Annual Conference of the PHM Society

Evaluation of the Use of the Angular Domain and Order Domain in a Bearing Fault Detection Framework using Deep Learning

Racquel Knust Domingues — 2024-11-05

Bearing failures are very common in the industrial environment, requiring effective fault detection methods, which can be categorized into physics-based, knowledge-based and data-driven types. Data-driven methods are efficient in differentiating healthy conditions from faulty conditions by characterizing machine signals, involving stages of data acquisition, feature extraction, and condition determination. Traditionally, feature extraction and condition determination were manual, but advances in artificial intelligence and machine learning, especially deep learning, have automated this process. Although deep learning automatically learns the best features from the input data, the signal domain can influence the model's performance. Time and frequency domain representations are widely used in fault detection methodologies using vibration signals, while angular and order domains are more common in variable operating conditions, but direct use of these domains with deep learning is still rare in the literature. Considering this, this study evaluates a bearing fault detection methodology using vibration signals in different domains (time, frequency, angular, and order) under various rotational conditions. Three distinct approaches were tested to assess the effectiveness of these representations. The results indicated that the frequency domain representation had the best overall performance and the study concluded that the angular and order domains do not offer significant advantages compared to the frequency domain. Nonetheless, it is recommended to conduct a more in-depth analysis with more diverse datasets, especially those containing early-stage bearing fault signals.

A Multidisciplinary Framework for Vibration-Based Gear Fault Diagnosis Using Experiments, Modeling, and Machine Learning

Lior Bachar — 2024-11-05

Vibration-based gear diagnosis is crucial for ensuring the reliability of rotating machinery, making the monitoring of gear health essential for preventing costly downtime and optimizing performance. This study proposes a multidisciplinary framework to enhance gear diagnosis, that aligns with the new era of digital twins by integrating experiments, dynamic modeling, physical preprocessing, and machine learning. Within this framework, we focus on three core procedures: domain adaptation to reduce discrepancies between measured data and synthetic data generated by dynamic models; physical preprocessing, grounded in in-depth investigations dictating signal processing and feature engineering techniques; and learning algorithms, encompassing the process of training AI-based models. We demonstrate this framework through a comprehensive case study of localized tooth fault diagnosis, using controlled-degradation tests and realistic simulations. First, we detect faults using unsupervised learning algorithms; then, we use zero-shot-learning for classifying between localized and distributed faults; finally, we adopt a one-shot-learning strategy for severity estimation. Above all, this hybrid framework bridges the gaps between physical-based and AI-based approaches by combining physical knowledge and advanced algorithmics with machine learning. This contributes to the PHM field by offering valuable insights into integrating different aspects of research, thereby enhancing performance in gear diagnosis tasks.

On the Construction of Energy Efficiency-based Degradation Indicator for Photovoltaic Solar Inverters

Jorge Ruiz Amantegui — 2024-11-05

Photovoltaic systems are essential in the renewable energy sector, addressing global energy needs. PV inverters, which convert DC from solar panels to AC for grid use, are the most failure-prone components in these systems. This study aims to develop a degradation indicator based on energy efficiency for PV inverters to enhance their reliability and lifetime management. Through analysis of data from 35 PV plants in central Europe, involving eight inverter brands and fourteen models, this research identifies trends in inverter efficiency degradation. Despite literature suggesting minimal efficiency impact over time, our findings demonstrate measurable efficiency degradation, providing a new key performance indicator for proactive maintenance and replacement strategies.

SPIKE-Dx

Chetan Kulkarni — 2024-11-05

Diagnostic systems are important for many aerospace systems, which are severly limited in available power, like cubesats or UAVs. Therefore, traditional diagnostics systems cannot be used due to their substantial footprint and constraints. In this paper, we present our very low power diagnostic tool SPIKE-Dx. to monitor critical systems with constrained computational and energy resources. This is made possible through spiking neural networks (SNNs), which are executable within optimized simulation environments and further implemented on on cutting-edge neuromorphic hardware.

Based upon FMEA (Failure Mode and Effect Analysis) framework, Diagnostic Bayesian Networks (DBNs) can be constructed that provide powerful means for diagnostic reasoning. In this paper, we describe such DBNs and a method to automatically translate the DBN into highly structured networks of spiking neurons for execution in SPIKE-Dx.

Temporal Convolutional Network-based Approach for Forecasting Fluctuations Differential Pressure in Reverse Osmosis Systems

The Son Pham — 2024-11-05

Providing forecasts of pressure fluctuations and changes will aid in selecting appropriate maintenance strategies to optimize efficiency and costs. This paper presents a deep-learning-based model to forecast the degradation evolution of membrane biological fouling in RO (Reverse Osmosis) systems. Although applying deep learning in forecasting still faces many challenges, applying convolutional operations in convolution 1D has yielded promising results for sequential data, particularly time series data. Thus, in this paper we study and develop the 1D convolution operation-based Temporal Convolutional Network (TCN) model to predict pressure dynamics at both ends of the RO vessel. In addition, since the deep learning technique has yet to be widely explored in this field, thus we also need to pre-process the data collected from the Carlsbad Desalination Plant in California, such as the proposed model can identify complex relationships between timestamps and pressure features. The experiment results were evaluated and compared with other existing models, such as LSTM, CNN & LSTM, and GRU. The obtain results show that the TCN-based prediction model had the slightest error in the test dataset.

Assumption-based Design of Hybrid Diagnosis Systems

Daniel Jung — 2024-11-05

Hybrid diagnosis systems combine model-based and data-driven methods to leverage their respective strengths and mitigate individual weaknesses in fault diagnosis. This paper proposes a unified framework for analyzing and designing hybrid diagnosis systems, focusing on the principles underlying the computation of diagnoses from observations. The framework emphasizes the importance of assumptions about fault modes and their manifestations in the system. The proposed architecture supports both fault decoupling and classification techniques, allowing for the flexible integration of model-based residuals and data-driven classifiers. Comparative analysis highlights how classical model-based and pure data-driven systems are special cases within the proposed hybrid framework. The proposed framework emphasizes that the key factor in categorizing fault diagnosis methods is not whether they are model-based or data-driven, but rather their ability to decuple faults which is crucial for rejecting diagnoses when fault training data is limited. Future research directions are suggested to further enhance hybrid fault diagnosis systems.

Unsupervised Fault Detection in a Controlled Conical Tank

Joaquín Ortega — 2024-11-05

Current trends in the Industrial Internet of Things (IIoT) have increased the sensorization of systems, thus increasing data availability to apply data-driven fault detection and diagnosis techniques to monitor these systems. In this work, we show the capabilities of an information-driven method for detecting and quantifying faults in a subsystem common among a broad range of industries: the conical tank. Our main experiment consists of using a simple black-box model (multi-layer perceptron -- MLP) to capture the dynamics of a PID-controlled conical tank built in Simulink and then induce pump failures of different severities; the derived data-driven indicators that we developed increase with the severity of the fault validating its usefulness in this controlled setting. A complementary experiment is carried out to enrich our analysis; this consists of simulating an open-loop discrete-time version of the conical tank to explore a range of fault severity and analyze the distribution of the indicators across this range. All our results show the applicability of the data-driven fault monitoring method in conical tanks subjected to either open- or closed-loop operation.

Physics-Informed Data-Driven Approaches to State of Health Prediction of Maritime Battery Systems

Azzeddine Bakdi — 2024-11-05

Battery systems are increasingly being used for powering ocean going ships, and the number of fully electric or hybrid ships relying on battery power for propulsion and maneuvering is growing. In order to ensure the safety of such electric ships, it is of paramount importance to monitor the available energy that can be stored in the batteries, and classification societies typically require that the state of health (SOH) of the batteries can be verified by independent tests – annual capacity tests. However, this paper discusses physics-informed data-driven approaches to online diagnostics for state of health monitoring of maritime battery systems based on a combination of physical knowledge, physic-based models, insights from extensive characterization tests and operational sensor data collected from the batteries during actual operation. This represents an alternative approach to the annual capacity tests for electric ships that is found to be sufficiently robust and accurate under certain conditions. Previous attempts with purely data-driven models, including both cumulative and snapshot models, semi-supervised learning and simple models based on the state of charge did not achieve the required reliability and accuracy for them to be utilized in a ship classification perspective, as presented at previous PHM conferences. However, preliminary results from the physics-informed data-driven method presented in this paper indicate that it can be relied on for independent verification of state of health as an alternative to physical tests. It has been tested on battery cells cycled in laboratory degradation tests as well as on field returns from batteries onboard ships in service. Notwithstanding, further validation and verification of the method is recommended to further build confidence in the model predictions.

Feature-weighted Random Forest with Boruta for Fault Diagnosis of Satellite Attitude Control Systems

Shaozhi Chen — 2024-11-05

The performance of random forest (RF) based satellite attitude control system (ACS) fault diagnosis methods is limited by uninformative features in high-dimensional data. To solve this problem, we proposed a feature-weighted random forest with Boruta (FWRFB) based fault diagnosis method is proposed for fault diagnosis of ACSs. Firstly, a Boruta feature selection algorithm is used to obtain a feature set and determine significant feature weights. Subsequently, a novel feature-weighted random forest (FWRF) algorithm is designed, which utilizes feature-weighted random sampling instead of simple random sampling to generate feature subsets in the RF. The FWRFB effectively utilizes the feature information while mitigating noise interference. Finally, a FWRFB-based diagnostic module is developed for online fault diagnosis of ACSs. The effectiveness of the proposed method is verified by the ACS data from a semi-physical simulation platform.

