Annual Conference of the PHM Society

Vibration Signal Decomposition using Dilated CNN

Eli Gildish — 2023-10-28

Vibration sensors have gained increasing popularity as valuable tools for Prognostics and Health Management (PHM) applications, enabling early detection of mechanical failures in industrial machines. Vibration signals comprise two main sources of information: periodic vibrations from components, phase-locked to the rotating speed (e.g., gears), and non-deterministic broadband vibrations associated with bearings, structure, and background noise. In PHM applications, it is important to decompose vibrations into these two sources to optimize the use of different diagnostic methods for each signal component. In practice, the decomposition should be cost-effective by working without supplementary information about system operating conditions and kinematics.

Existing methods of vibration source separation commonly rely on an auto-regression (AR) model of vibrations and employ adaptive filtering techniques to estimate its parameters. However, these methods suffer from degraded accuracy in complex geared vibrations containing numerous periodic components and requiring large filter length to promise high frequency resolution in component separation.

To address these challenges, we propose a new method that utilizes dilated Convolutional Neural Networks (CNNs) instead of adaptive filtering to improve the accuracy of decomposing complex vibration signals, all without the need for any supplementary information.

To evaluate the performance of the new method, we conducted experiments using both simulated signals and real-world vibrations. The simulation results demonstrate improved accuracy in signal decomposition when our method is used instead of adaptive filtering. Additionally, the new method applied to real vibrations, showcases significant enhancement in bearing failure detection through accurate isolation of bearing-related vibrations.

This study reveals the potential of our new method in various PHM applications requiring highly accurate diagnostics and prognostics in complex geared vibrations, particularly when supplementary information about operating conditions and system kinematics is unavailable.

A Grey-box Approach for the Prognostic and Health Management of Lithium-Ion Batteries

Francesco Cancelliere — 2023-10-26

The Lithium-Ion Batteries (LIB) industry is rapidly growing and is expected to continue expanding exponentially in the next decade. LIBs are already widely used in everyday life, and their demand is expected to increase further, particularly in the automotive sector. The European Union has introduced a new law to ban Internal Combustion Engines from 2035, pushing for the adoption of electric vehicles and increasing the need for more efficient and reliable energy storage solutions such as LIBs. As a result, the establishment of Gigafactories in Europe and the United States is accelerating to meet the growing demand and partially reduce dependencies on China, which is currently the main producer of LIBs.

To fully realize the potential of LIBs and ensure their safe and sustainable use, it is crucial to optimize their useful life and develop reliable and robust methodologies for estimating their state of health and predicting their remaining useful life. This requires a comprehensive understanding of LIB behavior and the development of effective prognostic and health management approaches that can accurately predict battery degradation, plan for maintenance and replacements, and improve battery performance and lifespan.

This work, funded by the GREYDIENT project, a European consortium aiming to advance the state of the art in the grey-box approach, combines physical modeling (white box) and machine learning (black box) techniques to demonstrate the grey-box effectiveness in the Prognostic and Health Management. The grey-box approach here proposed consist in a combination of a physical battery model whose degradation parameters are estimated online at every cycle by a Multi-Layer Perceptron Particle Filter (MLP-PF).

An electrochemical degradation model of a Lithium-Ion battery cell has been derived by use of Modelica. The model simulates the output voltage of the cell, while the degradation over time is simulate through the variation of 3 parameters: qMax (maximum number of Lithium-Ions available), R0 (Internal Resistance) and D (Diffusion Coefficient). To validate the model we resorted to the well-known NASA Battery Dataset, which has also been used to infer the optimal values of the three hidden degradation parameters at every cycle, to obtain their Run-to-Failure history. Then, the physical model is combined the MLP-PF: a MLP
Artificial Neural Network is firstly trained on the Run-to-Failure degradation processes of the model parameters, allowing the propagation of the parameters in the future and the corresponding estimation of the battery Remaining Use ful Life (RUL). The MLP is then updated online by a Particle Filter every time a new measurement is available from the Battery Management System (BMS), providing flexibility to this method, needed for the electrochemical nature of the batteries, and allowing the propagation of uncertainties.

A NanoDet Model with Adaptively Weighted Loss for Real-time Railroad Inspection

Jiawei Guo — 2023-10-26

Monitoring the railroad’s components is crucial to maintaining the safety of railway operations. This article proposes a novel, compact computational vision system that works on edge devices, designed to provide precise, instantaneous assessments of rail tracks. This model reconfigures the teacher-student guidance system inherent in NanoDet by incorporating an innovative adaptively weighted loss (AWL) in the learning phase. The AWL assesses the caliber of the teacher and student models, establishes the weightage of the student's loss, and dynamically adjusts their loss contributions, directing the learning procedure towards effective knowledge transfer and direction. In comparison with cutting-edge models, our AWL-NanoDet boasts a minuscule model size of less than 2 MB and a computational expense of 1.52 G FLOPs, delivering a processing time of less than 14 ms per frame (evaluated on Nvidia’s AGX Orin). Compared to the original NanoDet, it also significantly enhances the model's accuracy by nearly 6.2%, facilitating extremely precise, instantaneous recognition of rail track elements.

A Hybrid Model-Based and Data-Driven Framework for Automated Spacecraft Fault Detection

Eric Pesola — 2023-10-26

Traditional fault management can be an onerous task and robust automated solutions are increasingly necessary to accommodate the complexities of modern space systems and mission operations. The present work proposes a hybrid framework for performing automated spacecraft fault detection by leveraging the benefits of both model-based and data-driven approaches. The framework uses a system model to generate residual data that are subsequently fed into a data-driven residual analysis stage. The framework was verified by using data from a hardware-in-the-loop test campaign in which faults were injected into a spacecraft attitude control system, and successfully identified. The fault detection approach implemented using this framework outperformed results obtained from expert-tuned fault detection parameters. Overall, the proposed framework is a promising alternative for sustainable fault detection and mission operations suitable for complex space systems.

A State Machine-Based Approach for Estimating the Capacity Loss of Lithium-Ion Batteries

Ruth David — 2023-10-26

The use of Lithium-Ion Batteries (LIBs) have increased in recent years in many applications such as hybrid electrical vehicles (HEV), consumer electronic equipment, and electricity grid. The batteries undergo degradation during usage due to material aging and electrochemical processes, leading to efficiency reduction of battery-powered systems as well as catastrophic events. Several stress factors such as battery temperature, ambient temperature, and C-rate in the loading profiles influence the degradation. Therefore, predicting the health of the battery has gained attention. The service life can be extended or a system failure can be avoided by maintenance measures precisely matched to the function loss or by changing usage strategies. The State-of Health (SoH) condition of the battery can be determined by the application of lifetime models. Various health indicators such as remaining useful lifetime (RuL) and capacity fade are determined by the models based on the stress factors (utilization variables). For optimal use of the battery, it is helpful to develop an accurate lifetime model to represent the dynamic properties. However, models developed are less computationally efficient and unable to represent the non-linear degradation behavior well. The development of a precise model with correct parameterization is also costly. This is particularly true for models developed based on physical and chemical properties of the battery. In this contribution, an artificial neural network (ANN)-based state machine approach is introduced for capacity fade estimation. The degradation process is represented using three states modeling three different levels and the progression from the first state to the last. Capacity associated with each state is described using the non-linear auto regressive neural network with external input (NARX). The NARX is selected due to its ability to accurately model non linear behavior and time series data. Unlike known models, which are developed using analytical mathematical equations related to the battery properties, a combined machine learning approach is used here instead to learn the capacity behavior from historical data. Battery data sets from NASA are used for experimental verification. Based on the results, the estimated capacity fade show close proximity to actual capacity fade, with a low mean square error for different data sets. In addition, the estimated state progression follows the actual state progression.

A Case Study Comparing ROC and PRC Curves for Imbalanced Data

Dan Watson — 2023-10-26

Receiver operating characteristic curves are a mainstay in binary classification and have seen widespread use from their inception characterizing radar receivers in 1941. Widely used and accepted, the ROC curve is the default option for many application spaces. Building on prior work the Prognostics and Health Management community naturally adopted ROC curves to visualize classifier performance. While the ROC curve is perhaps the best known visualization of binary classifier performance it is not the only game in town. Authors from across various STEM fields have published works extolling various other metrics and visualizations in binary classifier performance evaluation. These include, but are not limited

to, the precision recall characteristic curve, area under the curve metrics, bookmaker informedness and markedness. This paper will review these visualizations and metrics, provide references for more exhaustive treatments on them, and provide a case study of their use on an imbalanced prognostic health management data-set. Prognostic health management binary classification problems are often highly imbalanced with a low prevalence of positive (faulty) cases compared to negative (nominal/healthy) cases. In the presented data-set, time domain accelerometer data for a series of run-to-failure ball-on-disk scuffing tests provide a case where the vast majority of data, > 94%, is from nominally healthy data instances. A condition indicator algorithm targeting the hypothesized physical system response is validated compared to less informed classifiers. Several characteristic curves are then used to showcase the performance improvement of the physics informed condition indicator.

A Sequential Hybrid Method for Full Lifetime Remaining Useful Life Prediction of Bearings in Rotating Machinery

koengeurts — 2023-10-26

Optimal scheduling of the maintenance of bearings in rotating machinery requires accurate remaining useful life (RUL) prediction during the entire lifetime of the bearing. For that reason, this paper proposes a sequential hybrid method that combines the strengths of statistical and data-driven approaches. A statistical model-based approach is preferred before a bearing fault is detected, and a data-driven approach once a bearing fault is detected from the vibration measurements. The method is tested and evaluated on an extensive dataset of accelerated lifetime tests of deep groove ball bearings. It is shown that the method, with a limited amount of training data, delivers accurate RUL predictions during both the healthy stage of the bearing lifetime, as well as during the final stages of increasing degradation under both constant and varying speed conditions.

Adaptive Prognostics: A reliable RUL approach

Nick Eleftheroglou — 2023-10-26

Prognostic methodologies have found increasing use the last decade and provide a platform for remaining useful life (RUL) predictions of engineering systems utilizing condition monitoring data. Of particular interest is the reliable RUL prediction of engineering assets that either underperform or outperform due to unexpected phenomena that might occur during the operational life. These assets are often referred as outliers and the prediction of their RUL is a challenging task. The challenge is to accurately predict the RUL of an outlier without taking into account outlier’s condition monitoring data in the training process but just in the testing process. As a result, the lifetime of the testing asset is shorter (left outlier) or longer (right outlier) than the training process’ lifetimes.

This study addresses this challenge by proposing a new adaptive model; the Similarity Learning Hidden Semi Markov Model (SLHSMM), which is an extension of the Non-Homogenous Hidden Semi Markov Model (NHHSMM). The SLHSMM uses a similarity function, such as Minkowski distances, in order firstly to quantify the similarity between the testing asset and each training asset and secondly to adapt the trained parameters of the NHHSMM. To demonstrate the effectiveness of the proposed adaptive methodology, composite structures have been used as a validation engineering asset. In particular, the training data set consists of strain data collected from open-hole carbon–epoxy specimens, which were subjected to fatigue loading only, while the testing data set consists of strain data collected from specimens that were subjected to fatigue and in-situ impact loading, which can be considered as an unexpected phenomenon and unseen event regarding the training process.

Utilizing the aforementioned strain data the SLHSMM RUL predictions and the NHHSMM RUL predictions were compared, so as to verify that the SLHSMM provides better prognostics than the NHHSMM. SLHSMM provides better predictions in comparison to the NHHSMM for all the test cases, demonstrating its capability to adapt to unexpected phenomena and integrate unforeseen data to the prognostics course.

An Energy Consumption Auditing Anomaly Detection System of Robotic Manipulators based on a Generative Adversarial Network

Ge Song — 2023-10-26

Unexpected anomalies pose significant risks to the health and security of intelligent manufacturing systems. This paper proposes a generative adversarial network (GAN)-based anomaly detection framework specifically for monitoring robotic manipulator operation using a side-channel energy auditing mechanism. To tackle the limitation arising from the lack of labeled data, the GAN model is trained by a semi-supervised learning approach that identifies anomalies during online operations as outliers. The overfitting is purposely utilized during the model training to enlarge the difference between normal energy consumption patterns used for training and anomalous profiles in real-time testing. In addition, the GAN model is modified to use multiple discriminators to analyze the individual energy profile associated with each joint or motor. The anomaly is detected by evaluating the mean and standard deviation values of anomaly scores' distribution, and both values are continuously updated by Welford's algorithm in real time to take into account the effect of environmental variations during operations. The detection performance on our custom dataset demonstrates the feasibility of the proposed pipeline. Specifically, for physical attacks, the framework can achieve an accuracy of approximately 0.93 for instant-wise detection and 0.84 for event-wise detection.

