This study utilized Principal Component Analysis (PCA) and feature optimization with Fast Fourier Transform (FFT) analysis using greedy algorithm to achieve fault diagnosis systems for spacecraft to replace the current operation based on the expert knowledge. By applying PCA to simulation data for the faults were successfully detected and their locations and root causes identified.

Study on the Estimation of Concrete Defects Volume on Dam Body Surface

Akira Ishii — 2023-09-04

To maintain the safety and functionality of dams over the long term, it is necessary to make inspections more labor-saving and efficient using the latest technology and to improve the sophistication of inspections based on data. Although dam inspections cover a wide range of items, this study focuses on the continuous monitoring of popouts, a phenomenon of concrete deterioration occurring on the surface of a dam body. It is difficult to predict whether a popout will occur from information on the body surface of the dam, owing to the generation mechanism of the popout. The number of popouts was monitored over time; however, no examples of shape changes were monitored over time. Advancements in various digital technologies are required to accurately evaluate changes in the dam body's surface over time; therefore, in this study, three-dimensional (3D) point-cloud data is created by the Structure from Motion (SfM) from images captured by a Unmanned Aerial Vehicle (UAV) of the concrete defect area due to the popout in an arch dam in the Tohoku region of Japan. The volume of concrete defects of a popout in each of two different periods was calculated by estimating the plane shape of the surface of the dam body. In addition, the shapes of two popouts were compared to confirm the possibility of predictive signs of change.

Time Shifting Data Augmentation to Alleviate Class-Imbalance Problem for Cross-Domain Bearing Fault Diagnosis

Donghwi Yoo — 2023-09-04

This paper presents a new cross-domain fault diagnostic method for rolling element bearings with class-imbalanced datasets. The key idea to alleviate the class imbalance problem is the incorporation of the data augmentation strategy. This study proposes a new data augmentation technique, namely, time shifting data augmentation (TS- DA). Synthetic data is generated to balance the number of normal and fault data. The validity of the proposed method is evaluated using a dataset from the bearing testbed. The results show that the proposed method augments different types of bearing fault data effectively and outperforms existing methods under the class imbalance problem.

Shipping inspection trial of quantum machine learning toward sustainable quantum factory

Takao Tomono — 2023-09-04

In recent years, the diversification of consumer values has led to an increase in the number of small-quantity, high-mix products. For many manufacturing companies, shipping inspections of such products are of great importance. As all products have the same value, good and defective products need to be efficiently identified. Now, a promising future application of quantum technology is considered to be quantum machine learning. We believe that the quantum classifier for SVMs using quantum kernels is one of the areas where quantum advantages can be demonstrated. At present, there are few examples of quantum classifiers applied to real problems in manufacturing processes. In this study, we aim to build a classifier that can demonstrate the quantum advantage and compare SVMs using classical and quantum kernels with conventional ResNet. Initially, a binarised image was generated after image pre-processing. After principal component analysis and dimensionality reduction were performed on the images, SVM with kernels was carried out. The kernel-based SVMs was then compared with the conventionally implemented Residual neural network (ResNet) using an evaluation index: F1-score. The results showed that the F1-score of SVMs using classical kernels was equivalent to that of Resnet. In addition, SVMs using quantum kernels showed higher F1-score than ResNet. In addition, the impact of the feature map and principal components of the quantum kernel was also investigated. It was found that when the feature map became more complex, conversely, circuit generation took more time. It was also found that the principal components are highly relevant to the image and cannot lead to simple results. In the future, we plan to accumulate more experimental data, look for scenes where quantum machine learning can be used and apply it to the manufacturing field.