Using Explainable Artificial Intelligence to Interpret Remaining Useful Life Estimation with Gated Recurrent Unit

Marcia L. Baptista — 2024-11-05

In engineering, prognostics can be defined as the estimation of the remaining useful life of a system given current and past health conditions. This field has drawn attention from research, industry, and government as this kind of technology can help improve efficiency and lower the costs of maintenance in a variety of technical applications. An approach to prognostics that has gained increasing attention is the use of data-driven methods. These methods typically use pattern recognition and machine learning to estimate the residual life of equipment based on historical data. Despite their promising results, a major disadvantage is that it is difficult to interpret this kind of methodologies, that is, to understand why a certain prediction of remaining useful life was made at a certain point in time. Nevertheless, the interpretability of these models could facilitate the use of data-driven prognostics in different domains such as aeronautics, manufacturing, and energy, areas where certification is critical. To help address this issue, we use Local Interpretable Model-agnostic Explanations (LIME) from the field of eXplainable Artificial Intelligence (XAI) to analyze the prognostics of a Gated Recurrent Unit (GRU) on the C-MAPSS data. We select the GRU as this is a deep learning model that a) has an explicit temporal dimension and b) has shown promising results in the field of prognostics and c) is of simplified nature compared to other recurrent networks. Our results suggest that it is possible to infer the feature importance for the GRU both globally (for the entire model) and locally (for a given RUL prediction) with LIME.

Expected First Occurrence Time of Uncertain Future Events in One-Dimensional Linear Systems

David E. Acuña-Ureta — 2024-11-05

The rapid advancement of machine learning algorithms has significantly enhanced tools for monitoring system health, making data-driven approaches predominant in Prognostics and Health Management (PHM). In contrast, model-based approaches have seen modest progress, as they are often constrained by the need for prior knowledge of specific governing equations, limiting their applicability to a wide range of problems. Recently, rigorous theoretical foundations have been established to extend dynamical systems theory by incorporating prognosis of uncertain events. This article leverages this formal framework to introduce and demonstrate a fundamental mathematical result for one-dimensional linear dynamical systems. The presented theorem offers an exact expression for calculating the expected time at which an event will first occur in the future. Unlike typical thresholds, this event is triggered by a hazard zone, defined as an uncertain event likelihood function over the system's state space. Applications of this theorem can be found in implementing real-time prognostic frameworks, where it is crucial to quickly estimate the magnitude of impending failures. Emphasis is placed on minimizing computational burden to facilitate prognostic decision-making.

Damage Diagnostics of Miter Gates Using Domain Adaptation and Normalizing Flow-Based Likelihood-Free Inference

Yichao Zeng — 2024-11-05

Miter gates are vital civil infrastructure components in inland waterway transportation networks. To provide risk-informed insights for decisions related to repair and maintenance, sensors have been installed on some miter gates for monitoring. Despite the monitoring system's ability in collecting a large volume of monitoring data, accurately diagnosing damage state in such large structures remains challenging due to the lack of labeled monitoring data, since these structures are designed with high reliability and for a long operation life. This paper addresses this challenge by proposing a damage diagnostics approach for miter gates based on domain adaptation. The proposed approach consists of two main modules. In the first module, Cycle-Consistent generative adversarial network (CycleGAN) is employed to map monitoring data of a miter gate of interest and other similar yet different miter gates into the same analysis domain. Subsequently, a normalizing flow-based likelihood-free inference model is constructed within this common domain using data from source miter gates whose damage states are labeled from historical inspections. The trained normalizing flow model is then used to predict the damage state of the target miter gate based on the translated monitoring data. A case study is presented to demonstrate the effectiveness of the proposed method. The results indicate that the proposed method in general can accurately estimate the damage state of the target miter gate in the presence of uncertainty.

Adversarial Attacks and Defenses in Multivariate Time-Series Forecasting for Smart and Connected Infrastructures

Pooja Krishan — 2024-11-05

The emergence of deep learning models has revolutionized various industries over the last decade, leading to a surge in connected devices and infrastructures. However, these models can be tricked into making incorrect predictions with high confidence, leading to disastrous failures and security concerns. To this end, we explore the impact of adversarial attacks on multivariate time-series forecasting and investigate methods to counter them. Specifically, we employ untargeted white-box attacks, namely the Fast Gradient Sign Method (FGSM) and the Basic Iterative Method (BIM), to poison the inputs to the training process, effectively misleading the model. We also illustrate the subtle modifications to the inputs after the attack, which makes detecting the attack using the naked eye quite difficult. Having demonstrated the feasibility of these attacks, we develop robust models through adversarial training and model hardening. We are among the first to showcase the transferability of these attacks and defenses by extrapolating our work from the benchmark electricity data to a larger, 10-year real-world data used for predicting the time-to-failure of hard disks. Our experimental results confirm that the attacks and defenses achieve the desired security thresholds, leading to a 72.41% and 94.81% decrease in RMSE for the electricity and hard disk datasets respectively after implementing the adversarial defenses.

Data-driven Detection of Engine Faults in Infrequently-driven Ground Vehicles

Ethan Kohrt — 2024-11-05

We investigated the detection of engine faults in infrequently driven ground vehicles using data-driven methods based on neural network autoencoders. Multivariate time-series data from the infrequently driven vehicles under investigation had limited coverage of operating conditions. Hence, a considerable part of this work focused on identifying suitable vehicles, relevant signals, and pre-processing the data. We trained autoencoder models on eight vehicles with known faults and detected faults in six. Four of the faults were detectable under idle conditions and four were detectable under driving conditions. Model evaluations required human inspection to distinguish fault detections from other anomalies. We detail our procedures for pre-processing, model development, and post-processing, and we include a discussion on our interpretations of the model results.

Health Indicator Development for Low-Voltage Battery Diagnostics and Prognostics in Electric Vehicles

Xinyu Du — 2024-11-05

Each electric vehicle (EV) requires a low-voltage (e.g., 12V) auxiliary battery to provide electric power to onboard electronic control units, lighting systems, and various sensors during power off. Therefore, when the low-voltage battery is in low state of health (SOH) or low state of charge (SOC), it may cause no-start events. The existing OnStar Proactive Alert service can effectively predict low SOC or low SOH events for low-voltage batteries in Internal Combustion Engine vehicles using cranking signals. However, it does not work for EVs since there is no cranking event. In this work, a diagnostic and prognostic solution for the low-voltage battery of EVs is proposed. Four novel health indicators (HIs) along with the decision-making system are developed based on equivalent circuit models. Furthermore, the selection process of appropriate HIs tailored to various operational states of the vehicle is described. The validation results based on GM test EV data have demonstrated the effectiveness and robustness of the proposed solution.

Comparing Feature and Trajectory-Based Remaining Useful Life Modeling of Electrical Resistance Heating Wires

Simon Mählkvist — 2024-11-05

Industrial heating significantly contributes to global greenhouse gas emissions, accounting for a substantial portion of annual emissions. The transition to fossil-free operations in the heating industry is closely linked to advancements in industrial electrical heating systems, especially those using resistance heating wires. In this context, Prognostics and Health Management is crucial for enhancing system reliability and sustainability through predictive maintenance strategies.

The integration of machine learning technologies into Prognostics and Health Management has significantly improved the precision and applicability of Remaining Useful Life modeling. This improvement enables more accurate predictions of component lifespans, optimizes maintenance schedules, and enhances operational efficiency in industrial heating applications. These developments are essential for reducing greenhouse gas emissions in the sector.

This paper serves as a guide for conducting Remaining Useful Life modeling for industrial batch processes. It evaluates and compares two methodologies: deep learning-based approaches using full time-series data, such as recurrent neural networks and their variants, and feature-engineering-based methods, including random forest regression and support vector machines. Our results show that the feature-oriented approach performs better overall in terms of predictive accuracy and computational efficiency. The study includes a detailed sensitivity analysis and hyperparameter estimation for each method, providing valuable insights into developing robust and transparent Prognostics and Health Management systems. These systems are crucial in supporting the heating industry’s move towards more sustainable and emission-free operations.

The findings reveal that feature-oriented methods are both performant and robust, particularly excelling in handling outliers. The random forest regression model, in particular, demonstrated the highest performance on the test dataset according to the chosen evaluation metrics. Conversely, trajectory-oriented methods exhibited less bias across varying levels of degradation, a helpful characteristic for Prognostics and Health Management systems. While feature-oriented methods tend to systematically underestimate Remaining Useful Life at high true values and overestimate it at low actual values, this issue is less pronounced in trajectory-oriented models. Overall, these insights highlight the strengths and limitations of each approach, guiding the development of more effective and reliable predictive maintenance strategies.

A Self-supervised Learning Approach for Anomaly Detection in Rotating Machinery

Fabrizio De Fabritiis — 2024-11-05

Early fault detection in rotating machinery needs careful expert analysis of vibration data for monitoring a component state. Online methods that automatically set a threshold and raise an alarm when the vibration signature is anomalous are meant to efficiently manage key assets in a preventive maintenance plan.

In recent years a focus has raised on data driven methods in parallel with the increasing attention towards machine learning and, particularly, deep learning models. In this regard, for rotating equipment components, an important aspect relates to labelled data scarcity for supervised training. On the other hand, the advent of the Internet of Things allows to gather data from multiple assets with relevant information on the asset state itself. Self-supervised learning methods in deep learning application are currently tackling this problem. Investigating Self-learning approaches to integrate domain knowledge and learn relevant features from unlabeled data is therefore important for condition monitoring applications.