Battery State-of-Health Aware Path Planning for a Mars Rover

Mariana Salinas-Camus — 2023-10-26

A rover mission consists of visiting waypoints to gather scientific samples based on set requirements. However, rovers face operational uncertainties during the mission, affecting the performance of its electrical and mechanical components and overall mission success. Hence, it is critical to have a decision-making framework that is aware of the health state of the components when planning the path of the vehicle. In particular, battery degradation, and consequently the battery State of Health (SOH), can affect the optimality of decisions made by the autonomous system in the long term. This paper presents a decision-making system that incorporates information on the energy drawn from the battery (based on the velocity of the vehicle), terrain conditions, and model-based prognostic modules to assess impact on the battery state of charge (SoC). The decision-making system was formulated as a Markov Decision Process (MDP) to reach the goal destination by sending commands in a determined amount of time, while maintaining the battery SoC within the policy stated. The MDP problem was programmed using the open-source framework POMDPs.jl, which has a variety of online and offline solvers. To solve the MDP problem online, we used Monte Carlo Tree Search (MCTS). Results from simulations demonstrate the effect that battery degradation and charging plans have on decision-making.

Context-aware machine learning for estimating the remaining useful life of bearings under varying speed operating conditions

Seyed Ali Hosseinli — 2023-10-26

Remaining useful life estimation is a crucial and complicated task in predictive maintenance in order to reduce downtime and avoid catastrophic breakdowns in industrial plants. Thanks to the recent advances in our machine learning era, deep learning models can effectively deal with modeling complex phenomena such as the bearing degradation process, specifically under varying operating conditions. However, obtaining large labeled datasets for training the data-dependent deep learning models is challenging and expensive. To overcome this limitation, a phenomenological model has been used in this study as an effective approach to creating synthetic run-to-failure datasets under varying operating conditions. The suggested methodology is able to adjust synthetic run-to-failure datasets to the different periodic speed profiles, including the speed ranges that pass the resonance frequency of the structure. A Context-aware Domain Adversarial Neural Network is proposed to remove the domain shift between the simulated signals and the real ones as well as enable the deep learning model to understand the varying speed operating conditions and the sequential order of the measurements. The simulated signals are used as the source domain and a limited number of the real signals are used as the unlabeled samples for the domain adaptation task. Context awareness is introduced to the network by integrating contextual information into the architecture of the Domain Adversarial Neural Network, leading to an improvement in the model performance and its generalization ability. A dataset captured in a bearing test rig is adopted to verify the proposed method. Results show that context awareness can result in better performance and also more robust predictions against major speed changes in varying speed scenarios compared to the non-context-aware models.

Co-design Model for Neuromorphic Technology Development in Rolling Element Bearing Condition Monitoring

Daniel Strombergsson — 2023-10-26

This paper presents an end-to-end condition monitoring co-design model, from vibration measurement to anomaly detection, where conventional signal processing principles are combined with neuromorphic sensing and computing concepts to enable investigations of the potential improvements offered by brain-like information processing technologies.

The use of machine learning in condition monitoring became increasingly popular for intelligent fault diagnosis in the last decade, taking advantage of the rapid developments in deep learning.

However, the high computational cost of training and using deep neural networks prevents the use of such solutions for analysing the bulk of data generated by the resource constrained edge devices, i.e., the condition monitoring sensor systems, as only a minor fraction of data can be transmitted to the cloud or edge servers for analysis.

There is an untapped potential to process this data and thereby improve intelligent fault diagnosis models using event-triggered sensing, spiking neural networks, and neuromorphic processors that substantially can improve the energy efficiency and capacity of embedded machine learning condition monitoring solutions.

The proposed co-design model is evaluated on two use-cases involving rolling element bearing failures, one based on a labelled laboratory environment dataset, and one based on a wind turbine drivetrain bearing failure representing a real-world scenario with stochastic changes of machine state and unknown labels of the bearing condition.

By adjusting co-design parameters, the resulting hybrid conventional/neuromorphic model show a comparable accuracy in detection performance for the laboratory dataset compared to the state-of-the-art reported in the literature.

Similarly, for the wind turbine drivetrain dataset a bearing fault detection time comparable to that in previous work is obtained.

This shows the successful implementation of a hybrid conventional/neuromorphic co-design model for condition monitoring applications, offering novel opportunities to investigate performance trade-offs and efficiency improvements enabled by neuromorphic technologies.

DAGGER: Data AuGmentation GEneRative Framework for Time-Series Data in Data-Driven Smart Manufacturing Systems

Nicholas Hemleben — 2023-10-26

As industries transition into the Industry 4.0 paradigm, the relevance and interest in concepts like 
Digital Twin (DT) are at an all-time high. DTs offer direct avenues for industries to make more 
accurate predictions, rational decisions, and informed plans, ultimately reducing costs, increasing 
performance and productivity. Adequate operation of DTs in the context of smart manufacturing relies 
on an evolving data-set relating to the real-life object or process, and a means of dynamically updating 
the computational model to better conform to the data. This reliance on data is made more explicit when 
physics-based computational models are not available or difficult to obtain in practice, as it's the 
case in most modern manufacturing scenarios. For data-based model surrogates to "adequately" represent 
the underlying physics, the number of training data points must keep pace with the number of degrees of 
freedom in the model, which can be on the order of thousands. However, in niche industrial scenarios 
like the one in manufacturing applications, the availability of data is limited (on the order of a few 
hundred data points, at best), mainly because a manual measuring process typically must take place for 
a few of the relevant quantities, e.g., level of wear of a tool. In other words, notwithstanding the 
popular notion of big-data, there is still a stark shortage of ground-truth data when examining, for 
instance, a complex system's path to failure. In this work we present a framework to alleviate this 
problem via modern machine learning tools, where we show a robust, efficient and reliable pathway to 
augment the available data to train the data-based computational models.

 Small sample size data is a key limitation in performance in machine learning, in particular with 
very high dimensional data. Current efforts for synthetic data generation typically involve either 
Generative Adversarial Networks (GANs) or Variational AutoEncoders (VAEs). These, however, are are 
tightly related to image processing and synthesis, and are generally not suited for sensor data 
generation, which is the type of data that manufacturing applications produce. Additionally, GAN 
models are susceptible to mode collapse, training instability, and high computational costs when used 
for high dimensional data creation. Alternatively, the encoding of VAEs greatly reduces dimensional 
complexity of data and can effectively regularize the latent space, but often produces poor 
representational synthetic samples. Our proposed method thus incorporates the learned latent space 
from an AutoEncoder (AE) architecture into the training of the generation network in a GAN. The 
advantages of such scheme are twofold: \textbf{(\textit{i})} the latent space representation created 
by the AE reduces the complexity of the distribution the generator must learn, allowing for quicker 
discriminator convergence, and \textbf{(\textit{ii})} the structure in the sensor data is better 
captured in the transition from the original space to the latent space. Through time statistics (up to 
the fifth moment), ARIMA coefficients and Fourier series coefficients, we compare the synthetic data 
from our proposed AE+GAN model with the original sensor data. We also show that the performance of 
our proposed method is at least comparable with that of the Riemannian Hamiltonian VAE, which is a 
recently published data augmentation framework specifically designed to handle very small high 
dimensional data sets.

Data Augmentation of Sensor Time Series using Time-varying Autoregressive Processes

Douglas Baptista de Souza — 2023-10-26

This work presents a novel data-centric solution for fault diagnostics and failure prognostics consisting of a data-augmentation method which is well suited for non-stationary mutivariate time-series data. The method, based on time-varying autoregressive processes, can be employed to extract key information from a limited number of samples and generate new artificial samples in a way that benefits the development of diagnostics and prognostics solutions. The proposed approach is tested based on three real-world datasets associated with failure diagnostics problems using two types of machine learning methods. Results indicate the proposed method improves performance in all tested cases.

Data-Driven Approaches to Diagnostics and State of Health Monitoring of Maritime Battery Systems

Erik Vanem — 2023-10-26

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 important to monitor the available energy that can be stored in the batteries, and classification societies typically require that the state of health (SOH) can be verified by independent tests. However, this paper addresses data-driven approaches to state of health monitoring of maritime battery systems based on operational sensor data. Results from various approaches to sensor-based, data-driven degradation monitoring of maritime battery systems will be presented, and advantages and challenges with the different methods will be discussed. The different approaches include cumulative degradation models and snapshot models. Some of the models need to be trained, whereas others need no prior training. Moreover, some of the methods only rely on measured data, such as current, voltage and temperature, whereas others rely on derived quantities such as state of charge (SOC). Models include simple statistical models and more complicated machine learning techniques. Different datasets have been used in order to explore the various methods, including public datasets, data from laboratory tests and operational data from ships in actual operation. Lessons learned from this exploration will be important in establishing a framework for data-driven diagnostics and prognostics of maritime battery systems within the scope of classification societies.

A Comparison of Residual-based Methods on Fault Detection

Chi-Ching Hsu — 2023-10-26

An important initial step in fault detection for complex industrial systems is gaining an understanding of their health condition. Subsequently, continuous monitoring of this health condition becomes crucial to observe its evolution, track changes over time, and isolate faults. As faults are typically rare occurrences, it is essential to perform this monitoring in an unsupervised manner. Various approaches have been proposed not only to detect faults in an unsupervised manner but also to distinguish between different potential fault types. In this study, we perform a comprehensive comparison between two residual-based approaches: autoencoders, and theinput-output models that establish a mapping between operating conditions and sensor readings. We explore the sensorwise residuals and aggregated residuals for the entire system in both methods. The performance evaluation focuses on three tasks: health indicator construction, fault detection, and health indicator interpretation. To perform the comparison, we utilize the Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) dynamical model, specifically a subset of the turbofan engine dataset containing three different fault types. All models are trained exclusively on healthy data. Fault detection is achieved by applying a threshold that is determined based on the healthy condition. The detection results reveal that both models are capable of detecting faults with an average delay of around 20 cycles and maintain a low false positive rate. While the fault detection performance is similar for both models, the input-output model provides better interpretability regarding potential fault types and the possible faulty components.

Design and Implementation of a Model Selection Pipeline for Prognostics and Health Management in the Operational Environment

Peter Bishay — 2023-10-26

Model selection is a crucial aspect of Prognostics and Health Management (PHM). However, many PHM models are de- veloped for specific datasets and lack flexibility to adapt to different datasets with varying data quality considerations. To address this gap, we propose a generalizable model selection pipeline for PHM. Our approach involves creating a pipeline for testing models that users can tune in various ways. We designed a sequential pipeline of steps for model selection with a focus on implementation considerations which include recommendations for handling environmental variables, ca- pabilities for remote and local work environments, and stor- age considerations of the serialized pipeline. Performance metrics are designed to consider data quality characteristics such as ambiguous labeling. We illustrate the generalizability of our approach through a case study of our model selection pipeline applied to a field dataset with ambiguous labels. Our design accommodates data characteristics commonly found in field data, such as ambiguous labels and data wrangling. Our contribution fills a gap in real-world implementations of PHM by offering technology considerations and recommen- dations for effective deployment.

Diagnostic Signal Method for Fault Identification of Electro-Hydraulic Servo Actuators

Zihan Liu — 2023-10-26

In this paper, we develop a fault identification approach for electro-hydraulic servo actuators based on injecting a pre-defined diagnostic signal into the system and then extracting fault-related features from the phase space topology. Next, we build regression models using an artificial neural network, which maps the feature space to fault space to identify the faults represented by the system’s parameters. The performance of the proposed fault identification approach is evaluated when the degradation of permanent armature occurs. The effect of parametric faults on the dynamics is studied and discussed. The different excitation of the system is considered, and the robustness of the proposed method under the condition of noise is also explored. The obtained results indicate the effectiveness of injected diagnostic signals in enriching the dynamics of the system and increasing the quality of extracted features and the accuracy of trained artificial neural networks.

