PHM Society European Conference

Application of Machine Learning Methods to Predict the Quality of Electric Circuit Boards of a Production Line

2022-06-19T19:54:07+00:00

For the data challenge of the 2022 European PHM conference, data from a production line of electric circuit boards is provided to assess the quality of the produced components. The solution presented in this paper was elaborated to fulfill the data challenge objectives of predicting defects found in an automatic inspection at the end of the production line, predicting the result of a following human inspection and predicting the result of the repair of the defect components. Machine learning methods are used to accomplish the different prediction tasks. In order to build a reliable machine learning model, the steps of data preparation, feature engineering and model selection are performed. Finally, different models are chosen and implemented for the different sub-tasks. The prediction of defects in the automatic inspection is modeled with a multi-layer perceptron neural network, the prediction of human inspection is modeled using a random forest algorithm. For the prediction of human repair, a decision tree is implemented.

Prediction of Production Line Status for Printed Circuit Boards

2022-06-19T18:48:28+00:00

This paper focuses on the problem of predicting production line status for Printed Circuit Boards (PCBs). The problem contains three prediction tasks regarding PCB producing process. Firstly, data exploration is carried out and it reveals several data challenges, including data imbalance, data noise, small sample size, and component difference. To predict production line status for components of PCBs using records of inspection on pins, we proposed two possible feature extraction methods to compress the pin-level data into component-level. A statistical feature extraction method, which retrieves descriptive statistics such as mean, standard deviation, maximum, and minimum of pins on the same component, is applied to Task 1, while a PinNumber-based feature extraction method, which keep original values for 2-pin components, is applied to Task3. In addition, a neural-net model with feeding imbalance control is established for Task 1. and a random forests model is applied for both Task 2 and Task 3. Moreover, a threshold moving technique is proposed to optimize the threshold selection. Finally, the result shows that our models achieved f1-scores of 0.44, 0.54, and 0.71 on the test set for the three tasks, respectively.

A Novel Methodology for Health Assessment in Printed Circuit Boards

2022-06-19T19:56:10+00:00

The demand for Printed circuit boards (PCBs) has increased due to the rapid change in technology in recent years. Consequently, PCBs health assessment and fault detection play an important role in improving productivity. This study proposed a novel method which focused on feature engineering for health assessment in PCBs. The performance of the proposed method has been validated using data obtained from PHM Europe 2022 data challenge. In this data challenge, PCBs health assessment needs to be performed with data from the Solder Paste Inspection (SPI) and the Automated Optical Inspection (AOI) machine. The challenge has three tasks: 1) Predict the labels of the AOI machine using the SPI data. 2) Using both the SPI and AOI machine data, predict the operator's verification that the AOI machine correctly detected a defect. 3) With the SPI and AOI data, predict the classification of the defective PCBs as either repairable or unrepairable. The component level features are extracted from the original SPI and AOI data which contain the pin level features to solve these tasks. Two machine learning-based classification models, i.e., Light Gradient Boosting Machine (LightGBM) and eXtreme Gradient Boosting (XGBoost), have been used for classification purposes. Training data given by the organizer was divided into 70% training and 30% validation. Based on the validation data, the highest F1-score was observed with LightGBM in Tasks 1 and 2, whereas, in Task 3, the highest F1-score was observed with the XGBoost model. Hence, the LightGBM model has been used in Tasks 1 and 2, and the XGBoost model was developed for Task 3.

A Hierarchical XGBoost Early Detection Method for Quality and Productivity Improvement of Electronics Manufacturing Systems

2022-06-19T17:02:48+00:00

This paper presents XGBoost classifier-based methods to solve three tasks proposed by the European Prognostics and Health Management Society (PHME) 2022 conference. These tasks are based on real data from a Surface Mount Technologies line. Each of these tasks aims to improve the efficiency of the Printed Circuit Board (PCB) manufacturing process, facilitate the operator’s work and minimize the cases of manual intervention. Due to the structured nature of the problems proposed for each task, an XGBoost method based on encoding and feature engineering is proposed. The proposed methods utilise the fusion of test values and system characteristics extracted from two different testing equipment of the Surface Mount Technologies lines. This work also explores the problems of generalising prediction at the system level using information from the subsystem data. For this particular industrial case: the challenges with the changes in the number of subsystems. For Industry 4.0, the need for interpretability is very important. This is why the results of the models are analysed using Shapley values. With the proposed method, our team took the first place, capable of successfully detecting at an early stage the defective components for tasks 2 and 3.

A Hierarchical XGBoost Early Detection Method for Quality and Productivity Improvement of Electronics Manufacturing Systems

2022-06-29T21:14:25+00:00

Application of Machine Learning Methods to Predict the Quality of Electric Circuit Boards of a Production Line

2022-06-29T21:45:23+00:00

A Novel Methodology for Health Assessment in Printed Circuit Boards

2022-06-29T21:56:06+00:00

Prediction of Production Line Status for Printed Circuit Boards

2022-06-29T21:35:29+00:00

PHME 2021 Management Team, Publisher Information and Table of Contents

2022-06-30T07:14:31+00:00

PHME 2022 Management Team, Publisher Information and Table of Contents

2022-06-29T08:54:32+00:00

Experiences of a Digital Twin Based Predictive Maintenance Solution for Belt Conveyor Systems

2022-06-23T13:23:09+00:00

Availability of belt conveyor systems is essential in production and logistic lines to safeguard production and delivery targets to customers. In this paper, experiences from commissioning, validation, and operation of an interactive predictive maintenance solution are reported. The solution and its development is formerly presented in Al-Kahwati et.al. (Al-Kahwati, Saari, Birk, & Atta, 2021), where the principles to derive a digital twin of a typical belt conveyor system comprising component-level degradation models,estimation schemes for the remaining useful life and the degradation rate, and vision-based hazardous object detection.

Furthermore, the validation approach of modifying the belt conveyor and thus exploiting the idler misalignment load (IML) for the degradation predictions for individual components (including long-lasting ones) together with the actionable insights for the decision support is presented and assessed. Moreover, the approach to testing and validation of the object detection and its performance is assessed and presented in the same manner. An overall system assessment is then given and concludes the paper together with lessons learned.

As pilot site for the study a belt conveyor system at LKAB Narvik in northern Norway is used.

A Case-study Led Investigation of Explainable AI (XAI) to Support Deployment of Prognostics in the industry

2022-06-23T07:56:22+00:00

Civil nuclear generation plant must maximise it’s operational uptime in order to maintain it’s viability. With aging plant and heavily regulated operating constraints, monitoring is commonplace, but identifying health indicators to pre-empt disruptive faults is challenging owing to the volumes of data involved. Machine learning (ML) models are increasingly deployed in prognostics and health management (PHM) systems in various industrial applications, however, many of these are black box models that provide good performance but little or no insight into how predictions are reached. In nuclear generation, there is significant regulatory oversight and therefore a necessity to explain decisions based on outputs from predictive models. These explanations can then enable stakeholders to trust these outputs, satisfy regulatory bodies and subsequently make more effective operational decisions. How ML model outputs convey explanations to stakeholders is important, so these explanations must be in human (and technical domain related) understandable terms. Consequently, stakeholders can rapidly interpret, then trust predictions better, and will be able to act on them more effectively. The main contributions of this paper are: 1. introduce XAI into the PHM of industrial assets and provide a novel set of algorithms that translate the explanations produced by SHAP to text-based human-interpretable explanations; and 2. consider the context of these explanations as intended for application to prognostics of critical assets in industrial applications. The use of XAI will not only help in understanding how these ML models work, but also describe the most important features contributing to predicted degradation of the nuclear generation asset.

Long Horizon Anomaly Prediction in Multivariate Time Series with Causal Autoencoders

2022-06-26T09:28:24+00:00

Predictive maintenance is essential for complex industrial systems to foresee anomalies before major system faults or ultimate breakdown. However, the existing efforts on Industry 4.0 predictive monitoring are directed at semi-supervised anomaly detection with limited robustness for large systems, which are often accompanied by uncleaned and unlabeled data. We address the challenge of predicting anomalies through data-driven end-to-end deep learning models using early warning symptoms on multivariate time series sensor data. We introduce AnoP, a long multi-timestep anomaly prediction system based on unsupervised attention-based causal residual networks, to raise alerts for anomaly prevention. The experimental evaluation on large data sets from detector health monitoring of the Hadron Calorimeter of the CMS Experiment at LHC CERN demonstrates the promising efficacy of the proposed approach. AnoP predicted around 60% of the anomalies up to seven days ahead, and the majority of the missed anomalies are abnormalities with unpredictable noisy-like behavior. Moreover, it has discovered previously unknown anomalies in the calorimeter’s sensors.

Experimental Validation of Multi-fidelity Models for Prognostics of Electromechanical Actuators

2022-06-23T09:21:48+00:00

The growing adoption of electrical energy as a secondary form of onboard power leads to an increase of electromechanical actuators (EMAs) use in aerospace applications. Therefore, innovative prognostic and diagnostic methodologies are becoming a fundamental tool to early identify faults propagation, prevent performance degradation, and ensure an acceptable level of safety and reliability of the system. Furthermore, prognostics entails further advantages, including a better ability to plan the maintenance of the various equipment, manage the warehouse and maintenance personnel, and a reduction in system management costs.

Frequently, such approaches require the development of typologies of numerical models capable of simulating the performance of the EMA with different levels of fidelity: monitoring models, suitably simplified to combine speed and accuracy with reduced computational costs, and high fidelity models (and high computational intensity), to generate databases, develop predictive algorithms and train machine learning surrogates. Because of this, the authors developed a high-fidelity multi-domain numerical model (HF) capable of accounting for a variety of physical phenomena and gradual failures in the EMA, as well as a low-fidelity counterpart (LF). This simplified model is derived by the HF and intended for monitoring applications. While maintaining a low computing cost, LF is fault sensitive and can simulate the system position, speed, and equivalent phase currents.