In this paper a methodology is proposed based on cycle consistency representation learning for training an embedder network on univariate unlabeled data. In order to learn a distance metric in the embedding space the original data are transformed to generate sequences of augmented inputs to enforce learnable pattern similarity in the augmented pairs. A differentiable cycle-consistency loss is chosen to maximize the numbers of augmented pairs in the learned embedding space that have minimum features distance. The pretext task in the described self-supervised setting aims to train a feature extractor for discriminating dissimilar samples in the embedding space by a distance metric and to provide a useful representation for down-stream tasks.

The paper analyzes the performance of the approach for anomaly detection in rotating machinery. The methodology is tested on vibration data provided by the Center for Intelligent Maintenance Systems (IMS), University of Cincinnati, considering different accelerated life test campaigns. The data were collected to monitor the fault development in bearings and the model shows how the learned embedding space discriminates effectively anomalous samples from normal ones in the degradation stages of the bearings.

AI+AR based Framework for Guided Visual Diagnosis of Equipment

Teresa Gonzalez Diaz — 2024-11-05

Automated solutions for effective support services, such as failure diagnosis and repair, are crucial to keep customer satisfaction and loyalty. However, providing consistent, high quality, and timely support is a difficult task. In practice, customer support usually requires technicians to perform onsite diagnosis, but service quality is often adversely affected by limited expert technicians, high turnover, and minimal automated tools. To address these challenges, we present a novel solution framework for aiding technicians in performing visual equipment diagnosis. We envision a workflow where the technician reports a failure and prompts the system to automatically generate a diagnostic plan that includes parts, areas of interest, and necessary tasks. The plan is used to guide the technician with augmented reality (AR), while a perception module analyzes and tracks the technician’s actions to recommend next steps. Our framework consists of three components: planning, tracking, and guiding. The planning component automates the creation of a diagnostic plan by querying a knowledge graph (KG). We propose to leverage Large Language Models (LLMs) for the construction of the KG to accelerate the extraction process of parts, tasks, and relations from manuals. The tracking component enhances 3D detections by using perception sensors with a 2D nested object detection model. Finally, the guiding component reduces process complexity for technicians by combining 2D models and AR interactions. To validate the framework, we performed multiple studies to:1) determine an effective prompt method for the LLM to construct the KG; 2) demonstrate benefits of our 2D nested object model combined with AR model.

SYSAI for System Health Management - a Statistical Framework for the Analysis of Diagnosis Systems

Yuning He — 2024-11-05

On-board failure diagnosis and health management systems (HMS) are crucial for the operation of complex autonomous aerospace systems. False alarms (false positives, FPs) or false negatives (FNs) can lead to lower system performance or even loss of mission or the autonomous vehicle. Therefore, a careful verification and validation (V&V) is important. Due to the high dimensionality of the system’s state space, however, exhaustive testing of the HMS is usually not possible.

In this paper, we present how our SYSAI (System Analysis for Systems with AI components) framework can support intelligent analysis and testing of HMS on the system level. SYSAI’s capabilities to efficiently explore high-dimensional state and parameter spaces and to identify diagnosability regions and their boundaries, makes a comprehensive analysis of the diagnosis system possible and can provide feedback to the designer. We will illustrate our approach using the ADAPT (Advanced Diagnostics and Prognostics Testbed) redundant power storage and distribution system.

Zero-shot Video Change Detection for Real-life Industrial Applications

Mahbubul Alam — 2024-11-05

Change detection is crucial for various industrial applications. Although image change detection datasets are abundant, the collection of labeled video data is time-consuming, expensive, and cumbersome. This scarcity of labeled data motivates the development of few-shot or zero-shot video change detection techniques which may generalize well to new situations. Existing video change detection methods require large amounts of labeled data, are task-specific, and difficult to generalize. Therefore, in this paper, we propose a zero-shot video change detection algorithm using pre-trained deep learning models and conventional image processing techniques. Our approach identifies matching frames from input videos, adjusts lighting conditions if necessary, and uses an existing object detection model to identify objects in both frames. The method is easily generalizable by making few changes. We evaluate our proposed method on the VDAO dataset collected in a cluttered industrial environment and demonstrate its effectiveness in detecting changes between pairs of videos containing single and multiple objects.

Assessing Aircraft Engine Wear through Simulation Techniques

Abdellah Madane — 2024-11-05

In the field of aeronautical engineering, understanding and simulating aircraft engine performance is critical, especially for improving operational safety, efficiency, and sustainability. At Safran Aircraft Engines, we were able to demonstrate the effectiveness of using time series collected from the engines after each flight to build a digital twin that provides a dynamic virtual model able to mirror the real engine’s state by using a transformer-based conditional generative adversarial network. This virtual representation allows for advanced simulations under diverse operational scenarios like flight conditions and controls, aiding in understanding the impact of different factors on engine health. It is, therefore, possible for us to provide virtual flights performed by our engines in their actual state of wear. This research paper presents a machine learning model that effectively simulates and monitors the state of aircraft engines in real-time, enabling us to track the evolution of the engines’ health over their life cycle. The model’s adaptability to incorporate new data ensures its applicability throughout the engine’s lifespan, marking a step forward in proactive aeronautic maintenance and potentially enhancing engine longevity through timely diagnostics and interventions.

CycleGAN-based Data Augmentation for Enhanced Remaining Useful Life Prediction Under Unsupervised Domain Adaptation

Dorian Joubaud — 2024-11-05

Predictive maintenance is crucial for enhancing operational efficiency and reducing costs in Prognostics and Health Management (PHM). One of the key tasks in predictive maintenance is the estimation of Remaining Useful Life (RUL) of machinery. In practice, the data for different machines is not always accessible in sufficient quantity or quality, therefore the machine learning models trained on machines in one domain often perform poorly when applied to other domains due to covariate shifts. As a solution, Domain Adaptation (DA) aims to tackle domain shifts by extracting domain-invariant features. However, traditional methods often fail to adequately address the complexity and variability of real-world data. We propose to address this challenge, using a Wasserstein CycleGAN with Gradient Penalty (W-CycleGAN-GP) to learn mappings between domains and generate augmented data in the target domain from data in the source domain. We use our approach to generate realistic augmented data that bridge domain gap coupled with recent work on adversarial-based and correlation alignment-based DA models to improve the performance of RUL prediction models in target domains without having access to labeled data. The experimental results on the C-MAPSS dataset demonstrate a significant improvement in the RUL prediction score and accuracy within the target domain.

Advancing Light Aircraft Health Monitoring with Flight Phase Clustering

Oguz Bektas — 2024-11-05

This paper addresses the clustering of flight phases of a light aircraft for health monitoring using vibration data. The aim is to improve diagnostic and prognostic functions. Grouping condition monitoring data under similar operating conditions is significant for predictive maintenance. Clustering also supports advanced analytics for fault detection and estimation of remaining life. The proposed framework uses self-organizing maps for flight phase clustering. The findings show that the algorithm can recognize and classify flight phases in various operational domains. Additionally, visualization of cluster maps uncovers complex patterns and non-linear relationships in sensor data under different flight conditions. As a followup, analyzing the vibration properties within these estimated clusters (regimes) provides insights from condition monitoring data behavior during flight phases. The results confirm the effectiveness of the method, but also confirm that determining light aircraft regimes requires more focus due to their unique flight patterns that are absent in commercial airliners. In this context, this research has dealt with these unique patterns and provided the foundation for a new model for clustering with an attempt to contribute valuable insights into improving the reliability and efficiency of light aircrafts.

Exploring the Nexus between Sensor Reliability and System Performance

Deepak Tripathy — 2024-11-05

In contemporary technological landscapes, sensors play a pivotal role in enabling diverse applications across industries, from healthcare to manufacturing. This paper undertakes a thorough investigation on system performance (reliability and availability of a system), focusing on the critical interplay between baseline performance, performance with integrated sensors and performance considering sensor reliability, recognizing the foundational importance of sensors in data-driven decision-making processes. The research employs a causation-based approach to systematically develop functional relations within the system. The failures identified of each component and functional relationships will then be analyzed using a simulation technique to understand the inherent performance of the engineering system. From here, a genetic algorithm is used to design a sensor set and tailor it for an engineering system, providing a foundation for conducting trade studies in the paper's subsequent sections. Through rigorous quantitative analysis and simulations, we compare the impacts of the performance of the sensor set design compared to the baseline performance. The paper then investigates the complexities of sensor reliability on overall system performance. Through advanced simulations, we elucidate the potential cascading effects that variations in sensor reliability can have on the system's performance. By exploring these ripple effects, we aim to provide a comprehensive understanding of how sensor reliability plays a crucial role in determining the success of complex systems. Beyond the immediate considerations of sensor characteristics, the paper analyses the maintenance aspects of sensors by performing a series of analyses to suggest maintenance aimed at improving the sensor and hence system reliability. Highlighting the relationship between sensor reliability and system performance, this section stresses the critical role of consistent maintenance practices in ensuring sustained data quality and system functionality. In conclusion, this paper aims to highlight the different perspectives that can be analyzed to understand the reality of system performance, considering facets such as sensor maintenance and reliability. It also aims to demonstrate various approaches that can be applied to engineering systems to uncover truths about sensor performance and reliability.