Differentiable Short-Time Fourier Transform Window Length Selection Driven by Cyclo-Stationarity

Douw Marx — 2023-10-26

The Short-Time Fourier transform is widely applied in the condition monitoring of rotating machinery. Even so, selecting the optimal window length for the Short-Time Fourier Transform remains a challenge. This work presents a procedure for adapting the Short Time Fourier Transform algorithm to be differentiable with respect to window length by using continuous window functions defined over the entire input signal duration. Thanks to this modification, a differentiable loss criterion can be defined to measure the Short-Time Fourier quality, and the gradient of the loss criterion with respect to window length can be computed. The optimal window length for a given loss criterion can then be efficiently solved for using a gradient-based optimization algorithm. Results from a simulated bearing dataset and three experimental bearing datasets are used to compare the optimal spectrograms obtained using different loss criteria. Specifically, a sparsity-based loss criterion is compared with two loss criteria inspired by the characteristic cyclo-stationarity machine of faults in rotating machinery. The results demonstrate the effectiveness of the differentiable window length selection method and highlight the importance of selecting appropriate loss criteria for defining STFT quality. Loss criteria that account for the cyclo-stationary nature of the signals are shown to be less likely to target single high-amplitude impulsive events compared to the sparsity-based loss criterion.

Ensemble Learning Based Convolutional Neural Networks for Remaining Useful Life Prediction of Aircraft Engines

Thambirajah Ravichandran — 2023-10-26

Remaining useful life (RUL) prediction is an essential task of Prognostics and Health Management (PHM) of aircraft engines performed utilizing the data collected from multiple sensors to ensure their safety. While many studies have been reported on RUL prediction for aircraft engines, only a few of them focus on ensemble learning based convolution neural network (CNN) models for RUL prediction. This paper proposes a new data-driven approach based on a multistage ensemble learning strategy for developing CNN models for RUL prediction of aircraft engines. The proposed approach places a major emphasis on generating diverse CNN models by exploring 2D CNN models and 1D CNN models with multiple channels and developing a multistage ensemble approach employing sparsity promoting model selection and weight learning methods to utilize only a subset of available models thus improving the RUL prediction performance. The effectiveness of the proposed approach is validated using the NASA C-MAPSS dataset for aircraft engines.

Explainable Predictive Maintenance is Not Enough: Quantifying Trust in Remaining Useful Life Estimation

Ripan Kumar Kundu — 2023-10-26

Machine learning (ML)/deep learning (DL) has shown tremendous success in data-driven predictive maintenance (PdM). However, operators and technicians often require insights to understand what is happening, why it is happening, and how to react, which these black-box models cannot provide. This is a major obstacle in adopting PdM as it cannot support experts in making maintenance decisions based on the problems it detects. Motivated by this, several researchers have recently utilized various post-hoc explanation methods and tools, such as LIME, SHAP, etc., for explaining the predicted RUL from these black-box models. Unfortunately, such (post-hoc) explanation methods often suffer from the \emph{disagreement problem}, which occurs when multiple explainable AI (XAI) tools differ in their feature ranking. Hence, explainable PdM models that rely on these methods are not trustworthy, as such unstable explanations may lead to catastrophic consequences in safety-critical PdM applications. This paper proposes a novel framework to address this problem. Specifically, first, we utilize three state-of-the-art explanation methods: LIME, SHAP, and Anchor, to explain the predicted RUL from three ML-based PdM models, namely extreme gradient boosting (XGB), random forest (RF), logistic regression (LR), and one feed-forward neural network (FFNN)-based PdM model using the C-MAPSS dataset. We show that the ranking of dominant features for RUL prediction differs for different explanation methods. Then, we propose a new metric \emph{trust score} for selecting the proper explanation method. This is achieved by evaluating the XAI methods using four evaluation metrics: fidelity, stability, consistency, and identity, and then combining them into a single \emph{trust score} metric through utilizing Kenny and Borda rank aggregation methods. Our results show that the proposed method effectively selects the most appropriate explanation method from a set of explanation methods for estimated RULs. To the best of our knowledge, this is the first work that attempts to address and solve the disagreement problem in explainable PdM.

Explainable Prognostics Method through Differential Evolved RVR Ensemble of Relevance Vector Machines

Miltiadis Alamaniotis — 2023-10-26

Operating experience from various mechanical components indicates that their operating performance depends on non-well known physical mechanisms, while it is likely that various unexpected factors will act as catalysts for reaching the failure point. Therefore, one way to overcome the partially knowledge of physical mechanisms is the use of data-driven methods that estimate the degradation patterns and predict the failure point. Thus, there is a growing need to design and develop new and more sophisticated prognostic technologies that can estimate the remaining useful life of a mechanical component. In this work, a new method for prognostics is proposed that not only provides a prediction over the failure point but also provides an explanation over the rationale behind that prediction. The proposed method utilized tools from artificial intelligence and more specifically relevance vector machines (RVM) and differential evolution (DE). The cornerstone of the method is the assembly of an ensemble comprised of multiple RVM equipped with different kernels, and the subsequent evolution of the ensemble using the differential evolution. DE will provide a set of values for the coefficients of the ensemble. Then based on the coefficients together with their associated RVMs are used to provide an explanation over the prediction. The explanation stems from the kernels themselves as each kernel models different set of properties. The presented method is tested on a set of real-world degradation data taken from a Gas Turbine (GT) propulsion plant.

Exploring Filter Banks and Spike Interval Statistics of Level-Crossing ADCs for Fault Diagnosis of Rolling Element Bearings

Ashwani Kumar — 2023-10-26

Nowadays, lots of data are generated in industries using vibration sensors to evaluate the equipment’s working condition and identify faults. A significant challenge is that only a small fraction of data can be transmitted for intelligent fault diagnosis and storage. The edge processing capacity is often insufficient for advanced analysis due to time and resource constraints. The neuromorphic signal encoding scheme efficiently reduces the data rate by encoding relevant signal changes into spike trains while discarding redundant information and noise, enabling energy-efficient neuromorphic processing. Due to the presence of dominant operational features and noise in the original measurements, signal pre-processing is required to extract the relevant features before spike coding and processing. The work investigates the effects of different filter banks (pre-processing methods) on the spike encodings for vibration measurements from bearings. This also includes bearing fault features diagnosis based on statistical analysis of generated spikes. The comparative analysis is made for benchmarking different signal pre-processing methods (e.g., envelope, empirical mode decomposition (EMD), and gammatone filter) on bearing vibration datasets. An event-triggered scheme, i.e., Level-crossing analog-to-digital converters (LC-ADCs) is applied to encode the vibration measurement to spikes. Inter-spike intervals (ISIs) statistics are analysed for fault diagnosis of bearings. The results obtained for CWRU bearing databases indicate a possible fault detection and diagnosis with significant data rate reduction and an opportunity for improved computational efficiency. With the developed approach, the envelope filter is found to be the most efficient of all. This work enables a new approach to improve the energy efficiency of condition monitoring systems and further sets a new course of research development in this area using neuromorphic technologies.

Few-shot Learning for Plastic Bearing Fault Diagnosis – An Integrated Image Processing and NLP Approach

David He — 2023-10-26

Plastic bearings have a wide range of industrial applications due to their many desirable properties such as lightweight, low friction coefficient, chemical resistance, and ability to operate without lubrication. Timely bearing fault diagnosis can prevent equipment failure and costly downtime. In recent years, developing machine learning based bearing fault diagnosis with few labelled data has attracted a lot of attentions as datasets with fault labels are rare in many industrial applications. One effective approach to meet the challenge is few-shot learning. Among many approaches, utilizing a good pre-trained deep learning model to achieve few-shot learning is an effective and efficient alternative. In this paper, a pre-trained deep learning model called CLIP that combines image processing and natural language processing (NLP) is adopted to few-shot learning for plastic bearing fault diagnosis. We explore the feasibility of leveraging CLIP model in the realm of bearing fault diagnosis via few-shot learning. Specifically, we tackle the challenges posed by CLIP's creation of requisite text prompt embeddings for the diagnosis of mechanical faults, within a few-shot learning framework. Our investigation illuminates the remarkable capability of CLIP to adapt to new tasks with minimal examples, a feature we exploit to devise a solution for plastic bearing fault diagnosis. The effectiveness of the few-shot learning method with CLIP is demonstrated using vibration data collected from plastic bearing seeded fault tests in the laboratory.

Graph neural networks for dynamic modeling of roller bearings

Vinay Sharma — 2023-10-26

Machine learning has paved the way for the real-time monitoring of complex infrastructure and industrial systems. However, purely data-driven methods have not been able to learn the underlying dynamics and generalize them to operating conditions that have not been covered by the training datasets. Therefore, they have not been able to predict the long-term evolution of the system state of physical systems. Physics-informed neural networks (PINNs) have recently shown promising results in predicting the system state evolution over extended periods of time, owing to the loss terms derived from the underlying partial differential equations governing the dynamics of the systems. However, PINNs have limited generalization ability, i.e., a model trained on one type of boundary condition cannot generalize to other conditions. Moreover, the governing equations used for describing the dynamics of physical systems are an approximation of reality, which can lead to differences between the predictions and the actual roll-out of the trajectory. Recently, graph neural networks (GNNs) have been applied to predict the evolution of system dynamics. Due to the encoded inductive bias, they generalize well to systems with varying configurations and boundary conditions. Message-passing GNN comprises two parts that learn the interaction between nodes: an edge network that takes the translational invariant features between two nodes (for e.g., the distance vector) and generates a message, and a node network that takes the aggregated messages from all the neighboring nodes and produces a new node state. This process is repeated several times until the final node state is decoded as a required output.

In the presented work, we propose to apply the framework of GNNs for predicting the dynamics of a rolling element bearing. The computational efficiency and generalizability of such a method enable the scalable use of a real-time digital twin to monitor the health state of a rotating machine. To this end, a GNN is used to mimic a dynamic spring-mass-damper model. Bearings consist of different interacting parts like the inner race, outer race, and multiple rolling elements. This interconnected and interacting architecture of a typical bearing is suitable to be modeled as a graph with nodes representing different components.

We use the dynamic spring-mass-damper model to generate the training data for the GNN, where bearing components such as rolling elements, and inner and outer raceway are modeled as discrete masses. A Hertzian contact model is used to calculate the forces between these components. We evaluate the learning and generalization capabilities of the proposed GNN framework by testing bearing configurations different from the training configurations and comparing the performance to that of the spring-mass model.

Fault Severity Estimation in Cracked Shafts by Integration of Phase Space Topology and Convolutional Neural Network

Utkarsh Andharikar — 2023-10-26

With the rapid advancement of industrial systems and the unavoidable complications and interconnectedness in systems, diagnostics of industrial machinery are achieving paramount importance. Accurate estimation of health condition of industrial machinery becomes more challenging due to the inherent nonlinearity, complexity, and uncertainty of the observations. Nonlinear dynamic analysis has proven to be a powerful tool for providing information about the health condition of a system that can be used for diagnostic applications. The current study particularly focuses on crack depth estimation using phase space analysis. Phase space provides a topological representation of the dynamics of the system and is highly informative about the health condition. The information suitable for diagnostics is employed by Convolutional Neural Networks, which are known to be powerful in extracting spatial information from maps. The proposed diagnostic method is evaluated on a Jeffcott rotor model with transverse crack in the rotating shaft to estimate the severity of the fault from the phase space topology as a case study.

Increasing Robustness of Data-Driven Fault Diagnostics with Knowledge Graphs

Maximilian-Peter Radtke — 2023-10-26

In the realm of PHM, it is common to possess not only process data but also domain knowledge, which, if integrated into data-driven algorithms, can aid in solving specific tasks.
This paper explores the integration of knowledge graphs (KGs) into deep learning models to develop a more resilient approach capable of handling domain shifts, such as variations in machine operation conditions.
We present and assess a KG-enhanced deep learning approach in a representative PHM use case, demonstrating its effectiveness by incorporating domain-invariant knowledge through the KG.
Furthermore, we provide guidance for constructing a comprehensive hierarchical KG representation that preserves semantic information while facilitating numerical representation.
The experimental results showcase the improved performance and domain shift robustness of the KG-enhanced approach in fault diagnostics.