These models have been validated using a dedicated EMA test bench, designed and implemented by authors. The HF model can simulate the operation of the actuator in nominal conditions as well as in the presence of incipient mechanical faults, such as a variation in friction and an increase of backlash in the reduction gearbox.

Comparing the preliminary results highlights satisfactory consistency between the experimental test bench and the two numerical models proposed by the authors.

An Analysis of Vibrations and Currents for Broken Rotor Bar Detection in Three-phase Induction Motors

2022-06-22T07:25:50+00:00

Selecting the physical property capable of representing the health state of a machine is an important step in designing fault detection systems. In addition, variation of the loading condition is a challenge in deploying an industrial predictive maintenance solution. The robustness of the physical properties to variations in loading conditions is, therefore, an important consideration. In this paper, we focus specifically on squirrel cage induction motors and analyze the capabilities of three-phase current and five vibration signals acquired from different locations of the motor for the detection of Broken Rotor Bar generated in different loads. In particular, we examine the mentioned signals in relation to the performance of classifiers trained with them. Regarding the classifiers, we employ deep conventional classifiers and also propose a hybrid classifier that utilizes contrastive loss in order to mitigate the effect of different variations. The analysis shows that vibration signals are more robust under varying load conditions. Furthermore, the proposed hybrid classifier outperforms conventional classifiers and is able to achieve an accuracy of 90.96% when using current signals and 97.69% when using vibration signals.

Online Flow Estimation for Condition Monitoring of Pumps in Aircraft Hydraulics

2022-06-23T08:41:31+00:00

Hydraulic systems in conventional civil aviation are currently monitored in a very rudimentary way. Normally, measured values are compared with a fixed threshold. If these measured values are outside the predefined limits, the entire hydraulic system is usually shut down. To overcome this deficit, a study regarding a novel prognostic health management method for aircraft hydraulic pumps, which allows a statement about the pump condition, is presented in this paper. The method is based on measuring differential pressure and temperature at a suitable resistance. In the first part of the study, the overall concept for monitoring the motor pump unit is analyzed. This is followed by a discussion of possible measurement methods and suitable resistors to determine the condition of the pump. In the second part of the study, the implementation for online monitoring of the pump is discussed. After a suitable approximation is found, the quality of the proposed method is evaluated with real hydraulic power generation and consumers.

Hybrid Fault Prognostics for Nuclear Applications: Addressing Rotating Plant Model Uncertainty

2022-06-21T14:12:15+00:00

Nuclear plant operators are required to understand the uncertainties associated with the deployment of prognostics tools
in order to justify their inclusion in operational decision making processes and satisfy regulatory requirements. Operational
uncertainty can cause underlying prognostics models to underperform on assets that are subject to evolving impacts
of age, manufacturing tolerances, operating conditions, and operating environment effects, of which may be captured
through a condition monitoring (CM) system that itself may be degraded. Sources of uncertainty in the data acquisition
pipeline can impact the health of CM data used to estimate the remaining useful life (RUL) of assets. These uncertainties
can disguise or misrepresent developing faults, where (for example) the fault identification is not achieved until it has
progressed to an unmanageable state. This leaves little flexibility for the operator’s maintenance decisions and generally
undermines model confidence.

One method to quantify and account for operational uncertainty is calibrated hybrid models, employing physics, knowledge
or data driven methods to improve model accuracy and robustness. Hybrid models allow known physical relations to
offset full reliance on potentially untrustworthy data, whilst reducing the need for an abundance of representative historical
data to reliably identify the monitored asset’s underlying behavioural trends. Calibration of the model then ensures
the model is updated and representative of the real monitored asset by accounting for differences between the physics or
knowledge model and CM data.

In this paper, an open-source bearing knowledge informed machine learning (ML) model and CM datasets are utilized
in an illustrative bearing prognostic application. The uncertainty incurred by the decisions made at key stages in the
development of the model’s data acquisition and processing pipeline are assessed and demonstrated by the resultant impact
on RUL prediction performance. It was shown that design decisions could result in multiple valid pipeline designs
which generated different predicted RUL trajectories, increasing the uncertainty in the model output.

Data-driven Prognostics based on Evolving Fuzzy Degradation Models for Power Semiconductor Devices

2022-06-23T08:08:31+00:00

The increasing application of power converter systems based on semiconductor devices such as Insulated-Gate Bipolar Transistors (IGBTs) has motivated the investigation of strategies for their prognostics and health management. However, physicsbased degradation modelling for semiconductors is usually complex and depends on uncertain parameters, which motivates the use of data-driven approaches. This paper addresses the problem of data-driven prognostics of IGBTs based on evolving fuzzy models learned from degradation data streams. The model depends on two classes of degradation features: one group of features that are very sensitive to the degradation stages is used as a premise variable of the fuzzy model, and another group that provides good trendability and monotonicity is used for the auto-regressive consequent of the fuzzy model for degradation prediction. This strategy allows obtaining interpretable degradation models, which are improved when more degradation data is obtained from the Unit Under Test (UUT) in real time. Furthermore, the fuzzy-based Remaining Useful Life (RUL) prediction is equipped with an uncertainty quantification mechanism to better aid decisionmakers. The proposed approach is then used for the RUL prediction considering an accelerated aging IGBT dataset from the NASA Ames Research Center.

State of Health and Lifetime Prediction of Lithium-ion Batteries Using Self-learning Incremental Models

2022-06-21T15:02:40+00:00

Lithium-ion batteries are key energy storage elements in the context of environmental-aware energy systems representing a crucial technology to achieve the goal of zero carbon emission. Therefore, its conditions must be monitored to guarantee the safe and reliable operation of the systems that use these components. Furthermore, lithium-ion batteries’ prognostics and health management policies must cope with the nonlinear and time-varying nature of the complex electrochemical dynamics of battery degradation. This paper proposes an incremental-learning-based algorithm to estimate the State of Health (SoH) and the Remaining Useful Life (RUL) of lithium-ion batteries based on measurement data streams. For this purpose, a two-layer framework is proposed based on incremental modeling of the SoH. In the first layer, a set of representative features are extracted from voltage and current data of partial charging and discharging cycles; these features are then used to train the proposed model in a recursive procedure to estimate the battery’s SoH. The second layer uses the capacity data for incremental learning of an Autoregressive (AR) model for the SoH, which will be used to propagate the battery’s degradation through time to make the RUL prediction. The proposed method was applied to two datasets for experimental evaluation, one from CALCE and another from NASA. The proposed framework was able to estimate the SoH of 8 different lithium-ion cells with an average percentage error below 1.5% for all scenarios, while the lifetime model predicted the cell’s RUL with a maximum average error of 25%.

Wrong Injection Detection in a Small Diesel Engine, a Machine Learning Approach

2022-06-20T15:12:49+00:00

In the last ten years, Machine Learning (ML) and Artificial Intelligence (AI) have overwhelmed every engineering research branch finding a broad variety of applications; anomaly detection and anomaly classification are two of the topics that have benefited mostly by data-driven methods’ insights. On the other side, in the small diesel engine domain, the current trend is to lean on traditional anomaly detection/classification procedures and do not foster the use of AI. The goal of this work is to detect anomalies in the in-cylinders injectors of a small diesel engine as soon as a wrong quantity of fuel is inputted into one or more cylinders by means of ML approaches. Part of the analysis aim to understand which measurements are the most relevant for the detection and to compare different techniques to select the most suitable one. Furthermore, a condition-based methodology for maintenance is proposed. After a brief review of the state-of-the-art, the case study scenario is presented grouping sensors accordingly to their degree of accessibility; then, the implemented techniques are explained, and results are discussed.

Novel Metrics to Evaluate Probabilistic Remaining Useful Life Prognostics with Applications to Turbofan Engines

2022-06-21T13:59:38+00:00

Well-established metrics such as the Root Mean Square Error or the Mean Absolute Error are not suitable to evaluate estimated distributions of the Remaining Useful Life (i.e., probabilistic prognostics). We therefore propose novel metrics to evaluate the quality of probabilistic Remaining Useful Life prognostics. We estimate the distribution of the Remaining Useful Life of turbofan engines using a Convolutional Neural Network with Monte Carlo dropout. The accuracy and sharpness of the obtained probabilistic prognostics are evaluated using the Continuous Ranked Probability Score (CRPS) and weighted CRPS. The reliability of the obtained probabilistic prognostics is evaluated using the α-Coverage and the Reliability Score. The results show that the estimated distributions of the Remaining Useful Life of turbofan engines are accurate, reliable and sharp when using a Convolutional Neural Network with Monte Carlo dropout. In general, the proposed metrics are suitable to evaluate the accuracy, sharpness and reliability of probabilistic Remaining Useful Life prognostics.

Filtering Misleading Repair Log Labels to Improve Predictive Maintenance Models

2022-06-23T14:06:00+00:00

One of the main challenges for predictive maintenance in real applications is the quality of the data, especially the labels. In this paper, we propose a methodology to filter out the misleading labels that harm the performance of Machine Learning models. Ideally, predictive maintenance would be based on the information of when a fault has occurred in a machine and what specific type of fault it was. Then, we could train machine learning models to identify the symptoms of such fault before it leads to a breakdown. However, in many industrial applications, this information is not available. Instead, we approximate it using a log of component replacements, usually coming from the sales or maintenance departments. The repair history provides reliable labels for fault prediction models only if the replaced component was indeed faulty, with symptoms captured by collected data, and it was going to lead to a breakdown.

However, very often, at least for complex equipment, this assumption does not hold. Models trained using unreliable labels will then, necessarily, fail. We demonstrate that filtering misleading labels leads to improved results. Our central claim is that the same fault, happening several times, should have similar symptoms in the data; thus, we can train a model to predict them. On the contrary, replacements of the same component that do not exhibit similar symptoms will be confusing and harm the ML models. Therefore, we aim to filter the maintenance operations, keeping only those that can be used to predict each other. Suppose we can train a successful model using the data before a component replacement to predict another component replacement. In that case, those maintenance operations must be motivated by the same, or a very similar, type of fault.