Physics and Data Collaborative Root Cause Analysis

Hao Huang — 2024-11-05

Data-driven tools for asset health management face significant challenges, including a lack of understanding of physical principles, difficulty incorporating domain experts’ experiences, and consequently low detection accuracy, leading to trustworthiness issues. Automatically integrating data-driven analysis with human knowledge and experience, as found in literature and maintenance logs, is critically needed. Recent progress in large language models (LLMs) offers opportunities to achieve this goal. However, there is still a lack of work that effectively combines pretrained LLMs with data-driven models for asset health management using industrial time series data as input. This paper presents a framework that integrates our recently proposed data-driven AI with pretrained LLMs to address root cause detection in industrial failure analysis. The framework employs LLMs to analyze outputs from our data-driven root cause analysis models, filtering out less relevant results and prioritizing those that align closely with physical principles and domain expertise. Our innovative approach leverages advanced data-driven analytics and a multi-LLM debate for collaborative decision-making, seamlessly merging data-driven insights with domain knowledge. Specifically, through our proposed self-exclusionary debates among multiple LLMs, biases inherent in single-LLM systems are effectively mitigated, enhancing reliability and stability. Crucially, the framework bridges the gap between data-driven models and physics-informed LLMs, accelerating the interaction between data and knowledge for more informed and realistic decision-making processes.

A Gamma Process Based Degradation Model with Fractional Gaussian Noise

Xiangyu Wang — 2024-11-05

In modern industrial and engineering systems, stochastic degradation models are widely used for reliability analysis and maintenance decision-making. However, due to imperfect sensors and environmental influences, it is difficult to directly observe the latent degradation states. Traditional degradation models typically assume that measurement errors have simple statistical properties, but this assumption often does not hold in practical applications. To address this issue, this paper constructs a degradation model based on the Gamma process (GP) and assumes that measurement noise can be characterized by the fractional Gaussian noise (FGN). Furthermore, this paper proposes a method combining Gibbs sampling with the stochastic expectation-maximization (SEM) algorithm to achieve efficient estimation of the model parameters and accurate inference of the latent degradation states. Simulation results indicate that the proposed model exhibits better generalizability compared to the GP model with Gaussian noise.

Modeling and Simulation of Thermal Effects on Electrical Behavior in Lithium-Ion Cells

Cristóbal Allendes — 2024-11-05

Thermal effects exert a crucial influence on the electrical behavior of lithium-ion batteries, significantly impacting key parameters such as the open circuit voltage curve, internal impedance, and cell degradation rate. Furthermore, these effects may give rise to electrolyte loss, resulting in a reduction in capacity. The cycling of batteries inherently generates internal heat, establishing a direct relationship between cell temperature and power demand. This article aims to provide a methodology to model electrothermal relations and temperature influence on electrical behavior in lithium-ion cells, as well as a simulation of extended cell operation under arbitrary power loads, presenting a novel approach not previously explored. It does this by considering three models: the Bernardi model for heat generation within the cell, a thermal lumped model for the cell’s temperature, and the Vogel-Fulcher-Tammann model for the capacity change as a function of temperature. These models are then connected to a state-of-the-art open circuit voltage model of a cell, providing a connection from the thermal world back into the electrical world. Experiments with different power demands occur on the simulation, including estimation of thermal parameters with relative errors under 1%, visualizing the effects of the integrated models and potential for real-cell applications.

Mean Variance Estimation Neural Network Particle Filter for Predicting Battery Remaining Useful Life

Francesco Cancelliere — 2024-11-05

Traditional remaining useful life (RUL) prediction methods based on particle filter (PF) require the manual tuning of hyperparameters, such as process or measurement noise, which poses challenges, particularly in real-life applications where external and operating conditions may change, potentially leading to large errors in the predictions. We address this issue by replacing the measurement equation of a PF with a mean variance estimation neural network that estimates the mean and the variance of the output distribution. As a result, the measurement noise is automatically estimated by the neural network and does not require manual setting. Through simulations and comparative analyses with state-of-the-art methods, the proposed mean variance estimation neural network particle filter (MVENN-PF) is shown to provide more stable and accurate RUL predictions, thereby potentially enhancing the robustness of battery health management systems based on it. Additionally, by eliminating the need to manually set a model hyperparameter (the measurement noise) the proposed method simplifies the modeling process, making it more accessible and adaptable to various battery systems.

Data-Driven Remaining Useful Life Estimation

Ark Ifeanyi — 2024-11-05

This paper explores the development and application of data-driven prognostic models for estimating the Remaining Useful Life (RUL) of Nuclear Power Plant (NPP) condensers experiencing tube fouling. Due to the unavailability of run-to-failure industry sensor data, we utilized simulated data generated by the Asherah Nuclear Power Plant Simulator (ANS), initially designed by the International Atomic Energy Agency (IAEA) and programmed in Simulink for cyber security simulations. ANS's adaptability allows it to simulate Pressurized Water Reactor (PWR) behaviors given a time series of operating conditions and to introduce degradation modules to mimic fouling effects. Our study compares two primary approaches applied to data generated by ANS: inference-based and direct prediction methods. The selected inference-based approach estimates the health state of the condenser using a pipeline formed by an Auto Associative Kernel Regressor and a Hidden Markov Model (HMM), which subsequently combines the state estimates with its parameters to predict the RUL. The direct prediction method employs a Gradient Boosting Regressor Decision Tree (GBRDT) to map input variables directly to RUL. Our findings demonstrate the efficacy and limitations of each method through the case study, providing valuable insights for the adoption of data-driven RUL estimation techniques in industrial and energy applications.

Global-Local Continual Transfer Network for Intelligent Fault Diagnosis of Rotating Machinery

Jipu Li — 2024-11-05

Existing fault diagnosis methods face three fundamental challenges when deployed in the dynamic environment, including insufficient continuous diagnostic capability, poor model generalization performance, and inadequate data privacy protection. To address these problems, we develop a novel continual fault diagnosis framework named Global-Local Continual Transfer Network (GLCTN) for fault classification of unlabeled target samples under different working conditions without any source samples. Specifically, a consistency loss and a mutual information loss are introduced in the proposed GLCTN to transfer the learned diagnostic knowledge. Moreover, a dual-speed optimization strategy is utilized to preserve the acquired diagnostic knowledge and to endow the model with the ability to acquire new knowledge. Validation experiments conducted on an automobile transmission dataset demonstrate that the proposed GLCTN achieves satisfactory diagnostic performance on multiple continuous transfer diagnostic tasks.

Development of PHM Algorithm of e-Latch to Prepare for the Era of Autonomous Driving

Mooseok Kwak — 2024-11-05

In self-driving vehicles linked with mobility electrification, system failures that occur suddenly in situations where customers are unaware of signs of failure are directly related to customer injuries. Securing the durability and safety of closure automation system is necessary to increase the customer's safety value so PHM technology makes it possible to predict failures and remaining life in advance during system operation. In addition, since not only various forms of new concept design styling but also innovative new handle designs are applied, it is obviously seen that e-Latch system is widely equipped in the mobility. Thus, in this paper, the study to predict the failure of e-Latch and closure system is implemented via data-driven and physics-driven method, and the algorithm for PHM to estimate remaining life of e-Latch system is also introduced.

Extracting Quantitative Insights from electrochemical impedance spectra using Statistical Methods

Pavle Boskoski — 2024-11-05

Statistical analysis of electrochemical impedance spec- troscopy data provides a systematic way of detecting changes in electrochemical energy systems. Applying concepts of divergence measures directly on electrochemical impedance spectroscopy data, one can reliably detect and quantify statistically significant changes. The results is a set of high- lighted frequency bands where the measured impedance characteristics differ statistically significantly from a ref- erence curve. The approach is evaluated on a solid-oxide electrolyser cell operated under different conditions and proves to be sensitive to even the smallest changes. The complete numerical implementation and corresponding experimental data are available as supplementary material at https://portal.ijs.si/nextcloud/s/xTa2cmtfxXn2jSz

An Advanced Diagnostic Model for Gearbox Degradation Prediction Under Various Operating Conditions and Degradation Levels

Hanqi Su — 2024-11-05

This study introduces a novel three-stage diagnostic methodology aimed at enhancing the prediction and classification of gearbox degradation under various operating conditions and multiple degradation levels, addressing the complexities encountered in real-world industrial settings. Leveraging the latest advancements in data-driven approaches, from similarity-based methods to residual-based deep convolutional neural networks (CNNs) and pseudo-labeling techniques, our approach systematically classifies data into known, unknown, and undetermined categories, predicts known degradation levels, and refines classification models with augmented pseudo-label data. The efficacy of our methodology is demonstrated through its remarkable performance using the data from the PHM North America 2023 Conference Data Challenge. It achieves scores of 600 / 800 on the testing data and 574 / 813 on the validation data, significantly surpassing the first-place scores of 463.5 and 472 in the competition, respectively, setting a new benchmark in the field of gear fault diagnosis.