Limitations and Opportunities in PHM for Offshore Wind Farms: A Socio-Technical-Ecological System Perspective

Arvind Keprate — 2023-10-26

The burgeoning importance of offshore wind farms (OWFs) in the transition to sustainable energy systems underscores the need for effective Prognostics and Health Management (PHM) strategies. While the current PHM framework demonstrates its prowess in enhancing the reliability and operational efficiency of OWFs, this paper contends that its potential remains largely untapped due to certain inherent limitations. This study casts a comprehensive spotlight on the limitations and untapped opportunities within the PHM framework for OWFs from a Socio-Technical-Ecological Systems (SETS) perspective.

The limitations, as identified, are threefold. First, the existing framework exhibits an over-reliance on technical factors, thus prioritizing maximization of Remaining Useful Life and cost minimization. This emphasis disregards crucial Non-Technological Factors (such as community impacts, stakeholder engagement, Human and Organization Factors (HOFs)) and uncertainty arising from them, which can exert significant influences on OWF’s health and performance. Second, the PHM approach often adopts a component-centric view, with focus on dominant degradation modes, thus undermining the intricate interdependencies among diverse components and failure modes. This lack of a System Level Perspective (SLP) and Multi-Modal Degradation (MMD) hampers a comprehensive understanding of how component degradation cascades through the entire system. Third, the current framework largely ignores the ecological considerations, despite compelling evidence that the current monitoring, assessment, and maintenance activities has significant ecological consequences.

By addressing the identified limitations and leveraging the opportunities together with AI, the PHM framework for OWFs can evolve into a more comprehensive, inclusive, and resilient approach. The proposed paradigm shift resonates deeply with the contemporary drive towards sustainability, not only in terms of technical efficacy but also in terms of social acceptance and ecological compatibility.

Labelling of Annotated Condition Monitoring Data Through Technical Language Processing

Karl Lowenmark — 2023-10-26

We propose a novel approach to facilitate supervised fault diagnosis on unlabelled but annotated industry datasets using human-centric technical language processing and weak supervision. Fault diagnosis through Condition Monitoring (CM) is vital for high safety and resource efficiency in the green transition and digital transformation of the process industry. Learning-based Intelligent Fault Diagnosis (IFD) methods are required to automate maintenance decisions and improve decision support for analysts. A major challenge is the lack of labelled industry datasets, limiting supervised IFD research to lab datasets. However, features learned from lab environments generalise poorly to field environments due to different signal distributions, artificial induction or acceleration of lab faults, and lab set-up properties such as average frequency profiles affecting learned features. In this study, we investigate how the unstructured free text fault annotations and maintenance work orders that are present in many industrial CM systems can be used for IFD through technical language processing, based on recent advances in natural language supervision. We introduce two distinct pipelines, one based on contrastive pre-training on large datasets, and one based on a small-data human-centric approach with unsupervised clustering methods. Finally, we showcase one example of the small-data fault classification implementation on a CM industry dataset with a SentenceBERT language model, kMeans clustering, and conventional signal processing methods. Fault class imbalance and time-shift uncertainty is overcome with weak supervision through aggregates of features, and human-centric clustering is used to integrate technical knowledge with the annotation-based fault classes. We show that our model can separate cable and sensor fault recordings from bearing-related fault recordings with an F1-score of 93. To our knowledge, this is the first system to classify faults in field industry CM data based only on associated unstructured fault annotations.

Operational Wheel Flat Detector in Railway Vehicles

Ibon Erdozain — 2023-10-26

Maintenance of railway systems is shifting from being based on scheduled interventions to a continuous regime based on the actual status of assets. This change is supported mainly on three pillars: the development of new sensors and signal processing techniques, the capability to store and analyze all the information gathered by this huge amount of new sensors, and the capability of modifying dynamically the maintenance plans. This paper presents a new wayside system for detecting flats whose development has been based on combining physical models with Machine Learning Techniques. Physical models are used to understand the phenomena, define the key indicators to characterize the phenomena and generate synthetic data to train Machine Learning algorithms. Subsequently, regression models are generated to relate the key parameters with the flat severity. The last part of the paper is focused on validating the proposed methodology in a real environment.

Parameters identification of the satellite attitude control system with large inertia rotating components

Xueqin Chen — 2023-10-26

This paper investigates the unbalance parameter identification of the large inertia rotating component of satellite. Firstly, the dynamics model with unbalance parameter of the large inertia rotating component is established. Then, based on the principle of parameter separation and decoupling, a modified two-stage exogenous Kalman filter (TSXKF) algorithm is proposed. This method works directly on nonlinear system, estimates the centroid position, the centroid velocity, attitude angular, attitude angular velocity, and identifies the nonlinear unknown static unbalance parameter, which is the centroid offset. Finally, the simulation results verify the effectiveness of the method.

Prognosis of Li-ion Batteries Under Large Load Variations Using Hybrid Physics-Informed Neural Networks

Kajetan Fricke — 2023-10-26

The development of new modes of transportation such as electric vertical takeoff and landing aircraft and the use of drones for package and medical delivery have increased the demand for reliable and powerful electric batteries. Therefore, accurately predicting the degradation of a battery’s state-ofhealth (SOH) and state-of-charge (SOC) is a crucial albeit still challenging task. There is a need for models that can accurately predict the SOH and SOC while taking into account the specific characteristics of a battery cell and its usage profile. While traditional physics-based and data-driven approaches are used to monitor the SOH and SOC, they both have limitations related to computational costs or that require engineers to continually update their prediction models as new battery cells are developed and put into use in battery-powered vehicle
fleets.
Battery capacity degradation can vary from battery to battery and can also be influenced by changes in load due to internal thermal stress. While sophisticated electrochemistry-based models can provide precise predictions of the SOC during a
discharge cycle when parameters are well-tuned, using highfidelity models for prognostics purposes can be computationally expensive. Those models also require tuning to specific battery types and at times to specific specimens, thus hindering generalization. In contrast, purely data-driven approaches can learn the relationship between input and output for SOC prediction based on load input, but they require a large and diverse training dataset and lack any physical or electrochemical understanding, making far-ahead predictions challenging if test loading conditions fall outside the training distribution. To address some of the drawbacks of the aforementioned modeling approaches, in this paper, we enhance a hybrid physics-informed machine learning version of a battery SOC model we presented in previous work to predict voltage drop
during discharge. The enhanced model captures the effect of wide variation of load levels, in the form of input current,
which causes large thermal stress cycles. The cell temperature build-up during a discharge cycle is used to identify temperature-sensitive model parameters. Additionally, we enhance an existing aging model built upon cumulative energy drawn by introducing the effect of the load level. We then map cumulative energy and load level to battery capacity with a Gaussian process model. To validate our approach we use a battery aging dataset collected on a self-developed testbed, where we used a wide current level range to age battery packs in accelerated fashion. Prediction results show that our model can be successfully calibrated and generalizes across all applied load levels.

Promoting Explainability in Data-Driven Models for Anomaly Detection: A Step Toward Diagnosis

Quentin Dollon — 2023-10-26

Anomaly detection has become a critical task in industry. Data-driven models are often used for anomaly detection due to their ability to learn patterns from data and identify behaviors that deviate from the learned patterns. Furthermore, they are simple to implement as they do not rely on complex physical models to make predictions. However, one major limitation of these models is their lack of explainability, which hinders the diagnosis of detected anomalies.

Explainability provides transparency and interpretability, allowing stakeholders to understand the reasons for the detected deviation. In the absence of explainability, it is challenging to determine why a particular instance was classified as abnormal. Without an understanding of the underlying reason for the anomaly, it becomes difficult to prescribe a reliable diagnostic. This can result in missed opportunities for preventing or mitigating damage caused by the anomaly. Explainability can also help in detecting false positives and false negatives, especially, to distinguish between abnormal behaviors and sensor failures.

Hydro-Quebec is the principal actor in electricity management in Quebec, Canada. The overwhelming majority of the production comes from hydroelectric generating units. Power grid sustainability then strongly depends on the efficient health supervision of these assets. In this study, we introduce a data-driven semi-supervised algorithm for anomaly detection, with emphasis on statistical explainability. This feature needs to be distinguished from the traditional explainable models, that build upon physics to interpret observations. Here, the purpose is to track the sources of deviations through statistics. This model does not belong to diagnosis tools, because its sole output is not sufficient to find the root causes of a problem. However, it makes a bridge toward such tools by providing clues about origin of failures.

The algorithm performs in two-stages. First a model is trained to learn the normal behavior of the generating unit for a given set of operating conditions. This part involves clustering for data reduction and kriging for regression. Second, it compares the multidimensional prediction with the actual realization. It quantifies the deviation of the asset to its expected behavior and provides an explainable indicator for anomaly detection.

After introducing the background foundations of the method, some examples are given that demonstrate the advantage of interpretability for support to operation and diagnosis. It will be shown how such an algorithm can be deployed in an operational environment and how it should be combined with other tools to improve assets health management.

Reconceptualizing the Prognostics Digital Twin for Smart Manufacturing with Data-Driven Evolutionary Models and Adaptive Uncertainty Quantification

Jack Murray — 2023-10-26

This work presents an integrated architecture for a prognostic digital twin for smart manufacturing 
subsystems. The specific case of cutting tool wear (flank wear) in a CNC machine is considered, using 
benchmark data sets provided by the Prognostics and Health Management (PHM) Society. This paper 
emphasizes the role of robust uncertainty quantification, especially in the presence of data-driven 
black- and gray-box dynamic models. A surrogate dynamic model is constructed to track the evolution 
of flank wear using a reduced set of features extracted from multi-modal sensor time series data. The 
digital twin's uncertainty quantification engine integrates with this dynamic model along with a 
machine emulator that is tasked with generating future operating scenarios for the machine. The 
surrogate dynamic model and emulator are combined in a closed-loop architecture with an adaptive 
Monte Carlo uncertainty forecasting framework that allows prediction of quantities of interest 
critical to prognostics within user-prescribed bounds. Numerical results using the PHM dataset are 
shown illustrating how the adaptive uncertainty forecasting tools deliver a trustworthy forecast by 
maintaining predictive error within the prescribed tolerance.

Rethinking Reliability in Terms of Margins

Diego Mandelli — 2023-10-26

Current reliability approaches were designed to assess and quantify the reliability associated with complex systems such as nuclear power plants (NPPs). These approaches are generally based on classical Boolean logic structures such as event trees (ETs) and fault trees (FTs) [Rausand, 2020]. The outcome obtained by combining FTs and ETs is the set of minimal cuts sets (MCSs), with each MCS representing a unique combination of BEs that leads to an undesired outcome (e.g., core damage). Probabilistic evaluation of a MCS is performed by evaluating the product of the probability values associated with each BE. A relevant factor here is that the probability values associated with BEs used in the plant models are updated at least every 4 years based on past operational experience and through the use of a Bayesian statistical process [Siu, 1998]. Hence, the probability value of a BE associated with a physical asset (e.g., a centrifugal pump or motor-operated valve) in no way reflects that asset’s actual condition and performance.

This fact plays a major role in the application of plant reliability models to support risk-informed decisions. With the particular goal of reducing operation and maintenance costs, existing NPPs are moving from corrective and periodic maintenance toward new types of predictive maintenance strategies [Agarwal, 2021]. This transition is designed such that maintenance is conducted only when the asset requires it (i.e., prior to undergoing imminent failure). And though these benefits cannot be achieved through actual reliability modelling methods and currently employed reliability data, they can be achieved by employing asset-monitoring sensors, automated data acquisition systems, data analysis methods, and improved decision-making processes. Combined, these resources can provide precise information on the health of an asset, track its degradation trends, and estimate its expected failure time. Based on such information, maintenance operations can be scheduled and performed for each asset on an as-needed basis. This dynamic context of predictive maintenance operations requires new methods of data analysis, the propagation of asset health information from the asset level to the system level, and the optimization of plant resources.