We test this approach on a real scenario using data from a fleet of sterilizers deployed in hospitals. The data includes sensor readings from the machines describing their operations and the service logs indicating the replacement of components when the manufacturing company performs the service. Since sterilizers are complex machines consisting of many components and systems interacting with each other, there is the possibility of faults happening simultaneously.

Physics-informed Lightweight Temporal Convolution Networks for Fault Prognostics Associated to Bearing Stiffness Degradation

2022-06-26T09:12:40+00:00

This paper proposes hybrid methods using physics-informed (PI) lightweight Temporal Convolution Neural Network (PITCN) for bearings’ remaining useful life (RUL) prediction under stiffness degradation. It includes three PI hybrid models: a) PI Feature model (PIFM) — constructing physics-informed health indicator (PIHI) to augment the feature space, b) PI Layer model (PILM)—encoding the physics governing equations in a hidden layer, and c) PI Layer Based Loss model (PILLM)—designing PI conflict loss, taking into account the difference before and after integration of the physics input-output relations involved module to the loss function. We simulated 200 different bearing stiffness degradations, using their discrete monitored vibration signals to verify the effectiveness of the proposed method. We also investigate their inference process through feature heat map analysis to interpret how the models melt physics knowledge to assist in capturing the degradation trend. The physics knowledge considered in this paper is the dynamic relationship between vibration amplitude and stiffness in a damped forced vibration model. The results show that all three PITCN models effectively capture degradation-related trend information and perform better than the vanilla lightweight TCN. Furthermore, the visualization of the feature channels highlights the important role of physics information in model training. Channels containing physics information demonstrate higher correlation with results as they significantly dominate the heat map compared to other channels.

Design and validation of scalable PHM solutions for aerospace onboard systems

2022-06-22T07:39:59+00:00

In recent years, Prognostic & Health Management (PHM) has become a topic of strong interest in the aerospace domain. Health assessment and remaining useful life estimation for on-board systems provide several advantages, mainly related to the increased analysis capabilities and the reduction of maintenance interventions (and, consequently, of operating costs). For this reason, it is of interest for the aerospace industry to identify and define efficient strategies both for the introduction of native PHM capabilities in new generation on-board systems and for the retrofit of existing ones. This paper proposes a strategy for the scalable deployment of PHM techniques for on-board systems, with particular focus on edge computing capabilities. Different reference scenarios (ranging from cloud-based processing to local-only processing) are presented, and an edge-focused PHM architecture is discussed in detail, with the relative challenges addressed. The design and validation of proposed edge-based solution is described, with specific reference to its support for an existing data analytics framework. The solution is then assessed against a reference aerospace use case involving a representative aircraft braking system, focusing on computational aspects to highlight the compatibility of the proposed deployment strategy with efficient on-board computations.

Sensor Fault/Failure Correction and Missing Sensor Replacement for Enhanced Real-time Gas Turbine Diagnostics

2022-06-21T08:54:40+00:00

Gas turbine sensors are prone to bias and drift. They may also become unavailable due to maintenance activities or failure through time. It is, therefore, important to correct faulty signal or replace missing sensors with estimated values for improved diagnostic solutions. Coping with a small number of sensors is the most difficult to achieve since this often leads to underdetermined and indistinguishable diagnostic problems in multiple fault scenarios. On the other hand, installing additional sensors has been a controversial issue from cost and weight perspectives. Gas path locations with difficult conditions to install sensors is also among other sensor installation related challenges. This paper proposes a sensor fault/failure correction and missing sensor replacement method. Auto-regressive integrated moving average models are employed to correct measurements from faulty and failed sensors. To replace additional sensors needed for further diagnostic accuracy improvements, neural network models are devised. The performance of the developed approach is demonstrated by applying to a three-shaft turbofan engine. Test results verify that the method proposed can well-recover measurements from faulty/failed sensors, no matter with small or major failures. It can also compensate key missing temperature and pressure measurements on the gas path based on the data from other available sensors.

Helicopter Bolt Loosening Monitoring using Vibrations and Machine Learning

2022-06-21T14:22:55+00:00

The existing helicopter Health and Usage Management Systems (HUMS) collect and process flight operational parameters and sensors data such as vibrations to provide health monitoring of the helicopter dynamic assemblies and engines. So far, structure-related mechanical faults, such as looseness in bolted structures, have not been addressed by vibration-based condition monitoring in existing HUMS systems. Bolt loosening was identified as a potential risk to flight safety demanding periodical visual monitoring, and increased maintenance and repair expenses. Its automatic identification in helicopters by using vibration measurements is challenging due to the limited number of known events and the presence of high-energy vibrations originating in rotating parts, which shadow the low-level signals generated by the bolt loosening.

New developed bolt loosening monitoring approach was tested on HUMS vibrations data recorded from the IAF AH-64 Apache helicopters fleet. ML-based unsupervised anomaly detection was utilized in order to address the limited number of faulty cases. The predictive power of health features was significantly improved by applying the Harmonic filtering differentiating between the high-energy vibrations generated by rotating parts compared with the low-energy structural vibrations. Different unsupervised anomaly detection techniques were examined on the dataset. The experimental results demonstrate that the developed approach enable successful bolt loosening monitoring in helicopters and can potentially be used in other health monitoring applications.

On the Integration of Fundamental Knowledge about Degradation Processes into Data-Driven Diagnostics and Prognostics Using Theory-Guided Data Science

2022-06-23T10:03:16+00:00

In Prognostics and Health Management, there are three main approaches for implementing diagnostic and prognostic applications. These approaches are data-driven methods, physical model-based methods, and combinations of them, in the form of hybrid methods. Each of them has specific advantages but also limitations for their purposeful implementation. In the case of data-driven methods, one of the main limitations is the availability of sufficient training data that adequately cover the relevant state space. For model-based methods, on the other hand, it is often the case that the degradation process of the considered technical system is of significant complexity. In such a scenario physics-based modeling requires great effort or is not possible at all. Combinations of data-driven and model-based approaches in form of hybrid approaches offer the possibility to partially mitigate the shortcomings of the other two approaches, however, require a sufficiently detailed data-driven and physics-based model.

This paper addresses the transitional field between data-driven and hybrid approaches. Despite the issues of formulating a physics-based model that provides a representation of the degradation process, basic knowledge of the considered system and of the laws governing its degradation process is usually available. Integration of such knowledge into a machine learning process is part of a research field that is either called theory-guided data science, (physics) informed machine learning, physics-based learning or physics guided machine learning. First, the state of research in Prognostics and Health Management on methods of this field is presented and existing research gaps are outlined. Then, a concept is introduced for incorporating fundamental knowledge, such as monotonicity constraints, into data-driven diagnostic and prognostic applications using approaches from theory-guided data science. A special aspect of this concept is its cross-application usability through the consideration of knowledge that repeatedly occurs in diagnostics and prognostics. This is, for example, knowledge about physically justified boundaries whose compliance makes a prediction of the data-driven model plausible in the first place.

Toward Runtime Assurance of Complex Systems with AI Components

2022-06-23T14:12:30+00:00

AI components (e.g., Deep Neural Networks) are increasingly used in safety-relevant aerospace applications. Rigorous Verification and Validation (V&V) is mandatory for such components, yet V&V techniques for DNNs are still in their infancy and can often only provide relatively weak guarantees. In this paper, we will present a runtime-monitoring architecture, which combines the advanced statistical analysis framework SYSAI (System Analysis using Statistical AI) with temporal and probabilistic runtime monitoring carried out by R2U2 (Realizable, Responsive, and Unobtrusive Unit). We will present initial results of our tool set and architecture on a case study, a DNN-based autonomous centerline tracking system (ACT).

Machine Learning Methods for Health-Index Prediction in Coating Chambers

2022-06-21T13:45:32+00:00

Coating chambers create thin layers that improve the mechanical and optical surface properties in jewelry production using physical vapor deposition. In such a process, evaporated material condensates on the walls of such chambers and, over time, causes mechanical defects and unstable processes. As a result, manufacturers perform extensive maintenance procedures to reduce production loss. Current rule-based maintenance strategies neglect the impact of specific recipes and the actual condition of the vacuum chamber. Our overall goal is to predict the future condition of the coating chamber to allow cost and quality optimized maintenance of the equipment. This paper describes the derivation of a novel health indicator that serves as a step toward condition-based maintenance for coating chambers. We indirectly use gas emissions of the chamber’s contamination to evaluate the machine’s condition. Our approach relies on process data and does not require additional hardware installation. Further, we evaluated multiple machine learning algorithms for a condition-based forecast of the health indicator that also reflects production planning. Our results show that models based on decision trees are the most effective and outperform all three benchmarks, improving at least 0.22 in the mean average error. Our work paves the way for cost and quality optimized maintenance of coating applications.

Approximate Bayesian Computation as a New Tool for Partial Discharge Analysis of Partial Discharge Data

2022-06-21T08:18:52+00:00

Partial Discharges are short breakdowns inside electrical equipment. As they indicate weaknesses of the insulation strength, they are seen as important precursors to a failure of the system. Therefore measurement and analysis of the patterns of instances in time and strength of the discharge are an important tool to analyze the insulation status of electric equipment, that has been addressed already using different methods in the past. In this work we explore how a physics-based stochastic process can be combined with Approximate Bayesian Computation (ABC) as a new way to analyze them. ABC is a method to infer probability distributions of model parameters in cases, where the likelihood is not tractable, but simulations can be done easily. As such it is of interest for complex phenomena or measurement systems, as often found in prognostics applications. Especially the ABC-SMC method was found to be useful here. Real Partial Discharge measurement data was used not only for parameter estimation, but also to do model comparison in order to compare different physical models proposed in the literature.