Investigation of the Dynamic Relationship between Oil Temperature and Bearing Gearbox Condition Indicator Values for the Bell 407 Helicopter Based on Cointegration Analysis

Oumayma Babay — 2024-11-05

The primary objective of this study is to investigate the dynamic relationship between oil temperature and the Bearing Gearbox Condition Indicator (BGCI) values of the Bell 407 helicopter. To achieve this goal, we employ robust econometric tools, such as unit root tests, cointegration tests, and Autoregressive Distributed Lag (ARDL) models for both, long-run and short-run estimates. Our findings indicate that variable temperature tends to converge to its long-run equilibrium path in response to changes in other variables. The results of the ARDL analysis confirm that spectral kurtosis, inner race, cage, and ball energy significantly contribute to the increase in temperature. Furthermore, we use the impulse response function (IRF) to trace the dynamic response paths of shocks to the system. The identification of a cointegrating relationship between oil temperature and BGCI values suggests a practical and significant connection that can potentially be used to predict hazardous changes in oil temperature using BGCI values, which is an important implication for enhancing the safety and reliability of helicopter operations.

This study presents a promising direction for condition monitoring (CM) in rotating aircraft machinery, emphasizing the potential of integrating temperature data to simplify the diagnostic process while still achieving reliable results.

A Deep Learning Solution for Quality Control in a Die Casting Process

Paula Mielgo — 2024-11-05

Industry 4.0 aims for a digital transformation of manufacturing and production systems, producing what is known as smart factories, where information coming from Cyber-Physical Systems (core elements in Industry 4.0) will be used in all the manufacturing stages to improve productivity. Cyber-physical systems through their control and sensor systems, provide a global view of the process, and generate large amounts of data that can be used for instance to produce datadriven models of the processes. However, having data is not enough, we must be able to store, visualize and analyze them, and to integrate induced knowledge in the whole production process. In this work, we present a solution to automate the quality control process of manufactured parts through image analysis. In particular, we present a Deep Learning solution to detect defects in manufactured parts from thermographic images of a die casting machine at an aluminum foundry.

Elastodynamics based Modelling of Acoustic Emission for Earlier Bearing Damage Detection

Anurag Bhattacharyya — 2024-11-05

It is crucial for many applications to detect bearing damage as early as possible to allow for scheduling of maintenance with lead times that minimize operational disruption. State of the practice is the detection of spalling but damage initiates prior to spalling as subsurface and surface cracks. Such damage is much harder to detect and to model. This study proposes a unique application of the nanofrictional Prandtl-Tomlinson model to predict macroscopic acoustic emission (AE) signals that occur at cracked interfaces under relative motion. The study integrates large deformation modelling of structures with elastodynamic simulations to investigate early AE signals generated under different bearing rotational speeds. Experimental studies are carried out to measure acoustic vibrations from metal-metal surface friction using fiber optic sensors and compared to those predicted by the model. Broad agreement of results highlights the validity of this framework.

Online and Offline Fault Detection and Diagnostics in a Nuclear Power Plant Condenser

Ark Ifeanyi — 2024-11-05

Nuclear power plants (NPPs) face significant financial pressures due to operational and maintenance costs. This research investigates Fault Detection and Diagnostics (FDD) techniques to optimize maintenance schedules and reduce expenses. The NPP condenser plays a critical role in converting turbine exhaust steam back into water for reuse. Condenser tube fouling, a prevalent fault mode, impedes heat transfer efficiency and can lead to decreased plant efficiency and safety risks. This study proposes an FDD framework that leverages raw signal analyses from temperature and pressure monitoring to detect and diagnose condenser tube fouling in both online and offline settings. The online approach facilitates close-to-real-time predictions, enabling proactive maintenance strategies. Additionally, the framework explores incorporating a condenser’s maintenance history for enhanced diagnostics. We employ a dataset obtained from a simulated nuclear power plant condenser using the Asherah Nuclear Power Plant Simulator (ANS). ANS replicates the operational dynamics of a pressurized water reactor (PWR) type NPP. The proposed methodology utilizes an encoder-decoder (E-D) structured 1DCNN model to analyze the raw signals. The research demonstrates consistent and accurate fault detection and diagnostics for condenser tube fouling in both online and offline scenarios. A high potential for generalization to unseen conditions was observed. However, online detection using small data windows necessitates caution due to potential false alarms around the transition points. Our findings pave the way for further exploration of robust diagnostics by accommodating a wider spectrum of fouling rates within degradation classes using ANS. This combined online and offline FDD approach offers a promising solution for promoting operational safety, efficiency, and cost-effectiveness in NPP condensers.

A Process for Real-Time Monitoring of a Turboshaft Engine

Eric Bechhoefer — 2024-11-05

For helicopters engaged in sling loads or heavy lift, there is a need to report current turboshaft engine health (e.g., margin) and contingency power available from the engine in realtime. Displaying this information allows the pilot in command of the aircraft to make more informed decisions about the safety of continuing a mission. For engine margin, when aircraft parameter data is recorded by a health and usage monitoring system (HUMS) or flight data monitoring system (FDM), this functionality allows maintainers to be notified of the engines’ degraded performance to initiate an inspection/maintenance action to restore the engine to its designed performance. However, this does not help the pilot make mission-critical decisions during the flight. The paper covers the method to use HUMS/FDM data to calculate, in real-time, the power available to the pilot.

Use of Tach from Vibration to Estimate Bearing Spall Length

Eric Bechhoefer — 2024-11-05

Vibration analysis is often used for bearing fault diagnostics. Envelope analysis or other cyclo-stationary processes can capture a fault feature or condition indicator that is correlated to the spall length. However, no study has defined a process for estimating spall length on real-world data. The problem is that the spall length is a time-domain property of the signal. This paper generates a synthetic tachometer signal from the fault itself. It is synchronous to the rolling element exit from the spall, allowing for a time-domain representation of waveform using the time synchronous average. From this, an estimate of the length of the bearing fault can be determined.

System-level Prognostics and Health Management for Complex Industrial Systems

Mattia Zanotelli — 2024-11-11

Prognostics and health management (PHM) has become essential to guarantee aware and safe system operation and to inform economic decision-making. However, due to the nature of detection, diagnostics, and prognostic methods, applications have mainly been limited to the component level. In practice, most industrial systems consist of multiple interacting components whose partial degradation could lead to the system's failure (or subsystems).
This research addresses the limitations of traditional component-level PHM techniques by proposing a novel system-level framework. By implementing a hierarchical structure of components and subsystems, we will select an optimal method for each subsystem to aggregate its component health assessments. The overall system health can then be estimated by further combining the obtained estimates. The research considers simplified and holistic modeling techniques, margin-based methods, and hybrid graphical models. This approach aims to provide reliable system health predictions and online components' sensitivity measures to enhance maintenance decision-making. We consider an application in the context of the nuclear industry, characterized by strict safety and economic requirements. Using a SIMULINK model to approximate a Pressurized Water Reactor (PWR) with real industrial inputs, we plan to add component degradation modules and use simulated sensor data and reliability information to test the proposed framework. Initial results on artificial case studies show the feasibility of integrating component-level health predictions.

A Graph Neural Network Approach to System-Level Health Index and Remaining Useful Life Estimation

Ark Ifeanyi — 2024-11-11

Current methods for predicting health index and remaining useful life (RUL) in complex systems struggle to account for performance dependencies between components, leading to inaccurate system-level estimates. This research proposes a novel approach utilizing graph neural networks (GNNs) to improve system-level health index and RUL estimation. GNNs excel at capturing complex interdependencies within a system, making them ideal for this task. The proposed methodology is designed for systems with synchronously sampled process data. To illustrate the application of the proposed approach, we will use the Condensate Extraction Subsystem (CES) of a nuclear power plant (NPP) as a case study. Sensor data like temperature, pressure, and flow rates will be used to train GNNs to predict the overall health and RUL of the CES over time. To evaluate the effectiveness of GNNs, a custom NPP simulator will be used to model the CES under various realistic fault modes across a variety of components. The GNN's performance will be verified and its robustness will be tested under diverse scenarios. This research aims to demonstrate the effectiveness and resilience of GNNs for system-level prognostics.
By providing valuable insights for maintenance decision-making, this approach can enhance operational reliability and safety in complex engineering systems.
The proposed framework has the potential to be applied across various industries, leading to advancements in predictive maintenance practices.

Digital Twin Generalization with Meta and Geometric Deep Learning

Raffael Theiler — 2024-11-11

Deep digital twins (DDTs) are deep neural networks that encode the behavior of complex physical systems. DDTs are excellent system representations due to their ability to continuously adapt to operational changes and their capability to capture complex relationships between system components and processes that cannot be explicitly modeled. For this challenge, DDTs benefit greatly from recent success in geometric deep learning (GDL) which allows the integration of information from multiple systems based on schematic representations. A major challenge in training DDTs is their dependence on the quality and representativeness of training data, especially under the dynamic conditions typical in prognostics and health management (PHM). Recent developments in differentiable simulation present new opportunities for optimizing the training data representativeness. In this thesis, we propose a novel meta-learning framework that trains DDTs using the output from differentiable simulators. This setup enables active optimization of training data sampling through gradient computation, enhancing training speed, robustness, and data representativeness. We extend this framework to address challenges in multi-system data integration in power grids and fault detection in railway traction networks. By applying our framework, we aim to tackle significant challenges in forecasting, anomaly detection and sensor-fault analysis using advanced data fusion techniques. Our approach promises substantial improvements in DDT robustness and operational efficiency, with its effectiveness to be demonstrated through empirical studies on both simple and complex case studies within the power systems domain.