This paper provides an alternative reliability approach designed for a predictive maintenance context in which a direct link is created between monitoring data and decision-making. Rather than thinking of reliability in terms of system/asset probability of failure, we propose a reliability mindset based on the concept of margin [Mandelli, 2023]. An asset’s health is quantified by determining its margin, based on the asset’s current and historical monitoring data. The margin values of the monitored asset are then propagated through system reliability models (e.g., FTs or reliability block diagrams) to identify the assets that are more critical to guarantee system operation. We show how a margin-based approach can be used assess asset health, based solely on current and historic monitoring data (e.g., condition-based, anomaly detection, diagnostic, and prognostic data) [Xingang, 2021]. A margin-based approach directly addresses the limitations of classical reliability modelling approaches and provides a snapshot of system health—given the availability of monitoring data. These two different approaches are designed to address different types of decisions: classical reliability models support static decisions (e.g., a set frequency of periodic maintenance or surveillance operations) based on past operational experience, whereas a margin-based approach directly supports dynamic decisions involving maintenance operations that should only be performed when necessary, based on monitoring data (i.e., a predictive maintenance context).

Sensitivity enhanced method for fault detection and prediction of elevator doors using a margin maximized hyperspace

Minjae Kim — 2023-10-26

This paper proposes novel fault classification and prediction method by addressing a margin maximized hyperspace (MMH) to solve the problems absent of any label at highly imbalanced dataset, which is a frequent but challenging problem in real-world industries. The proposed method features three characteristics. First, knowledge-based feature manipulation is executed using reference and feedback physical properties and the manipulated features are used for training the proposed neural network because the features contain rich information for classifying and predicting faults of the system of interest. Second, VAE transforms high-dimensional input features to a low-dimensional feature space. This nonlinear space transformation reduces the complexity of the classification securing high accuracy and robustness of fault classification in the MMH. Third, the acquired MMH through VAE with Bayesian optimization statistically allocates two extremes of major (normal) and minor (faulty) clusters at origin and unity at the feature space, indicating that sensitivity of fault prediction is maximized. The method would be highly effective in that the model only focuses on separating major and minor clusters deciding each health condition but ignores minor differences within the clusters which confuse users. The effect of the method is demonstrated with field measurements of an elevator door stroke dataset comprising normal, degradation, and faulty states in open and close strokes. The systematic analysis shows that these characteristics contribute to improve accuracy and robustness for fault classification. Specifically, knowledge-based feature manipulation improves the accuracy, and VAE enhances sensitivity on separating each cluster and locational constancy. Moreover, the MMH is effective to predict potential fault without any label for a highly imbalanced dataset. The proposed method provides remaining useful lifetime (RUL) using distances from normal and faulty clusters at the MMH, which enables to quantitatively provide RUL of the system without any definition of RUL. Considering that many systems deployed on fields lack information for fault life or residual useful life, the proposed method would be practical and effective for real world applications.

Unsupervised Physics-Informed Health Indicator Estimation for Complex Systems

Kristupas Bajarunas — 2023-10-26

Developing Health Indicators (HI) is a crucial aspect of prognostics and health management of complex systems. Previous research has demonstrated the benefits of accurately determining the HI, which can lead to better performance of prognostic models. However, the existing methodologies for determining HI in complex systems are mostly semi-supervised and rely on assumptions that may not hold in real-world scenarios. The existing methods usually involve using a reference set of healthy sensor readings or run-to-failure data to infer HI. But the unsupervised inference of HI from sensor readings, which is challenging in scenarios where diverse operating conditions can mask the effect of degradation on sensor readings, has not been extensively researched. In this paper, we propose a novel physics-informed unsupervised model for determining HI. Unlike previous methods, constrained by assumptions, the proposed method uses prior general knowledge about degradation to infer HI, thereby eliminating the need for a reference set of healthy sensor readings. The proposed unsupervised model is an Autoencoder that incorporates constraints on its latent space to ensure consistency with knowledge about degradation. We assess the efficacy of the proposed model by analyzing a prevalent prognostic case study, specifically the turbofan engine dataset (N-CMAPSS). Our analysis considers the model's sensitivity to data availability and the resulting Health Index's quality, including trendability and monotonicity. Additionally, we investigate the impact of incorporating the Health Index in predicting Remaining Useful Life (RUL). We demonstrate that our proposed method generates a Health Index that exhibits greater monotonicity and trendability than the current state-of-the-art semi-supervised approach. Moreover, our approach for identifying the Health Index leads to enhanced prognostic performance compared to the existing semi-supervised approach.

Underlying Probability Measure Approximated by Monte Carlo Simulations in Event Prognostics

David Acuña-Ureta — 2023-10-26

The prognostic of events, and particularly of failures, is a key step towards allowing preventive decision-making, as in the case of predictive maintenance in Industry 4.0, for example. However, the occurrence time of a future event is subject to uncertainty, so it is natural to think of it as a random variable. In this regard, the default procedure (benchmark) to compute its probability distribution is empirical, through Monte Carlo simulations. Nonetheless, the analytic expression for the probability distribution of the occurrence time of any future event was presented and demonstrated in a recent publication. In this article it is established a direct relationship between these empirical and analytical procedures. It is shown that Monte Carlo simulations numerically approximate the analytically known probability measure when the future event is triggered by the crossing of a threshold.

A Fine-grained Semi-supervised Anomaly Detection Framework for Predictive Maintenance of Industrial Assets

Xiaorui Tong — 2023-10-26

Reliable operation of industrial assets is of high priority for businesses where productivity determines the ability to deliver safety-critical products of high quality in a timely manner. The aerospace industry leads the demand for predictive maintenance (PdM). In the manufacturing space, unscheduled down time causes production delay and increases operational costs while introducing potential risks in product quality and on-time delivery. In field application of these products, unexpected breakdown of critical components can result in safety-critical events. Failure events are, therefore, extremely rare in industrial settings. Diverse operating conditions in the manufacturing environment and field applications contribute to the heterogeneous nature of data collected from these assets. This work presents an anomaly detection framework for PdM of industrial assets to address the practical challenges of scarce failure data sources and heterogeneous data across assets. We introduce a fine-grained modeling approach that efficiently accounts for individual asset differences in a semi-supervised fashion which requires only normal operation data for model training. The framework is demonstrated with an industrial 4.0 use case. Vibration sensor data from pumps in one of our manufacturing facilities is ingested to enable PdM with 2 weeks lead time using the proposed framework. This transforms unexpected breakdown time to scheduled maintenance, thereby reducing cost of delays and operation interruptions. The systematic implementation of the framework in the case study covers the practical production aspects including data quality evaluation, model training, optimization and daily serving of predictions. Furthermore, implementation challenges and recommendations are discussed based on the end-to-end solution implementation experiences.

Accelerated Degradation Test on Electric Scroll Compressor Using Controlled Continuous Liquid Slugging

Hadyan Ramadhan — 2023-10-26

Refrigerant-based electric scroll compressors are known for their reliability, efficiency, and quiet operation. They are often used in heat pump systems due to their ability to efficiently handle varying levels of load conditions, both for heating and cooling modes of operation. As electric compressors are considered the heart of the heat pump system, being able to determine degradation of compressors prior to failure is of paramount importance for the health of this system. Typical failures for electric scroll compressors range from electrical faults, refrigerant leaks, to mechanical failures and overheating. Specifically, one of the primary failure modes for an electric scroll compressor is mechanical damage due to the high stress effects of refrigerant liquid slugging. These stresses are due to excessively high internal pressures exhibited on the compressor scrolls, which are generated by compressing liquid refrigerant at the suction side of the compressor. This paper provides a new testing methodology that introduces liquid slugging at various degrees of refrigerant quality to degrade a compressor to near the end of useful life. Furthermore, this test aims to determine specific operating conditions and signals that can indicate early compressor degradation. This fault injection configuration consists of a modified heat pump system with the addition of two low pressure heat exchangers added in parallel (with respective electronically controlled expansion valve for each heat exchanger) used to control the refrigerant quality during compressor operations. For a given refrigerant quality, the heat pump system was operated at a fixed compressor performance conditions to sustain liquid slugging for a fixed duration. Afterwards, refrigerant was controlled to be pure vapor at the compressor suction side and the compressor was controlled at several different performance conditions (i.e., fixed compressor suction superheat temperature and compressor pressure ratios, at various compressor speeds), so as to duplicate conditions known to us from the compressor component data sheet for an ideal electric scroll compressor. Through these tests, the results show that the severity of scroll failures depend heavily on the refrigerant quality and the amount of liquid slugging exposure time. Furthermore, symptoms of compressor degradation are detected using the following signals: i) temperature and pressure at the compressor suction side, ii) temperature and pressure at the compressor discharge side, and iii) electric compressor speed and power consumption. To further aid in determining the compressor degradation ground truth, complete compressor teardown was performed to identify sections within the compressor that exhibited significant amounts of wear as compared to a stock compressor.

An Introductory Approach to Time-Series Data Preparation and Analysis

Edward Baumann — 2023-10-26

Machine learning (ML)/Artificial Intelligence (AI) has widespread applications and has revolutionized many industries due to advanced and matured sensor technology, as well as large-scale data collection efforts. One of the key tasks for effective ML/AI operations is the extraction and identification of useful and usable data to identify complex interrelationships and solve problems efficiently. The usefulness of the data is the value and meaning of the data within the desired model, while the usability of the data refers to the ease of use of data in a model. Complex supervised and unsupervised ML models, which used to be the domain of cutting-edge scientists and academics, can now be invoked as a basic function calls in public domain packages within Python, R, MATLAB, and other languages. While these functions require effective data preprocessing to overcome the unpredicted impacts of data quality in the real world (e.g. missing data, environmental noise, synchronizing at different sampling rates, etc.), their ease of use means they are often called with little to no understanding of the underlying math or ways to efficiently work through the data set. The approachability provided by the packages enables users to dive into complex problem sets with little advance preparation. However, in doing so there is a lack of understanding which will inevitably cause problems, skew results, or force the user to take a less efficient path to get to a similar answer. Each package provides relatively simple examples that deal with specific public data sets, yet not many provide the background knowledge and comprehensive methods required for building the inputs for extensive and effective time-series data modeling. Typically, the complex nature of time-series data requires an in-depth understanding of signals analysis and domain subject expertise to use in ML/AI predictive models. This paper will provide the reader an overview of the problems associated with time-series data modelling, propose a common set of preprocessing steps to follow, demonstrate a taxonomy classification for time series data, provide introductory reasoning regarding the underlying process, and discuss the models that would benefit from such a methodology. This is done here with the goal of equipping non-knowledge-domain experts with updated and approachable techniques to find which features to focus on while preprocessing for their time-series data preparation efforts.

Cooling Fan Failure Modes to Enable Development of Automotive ECU Fan Health Monitoring System

Alaeddin Banimilhim — 2023-10-26

Electronic control units (ECUs) are widely used in the automotive industry. Recent efforts to enable enhanced and automated driving requires these ECUs to process and execute computationally expensive algorithms. With these developments, the ECUs now have a higher computing power and thus are at a greater risk of overheating. This may limit the availability of the essential functionalities in the vehicle. Currently, high operating temperatures are mitigated using passive cooling, which allows heat to dissipate without expelling any energy; however, more robust methods are required to enable this new technology. A cooling fan system is one of the desired methods for ECU thermal management, as this type of system draws cooler air from outside and expels the warm air from within. Therefore, the fan health status is critical to ensure ECU availability and reliability for vehicle operation, as when the fans become degraded, they cannot maintain the required airflow to minimize the ECU operating temperature. Traditionally, fan failures are detected by monitoring the fan speed versus the commanded duty cycle, thus it is desired to develop a robust health monitoring method for the fans. Fan failure mode study and fault injection can be used to enable the development of prognostics. Investigating the fan failure modes results in two main categories, which are internal and external. External fan failures include degradation and cracking of the outer casing, while internal failures include motor and ball bearing issues. Fault injection methods were developed based on these failure modes while considering potential operating conditions. For example, the fans were exposed to multiple environmental conditions, such as dust, humidity, and heat. These conditions can potentially trigger both internal and external failures. The data collection was conducted with the fans running in a standalone setup, being controlled by external equipment to ensure that the electronic input values were known. After running tests for 30 days, sufficient data was collected to enable degradation modelling. The data will contribute to the development of a predictive algorithm which will estimate the state of health of the fan based on its performance over time. This paper will discuss the failure modes and the data generated through simulation and fault injection.