Unsupervised Prognostics based on Deep Virtual Health Index Prediction

2022-06-23T13:55:44+00:00

Prediction of the Remaining Useful Life (RUL) for industrial systems has been facilitated by the acquisition of large amounts of real-time data and the use of deep learning methods. However, the vast majority of these methods rely on the availability of extensive RUL-labeled data, which is not the case for most of real industrial applications. The goal of this paper is to show how unsupervised learning can provide alternative ways to address this issue. The proposed method is essentially made of two steps. First, a Virtual Health Index (VHI) is extracted in an unsupervised manner from the raw sensor data using a Deep Convolutional Neural Network (CNN) autoencoder. Secondly, an Long-Short Term Memory (LSTM) Encoder-Decoder predicts the future values of the VHI, until an End-of-Life (EOL) pattern is recognized (using a sliding window DTW algorithm). The suggested method is tested on the C-MAPSS dataset and offers promising results with a great potential to be applicable on real-life use cases.

Autoencoder Based Anomaly Detection and Explained Fault Localization in Industrial Cooling Systems

2022-06-23T09:39:12+00:00

Anomaly detection in large industrial cooling systems is very challenging due to the high data dimensionality, inconsistent sensor recordings, and lack of labels. The state of the art for automated anomaly detection in these systems typically relies on expert knowledge and thresholds. However, data is viewed isolated and complex, multivariate relationships are neglected. In this work, we present an autoencoder based end-to-end workflow for anomaly detection suitable for multivariate time series data in large industrial cooling systems, including explained fault localization and root cause analysis based on expert knowledge. We identify system failures using a threshold on the total reconstruction error (autoencoder reconstruction error including all sensor signals). For fault localization, we compute the individual reconstruction error (autoencoder reconstruction error for each sensor signal) allowing us to identify the signals that contribute most to the total reconstruction error. Expert knowledge is provided via look-up table enabling root-cause analysis and assignment to the affected subsystem. We demonstrated our findings in a cooling system unit including 34 sensors over a 8-months’ time period using 4-fold cross validation approaches and automatically created labels based on thresholds provided by domain experts. Using 4-fold cross validation, we reached a F1-score of 0.56, whereas the autoencoder results showed a higher consistency score (CS of 0.92) compared to the automatically created labels (CS of 0.62) – indicating that the
anomaly is recognized in a very stable manner. The automatically created labels, however, detected anomaly earlier. The main anomaly was found by the autoencoder and automatically created labels, and was also recorded in the log files. Further, the explained fault localization highlighted the most affected component for the main anomaly in a very consistent manner.

Joint Autoencoder-Classifier Model for Malfunction Identification and Classification on Marine Diesel Engine Diagnostics Data

2022-06-23T07:48:30+00:00

There has been an increasing demand on marine transportation and traveling, since the voyage of the ships are more economical and efficient than air or land-based alternatives. The propulsion of a ship is provided by a main engine system which includes the shaft, the propellers, and other auxiliary equipment. Marine diesel engine is a complex structure that the faults within these machines can cause malfunction of the whole system, which in turn inhibits the ship’s mission. It is crucial to monitor the engine and other auxiliary systems during the operation and infer their condition from their diagnostic data. In this study, we analyze monitoring data of a crude oil tanker for different ship loads and conditions. Our primary analysis includes main engine fault detection and classification for which we propose an end-to-end joint autoencoder-classifier model that contains a convolutional autoencoder, and a long-short term memory regressor connected to the latent space. Genetic algorithms optimized models gave us 93.61% accuracy for fault classification task. Further investigation on feature’s contributions to the model, we increased the accuracy up to 96%. One concern about marine transportation is the pollution of the air with greenhouse effect gases. In this study, we have developed NOx and SOx emission estimators for different faults and working conditions. Leveraging ship load, working conditions and engine faults in the models helped us to achieve 50% better estimation performance. Although there are other studies regarding gases emissions in the literature, this is the first study that took engine faults into account. We believe that the joint autoencoder-classifier model will be useful for other time series estimation task on other domains, especially where the operating condition plays a role in the process.

Physics Informed Neural Network for Health Monitoring of an Air Preheater

2022-06-23T08:53:17+00:00

Air Preheater (APH) is a regenerative heat exchanger employed in thermal power plants to save fuel by improving their thermal efficiency. Monitoring the health of APH vis-a-vis its fouling is critical because fouling often results in forced outages of the power plant, incurring huge revenue losses. APH fouling is a complex thermo-chemical phenomenon governed by flue gas composition, operating temperatures, fuel type and ambient conditions. Absence of sensors within the APH make it difficult to estimate the level of fouling and its progression even for an experienced operator. Attempts to estimate APH fouling in real-time via modeling are scarce. Here we present a physics-informed neural network (PINN) that tracks the health of an APH by real-time estimation of fouling conditions within the APH as a function of real-time sensor measurements. To account for multi-fluid operation in a multi-sector design of APH, the domain is decomposed into several sub-domains. PINN is applied to each sub-domain and the overall solution is ensured by applying continuity conditions at the sub-domain interfaces. The model predicts the interior temperatures and fouling zones within the APH using external sensor measurements such as air temperature and gas composition. The model predictions are consistent with physics and yet computationally efficient in run-time. The model does not need sensor data but can be improved further by accommodating available sensor data. The real-time predictions by the model improve operator’s visibility in fouling. The predictions can be used further for estimating the remaining useful cycle life of the APH, thereby avoiding forced outages. The model can easily be integrated with the digital twin of an APH for its predictive maintenance.

A Health Index Framework for Condition Monitoring and Health Prediction

2022-06-21T15:09:37+00:00

In the field of Maintenance, Repair and Overhaul (MRO), stakeholders such as operators or service providers have to keep track of the health status of fleets of complex systems. The ability to estimate the future health status of these systems and their components becomes more pivotal when seeking to efficiently operate and maintain these systems. Today, these stakeholders have access to a lot of different data sources regarding fleet, operation schedule, ambient condition, system and component information. Many different prognostic methods from different disciplines are available and will further improve henceforward. In many cases these data sources and methods function as isolated methods in their own field. This fragmentation makes a holistic prognosis very challenging in many cases. Therefore, stakeholders need information integrating methods and tools to gain an exhaustive insight into the health status development of the complex assets they are operating or maintaining, in order to make well-founded decisions regarding operation or maintenance planning. In this paper, a Python-based health index framework is presented. It enables users to integrate operation schedules of different detail levels with enriching data sources such as ambient condition data. Furthermore, it provides methods to design complex asset systems which are linked via their construction, function or degradation mechanisms/ health indices via transfer relations. It allows to monitor the asset’s condition based on operation data and to simulate different operation scenarios regarding the health index development.

Tool Compatibility Index: Indicator Enables Improved Tool Selection for Well Construction

2022-06-23T09:13:53+00:00

In the area of well construction, the tool reliability and the field environment are two contributing factors that influence drilling job efficiency and success. Either using high specification tools in low-risk environmental or applying tools of low reliability in harsh environments is inadvisable. Thus, how to select a suitable tool fitting the environment of an approaching drilling job is of great significance for tool planning. However, today, the tool selection decision is not optimized because it is often based on partial data availability and understanding.

This paper presents an indicator called tool compatibility index, which can support improved tool selection decision making. This index takes part reliability, part criticality, and field environment into consideration, and gives a score indicating the compatibility of the tool to a specific environment. Moreover, the tool compatibility index is computed based on a weighted average method, which is computation simple and can be easily deployed. This work is part of a long-term project aiming to construct a risk-based decision advisor for drilling and measurement tools.

An End-to-End Pipeline for Uncertainty Quantification and Remaining Useful Life Estimation: An Application on Aircraft Engines

2022-06-21T13:28:25+00:00

Estimating the remaining useful life (RUL) of an asset lies at the heart of prognostics and health management (PHM) of many operations-critical industries such as aviation. Modern methods of RUL estimation adopt techniques from deep learning (DL). However, most of these contemporary techniques deliver only single-point estimates for the RUL without reporting on the confidence of the prediction. This practice usually provides overly confident predictions that can have severe consequences in operational disruptions or even safety. To address this issue, we propose a technique for uncertainty quantification (UQ) based on Bayesian deep learning (BDL). The hyperparameters of the framework are tuned using a novel bi-objective Bayesian optimization method with objectives the predictive performance and predictive uncertainty. The method also integrates the data pre-processing steps into the hyperparameter optimization (HPO) stage, models the RUL as a Weibull distribution, and returns the survival curves of the monitored assets to allow informed decision-making. We validate this method on the widely used C-MAPSS dataset against a single-objective HPO baseline that aggregates the two objectives through the harmonic mean (HM). We demonstrate the existence of trade-offs between the predictive performance and the predictive uncertainty and observe that the bi-objective HPO returns a larger number of hyperparameter configurations compared to the single-objective baseline. Furthermore, we see that with the proposed approach, it is possible to configure models for RUL estimation that exhibit better or comparable performance to the single-objective baseline when validated on the test sets.