Adaptable and Generic Methods for Monitoring and Prognostics of Energy Assets

Mohammad Badfar — 2024-11-11

Monitoring and prognostics of energy assets are crucial for maintaining their reliability and efficiency. Effective monitoring ensures that potential issues are identified early, preventing unexpected failures and optimizing maintenance schedules. However, several challenges complicate this process in real-world scenarios, including poor data quality, low-fidelity and sparse data, the influence of external environmental factors, and diverse operating conditions and asset types. These challenges highlight the need for adaptable and generic solutions that can handle variability and complexity across different energy systems. This Ph.D. project aims to address these challenges by developing scalable, data-driven approaches for monitoring and prognostics. By focusing on creating adaptable and generic frameworks, the research seeks to provide robust solutions for real-world monitoring and prognostic problems for energy assets.

Diagnostics and Prognostics with High Dimensional Spatial-Temporal Data: From Structures to Human Brains

Yan Xue — 2024-11-11

Diagnostics and prognostics with high-dimensional spatial-temporal data require innovative methodologies due to the inherent complexity of such datasets. This thesis explores the challenges of diagnostics and prognostics in high-dimensional spatial-temporal data, extending from physical structures to complex human brain analyses through resting-state functional magnetic resonance imaging (rs-fMRI). Drawing an analogy to engineering structural health monitoring using spatial-temporal vibration data, the approach leverages techniques from engineering diagnostics and prognostics data analytics to handle clinical problems with similar characteristics. A pioneering approach is developed to analyze multimodal datasets that not only include advanced rs-fMRI features—Amplitude of Low-frequency Fluctuations (ALFF), Regional Homogeneity (ReHo), Euler Characteristics (EC), and Fractal Analysis—but also encompass a wide array of clinical data. This integration includes infant developmental metrics such as birth weight and gestational age, maternal health factors like BMI and fat mass, and environmental influences including dietary intake and mental health during pregnancy. The study establishes a robust computational framework that uses advanced machine learning algorithms to analyze the interplay of these diverse data types, enhancing the precision and predictive power of our models for early childhood development. Initial validations have demonstrated the effectiveness of this comprehensive approach in identifying ADHD, with ongoing efforts aimed at expanding the methodology to address a broader range of developmental disorders. This work not only advances the diagnostic and prognostic capabilities in medical imaging but also significantly contributes to the field of Prognostics and Health Management (PHM). By providing a solid foundation for managing and understanding high-dimensional and multimodal spatial-temporal data across various disciplines, it bridges the gap between engineering and clinical diagnostics, demonstrating the potential for cross-disciplinary innovation.

Multimodal sensor-to-machined surface image diffusion for defect detection in industrial processes

Jae Gyeong Choi — 2024-11-11

Generative models, particularly diffusion-based approaches, have gained significant attention recently due to their ability to create realistic outputs. Despite their potential, the application of these models in manufacturing remains largely unexplored. This work presents a framework that addresses this gap by generating machined surface images guided by multiple sensor inputs in manufacturing. The proposed model integrates information from multiple sensors with varying sampling rates using multimodal embedding and employs a latent diffusion model to translate the fused sensor embedding into an image embedding, which is then converted into a machined surface image. The effectiveness of the framework is validated using real-world time-series data, including force, torque, acceleration, and sound, collected from various industrial processes, such as a carbon-fiber-reinforced plastic drilling process. The results demonstrate the model’s ability to predict defects from the generated machined surface images. The proposed approach can potentially revolutionize prognostics and health management (PHM) in smart manufacturing by enabling sensor-guided visual inspection, defect detection, process monitoring, and predictive maintenance.

Development of a methodology for diagnosing faults in bearings operating under variable operating conditions based on self-supervised learning

Racquel Knust Domingues — 2024-11-11

Predictive maintenance plays a crucial role in ensuring the efficiency and availability of industrial assets. Bearings, essential components in rotating machinery, are subject to diverse and complex operating conditions, necessitating advanced fault diagnosis methods. Traditional diagnostic approaches often rely on supervised learning, which requires extensive labeled datasets, a process that is both costly and impractical under varying conditions. This work proposes a novel methodology for diagnosing bearing faults using self-supervised learning, which leverages unlabeled data to generate useful representations for fault detection. The proposed approach aims to develop an end-to-end system that processes raw vibration signals to accurately diagnose the current state of bearings, including fault detection, localization, and severity assessment. The methodology is validated using experimental data from a test rig simulating various fault conditions and will be further tested on real industrial machinery. This research contributes to the development of more efficient and generalizable diagnostic tools for rotating machinery, particularly under variable operational conditions.

A Two-Step Framework for Predictive Maintenance of Cryogenic Pumps in Semiconductor Manufacturing

Sanjoy Kumar Saha — 2024-11-11

Semiconductor manufacturing involves many critical steps, wherein maintaining an ultra-high vacuum is mandatory. To this end, cryogenic pumps are used to create a controlled ultra-low-pressure environment through the use of cryogenic cooling. However, a sudden pump malfunction leads to contamination in the processing chamber, disrupting production. The primary focus of this study is preventing unplanned shutdowns of cryogenic pumps. The data was collected from various pump sensors also known as status variable identification (SVID) that reveals current behavior of the pump. A comprehensive framework is presented here to develop a condition monitoring and fault detection. In the proposed framework, a drift detection method is used for condition monitoring of the pump to locate gradual and abrupt drifts. Additionally, during regeneration (or maintenance) phase, intrinsic features are extracted to distinguish between normal and abnormal regeneration, achieving an accuracy of 90.91% and a precision of 66.67%. Utilizing the proposed system, cryo-pump operators can be given maintenance guidelines and warnings about potential health degradation of the pumps.

Remaining Useful Life Prognostics of Rolling Element Bearings Based on State Estimation Techniques

Zhen Li — 2024-11-11

Rolling element bearings (REBs) are key components in rotating machines. 40% of the failures in electrical motors occur due to bearing faults. Consequently, monitoring the health stage and estimating the remaining useful life (RUL) of the REBs is essential. Additionally, maintenance of rotating machines can be scheduled based on the RUL estimation, which will mitigate potential time wasting, economic losses and hazards. In this summary, several issues that exist in current research are highlighted. Then a series of preliminary explorations have been performed, and some results have been already obtained. Finally, a systematic methodology is expected to be proposed by combining the state estimation techniques and physical models to facilitate the development of the PHM.

Uncertainty-Aware Prediction of Remaining Useful Life in Complex Systems

Weijun Xu — 2024-11-11

Accurate prediction of the remaining useful life (RUL) of industrial systems is critical to ensuring smooth operation and safety. Various prognostic methods have been developed, but significant challenges remain for field applications. While many methods may achieve high accuracy, they often fall short in quantifying the uncertainty of their predictions. Without uncertainty quantification, it is difficult to assess the confidence level of the prognostic results. Therefore, it is essential to transparently present the uncertainty levels in the predicted results. This Ph.D. project aims to develop novel uncertainty-aware methods for RUL prediction of complex systems. The project will address the following situations where it is more and more uncertain: (a) propose a general framework for data-driven RUL methods to quantify uncertainty and generate adaptive confidence intervals under a single fault mode and a single operating condition; (b) consider both epistemic and aleatoric uncertainties in scenarios with multiple fault modes and multiple operating conditions and then calibrate uncertainty to enhance their accuracy; (c) explore how to predict RUL and quantify uncertainty when there are no run-to-failure data and RUL labels in practice; (d) handle uncertainty propagation from the component level to the system level. Through this research, the project will provide more reliable and comprehensive solutions for RUL prediction in complex systems.

PHM-Based Modeling for Cyberattack Classifier Performance

Priscila Silva — 2024-11-11

This research implements Prognostics and Health Management (PHM) using multiple linear regression and multivariate time series models to monitor and predict when the performance of a Machine Learning-based cyberattack classifier might degrade to an unacceptable level, enabling preemptive maintenance strategies.

Improving Freight Train Wheel Monitoring with Smart Sensors

Igor Makienko — 2024-11-12

Onboard monitoring of freight car axleboxes enhances safety, reduces maintenance costs, and improves track conditions by preventing secondary damage. Installing wireless sensors on freight cars without a nearby power source should be cost-effective, given the large quantities involved. To address this, a new wireless smart sensor node has been deployed. The sensor automatically recognizes stable operating conditions, detects wheel rotational speed from vibrations, performs real-time condition monitoring, and transmits the results to the cloud. This study outlines the smart sensor concept and the pilot field test conducted with real freight cars. The results demonstrate the ability to estimate wheel rotational speed from vibrations and the potential for detecting wheel out-of-roundness (OOR) using a newly developed condition indicator for low-power real-time operations.

Proactive Aircraft Engine Removal Planning with Dynamic Bayesian Networks

Bharath Pidaparthi — 2024-11-05

Aircraft engine removal for maintenance is an expensive ordeal, and planning for it while balancing fleet stability objectives is a complex multi-faceted challenge. This is further compounded by uncertainties associated with usage or condition-based maintenance approaches that are becoming prevalent. Engine removal decisions rely on accurate estimation of damage growth or remaining useful life of critical components and a framework for aggregating these component-level estimates (and their uncertainties) into an engine-level removal forecasting model. An approach to this planning challenge is to develop probabilistic prognostic digital twins tailored to engine-specific operations and calibrate/update them with inspection data from the field. To this end, this work outlines a framework involving: 1) building component-level probabilistic models capable of forecasting damage growth or remaining useful life, 2) aggregating the outputs of these component-level models into a system-level view using a Dynamic Bayesian Network (DBN), and 3) updating the states of the DBN with inspection information as and when they become available.