Failure Mode Investigation to Enable LiDAR Health Monitoring for Automotive Application

Fred Chang — 2023-10-26

Light Detection and Ranging (LiDAR) sensors are critical components of the perception system and play a significant role in enabling fully autonomous driving. Given that LiDARs have a higher failure rate than other sensors, such as camera and radar, it is crucial to monitor the health of this component to increasing the availability of autonomous driving features. Such a health monitoring system can additionally provide cost-effective maintenance for retail and fleet, improve the service experience of retail customers, and ensure the fidelity of the data produced by the LiDAR for engineering development. Since LiDAR is a relatively new technology, there is currently limited work in the area of LiDAR health monitoring. The failure modes and degradation behavior of these components have not been thoroughly studied in the literature for automotive applications. Therefore, this paper reviews LiDAR external and internal failure modes and their impacts on the perception performance. The external failure modes are categorized into multiple fault classes such as sensor blockage due to a layer of debris on the sensor, mechanical damage to the sensor cover, and mounting issues. The internal faults corresponding to LiDAR subcomponents such as transmitter, receiver or scanning mechanism, are explored for these LiDAR types: mechanical spinning, flash LiDAR, Micro-opto-electromechanical mirror LiDAR, and micromotion technology LiDAR. The failure modes of each subcomponent are also investigated to determine if they can be categorized as slow degradation or sudden failure. It is concluded that mechanical spinning LiDARs are expected to have higher failure rates than solid-state LiDARs. Both internal and external LiDAR failure modes can lead to reduced accuracy and reliability in detecting objects and obstacles, compromising the safety of autonomous driving systems, and increasing the possibility of collisions.

Fusion and Comparison of Prognostic Models for Remaining Useful Life of Aircraft systems

Shuai Fu — 2023-10-26

Changes in the performance of an aircraft system will straightforwardly affect the safe operation of the aircraft, and the technical requirements of Prognostics and Health Management (PHM) are highly relevant. Remaining Useful Life (RUL) prediction, part of the core technologies of PHM, is a cutting-edge innovation being worked on lately and an effective means to advance the change of upkeep support mode and work on the framework's security, unwavering quality, and economic reasonableness. This paper summarizes a detailed preliminary literature review and comparison of different prognostic approaches and the forecasting methods' taxonomy, the methodology's details, and provides its application to aircraft systems. It also provides a brief introduction to the predictive maintenance concept and condition-based maintenance (CBM). This article uses several predictive models to predict RUL and classifies conventional regression algorithms according to the similarity in function and form of the algorithms. More classical algorithms in each category are selected to compare the prediction results, and finally, the combined effects of the RUL prediction are obtained by weighted fusion, accuracy, and compatibility. The performance of the proposed models is assessed based on evaluations of RUL acquired from the hybrid and individual predictive models. This correlation depends on the most current prognostic metrics. The outcomes show that the proposed strategy develops precision, robustness, and adaptability. Hence, the work in this paper shall enrich the advancement of predictive maintenance and modern innovation of prognostic development.

Interpolate and Extrapolate Machine Learning Models using An Unsupervised Method

Peng Liu — 2023-10-26

The 2023 PHM North America Data Challenge is intriguing because it requires one to predict outcomes and use data patterns that training models do not see. Modern machine learning models based on gradient boosting and neural networks are not designed to address such issues in usually circumstances. Our final approach to address the challenge consists of five steps. In our approach, we use an unsupervised method besides machine learning models to address the challenge.

System-based Monitoring of Muscular Fatigue in Lower-Extremity Movement

Samuel Bertelson — 2023-10-26

Physical fatigue accounts for many injuries in the workplace, sports arena, or battlefield. The traditional approaches to monitor fatigue rely on detecting and measuring shifts in the person’s muscular surface electromyography (sEMG) signals. However, assessing neuromuscular fatigue based purely on sEMG signals fails to account for the changing muscle dynamics during long dynamic physical tasks. To combat this dilemma, a system-based methodology has been recently developed and applied to several upper-extremity tasks. In this paper, we validate the efficacy of this novel methodology on the lower extremities during a dynamic activity. Specifically, the system-based monitoring methodology was applied to a cycling endurance task. It was statistically demonstrated that the system-based methodology resulted in a more-sensitive and less noisy metric, in comparison with an EMG-based methodology. The efficacy of the methodology was further illustrated by analyzing the inter-segmental recovering and fatiguing trends, which aligned with each muscle’s expected inter-muscle synergistic relationship.

Using Charge Determination Design of Experiments to Develop A Refrigerant Charge Health Status Model for Heat Pump Systems

Hong Wong — 2023-10-26

Refrigerant based heat pump systems are becoming an integral system in electric vehicle architectures due to their high efficiencies in providing heating and cooling to people and components within the car. An important component in heat pump systems that determines optimal efficiency is the amount of refrigerant. As such, the capability to model refrigerant charge helps quantify the health status of the heat pump system, whereby the lack or over abundance of refrigerant in a heat pump refrigerant system leads to various other component failures, e.g., liquid slugging, compressor overheating, material fatigue in heat exchanger, and degraded/stuck expansion valves. In designing a heat pump system, engineers need to perform a set of design of experiments to determine an optimal refrigerant charge based on a set of performance metrics in the presence of certain noise factors. This optimal refrigerant charge provides conditions where the heat pump system operates efficiently in both heating and cooling, in addition to facilitating operational conditions that will not lead to secondary component degradation or damage. The search for optimal refrigerant charge is classified as refrigerant charge determination, whereby engineers incrementally increase the refrigerant in the heat pump system in operation of heating/cooling and collect data about performance metrics. Some of the key performance metrics used to determine efficiency of a heat pump system include i) compressor inlet superheat temperature, ii) condenser outlet subcool temperature, iii) compressor high side pressure, iv) compressor low side pressure, v) condenser outlet pressure, and vi) condenser quality estimate. Furthermore, this process follows design of experiments concepts and is performed for both heating and cooling modes of operation. In this paper, we leverage refrigerant charge determination as a training data source to develop refrigerant charge models, where several performance metrics are health indicators used as model inputs and the amount of refrigerant added to the heat pump system are ground truth refrigerant charges used as model outputs. In this paper we develop regression models to estimate the total refrigerant charge, which is used to classify different health states of refrigerant based on levels of performance degradation corresponding to specific refrigerant charge thresholds. We trained a robust linear regression model using this charge determination data and found that the worst case estimation error was less than 10% with respect to the refrigerant charge grouth truth.

An Introduction to 2023 PHM Data Challenge: The Elephant in the Room and an Analysis of Competition Results

Yongzhi Qu — 2023-10-27

The trend in diagnostics and prognostics for PHM is shifting toward explainable data-driven models. However, complex engineered systems are typically challenging to develop entirely explainable models for, whether they are grounded in physics or data-driven techniques. Consequently, the development of machine learning models, including hybrid variants capable of both interpolation and extrapolation, holds significant promise for enhancing the practicality of system simulation, analysis, modeling, and control in industry. The primary objective of this data challenge is to encourage contributions that expand the scope of model generalization beyond the training domain. The second aim of this data challenge is to quantify model uncertainty and methods to incorporate it into predictions. For most PHM tasks, clear guidance of the required action is ideal. To issue a definitive guidance to end users, it is useful to quantify uncertainty for the whole model. This data challenge addresses both estimation and uncertainty.

Anomaly Detection and Fault Classification in Multivariate Time Series Using Multimodal Deep Models

Gunwoo Ryu — 2023-10-26

In the realm of gear fault diagnosis, where various analytical methods often require extensive domain expertise, automation remains challenging due to diverse fault diagnosis tasks. To address these limitations, we propose a novel PHM algorithm integrating out-of-distribution detection and representation learning. Initial steps involve feature extraction using envelopes and fast Fourier transform (FFT). Representation Learning employs Transformers and Self-supervised learning for meaningful representations. The latent space values are then utilized for Out-of-Distribution Detection through kNN and classification, achieving a remarkable 99% accuracy. Our approach significantly enhances gear fault diagnosis automation, proving effective across diverse, unencountered problems.

Gearbox Degradation Prediction through Deep CNN and Bayesian Optimization

Kai Shen — 2023-10-26

The original dataset, recorded at a high sampling rate of 20,480 Hz, presents a considerable volume of data. To facilitate efficient processing and analysis, we implement a data reduction strategy, which involves downsampling the signal to a more manageable frequency of 2,048 Hz.

Gear Pitting Fault Diagnosis Using Domain Generalizations and Specialization Techniques

Fan Chu — 2023-10-26

Gear pitting is a common gear fault, which has been an important subject to industry and research community, In the past, the diagnosis of gear pitting faults was all based on fixed operating conditions and the fixed gear health state, which is a in-set detection, However, in real industrial scenarios, gear pitting fault diagnosis is always an open-set detection, in which the working conditions and the gear health state are commonly not known in advance. In order to deal with this open-set detection problem, we proposed a three-stage diagnosis method. In the first stage, we built an in-set health state classification model based on Domain2Vec to solve the domain generalization problem caused by different operating conditions. In the second stage, we modify the classification model to a regression model to predict the out-of-set health state sample in the dataset. In the third stage, we used KNN algorithm to correct the wrong model in the second stage and further improve the accuracy of classification. Our proposed method achieved scores of 463.5 and 472 on the test set and validation set, respectively, and ranked first in the 2023 PHM Conference Data Chanllenge.

Predicting pitting severity in gearboxes under unseen operating conditions and fault severities using convolutional neural networks with power spectral density inputs

Rik Vaerenberg — 2023-10-26

The PHM North America 2023 Data Challenge tasked participants to diagnose the pitting fault severity of a gearbox from a three-channel vibration signal.
This work summarizes the authors' proposed diagnostics solution which consists of a convolutional neural network with an ordinal loss criterion, trained on the power spectral density of the signal.
This method is selected based on a rigorous evaluation using three dedicated validation sets, designed to evaluate the model's ability to generalize to unseen operation conditions and fault severities.
Ultimately, the proposed approach achieved a competition validation score of $282.2$ and a test score of $213.3$.

A Closer Look at Bearing Fault Classification Approaches

Harika Abburi — 2023-10-26

Rolling bearing fault diagnosis has garnered increased attention in recent years owing to its presence in rotating machinery across various industries, and an ever increasing demand for efficient operations. Prompt detection and accurate prediction of bearing failures can help reduce the likelihood of unexpected machine downtime and enhance maintenance schedules, averting lost productivity. Recent technological advances have enabled monitoring the health of these assets at scale using a variety of sensors, and predicting the failures using modern Machine Learning (ML) approaches including deep learning architectures. Vibration data has been collected using accelerated run-to-failure of overloaded bearings, or by introducing known failure in bearings, under a variety of operating conditions such as rotating speed, load on the bearing, type of bearing fault, and data acquisition frequency. However, in the development of bearing failure classification models using vibration data there is a lack of consensus in the metrics used to evaluate the models, data partitions used to evaluate models, and methods used to generate failure labels in run-to-failure experiments. An understanding of the impact of these choices is important to reliably develop models, and deploy them in practical settings. In this work, we demonstrate the significance of these choices on the performance of the models using publicly-available vibration datasets, and suggest model development considerations for real world scenarios. Our experimental findings demonstrate that assigning vibration data from a given bearing across training and evaluation splits leads to over-optimistic performance estimates, PCA-based approach is able to robustly generate labels for failure classification in run-to-failure experiments, and $F$ scores are more insightful to evaluate the models with unbalanced real-world failure data.

A Framework for Rapid Prototyping of PHM Analytics for Complex Systems using a Supervised Data-Driven Approach

Katarina Vuckovic — 2023-10-26

Prognostic and Health Management (PHM) solutions are becoming increasingly popular in industries that rely on large
systems such as aircraft, spacecraft, and power plants. PHM analytic solutions are designed to monitor the health of each
subsystem and component and apply predictive analytic to improve system reliability and safety, reduce the cost and decrease time spent on unscheduled maintenance. However, identifying correlations between different components and
associated monitors in these large systems can be challenging. To address this issue and achieve maximum utilization of available monitoring signals, a methodology is required that can identify correlations between degraded or failed components and the features engineered from the monitors and sensors. This paper introduces a framework that enables rapid prototyping of analytics, allowing users to seamlessly move from designing and discovering features to developing models for a specific event or component of interest. The framework has three main components: feature exploration, data
preparation, and model development. Feature exploration focuses on feature engineering using raw monitor data from all available monitors. Data preparation purges the data, and down-selects relevant features based on correlation defined in the feature exploration part. The data preparation step also creates a training dataset. Model development enables
quick testing and comparison of multiple supervised Machine Learning (ML) models. To demonstrate the framework, this paper presents an example of a remaining useful life model for an aircraft component. While the examples and simulations are aircraft-focused, the principles behind the framework can be applied to other large systems.