Fault Detection in a Wind Turbine Hydraulic Pitch System Using Deep Autoencoder Extracted Features

2022-06-22T07:16:49+00:00

A wind turbine is equipped with lots of sensors whose measurements are recorded by the supervisory control and data acquisition (SCADA) system and stored every 10 minutes. The pitch subsystem of a wind turbine is of critical importance as it presents the highest failure rate. Thus, selecting the most essential features from the SCADA system is performed in order to detect faults efficiently. In this study, a feature space of 49 features is available, referring to the condition of a hydraulic pitch system. The dimensionality of this feature space (original input space) is reduced using a Deep Autoencoder in order to extract latent information. The architecture of the Autoencoder is investigated regarding its efficiency on fault detection task. This way, effect of new extracted features on the performance of the classifier is presented. A Support Vector Machine (SVM) classifier is trained using a set of healthy (fault free) and faulty data, representing different kind of pitch system failures. The data are acquired from a wind farm of five 2.3MW fixed-speed wind turbines. The performance metric used to evaluate their effect on data is F1-score. Results show that SVM using new extracted feature by Autoencoder outperforms SVM classifier using the original feature set, underlining the power of Autoencoders to unveil latent information.

iVRIDA: intelligent Vehicle Running Instability Detection Algorithm for High-speed Rail Vehicles using Temporal Convolution Network – A Pilot Study

2022-06-23T09:00:16+00:00

Intelligent fault identification of rail vehicles from onboard measurements is of utmost importance to reduce the operating and maintenance cost of high-speed vehicles. Early identification of vehicle faults responsible for an unsafe situation, such as the instable running of highspeed vehicles, is very important to ensure the safety of operating rail vehicles. However, this task is challenging because of the nonlinear dynamics associated with multiple subsystems of the rail vehicle. The task becomes more challenging with only accelerations recorded in the carbody where, nevertheless, sensor maintenance is significantly lower compared to axlebox accelerometers. This paper proposes a Temporal Convolution Network (TCN)-based intelligent fault detection algorithm to detect rail vehicle faults. In this investigation, the classifiers are trained and tested with the results of numerical simulations of a high-speed vehicle (200 km/h). The TCN based fault classification algorithm identifies the rail vehicle faults with 98.7% accuracy. The proposed method contributes towards digitalization of rail vehicle maintenance through condition-based and predictive maintenance.

Remaining-Useful-Life prognostics for opportunistic grouping of maintenance of landing gear brakes for a fleet of aircraft

2022-06-21T09:03:39+00:00

Several studies have proposed Remaining-Useful-Life (RUL) prognostics for aircraft components in the last years. However, few studies focus on integrating these RUL prognostics into maintenance planning frameworks. This paper proposes an optimization model for opportunistic maintenance scheduling of aircraft components that integrates RUL prognostics and that groups the maintenance of these components to reduce costs. We illustrate our approach for the maintenance of a fleet of aircraft, each equipped with multiple landing gear brakes. RUL prognostics for the landing gear brakes are obtained using a Bayesian regression model. Based on these RUL prognostics, we group the replacement of brakes using an integer linear program. As a result, we obtain a cost-optimal RUL-driven opportunistic-maintenance schedule for the brakes of a fleet of aircraft. Compared with traditional maintenance strategies, our approach leads to a reduction of up to 20% of the total maintenance costs.

Novel Graph-Based Features for Bearing Fault Diagnosis: Two Aspects of Time Series Structure

2022-06-21T08:28:55+00:00

The feature-based methods for bearing fault diagnosis in prognostics and health management have been achieved satisfactory performances because of their robustness to noise and reduced dimension by pre-defined features. However, widely employed time- and frequency-domain features are insufficient to recognize the global pattern that indicates the structure of a time-series instance. In this paper, we propose two novel graph-based features which reflect the connection strength and degree of time series, respectively. First, we construct a graph of which an edge is defined as the Euclidean distance between each pair of time steps to measure the strengths of connections between the nodes. The other graph is constructed by the visibility algorithm, which converts a time series into a complex network to reflect the degrees of connections. Then, we calculate the Frobenius norms of the adjacency matrices of both graphs and use them as features for bearing fault diagnosis. To verify the proposed features, we performed several experiments with both synthetic and real datasets. From the synthetic datasets, it is observed that the changes in amplitudes and frequencies are detected by the features for the connection strength and degree, respectively. In addition, the proposed features also well-separate the distributions of each bearing state, including normal and several fault types, and show significant performance improvement as applied to the fault diagnosis task.

Certainty Groups: A Practical Approach to Distinguish Confidence Levels in Neural Networks

2022-06-22T07:21:15+00:00

Machine Learning (ML), in particular classification with deep neural nets, can be applied to a variety of industrial tasks. It can augment established methods for controlling manufacturing processes such as statistical process control (SPC) to detect non-obvious patterns in high-dimensional input data. However, due to the widespread issue of model miscalibration in neural networks, there is a need for estimating the predictive uncertainty of these models. Many established approaches for uncertainty estimation output scores that are difficult to put into actionable insight. We therefore introduce the concept of certainty groups which distinguish the predictions of a neural network into the normal group and the certainty group. The certainty group contains only predictions with a very high accuracy that can be set up to 100%. We present an approach to compute these certainty groups and demonstrate our approach on two datasets from a PHM setting.

Processing of Condition Monitoring Annotations with BERT and Technical Language Substitution: A Case Study

2022-06-23T13:33:20+00:00

Annotations in condition monitoring systems contain information regarding asset history and fault characteristics in the form of unstructured text that could, if unlocked, be used for intelligent fault diagnosis. However, processing these annotations with pre-trained natural language models such as BERT is problematic due to out-of-vocabulary (OOV) technical terms, resulting in inaccurate language embeddings. Here we investigate the effect of OOV technical terms on BERT and SentenceBERT embeddings by substituting technical terms with natural language descriptions. The embeddings were computed for each annotation in a pre-processed corpus, with and without substitution. The K-Means clustering score was calculated on sentence embeddings, and a Long Short-Term Memory (LSTM) network was trained on word embeddings with the objective to recreate the output from a keywordbased annotation classifier. The K-Means score for SentenceBERT annotation embeddings improved by 40% at seven clusters by technical language substitution, and the labelling capacity of the BERT-LSTM model was improved from 88.3 to 94.2%. These results indicate that the substitution of OOV technical terms can improve the representation accuracy of the embeddings of the pre-trained BERT and SentenceBERT models, and that pre-trained language models can be used to process technical language.

A Design Methodology for Robust Model-Based Fault Diagnosis Schemes and its Application to an Aircraft Hydraulic Power Package

2022-06-23T08:14:17+00:00

In a system’s design phase, where knowledge about the actual behavior of the system is shallow, the design of an efficient and robust system diagnostics is a complex task. In order to simplify this process, this paper presents a modelbased methodology for the design of fault diagnosis schemes. The methodology analyzes the structure of available behavioral models of the system and proposes minimal sets of sensors to fulfill diagnostic requirements. In order to choose an optimal set of sensors, the method evaluates the sets in terms of costs and diagnostic robustness. The proposed fault detection, isolation and identification schemes rely on the robust evaluation of model-based residuals using Monte-Carlo methods and highest density regions to account for measurement and parameter uncertainty. To show the design capabilities, the presented method is applied to an aircraft hydraulic power package and the resulting schemes are tested on a real test rig.

Prognosis of Wear Progression in Electrical Brakes for Aeronautical Applications

2022-06-23T10:09:28+00:00

The evolution towards “more electric” aircrafts has seen a decisive push in the last decade, due to the growing environmental concerns and the development of new market segments (flying taxis). Such push interested both the propulsion components and the aircraft systems, with the latter seeing a progressive trend in replacing the traditional solutions based on hydraulic power with electrical or electromechanical devices. Although more attention is usually devised towards the flight control actuation, an interesting and fast-developing application field for electro-mechanical systems is that of the aeronautical brakes. Electro-mechanical brakes, or E-Brakes hereby onwards, would present several advantages over their hydraulic counterparts, mainly related to the avoidance of leakage issues and the simplification of the system architecture. The more difficult heat dissipation, associated with the thermal issues that usually constitute one of the most significant sizing constraints for electromechanical actuators, limits so far, their application (or proposal of application) to light-weight vehicles. Within this context, the development of PHM solutions would align with the need for an on-line monitoring of a relatively unproven component. This paper deals with the preliminary stages of the development of such PHM system for an E-Brake to be employed on a future executive class aircraft, where the brake is actuated through four electro-mechanical actuators. Since literature on fault diagnosis and prognosis for electrical motors is fairly extensive, we focused this preliminary analysis on the development of PHM techniques suitable to monitor and prognose the evolution of the brake pads wear instead. The paper opens detailing the system architecture and continues presenting the high-fidelity dynamic model used to build synthetic data-sets representative of the possible operating conditions faced by the E-Brake within realistic operative scenarios. Such data are then used to foster a preliminary feature selection process, where physics-based indexes are compared and evaluated. Simulated degradation histories are then used to test the application of data-driven fault detection algorithm and the possible application of particle-filtering routines for prognosis.

Domain Knowledge Informed Unsupervised Fault Detection for Rolling Element Bearings

2022-06-23T09:27:38+00:00

Early and accurate detection of rolling element bearing faults in rotating machinery is important for minimizing production downtime and reducing unnecessary preventative maintenance. Several fault detection methods based on signal processing and machine learning methods have been proposed. Particularly, supervised, data-driven approaches have proved to be very effective for fault detection and diagnostics of rolling element bearings. However, supervised methods rely heavily on the availability of failure data with volume, variety and veracity, which is mostly unavailable in industry. As an alternative data-driven strategy, unsupervised methods are trained on healthy data only and do not require any failure data.

In contrast to supervised and un-supervised data-driven models, physics-based and phenomenological models are based on domain knowledge and not on historical data. Although these models are useful for studying the way in which damage is expected to manifest in a measured signal, they are difficult to calibrate and often lack the fidelity required to model reality. In this paper, an unsupervised data-driven anomaly detection method that exploits informative domain knowledge is proposed. Hereby, the versatility of unsupervised data-driven methods are combined with domain knowledge.

In this approach, supplementary training data is generated by augmenting healthy data towards its possible future faulty state based on the characteristic bearing fault frequencies. Both healthy and augmented squared envelope spectrum data is used to train an autoencoder model that includes regularisation designed to constrain the latent features at the autoencoder bottleneck. Regularisation in the autoencoder loss enforces that the expected deviation of the healthy latent representation towards the augmented latent representation at dam aged conditions, is constrained to be maximally different for different fault modes. Consequently, the likelihood of a new test sample being healthy can be evaluated based on the projection of the sample onto an expected failure direction in the latent representation.

A phenomenological and experimental dataset is used to demonstrate that the addition of augmented training data and a specialized autoencoder loss function can create a separable latent representation that can be used to generate interpretable health indicators.