Diagnostics-LLaVA

Aman Kumar — 2024-11-05

The recent advancements in the area of Large language models (LLMs) has opened horizons for conversational assistant-based intelligent models capable of interpreting images, and providing textual response, also known as Visual language models (VLMs). These models can assist equipment operators and maintenance technicians in the complex Prognostics and Health Management (PHM) tasks such as diagnostics of faults, root cause analysis and repair recommendation. Significant open source contributions in the area of VLMs have been made. However, models trained on general data fail to perform well in complex tasks in specialized domains such as diagnostics and repair of industrial equipment. Therefore, in this paper we discuss our work on development of Diagnostics-LLaVA, a VLM suitable for interpreting images of specific industrial equipment and provide better response than existing open source models in PHM tasks such as fault diagnostics and repair recommendation. We introduce Diagnostics-LLaVA based on the architecture of LLaVA, and created one instance of the Diagnostics-LLaVA for the automotive repair domain, referred to as Automotive-LLaVA. We demonstrate that our proposed Automotive-LLaVA model performs better than the state-of-the-art open source visual language models such as mPlugOWL and LLaVA in both qualitative and quantitative experiments.

Data Augmentation of Multivariate Sensor Time Series using Autoregressive Models and Application to Failure Prognostics

Douglas Baptista de Souza — 2024-11-05

This work presents a novel data augmentation solution for non-stationary multivariate time series and its application to failure prognostics. The method extends previous work from the authors which is based on time-varying autoregressive processes. It can be employed to extract key information from a limited number of samples and generate new synthetic samples in a way that potentially improves the performance of PHM solutions. This is especially valuable in situations of data scarcity which are very usual in PHM, especially for failure prognostics. The proposed approach is tested based on the CMAPSS dataset, commonly employed for prognostics experiments and benchmarks. An AutoML approach from PHM literature is employed for automating the design of the prognostics solution. The empirical evaluation provides evidence that the proposed method can substantially improve the performance of PHM solutions.

Large Language Model Agents as Prognostics and Health Management Copilots

Sarah Lukens — 2024-11-05

Amid concerns of an aging or diminishing industrial workforce, the recent advancement of large language models (LLMs) presents an opportunity to alleviate potential experience gaps. In this context, we present a practical Prognostics and Health Management (PHM) workflow and self-evaluation framework that leverages LLMs as specialized in-the-loop agents to enhance operational efficiency without subverting human subject matter expertise. Specifically, we automate maintenance recommendations triggered by PHM alerts for monitoring the health of physical assets, using LLM agents to execute structured components of the standard maintenance recommendation protocol, including data processing, failure mode discovery, and evaluation. To illustrate this framework, we provide a case study based on historical data derived from PHM model alerts. We discuss requirements for the design and evaluation of such “PHM Copilots” and formalize key considerations for integrating LLMs into industrial domain applications. Refined deployment of our proposed end-to-end integrated system may enable less experienced and professionals to back-fill existing personnel at reduced costs.

Towards a Fault Management Analysis Tool for Model Centric Systems Engineering

Ksenia Kolcio — 2024-11-05

In an effort to effectively develop more complex spacecraft fault management (FM) systems new technologies are sought to enable rapid diagnostic model generation and validation, and provide tools to perform FM analyses Model-based Systems Engineering approaches to FM system development are uniquely suited to be combined with model-based tools currently utilized in the design of other parts of flight systems. Combined tools utilizing information from a common system model can reduce design inconsistencies and gaps in analyses. Tighter integration of FM with other system-level and subsystem-level hardware/software development activities allows crucial redundancy and sensor placement trades to be performed earlier and throughout the mission lifecycle.

Our work has been towards the integration of a model-based fault management tool suite called MONSID®, with JPL’s Computer Aided Engineering for Systems ARchitecture (CAESAR ) platform as a way to improve FM system modeling and analysis. MONSID relies on application-specific models of the system being monitored. MONSID models consist of interconnected elements representing system hardware and measurement/command input points, called the topology. Model topology design is currently a manual process and often relies heavily on paper documentation such as hardware/software specs, engineering drawings, and interface control documents. CAESAR is a semantically- driven toolchain for model-based system engineering. At the core is a system model expressed in the Ontological Modeling Language (OML). It is intended to support semantic modeling, consistency validation, and continuous integration.

A goal of the combined toolset is to automate FM model development by directly extracting models from CAESAR and then analyzing them in MONSID. Analyses currently available in MONSID include model topology inspection and validation and fault isolation capability based on sensor placement. While we have focused on two specific tools, the integration approaches can be leveraged by other semantically driven model-centric platforms and tools.

This paper describes the evolution of our integration approaches as we evaluated different insertion points in the CAESAR toolchain with respect to MONSID model requirements. The MONSID-CAESAR tool is demonstrated on a simplified example of a spacecraft heat reclamation system. Results of the generated MONSID model are discussed, including levels of automation achieved and information surfaced to the users about the extracted model topology.

Health Assessment of Pump Stations using Time Series Anomaly Detection

Abhishek Murthy — 2024-11-05

Industrial operators aim to adopt proactive asset management strategies. However, changes in workplace tenure have led to reduced operator expertise, hindering this goal. Shifting from manually detecting equipment problems, to using automated technology with predictive analysis can help address this challenge. However, legacy systems, like the Supervisory Control and Data Acquisition (SCADA), are not well-suited to scalable predictive approaches.

SCADA is widely used to manually monitor and manage distributed physical assets. Supporting infrastructure was designed and optimized for that need. Specialized communication protocols are utilized for applications which span large geographical deployments. These protocols ensure data robustness and consistency in variable-quality network environments. However, the resulting data, while forming enterprise data pipelines, lacks granularity and has irregular time spacing, making it unsuitable for machine learning applications.

We present a hybrid cloud-to-edge health monitoring solution for assets connected to SCADA or other legacy control systems. Our solution uses a modbus-based polling system on the edge, to collect data at a much higher granularity than the adjacent SCADA system, letting us detect even subtle and acute patterns in the data. Note that no new sensors are needed, as we connect to the same registers as the existing SCADA system. The high granularity data is assessed at the edge for anomalies, using time series anomaly detection algorithms. We then synthesize the prediction into a health index that quantifies the recency and the frequency of the detected anomalies for the asset. The health index is then transmitted to a web-based application, where the user can configure thresholds for generating alerts based on the criticality of the asset.

We demonstrate our solution in a case study, where the application was deployed using Schneider Electric's Customer First Digital Hub, to monitor a sewage pump station for blockages and other subtle deviations in operating patterns.

Investigating Model Form Error Estimation for Sparse Data

Kyle D. Neal — 2024-11-05

Computational simulations of dynamical systems involve the use of mathematical models and algorithms to mimic and analyze complex real-world phenomena. By leveraging computational power, simulations enable researchers to explore and understand systems that are otherwise challenging to study experimentally. They offer a cost-effective and efficient means to predict and analyze the behavior of physical, biological, and social systems. However, model form error arises in computational simulations from simplifications, assumptions, and limitations inherent in the mathematical model formulation. Several methods for addressing model form error have been proposed in the literature, but their robustness in the face of challenges inherent to real-world systems has not been thoroughly investigated. In this work, a data assimilation-based approach for model form error estimation is investigated in the presence of sparse observation data. Including physics-based domain knowledge to improve estimation performance is also explored. The Lotka-Volterra equations are employed as a simple computational simulation for demonstration.

United Airlines In-Flight Wi-Fi Health Management: Revolutionizing Aircraft Connectivity through Real-time Prognostics and Big Data Analysis

Ehsan Rahimi — 2024-11-05

This paper highlights an innovative initiative, focusing on Prognostics and Health Management (PHM) to enhance in-flight Wi-Fi performance by proactively identifying aircraft component failures. We propose a novel metric, the Normalized Wi-Fi Health Score (NWiHS), alongside a corresponding alerting mechanism, which together represents a significant advancement in the evaluation and improvement of in-flight Wi-Fi connectivity. To achieve this goal, we utilized big data consisting of millions of historical Wi-Fi heartbeats (HBs) received from each aircraft over the past three years. These HBs refer to periodic data packet transmissions sent from United’s aircraft to ground stations, providing crucial real-time insights into the Wi-Fi system’s status. Leveraging that data, we utilized advanced statistical methods to estimate a NWiHS - a robust indicator of aircraft-level connectivity performance, which quantifies the percent of missing Wi-Fi HBs normalized to exclude the effect of Wi-Fi provider performance and global coverage.

Estimating the health of turbine engine based on the relationship between torque margin and density altitude

Kosei Ozeki — 2024-11-06

We present an anomaly detection method developed for the PHM North America 2024 Conference Data Challenge. This competition is aimed at estimating the health of helicopter turbine engines (PHM Society, 2024). The task includes the estimation of the torque margin (regression) and the health state (binary classification) of turbine engines. We developed an estimation model using a hybrid algorithm that combines data-based machine learning and domain knowledge-based processing. Our method achieved scores over 0.99 for both the testing and validation datasets. based on the calculation rules provided by PHM Society. These results were ranked first among all the participating teams.