A Hypothesis testing approach to Zero-Fault-Shot learning for Damage Component Classification

Eric Bechhoefer — 2023-10-26

Often, in condition monitoring, datasets are asymmetric. That is, for most machines being monitored, there is no labeled fault data, only nominal data (hence, the dataset is asymmetric). Deep Learning and other neural network-based mechanization have difficulty solving this type of problem, as they typically require a full set of labeled data, both nominal and faulted. Zero-Fault Shot learning is a class of machine learning problems with no labeled fault training data. In this class of problems, only nominal data is used for knowledge transfer. In this paper, a mixed hypothesis testing and Bayes classifier it used to provide both inferences to the type of fault and also provide information as to when maintenance should be provided. This is done without any fault data and demonstrates knowledge transfer from a set of nominal components, greatly reducing the cost of implementation and fielding of a system.

A Novel Operations-Based Application of Natural Language Processing to Enhance Aircraft System Troubleshooting

Jamie Asbach — 2023-10-26

Troubleshooting an aircraft system is difficult. With flights often logging hundreds, or even thousands, of codes, the task of isolating the root cause of an issue is a complex undertaking. By leveraging Natural Language Processing techniques such as Word2Vec, artificial intelligence can be used to extract patterns from the context of these faults. Treating the fault codes issued by the on-board system in an aircraft as the “words” which make up a body of text, a model can be trained to understand the patterns of this language in a similar approach to how natural language is processed by computers to discretize the order and structure of human language. By assessing the cosine similarity of vectorized fault sequences used to train the model, faults occurring in similar sequences can be extracted, resulting in improved troubleshooting. The result of this effort is a tool to aid maintainers in isolating faults by quantifying the relations between the different codes and analyzing the patterns in which they occur. The benefits of such a tool include significant reduction in time and cost in aircraft maintenance by avoiding unnecessary exploratory maintenance.

Automotive Electronic Control Unit Ground Line Health Monitoring Method

Alaeddin Milhim — 2023-10-26

Electronic Control Units (ECUs) have been used in the automotive industry for decades to control one or more of the vehicle subsystems. The ECUs communicate primarily using the in-vehicle Controller Area Network (CAN) communication protocol. The recent rapid development of connected, electric, and autonomous vehicles expands the number of ECUs and complexity of the CAN network required to integrate vehicle systems and deliver the desired functionalities. This demands increased reliability of the ECUs to ensure for robust vehicle performance. One of the most common ECU failure modes is the ECU ground fault. A ground fault occurs when the ground path in the ECU circuit is corroded, which is usually developed slowly over time. Such failure usually results in various symptoms including ECU incapable of functioning and further impacts the vehicle functionalities negatively. This type of fault can be difficult to detect prior to vehicle functionality loss. It usually involves routinely testing the resistance of the ground circuit, visually inspecting the connectors and wirings, and checking the voltage drop across the ground circuit. Therefore, it is highly desirable to continuously monitor the ECU ground line health status to predict any degradation and thus prevent vehicle functionality losses.

This paper presents a novel method to monitor the health status of ECU ground line. The method leverages measured CAN voltage data to estimate the ECU ground state of health. The CAN voltage measurements are preprocessed and fed into a real-time data buffer of predefined size. Statistical moments are calculated from the buffered data to generate health indicators, which are then combined to form a fused health indicator. The fused health indicator is used to determine the health stage of ECU ground line. The health stage is classified based on the relationship between ground line degradation level and the ECU communication loss status. The method was developed and validated using actual vehicle data.

Deep Regression Network with Prediction Confidence in Time Series Application for Asset Health Estimation

Hao Huang — 2023-10-26

Many works have been focused in developing detection, monitoring and prediction routines for asset health estimation system. Classic machine learning based models benefit from quality of physics-informed features available from domain knowledge. This, however, can be labor intensive and is limited by quality of features developed through available knowledge. Deep learning based approach, if successful, can alleviate this laborious step. On the other hand, users often need to decide whether or not to trust an algorithmic prediction while the true error in the prediction is unknown. In this work, we propose a deep learning based regression network that output both prediction value and confidence score for asset health estimation in short intermittent transients time series application. In the experimental study, we show that our model has low prediction error given short intermittent transients multivariate time series as input. Furthermore, our model also provides a confidence score for each prediction that is highly negatively correlated with true prediction error. Experiments show that by setting an acceptance threshold on confidence score, our model can reach an averaged improvement of 20% on the prediction quality with 90% coverage.

Evaluating the Performance of ChatGPT in the Automation of Maintenance Recommendations for Prognostics and Health Management

Sarah Lukens — 2023-10-26

Until now, automation of maintenance recommendations for Prognostics and Health Management (PHM) has been a domain-specific technical language processing (TLP) task applied to historical case data. ChatGPT, Bard, GPT-4 and Sydney are a few examples of generative large language models (LLMs) that have received significant media attention for their proficiency in natural language tasks across a variety of domains. Preliminary exploration of ChatGPT as a tool for generating maintenance recommendations has shown promise in its ability to generate and explain engineering concepts and procedures, but the precise scope of its capabilities and limitations remains uncertain. Currently we know of no performance criteria related to formally measuring how well ChatGPT performs as a tool for industrial use cases. In this paper, we propose a methodology for the evaluation of the performance of LLMs such as ChatGPT for the task of automation of maintenance recommendations. Our methodology identifies various performance criteria relevant for PHM such as engineering criteria, risk elements, human factors, cost considerations and corrections. We examine how well ChatGPT performs when tasked with generating recommendations from PHM model alerts and report our findings. We discuss the various strengths and limitations to consider in the adoption of LLM’s as a computational support tool for prescriptive PHM as well as the different risks and business case considerations.

Enhancing Realistic Remaining Useful Life Prediction using Multi-Fidelity Physic-Informed Neural Network Approach

Yoojeong Noh — 2023-10-26

The prediction of remaining useful life (RUL) for a component requires a certain failure threshold to be reached. However, using only monitoring data available up to the current time in prognosis can lead to unrealistic RUL predictions. To address this issue, this study proposes a multi-fidelity approach that uses discrepancy predictions to inform a physical model for RUL estimation. Discrepancy predictions are obtained by training the difference between low- and high-fidelity models using a neural network. The low- and high-fidelity models are constructed using the exponential function and monitoring data from experimental work, respectively. As the exponential function has a monotonically increasing trend, a multi-fidelity model can lead to realistic RUL predictions. This proposed method was tested on several failure cases involving rotating components, such as bearings and unbalanced cooling fans. The results show that the proposed method yields realistic and accurate RUL predictions, despite the lack of available monitoring data.

Fault Detection and Diagnosis in Tennessee Eastman Process with Deep Autoencoder

Zhongying Xiao — 2023-10-26

Data-driven modeling has been considered as an attractive approach for fault detection in chemical processes. Of special interest to industry are methods that represent nonlinear phenomena and detect complex faults. In this paper, a semi-supervised deep learning method - deep autoencoder for fault detection in Tennessee Eastman Process (TEP) is proposed. The TEP process is a simulated benchmark for evaluating process control and monitoring methods. The performance of the proposed method is evaluated and compared to Principal Component Analysis (PCA). The experimental results demonstrate that the proposed optimized five-layers DAE model for fault detection outperforms the standard PCA. Of special importance to real-world applications is its capability for automatic variable selection. In comparison to PCA it demonstrated higher prediction accuracy for most of the generated faults. Deep autoencoder has the potential to become an excellent approach for process monitoring and fault detection in chemical processes.

Identifying Key Factors in Turbofan Engine Health Degradation using Functional Analysis

Declan Mallamo — 2023-10-26

A method is presented for predicting the health of turbofan engines using data and simulations from NASA. The method involves estimating engine health using k-nearest neighbors’ regression and fitting a remaining useful life model that considers engine usage. A matching pursuit algorithm identifies key parameters, while functional principal components provide insight into degradation precursors. Model performance is evaluated using root mean square error and future research and applications are discussed.

Joint feedback control and fault diagnosis disambiguation

Ion Matei — 2023-10-26

This paper proposes a model-based diagnosis approach to detect and isolate intermittent faults in complex systems that operate under feedback control. The feedback control attempts to compensate for model uncertainties and deviations from nominal behavior, but these uncertainties are crucial for accurate fault diagnosis. We focus on faults that are observable only in a particular region of the state space, which is rarely reached in nominal behavior. To address this, we present an approach that considers both control requirements and diagnosis uncertainty in an optimization problem similar to model-predictive control. We compute perturbations on control signals that forces the system to reach states where faults are detectable. We apply our approach to a quadrotor system under motion feedback control, demonstrating the effectiveness of our method. Our approach has the potential to improve the resilience of complex systems by quickly detecting and recovering from disruptive events.

OSPtk: Cost-aware Optimal Sensor Placement Toolkit Enabling Design-for-PHM in Critical Industrial Systems

Liang Tang — 2023-10-26

Performance of a Prognostics and Health Management (PHM) system in a fielded application depends on observability from existing monitoring equipment and sensing, which get determined at the design phase. Although various technologies have been proposed in the literature, there is currently a lack of known generic tools specifically designed for performing design stage sensor placement analyses from a PHM perspective. This leads to PHM observability being an afterthought and resulting PHM designs being sub-optimal. This paper describes a new Optimal Sensor Placement (OSP) framework, its implementation as a toolkit and the experience with applying it to a new product design in the context of a Small Modular Reactor (SMR). The formulation adds multiple important features that are critical to PHM applications. Firstly, it establishes a direct link to PHM performance requirements with intent to reduce operational and maintenance costs. Moreover, it acknowledges and accounts for the costs and risks of errors that PHM system will incur, and simultaneously considers operational requirements on sensing for performance, control and/or regulatory requirements. The toolkit described here implements formulations of a large number of requirements scenarios applicable in a generic industrial product development setting.

Process for Turboshaft Engine Performance Trending

Eric Bechhoefer — 2023-10-26

Turboshaft engines are ubiquitous in aerospace applications where high power and reliability are needed in a low-weight package. Most all helicopters incorporate turboshaft engines. All turboshaft-equipped aircraft have power assurance checks to ensure the engine can achieve the minimum specification for power. However, these checks seldom are automatically collected, nor do they trend the engine health over time to better assess vehicle health. Engines degrade over time, and the ability to assess when maintenance is required is accentual for the safe and efficient operation of the aircraft. This paper covers a process to evaluate a turboshaft engine's state of health using a model-based assessment of the engine’s performance margin over time.

Predictive Analytics for Hydropower Fleet Intelligence

Yigit Yucesan — 2023-10-26

A primary challenge in hydropower industry is the ability to maintain cost-competitiveness, reliability, and security of hydropower assets through evolving power system contexts and aging of the fleet. Maintaining cost-effective and reliable operations under these conditions is expected to require new modernization and maintenance paradigms for changing contexts. Changes in existing practices for O&M will require an understanding of the current state and health of hydropower assets, and the impact of changing paradigms on asset health and reliability. The Hydropower Fleet Intelligence project is developing and evaluating standardized methodologies and analysis tools for data-driven asset reliability and management technologies for hydropower, leading to eventual predictive maintenance planning, repair/replacement decision making, and asset-reliability and cost-optimized operations. A key question is the feasibility of using existing data sets at hydropower facilities to perform assessments of asset reliability. This document uses data from hydropower facilities to assess the potential for using available analytics methods for asset reliability estimates. In addition to reliability assessments, the feasibility of using existing analytics techniques for several other potential applications is discussed. Finally, a case study that a data-driven model is trained to learn nominal operations via vibration data from an asset of a certain plant, and then utilized to identify anomalies on a similar asset from a different plant, highlighting the generic use of proposed Prognostics and Health Management (PHM) approaches.