Estimation of Wind Turbine Performance Degradation with Deep Neural Networks

2022-06-22T07:02:00+00:00

In this paper, we estimate the age-related performance degradation of a wind turbine working under Norwegian environment, based on a deep neural network model. Ten years of high-resolution operational data from a 2 MW wind turbine were used for the analysis. Operational data of the turbine, between cut-in and rated wind velocities, were considered, which were pre-processed to eliminate outliers and noises. Based on the SHapley Additive exPlanations of a preliminary performance model, a benchmark performance model for the turbine was developed with deep neural networks. An efficiency index is proposed to gauge the agerelated performance degradation of the turbine, which compares measured performances of the turbine over the years with corresponding bench marked performance. On an average, the efficiency index of the turbine is found to decline by 0.64 percent annually, which is comparable with the degradation patterns reported under similar studies from the UK and the US.

Weighted-QMIX-based Optimization for Maintenance Decision-making of Multi-component Systems

2022-06-21T13:52:36+00:00

It is well-known that maintenance decision optimization for multi-component systems faces the curse of dimensionality. Specifically, the number of decision variables needed to be optimized grows exponentially in the number of components causing computational expensive for optimization algorithms. To address this issue, we customize a multi-agent deep reinforcement learning algorithm, namely Weighted QMIX, in the case where system states can be fully observed to obtain cost-effective policies. A case study is conducted on a 13- component system to examine the effectiveness of the customized algorithm. The obtained results confirmed its performance.

Data Driven Seal Wear Classifications using Acoustic Emissions and Artificial Neural Networks

2022-06-22T06:53:44+00:00

The work presented in this paper is built on a series of experiments aiming to develop a data-driven and automated method for seal diagnostics using Acoustic Emission (AE) features. Seals in machineries operate in harsh conditions, and seal wear in hydraulic cylinders results in fluid leakage, and instability of the piston rod movement. Therefore, regular inspection of seals is required using automated approaches to improve productivity and to reduce unscheduled maintenance. In this study, we implemented a data-driven diagnostics approach which utilizes AE measurements along with light weight Artificial Neural Networks (ANN) as a classifier to investigate the performance and resources (hardware & software) required for implementing a real-time soft sensor unit for monitoring seal wear condition. We used a feedforward multilayer perceptron ANN (Scaled Conjugate Gradient- SCG algorithm) that is trained with the back propagation algorithm, which is a popular network architecture for a multitude of applications (automotive, oil and gas, electronics). We benchmark the developed method against previous work conducted based on Support Vector Machine (SVM), and we compare ANN performance in classifying the running condition of seals in hydraulic cylinders by applying it to both raw (full frequency spectrum) and down sampled frequency measurements. The experiments were performed at varying pressure conditions on a hydraulic test rig that can simulate fluid leakage conditions like that of hydraulic cylinders. The test cases were generated with seals of three different conditions (unworn, semi-worn, worn). From the AE spectrum, the frequency bands were identified with peak power and by heterodyning the signal. This technique results in 10X down sampling without losing the information of interest. Further, the signal was divided into smaller “snapshots” to facilitate rapid diagnosis. In these tests, the diagnosis was made on short-time windows, as low as 0.3 seconds in length. A general set of 16 time and frequency domain features were designed. Then a training set was developed using relevant set of features (4, 5, and 16 features). The data was used to train the ANN (70% training – 30% test & validation) and SVM (60 % training - 40% test and validation). Classification of down sampled measurements, both ANN and SVM were able to accurately classify the status irrespective of the pressure conditions, with an accuracy of ~99% with execution time less than seconds. Therefore, the proposed approach can be applied as part of an automated seal wear classification technique based on AE and ANN/SVM and can be used for real-time monitoring of seal wear in hydraulic cylinders.

Severity Estimation of Faulty Bearings Based on Strain Signals From Physical Models and FBG Measurements

2022-06-23T08:01:59+00:00

Condition based maintenance (CBM) is the preferred approach in rotating machinery and aim to replace the commonly used approach of maintenance based on service time. To achieve an effective CBM, different types of sensors should be placed in the system for condition monitoring to detect the location of the fault and its severity. In this research, a Fiber Bragg Grating (FBG) has been used for condition monitoring on spalls in deep grove ball bearings. The motivation for using these sensors is the ability to get a high-noise signal (SNR) ratio. The usage of FBG sensors is relatively new for health monitoring systems of rotating machinery. Therefore, there is not enough understanding of the strain signature measured by the FBG. To examine the phenomena in the strain signals, a physics-based model of the strain signature has been developed. In this model, two complementary models were integrated, a finite element (FE) model and a dynamic model . The strain model describes the interaction between the rolling elements (REs) and the bearing housing and simulates the strain behavior measured on the bearing housing. The simulation results are validated with strain signals measured by the FBG sensor at different stages of an endurance test. The model allows simulation of a wide range of spall lengths and describes the behavior of the strain signals for different levels of misalignment. The insights from the model enabled the development of an automatic algorithm that assess the severity of the defect and to track spall length during bearing operation, based on strain signals.

A Comparative Study of Health Monitoring Sensors based on Prognostic Performance

2022-06-23T09:49:31+00:00

In the safety critical systems such as industrial plants or aircraft, failure occurs inevitably during the operation, and it is important to prevent this while maintaining high availability. Therefore, a lot of efforts are being directed toward developing advanced prognostics algorithms and sensing techniques as an enabler for predictive maintenance. The key for reliable and accurate prediction not only relies on the prognostics algorithms but also based on the collection of sensor data. However, there is not much in-dept studies toward evaluating the varying sensing techniques based on the prediction performance and inspection scheduling. It would be more reasonable for practitioner to select different cost of sensors based on the sensors’ contribution on reducing the cost on unnecessary inspection or measurement while maintaining its prognosis performance. Thus, the authors try to thoroughly evaluate the cost-effectiveness of the different sensor with respect to sensor resistance to noise. The simulation is conducted to analyze the prediction performance with varying measurement interval and different level of noise during degradation. Then real run-to-fail (RTF) dataset acquired from two different sensors are analyzed to design optimal measurement system for predictive maintenance.

Forecasting Piston Rod Seal Failure Based on Acoustic Emission Features in ARIMA Model

2022-06-22T06:46:56+00:00

Fluid leakage due to piston rod seal failure in hydraulic cylinders results in unscheduled maintenance, machine downtime and loss of productivity. Therefore, it is vital to understand the piston rod seal failure at initial stages. In literature, very few attempts have been made to implement forecasting techniques for piston rod seal failure in hydraulic cylinders using acoustic emission (AE) features. Therefore, in this study, we aim to forecast piston rod seal failure using AE features in the auto regressive integrated moving average (ARIMA) model. AE features like root mean square (RMS) and mean absolute percentage error (MAPE) were collected from run-to-failure (RTF) tests that were conducted on a hydraulic test rig. The hydraulic test rig replicates the piston rod movement and fluid leakage conditions similar to what is normally observed in hydraulic cylinders. To assess reliability of our study, two RTF tests were conducted at 15 mm/s and 25 mm/s rod speed each. The process of seal wear from unworn to worn state in the hydraulic test rig was accelerated by creating longitudinal scratches on the piston rod. An ARIMA model was developed based on the RMS features that were calculated from four RTF tests. The ARIMA model can forecast the RMS values ahead in time as long as the original series does not experience any large shifts in variance or deviates heavily from the normal increasing trend. The ARIMA model provided good accuracy in forecasting the seal failure in at least two of four RTF tests that were conducted. The ARIMA model that was fitted with 15 pre-samples was used to forecast 10 out of sequence samples, and it showed a maximum moving absolute percentage error (MAPE value) of 28.99 % and a minimum of 4.950 %. The forecasting technique based on ARIMA model and AE features proposed in this study lays a strong basis to be used in industries to schedule the seal change in hydraulic cylinders.

Improved Time-Frequency Representation for Non-stationary Vibrations of Slow Rotating Machinery

2022-06-26T08:56:03+00:00

The short-time Fourier transform (STFT) is a staple analysis tool for vibration signal processing due to it being a robust, non-parametric, and computationally efficient technique to analyze non-stationary signals. However, despite these beneficial properties, the STFT suffers from high variance, high sidelobes, and a low resolution. This paper investigates an alternative non-parametric method, namely the sliding-window iterative adaptive approach, to use for time-frequency representations of non-stationary vibrations. This method reduces the sidelobe levels and allows for high resolution estimates. The performance of the method is evaluated on both simulated and experimental vibration data of slow rotating machinery such as a multi-megawatt wind turbine gearbox. The results indicate significant benefits as compared to the STFT with regard to accuracy, readability, and versatility.

Towards Data Reliability Based on Triple Redundancy and Online Outlier Detection

2022-06-26T09:20:12+00:00

Today, air quality monitoring is a global concern. The World Health Organization (WHO) defined standards for each pollutant and each member state is committed to monitoring them continuously and reliably to protect the population. This responsibility is delegated to air quality monitoring associations. To achieve the objectives of reliable, accurate, and continuous measurements, these associations rely on conventional measuring stations with demanding specifications to serve as scientific references and decision supports for the authorities. However, because of heavy investments and required qualified staff, there are few stations and the coverage is coarse for territories of several thousand km2. To circumvent this difficulty, measurement network architectures using Low-Cost Sensors (LCS) have been deployed. Low cost and requiring less qualification, This alternative technology to conventional measuring stations makes it possible to target local pollution that could not otherwise be detected. Although it is more accurate on the spatial dimension, this technology has several drawbacks, notably in terms of measurement repeatability and hardware quality. It is also subject to measurement drifts over time. To overcome these drawbacks, a resilient and reliable architecture based on LCS and triple redundancy has been proposed. The basic principle is based on the implementation of three smart sensors (SmS) using LCS measuring the same parameters on the same perimeter. These SmS communicate with an Aggregator that aggregates the data sent by SmS. The aggregator includes also detection and voting tasks allowing to compare, cross the data, detect faults of LCS online, and provide data that are ready for processing. In this paper, a pre-processing algorithm in four steps is presented. It identifies hardware faults from one or more LCS and reports outliers for verification by an expert. It is configurable and can identify failure behaviors (LCS or air quality). Finally, the proposed algorithm excludes the outliers data from faulty LCS and presents only reliable ones.