Estimating The Health of Helicopter Turbine Engines by Means of Regression and Classification Using a Probabilistic Neural Network

Tyler Romano — 2024-11-06

This paper presents team Mad SoftMax’s approach to competing in the Conference of the Prognostics and Health Management Society 2024 Data Challenge. The competition tasked participants with estimating the health of helicopter turbine engines by means of regression and classification. Classification was used to categorize each observation within the dataset as belonging to a healthy or faulty engine while probabilistic regression was employed to estimate the torque margin at each measurement. Additionally, teams were challenged to provide a confidence metric to each of their estimations. These metrics reflected the trustworthiness of solutions and added a risk versus reward element to the competition. While the complete dataset for the challenge contained seven engines, four were provided in a training set to encourage robustness of solutions to handle the three withheld engines’ data during the testing and validation phase. The data was scrambled and identifiers to specific engines were removed to not give away patterns specific to a given engine. Open-source libraries such as TensorFlow were utilized to develop classification and regression models and the following paper opines on the process of understanding the data, data cleaning and model evaluation.

Robust Health Condition Prediction of Helicopter Turboshaft Engines Using Ensemble Machine Learning Models

Zihan Wu — 2024-11-06

This paper presents a novel ensemble approach that combines multiple machine-learning algorithms to deliver robust predictions of helicopter turboshaft engine health status (nominal or faulty) using operational data. Engine health is evaluated through the torque margin, defined as the percentage difference between the measured and target torque values. A Gaussian process model is used to estimate the torque margin as a probability distribution function (PDF), and these predictions are incorporated as features into various machine-learning models. These models are then employed to perform binary classification, determining the engine's health state. To enhance performance, a reference set is defined for each unseen data point, allowing a comparison of the relative performances of the models, with the best performer selected for the final prediction. Our ensemble method achieves high accuracy in health classification while providing precise torque margin estimates. The results demonstrate that ensemble models offer superior generalization and reliability compared to individual machine-learning algorithms, especially when applied to complex, multivariate datasets like those from helicopter turboshaft engines.

A Comprehensive Approach to Fault Classification of Helicopter Engines with Adaboost Ensemble Model

Peeyush Pankaj — 2024-11-06

This work is based on the PHM North America 2024 Conference Data Challenge’s datasets of Helicopter turbine engine performance measurements. These datasets were large and moderately imbalanced. For dealing with these challenges, we demonstrate significant tools covering feature engineering, augmentation and selection, model exploration, visualizations, model explainability and confidence margin estimation. This work was performed in its entirety using MATLAB. All these tools will be generally applicable to data-driven modeling and prediction of health to real life applications.

Initially, we explored the 742k observations in the training set, noting a 60-40 split between healthy and faulty labels, and identified two major operational clusters within the data. We enhanced the dataset by removing duplicates and engineered new features based on domain knowledge, expanding the feature set to 242 dimensions.

However for the torque margin estimation, we trained a regression model on a limited subset (18 features), which includes engineered features using domain knowledge, quadratic terms and linear interaction between all the terms. For the final submission, we utilized a stepwise linear regression model to optimize feature selection. This approach achieved a perfect regression score on test data, validated by a consistent torque margin residual range of +/- 0.5%. The model's RMSE and MAE metrics were optimal for employing a normal distribution probability density function.

For , we reduced the feature set to 58 using dimensionality reduction techniques and balanced the data with upsampling and down-weighing the minority class. We employed ASHA (Asynchronous Successive Halving Algorithm) in conjunction with AutoML to efficiently determine the most suitable model family, significantly saving compute time. Subsequently, we trained ensemble models, including bagged tree and AdaBoost (Adaptive Boosting), which minimized false negatives and positives, achieving robust classification performance. This was particularly critical given the high penalty for false negatives in the data challenge. The MathWorks team score on Testing Data was 0.9686 at the close of competition. This was further improved to 0.9867.

Our approach demonstrates the effectiveness of combining strategic data processing, feature engineering, and model selection to enhance predictive accuracy in complex operational datasets.

Intelligent Helicopter Turbine Engine Fault Diagnosis Using Multi-Head Attention

Yong Hun Park — 2024-11-06

A turbine engine provides power to the helicopter, enabling the helicopter to travel and hover in the air. Since the rotorcraft operates at high altitudes, ensuring safety and maintaining a healthy operational status are crucial at all times. Therefore, a prognostics and health management (PHM) system for the turbine engine must be implemented to predict any anomalies or faults to prevent catastrophic accidents. This research proposes a novel fault diagnosis method for helicopter turbine engines based on operational data acquired from actual aircraft. First, the proposed method predicts engine torque using other operational data while accounting for uncertainty. A Bayesian regression approach is employed to predict the engine torque. The torque margin, defined as the difference between the actual torque and the estimated torque, is then used to diagnose engine faults. Specifically, a multi-head attention mechanism is incorporated to capture interactions between various engine parameters. Additionally, domain adaptation techniques are applied to enhance the model's generalization performance, ensuring robustness across diverse operating conditions. The proposed method is validated using seven different datasets, each acquired from a helicopter engine. Four datasets were used for training, while the remaining three were allocated for testing and validation. The results indicated that the proposed method accurately predicted torque. Furthermore, the fault diagnosis showed promising results, leading to a 3rd-place finish in the 2024 PHM Society Data Challenge in terms of validation score.

A Design Science Approach Comparing Ensemble Learning and Artificial Neural Networks for Uncertainty-Aware Helicopter Turbine Engines Health Monitoring

Victor Henrique Alves Ribeiro — 2024-11-06

This work presents the development of an uncertainty-aware health monitoring system for helicopter turbine engines, focusing on improving operational availability and reducing maintenance costs. We address the critical issue of uncertainty quantification in data-driven fault detection and prognostics, essential for increasing system reliability. The project follows an iterative development cycle, incorporating multiple techniques for data processing, such as polynomial feature generation and data cleansing, and model development, including ensemble learning and artificial neural networks. Evaluation is performed using K-fold and group-fold cross-validation. The final solution consists of a cascaded ensemble learning model combining bagged linear regression built on polynomial features and random forest. This model demonstrates robust performance, achieving a test score of 0.955719 and a validation score of 0.886953, showcasing the effectiveness of uncertainty-aware machine learning methods in health monitoring systems.

Assessing Helicopter Turbine Engine Health: A Simple Yet Robust Probabilistic Approach

Peihua Han — 2024-11-06

This paper presents a data-driven approach for assessing the health of helicopter turbine engines, developed for the PHM North America 2024 Conference Data Challenge. The task involves both regression and classification to estimate the torque margin and classify engine health as either nominal or faulty. To quantify the reliability of predictions, probabilistic outputs are generated. We employ a two-stage model where the predicted torque margin serves as an input feature for health classification. For probabilistic torque margin estimation, we introduce an empirical error sampling method to generate torque margin samples, followed by a rule-based distribution selection scheme to evaluate the resulting distributions. For fault classification, logistic regression is used to provide confidence estimates, and we incorporate a score-optimized loss function during training to mitigate penalties for false negatives. Our approach demonstrates strong generalization to unseen assets, ranking 2nd in the competition with a score of 0.94, demonstrating its effectiveness in predicting health conditions and uncertainty for more informed helicopter engine management.

Accepting Technology in Aviation Safety Risk Management: an extension of the technology acceptance model

Washington Mhangami — 2024-11-11

Aviation safety is paramount, and advancements in technology play a pivotal role in mitigating risks and enhancing operational efficiency. The Technology Acceptance Model (TAM) has been widely utilised to understand the adoption of various technologies across industries. However, its application within the context of aviation risk assessment requires nuanced considerations due to the unique operational environment and stringent safety requirements. This paper critically reviews existing literature on TAM and its adaptations in aviation risk assessment, identifying limitations and gaps. Drawing from interdisciplinary insights in psychology, human factors, and aviation safety, this paper proposes an enhanced framework for TAM tailored specifically to the aviation industry. The proposed model integrates key constructs such as perceived usefulness of technology in this area, trust in technology, system complexity, and organizational factors to provide a comprehensive understanding of technology acceptance within aviation risk assessment practices. By enhancing the TAM framework, this paper aims to offer valuable insights for researchers, practitioners, and regulators involved in aviation safety management and technology integration efforts.

Frequency domain tensor-based 1D-convolutional neural network and multilinear principal component analysis for machinery fault detection

Ayantha Senanayaka Mudiyanselage — 2024-11-05

Challenges in detecting machinery faults, particularly in multivariate sensor environments, necessitate advanced feature extraction and classification techniques. This study introduces a novel approach that combines Multilinear Principal Component Analysis (MPCA) with a 1D-Convolutional Neural Network (1D-CNN) for efficient fault detection. By constructing Frequency Domain (FD) tensors from multivariate sensor data and applying MPCA for dimensionality reduction, our methodology enhances the capabilities of a 1D-CNN in feature learning and fault classification. The efficacy of this approach is validated through experiments on a Machinery Fault Simulator (MFS) with acoustic and vibration sensors, demonstrating notable improvements in fault detection accuracy compared to benchmark methods. The study results demonstrate that the proposed approach exhibits high accuracy in identifying machine fault conditions and outperforms the benchmark methods. The findings of this study have significant inferences for machine fault detection and fill the gap of more effective and reliable techniques in this domain.

Bearing Fault Detection in Conveyor Belt Drums Using Machine Learning

Victor Bauler — 2024-11-05

In recent years, the application of machine learning techniques in condition monitoring has significantly advanced the precision and efficiency of fault detection processes. In particular, detecting bearing faults in conveyor belt drums is critical in the mining industry for maintaining operational reliability and productivity. This paper presents a case study using vibration signals and diagnostic reports provided by the company Dynamox. After meticulous data cleaning, preprocessing, and feature extraction employing advanced signal processing techniques and statistical features, several machine learning models were trained, optimized and evaluated, with the best models providing very promising results.