Servomotor Dataset: Modeling Health in Mechanisms with Typically Intermittent Operation

Arun Subramanian — 2023-10-26

Servomotors are used in a variety of industrial applications where precise movements are of critical importance. Degradation mechanisms in servomotors have been mostly studied and modeled for systems with long duration steady state modes. However, some specialized applications require health estimation from very short duration intermittent operations, which require different analysis techniques. With such applications in mind, a simulated dataset for servomotor health modeling and prediction is described and made available for public use. The application scenario is motivated by a fine motion control rod drive (FMCRD) mechanism used for intermittent, and typically infrequent, fine motion (insertion or withdrawal) adjustment of control rods in some nuclear reactor designs. Though the drives do not run continuously, servomotor and associated linear motion mechanisms do show wear and damage during its operational lifetime. Specifically, in FMCRD such degradations may be caused by internal as well as external damage to the system. While the causes of such damage can be diverse, in simulation we model the impact of cumulative damage as an external opposing load which resists the movement of the motor shaft. Such scenarios represent effects of rod-binding and debris in the fuel channels. The dataset includes measurements such as motor currents and rotor speed which would be part of the instrumentation in a typical deployments of rotating machinery. These observable measurements can be used to predict the health state of the servomotor. Also presented are baseline results on health state estimation, formulated as classification and regression problems, which can be used by the larger PHM community for performance comparisons. This dataset is hosted at the PHM Society Data repository [https://data.phmsociety.org/servomotor_dataset/].

Signal pre-processing techniques for fault signature enhancement in a bearing health monitoring system used in the automotive industry

Ehsan Jafarzadeh — 2023-10-26

Traditional internal combustion engine vehicles have low transmission bearing failure rates in their lifespans. However, the prolonged lifespan of electric and autonomous vehicles can surpass the reliable life of bearing designs, which poses a risk of bearing failure and loss of propulsion. Compared to replacing bearings on a fixed schedule to ensure reliability, a bearing health monitoring system is a more cost-effective solution. Despite extensive research on bearing condition monitoring, implementing well-known methods such as vibration spectrum analysis in vehicles can be challenging due to vibrations from vehicle components and the road. This paper explores and compares the effect of various pre-processing techniques on the spectrum of a faulty bearing with various fault levels. To achieve this objective, faults with the width size of 0.1 mm (mild), 0.5 mm (moderate) and 2 mm (severe) were injected into the inner race of a ball bearing. A bench setup was then used to capture the vibrations of multiple vehicle components including the faulty ball bearing under various speed/ load conditions. Phase domain transform, envelope and Fourier transform were used as the core signal processing steps, and advanced signal processing methods for removing discrete frequencies from other components and enhancing the fault signature were explored. 4 health indicators were then developed from the vibration spectrum of the vibration signals and calculated for the captured data. Next, for each fault level, the area under Receiver operating characteristic (ROC) curve was calculated and used as a metric to compare the performance of our health monitoring system for classification of faulty and healthy bearings. For our best health indicator, the results show that applying minimum entropy deconvolution, and spectral kurtosis-based band pass filtering increases the ROC area from 0.40, 0.99, 1.0 to 0.86, 1.0 and 1.0 for the mild, moderate, and severe inner race faults, respectively. This implies that although applying only phase domain transform, envelope and Fourier transform might be enough for moderate and severe faults, advanced signal processing is needed to enhance the fault signature for early detection of mild faults.

Unsupervised Causal Deep Learning-Based Anomaly Detection in Nuclear Power Plant Applications

Abhinav Saxena — 2023-10-26

Nuclear power generation will be key to meeting carbon free energy transition goals. However, nuclear power must provide agility and flexibility to fluctuating power demands when other sources of carbon-free energy like solar and wind may not be as flexible. This has led to small modular reactor (SMR) development where nuclear power will be generated from a distributed fleet of smaller reactors where units can be brought online or offline as needed. Due to its high operational and maintenance (O&M) costs as it is, a distributed fleet will put additional cost burden if remote monitoring and crew sharing is not enabled. This requires prognostics and health management (PHM) capabilities such as early warning, diagnostics, and prognostics to enable predictive maintenance with high accuracy. Typically, monitoring solutions are developed on component and subsystem levels targeting specific failure modes. However, it is argued that a systemwide monitoring, in addition to specific targeted analytics, would be of key importance. This paper presents a deep-causal unsupervised anomaly detector that has been successfully applied in various aerospace and renewable energy applications. In this paper we share our experience applying this method on a nuclear power plant (NPP) application. Specifically, we share how we dealt with practical challenges of data quality, ground truth labeling, performance evaluation and field validation in an unknown-unknown setting where prior knowledge of failures and failure modes were not available to begin with.

A Physics-informed, Transfer Learning Approach to Structural Health Monitoring

Trent Furlong — 2023-10-26

One of the main challenges for structural health monitoring (SHM) is a lack of failure data to make accurate health predictions. Obtaining desirable failure data is generally very expensive, given the required testing needed to measure all types of system failures, which may be unfeasible in many health monitoring applications. Machine learning has helped to improve health monitoring performance but is still limited by the availability, relevance, and quality of the training data. This data dependence impedes data-driven models from generalizing to unseen data, which is problematic for datasets lacking failure data. Physics-driven models, like finite-element models, are powerful tools for predicting structural responses when the governing physics are not clearly defined. These models can generate simulated fault data to address the data limitation without having to physically damage a structure, but are computationally expensive and susceptible to modeling errors that can prevent the data from being statistically comparable to experimental data.

A new trend has been to develop physics-guided machine learning models (PGML), a hybridization of the two aforementioned models that have been shown to improve generalization of, and even outperform, pure data-driven models while using less training data. These PGML models can take many forms, but generally embed some form of physics into a data-driven model as physically relevant constraints. Our research plan is to utilize PGML to improve neural network capabilities to predict structural damage. The proposed PGML model will follow a neural network architecture found in related literature consisting of feature extraction, physics-informed, and label prediction layers. The physics-informed layer will consist of an aggregate of sub-networks trained from simplified structure models which have known governing equations and can be used to generate simulated training data. The full PGML model will use transfer learning to bridge the connections between the untrained layers to the physics-informed layer using experimental data from more complex structures. We will verify our model using publically available SHM datasets used in a variety of past literature experiments.

A Neural Network Framework for Predicting Durability and Damage Tolerance of Polymer Composites under Combined Hygrothermal-mechanical Loading

Partha Pratim Das — 2023-10-26

Fiber-reinforced polymer (FRP) composites are used in crucial structures which are susceptible to a combination of mechanical (static/dynamic) and hygrothermal (moisture absorption and temperature) loads. This research presents a novel artificial neural network (ANN) framework that employs the dielectric permittivity response of FRP composites under combined mechanical-hygrothermal loading to predict the extent of moisture absorption, fatigue life, and remaining useful life. The proposed framework is based on the phenomenological and data-driven study of the effects of static and dynamic mechanical loads along with moisture absorption in the dielectric characteristics of these composites.

Generic Hybrid Models for Prognostics of Complex Systems

Kristupas Bajarunas — 2023-10-26

Hybrid models combining physical knowledge and machine learning show promise for obtaining accurate and robust prognostic models. However, despite the increased interest in hybrid models in recent years, the proposed solutions tend to be domain-specific. As a result, there is no compelling strategy of what, where, and how physics-derived knowledge can be integrated into deep learning models depending on the available representation of physical knowledge and the quality of data for the development of prognostic models for complex systems. This Ph.D. project aims to develop a general strategy for hybridizing prognostic models by exploring multiple methods to incorporate physical knowledge at various stages of the learning algorithm. The project will prioritize expert knowledge as the primary source of information, while domain-specific knowledge will serve as an additional feature when applicable.

Integrating Advanced Prognostic Methods for Accurate Remaining Useful Life Prediction in Industrial Systems

Hyung Jun Park — 2023-10-26

Accurate remaining useful life (RUL) prediction of industrial system is critical to ensure smooth operation and its safety. Various prognostic methods have been developed but there still exist critical challenges for field applications. One challenge is the unhealth degradation exhibiting the change of state from those of normal degradation. Another is the prediction in the face of severe noise with limited data (i.e., early prediction) using empirical models. Final challenge is the prediction under varying operating conditions, which occurs in practice in various industrial applications. To overcome these challenges, this research proposes advanced prognostics methods with different recipes featured by high adaptability, physical constraints, and monotonic health indicator (HI). The developed methods are validated with specific case studies involved with the challenges.

Information Fusion and Data Augmentation for Risk-based Maintenance Optimization of Hydrogen Gas Pipelines

Kaushik Kethamukkala — 2023-10-26

Demand for energy is increasing every year and hydrogen is being seen as a good alternative to conventional natural gas. The current focus is on the use of existing pipeline infrastructure for the transport of hydrogen gas, and it is necessary for us to ensure the safe and efficient operation of the pipeline infrastructure given the risks posed by hydrogen. Pipeline integrity management is critical for hydrogen transport and there are knowledge gaps for the impact of hydrogen on the pipeline integrity and operational considerations, thus hindering the pipeline operators from adopting hydrogen into their networks. To realize the concept of transporting hydrogen through existing pipeline systems, it is necessary to have reliable risk assessment and maintenance optimization frameworks in place. A Bayesian network methodology is proposed to fuse information from multiple sources obtained by multimodality diagnosis of pipe materials and Bayesian updating will be incorporated to reduce the uncertainty arising from different random variables. Risk assessment of the pipeline systems will be carried out based on the posterior distributions of the random variables. Given the predicted risk level, we then propose a risk-based maintenance optimization framework to minimize the maintenance costs while ensuring the safe operation of the pipeline systems.

Mission-Specific Prognosis of Li-ion Batteries using Hybrid Physics-Informed Neural Networks

kajetanfricke — 2023-10-26

New transportation modalities such as electric powered vertical
takeoff and landing aircraft and logistic applications
like delivery of packages with drones require highly reliable
and powerful electric batteries for operation. A challenging
but very important task hereby is the precise forecasting of
the degradation of battery state-of-health (SOH) and stateof-
charge (SOC). While high-fidelity electrochemistry based
models can provide precise predictions of the SOC, they can
be computationally expensive. On the other hand, purely datadriven
approaches require a large amount of training data in
order to learn the input to output relation. In this research an
improved hybrid physics-informed machine learning model
is introduced, that conserves the electrochemistry based laws
and is implemented with data-driven layers to compensate for
unknown portions of internal voltage drop during discharge.
Preliminary results indicate that the model can predict discharge
for a large variety of loads, accurately predicts capacity
degradation over age and can be enhanced through extracting
information from cell temperature data as surrogate for aging.

Physics-Informed Deep Learning-Based Approach for Probabilistic Modeling of Degradation

Hamidreza Habibollahi Najaf Abadi — 2023-10-26

Deep learning (DL) models have gained significant popularity for the prognostics of systems experiencing degradation. However, there are two major concerns with such models. Firstly, they require a substantial amount of training data due to their large number of parameters. Secondly, they disregard the underlying physics and solely fit the available data, leading to potentially weak generalization capabilities when faced with unseen out-of-distribution data in the field. This study aims to tackle these challenges by incorporating the underlying physics of degradation into DL models. The objective is to develop a novel DL-based approach in conjunction with Bayesian filtering, enabling physics-informed probabilistic life prediction for systems subject to environmentally induced degradation. The proposed framework consists of two main components: physics discovery and degradation prediction. The former involves identifying the dominant stress agents and formulating the underlying physics of degradation. The latter predicts the degradation of the system by incorporating the discovered physics into a DL model. It is expected the results indicate that by combining data-driven DL with physics-based insights, more robust and reliable life predictions can be achieved, addressing the limitations of DL approaches. This framework holds promise for enhancing decision-making processes related to maintenance strategies in various industries.

Vehicle State Monitoring and Fault Detection System for Unmanned Ground Vehicles (UGV) using Markov Models

Kalpit Vadnerkar — 2023-10-26

This research presents a novel fault detection and diagnostics system for unmanned ground vehicles (UGVs) by combining Markov models representing the vehicle's navigation, kinematic behavior, and vehicle dynamics systems. Existing studies do not specifically address the challenges related to UGVs and their complex subsystems or the incorporation of weather and environmental condition data. The proposed system leverages environmental and weather condition data to monitor the UGV's state and detect anomalies in its behavior. By predicting the probability of faults such as collisions, sensor damage, and other malfunctions, the system aims to enhance the safety, reliability, and performance of UGVs. The research will demonstrate the effectiveness of the proposed methodology through case studies and performance evaluation, highlighting its potential application in various real-world scenarios. This work contributes to the ongoing research in prognostics and health management, particularly for autonomous systems, by providing a new approach to fault detection and diagnostics in UGVs.