Expert Knowledge Induced Logic Tensor Networks: A Bearing Fault Diagnosis Case Study

2022-06-22T07:08:43+00:00

In the recent past deep learning approaches have achieved some remarkable results in the area of fault diagnostics and anomaly detection. Nevertheless, these algorithms rely on large amounts of data, which is often not available, and produce outputs, which are hard to interpret. These deficiencies make real life applications difficult. Before the broad success of deep learning machine faults were often classified using domain expert knowledge based on experience and physical models. In comparison, these approaches only require small amounts of data and produce highly interpretable results. On the downside, however, they struggle to predict unexpected patterns hidden in data. Merging these two concepts promises to increase accuracy, robustness and interpretability of models. In this paper we present a hybrid approach to combine expert knowledge with deep learning and evaluate it on rolling element bearing fault detection. First, we create a knowledge base for fault classification derived from the expected physical attributes of different faults in the envelope spectrum of vibration signals. This knowledge is used to derive a similarity function for comparing input signals to expected faulty signals. Afterwards, the similarity measure is incorporated into different neural networks using a Logic Tensor Network (LTN). This enables logical reasoning in the loss function, in which we aim to mimic the decision process of an expert analyzing the input data. Further, we extend LTNs by weight schedules for axiom groups. We show that our approach outperforms the baseline models on two bearing fault data sets with different attributes and directly gives a better understanding of whether or not fault signals are influenced by other effects or behave as expected.

Domain Adaptation in Predicting Turbocharger Failures Using Vehicle’s Sensor Measurements

2022-06-23T08:35:59+00:00

The discrepancy in the distribution of source and target domains is usually referred to as a domain shift. It is one of the reasons for the inferior performance of machine learning solutions at deployment. We illustrate that the domain shift issue is pertinent to the readings of the vehicles’ operational sensors. This is due to the fact that these measurements are collected over a period of time and are susceptible to various changes that happen in the meantime. Examples of these changes are usage pattern variations, aging of the vehicles, seasonal shifts, and driver changes. However, domain adversarial neural networks (DANN) have shown promising results to reduce the negative impact of the domain shift. The present study investigates domain adaptation (DA) in the predictive maintenance field by estimating the remaining useful life (RUL) of turbochargers. The devices are operating on a fleet of VOLVO trucks, and the information about their services is collected over four years between 2016 and 2019. The input features to the model are a set of bi-weekly collected measurements called logged vehicle data (LVD). The contributions of this paper are two-fold. First, we propose a new approach for detecting domain (covariate) shift using an autoencoder. Second, we adapt domain adversarial neural networks to the specific application of predicting turbocharger failures. Finally, we deploy a recurrent feature extraction layer in the DANN architecture to incorporate temporal aspect of the data. The experimental results demonstrate the superiority of the proposed method over the traditional approach.

Experimental Assessment of a Broadband Vibration and Acoustic Emission Sensor for Rotorcraft Transmission Monitoring

2022-06-26T09:33:48+00:00

Modern rotorcrafts rely on Health and Usage Monitoring Systems (HUMS) to enhance their availability, reliability, and safety. In those systems, data related to the health of key mechanical components is acquired, in addition to typical flight condition history data such as speed and torque. Commercial HUM systems usually rely on vibration measurements to assess the condition of shafts, gears, and bearings; using techniques such as spectral analysis, harmonic analysis, vibration trend and others. Recent research has shown that acoustic emissions (AE) can be advantageous in the detection of mechanical faults, in particular detecting very early small defects on bearings and gears, providing extra time for maintenance planning. However, the addition of extra sensors adds complexity and weight to the HUMS system, which is undesirable. This research is an experimental study to assess the monitoring capabilities of a broadband sensor, able to cover both low frequency vibration components as well as ultrasonic events, hence combining the benefits of both in a single compact sensing unit. The experimental results obtained from an instrumented rig using healthy components as well as seeded faults show the ability of the sensor to detect high frequency events, and compares the performance of the sensor in the low frequency range with a commercial accelerometer.

Optical Cutting Tool Wear Monitoring by 3D Geometry Reconstruction

2022-06-23T09:56:52+00:00

Cutting tool wear needs to be monitored closely to ensure good quality of machined parts. However, manual inspection is both expensive and time consuming, therefore there is a need for automated monitoring methods. We present a technique that can reconstruct the cutting tool surface in 3D, allowing a spatial estimation of the tool wear with high accuracy. The reconstruction allows an automated direct monitoring method that estimates at any time the cutting tool condition, avoiding conversion work and major quality issues. The optical measurement setup consists of a hardware triggered line scan camera that registers the spinning cutting tool’s shadow inflicted by a collimated backlight. We show how to leverage the 1D line scan signal acquired at varying cutting heights of the tool into a full 3D reconstruction. The progression of tool wear may thus be monitored by comparing the reconstructed shape to previous measurements. To this end we show a methodology for tool wear quantification. Additionally, to assess the measurement technique, an accuracy analysis with ground truth geometry was performed. The technique was applied to multiple degrading drilling tools. By automation of the cutting tool health monitoring, retrofitting this technology on a conventional machining center would transform it into an Industry 4.0 compatible (smart) machining center utilizing off-the-shelf optical equipment with moderate costs.

Data-Driven Fault Detection for Transmitter in Logging-While-Drilling Tool

2022-06-26T08:50:28+00:00

Logging tools widely used in the oil and gas industry are exposed to demanding environmental conditions that can lead to faster degradation and unexpected failures. These events can reduce productivity, delay deliverables, or even bring entire drilling operations to an end. However, such accidents can be avoided using a prognostics and health management approach. This paper presents a data-driven fault detection method for transmitter in logging-while-drilling tool adopting a support vector machine classifier. The health analyzer determines the component’s physical condition in just a few minutes, demonstrating an exceptional value for both field and maintenance engineers. This work is part of a long-term project aimed at constructing a digital fleet management system for downhole testing tools.

Autonomous Bearing Tone Tracking Algorithm

2022-06-26T09:04:44+00:00

To date, much of the research done in the field of condition monitoring of rotating machinery is conducted in the frequency domain. The frequency domain analysis specifically for bearings is based on extracting features from the spectrum of the vibration signature. These features are mostly based on the amplitude at the bearing tones along with their sidebands and high order harmonics. Therefore, it is important to determine the location of the mentioned bearing tones in the spectrum accurately and automatically. For the case of ball bearings this process can be problematic due to slippage of the rolling elements and variations in the angle of contact. These may cause the bearing tone to deviate from its nominal value.

To this day, the common practice for bearing diagnostics is based on the vibration level at the analytical bearing tones or involvement of experts to identify the true location of the bearing tone. In this research an autonomous algorithm for bearing tone extraction, based on pattern matching, was developed. The proposed algorithm is based on the common assumption that the spectrum of a faulted bearing contains a certain known pattern of prominent peaks. The algorithm “scans” the entire spectrum and determines the frequency value which has the highest correlation to the mentioned pattern.

The proposed algorithm was validated and its capabilities are illustrated using experimental data. This algorithm is able to assist any diagnostic approach towards automatic and reliable feature extraction process, both for physics based and data driven approaches.

Noise-Robust Representation for Fault Identification with Limited Data via Data Augmentation

2022-06-23T07:42:22+00:00

Noise will be unavoidably present in the data collected from physical environments, regardless of how sophisticated the measurement equipment is. Furthermore, collecting enough faulty data is a challenge since operating industrial machines in faulty modes not only has severe consequences to the machine health, but also may affect collateral machinery critically, from health state point of view. In this paper, we propose a method of denoising with limited data for the purpose of fault identification. In addition, our method is capable of removing multiple levels of noise simultaneously. For this purpose, inspired by unsupervised contrastive learning, we first augment the data with multiple levels of noise. Later, we construct a new feature representation using Contrastive Loss. The last step is building a classifier on top of the learned representation; this classifier can detect various faults in noisy environments. The experiments on the SOUTHEAST UNIVERSITY (SEU) dataset of bearings confirm that our method can simultaneously remove multiple noise levels.

Automating Critical Surface Identification and Damage Detection Using Deep Learning and Perspective Projection Methods

2022-06-23T09:06:55+00:00

With an increased collection of data, assessing the health of an asset and designing recommendations or executing response actions via prognostics and health management (PHM) has made great advances. These actions can be corrective or preventive depending upon the risk of failure or the cost of repair. As downhole testing tools operate in extreme environments, they are subjected to conditions like elevated temperature, shocks, vibrations, and pressures. The dump mandrels used in the process are prone to wear and tear like scratches, pits, and corrosion, which may cause operational failure. If these damages and their degree goes undetected and no remedial actions are taken, possibilities of non-productive time (NPT) and financial losses increase drastically. This paper aims to develop a fitness inspector which uses Computer Vision and Deep Learning to identify critical surfaces of these tools and the damage within them. This will help the Subject Matter Experts (SMEs) by replacing the qualified workforce provided by them and reducing the time consumed to gauge the health status of all the tools as the diagnosis can be made in real-time.

State of Health Forecasting of Heterogeneous Lithium-ion Battery Types and Operation Enabled by Transfer Learning

2022-06-20T15:20:29+00:00

Due to the global transition to electromobility and the associated increased use of high-performance batteries, research is increasingly focused on estimating and forecasting the state of health (SOH) of lithium-ion batteries. Several data-intensive and well-performing methods for SOH forecasting have been introduced. However, these approaches are only reliable for new battery types, e.g., with a new cell chemistry, if a sufficient amount of training data is given, which is rarely the case. A promising approach is to transfer an established model of another battery type to the new battery type, using only a small amount of data of the new battery type. Such methods in machine learning are known as transfer learning. The usefulness and applicability of transfer learning and its underlying methods have been very successfully demonstrated in various fields, such as computer vision and natural language processing. Heterogeneity in battery systems, such as differences in rated capacity, cell cathode materials, as well as applied stress from use, necessitate transfer learning concepts for data-based battery SOH forecasting models. Hereby, the general electrochemical behavior of lithium-ion batteries, as a major common characteristic, supposedly provides an excellent starting point for a transfer learning approach for SOH forecasting models. In this paper, we present a transfer learning approach for SOH forecasting models using a multilayer perceptron (MLP). We apply and evaluate it on the method presented by von Bülow, Mentz, and Meisen (2021) using five battery datasets. In this regard, we investigate the optimal conditions and settings for the development of transfer learning with respect to suitable data from the target domain, as well as hyperparameters such as learning rate and frozen layers. We show that for the transfer of a SOH forecasting model to a new battery type it is more beneficial to have data of few old batteries, compared to data of many new batteries, especially in the case of superlinear degradation with knee points. Contrarily to computer vision freezing no layers is preferable in 95% of the experimental scenarios.

Failures Mapping for Aircraft Electrical Actuation System Health Management

2022-06-23T13:15:09+00:00

This paper presents the different types of failure that may occur in flight control electrical actuation systems. Within an aircraft, actuation systems are essential to deliver physical actions. Large actuators operate the landing gears and small actuators adjust passenger seats. As developing, aircraft systems have become more electrical to reduce the weight and complexity of hydraulic circuits, which could improve fuel efficiency and lower NOx emissions. Electrical Actuation (EA) are one of those newly electrified systems. It can be categorized into two types, Electro-Hydraulic Actuation (EHA) and Electro-Mechanical Actuation (EMA) systems. Emerging electric and hydrogen fuel aircraft will rely on all-electric actuation. While electrical actuation seems simpler than hydraulic at the systems level, the subsystems and components are more varied and complex. The aim of the overall project is to develop a highly representative Digital Twin (DT) for predictive maintenance of electrical flight control systems. A comprehensive understanding of actuation system failure characteristics is fundamental for effective design and maintenance. This research focuses on the flight control systems including the ailerons, rudders, flaps, spoilers, and related systems. The study uses the Cranfield University Boeing 737 as the basis to elaborate the different types of actuators in the flight control system. The Aircraft Maintenance Manual (AMM) provides a baseline for current maintenance practices, effort, and costs. Equivalent EHA and EMA to replace the 737 systems are evaluated. In this paper, the components and their failure characteristics are elaborated in a matrix. The approach to model these characteristics in DT for aircraft flight control system health management is discussed. This paper contributes to the design, operation and support of aircraft systems.

An Approach to Condition Monitoring of BLDC Motors with Experimentally Validated Simulation Data

2022-06-23T13:41:08+00:00

Due to their compact design and low number of wear parts, Brushless Direct Current (BLDC) motors are ideally suited for use in unmanned aerial vehicles (UAVs). In view of the growing areas of application and the increasing complexity of unmanned flight missions, the need for suitable safety mechanisms for the operation of technical components, such as BLDC motors, in unmanned aircraft drive trains is also increasing. The integration of redundant components analogous to manned aviation is often not possible for smaller unmanned aerial vehicles for weight reasons. Therefore, online-capable dynamic diagnosis and prognosis methods for monitoring safety-critical components of unmanned aircraft are subject of ongoing research.

One major challenge in the development of data based condition monitoring approaches for safety critical components is the availability of operational data of degraded components. This often leads to an unbalanced database without sufficient information on components’ degradation behavior.

In the presented work, this problem is approached by combining bench testing and simulation models. On a test rig, common degradation effects are recreated by targeted manipulation. This allows for a safe and expressive data acquisition of the components’ behavior. In order to reduce the material and time required to build up a sufficient database for condition monitoring with experimental data, the observable effects are replicated in a simulation. This provides the opportunity to create a large database with slight variations in simulation parameters and incorporated noise in the simulation.

The BLDC motor manipulation on the test rig includes mechanical, electrical and magnetic manipulation. The effects of the manipulation are analyzed and their representation by parameters in the corresponding simulation is derived. The model is built in MATLAB Simulink and replicates both the electrical and physical behavior of the motor, as well as its commutation behavior.

The established simulation data shall be used as a balanced dataset on which condition monitoring algorithms can be trained. This will allow for the comparison of various data based condition monitoring methods in the future. A remaining challenge lies in the time behavior of the analyzed degradation, which has not yet been explored in depth. The proposed approach might also be applied to further unmanned aerial vehicle components, such as servo motors.

Uncertainty Informed Anomaly Scores with Deep Learning: Robust Fault Detection with Limited Data

2022-06-23T08:47:49+00:00

Quantifying the predictive uncertainty of a model is an important ingredient in data-driven decision making. Uncertainty quantification has been gaining interest especially for deep learning models, which are often hard to justify or explain. Various techniques for deep learning based uncertainty estimates have been developed primarily for image classification and segmentation, but also for regression and forecasting tasks. Uncertainty quantification for anomaly detection tasks is still rather limited for image data and has not yet been demonstrated for machine fault detection in PHM applications.

In this paper we suggest an approach to derive an uncertaintyinformed anomaly score for regression models trained with normal data only. The score is derived using a deep ensemble of probabilistic neural networks for uncertainty quantification. Using an example of wind-turbine fault detection, we demonstrate the superiority of the uncertainty-informed anomaly score over the conventional score. The advantage is particularly clear in an ”out-of-distribution” scenario, in which the model is trained with limited data which does not represent all normal regimes that are observed during model deployment.

Deep Learning Representation Pre-training for Industry 4.0

2022-06-14T08:52:09+00:00

Deep learning (DL) approaches have multiple potential advantages that have been explored in various fields, but for prognostic and health management (PHM) applications, this is not the case due to the lack of data in particular applications and also due of the absence of multiple DL-oriented benchmarks as in other fields, which limits the research in this area even though these types of applications will have a strong impact on the industrial world. To introduce the benefits of DL in this area, we should be able to develop models even when we have small data sets, to verify whether or not this is possible, in this thesis we explore the research direction of few shot learning in the context of equipment PHM.

Physics Informed Self Supervised Learning For Fault Diagnostics and Prognostics in the Context of Sparse and Noisy Data

2022-06-14T06:34:29+00:00

Sparse & noisy monitoring data leads to numerous challenges in prognostic and health management (PHM). Big data volume but poor quality with scarce healthy states information limits the performance of training machine learning (ML) and physics-based failure modeling. To address these challenges, this thesis aims to develop a new hybrid PHM framework with the ability to autonomously discover and exploit incomplete implicit physics knowledge in sparse & noisy monitoring data, providing a solution for deep physics knowledge-ML fusion by physics-informed machine learning algorithms. In addition, the developed hybrid framework also applies the self-supervised learning paradigm to significantly improve the learning performance under uncertain, sparse, and noisy data with lower requirements for specialist area knowledge. The performance of the developed algorithms will be investigated on the sparse and noise data generated by simulation data sets, public benchmark data sets, and the PHM platform to demonstrate its applicability.

A Novel Way to Apply Transfer Learning to Aircraft System Fault Diagnosis

2022-06-14T12:52:21+00:00

In recent years, transfer learning as a method that solves many issues limiting the real-world application of conventional machine learning methods has received dramatically increasing attention in the field of machine fault diagnosis. One major finding from an initial literature review shows that the majority of the existing research only focus on the transfer of diagnostic knowledge between various conditions of the same machine or different representation of similar machines. The primary goal of the current work is to seek a way to apply transfer learning to distinct domains, thereby expanding the boundary of transfer learning in the fault diagnosis field. In particular, attempts will be made to explore ways of transferring knowledge between diagnostic tasks of different aircraft systems. One promising method to help achieving this goal is transfer learning by structural analogy, since this method is capable of extracting high-level structural knowledge to apply transfer learning between seemingly unrelated domains, similar to the scenarios of transfer between different aircraft systems.

The Application, Utility and Acceptability of Data Analytics in Safety Risk Management of Airline Operations

2022-06-12T19:26:02+00:00

One area the aviation industry is grappling with is the quantification of the probability of occurrence of safety incidents. Currently, aviation professionals involved in safety risk management mostly rely on collective experience to determine probability of incident occurrences and apply it to the International Civil Aviation Organisation (ICAO) matrix or equivalent to evaluate the risk. A number of limitations linked to the use of risk matrices will be explored in this paper. It is the aim of this paper to explore statistical methods that can be used to determine the probability of safety occurrences and come up with an algorithm that can be used by airlines using available safety data. The novelty of this research is that it combines the exploration of use of statistical techniques to quantitatively assess risk using Flight Data Monitoring (FDM) and other data, with acceptability of Safety Risk Management (SRM) data analytics by operational personnel. The paper also explores the contributory factors leading to the reluctance of operational personnel to use data analytics to inform their risk assessments despite the increasing availability of operational data and advancement in technology.

Diagnosis and Fault-Tolerant Control for a Multi-Engine Cluster of a Reusable Launcher with Sensor and Actuator Faults

2022-06-12T20:14:38+00:00

A possible way to increase the reliability and availability of a system is to apply an Active Fault Tolerant Control (AFTC) algorithm. This thesis aims to use this algorithm in a multiengine propulsive cluster with sensor and actuator faults. First, a Health Monitoring System (HMS) will be developed to monitor the entire propulsive cluster. The HMS will use model-based fault diagnosis techniques. Then, in case of actuator faults, the cluster will be reconfigured to minimize its effects. The reconfiguration can be made by using control allocation or modifying the control law of the engine. A simulation model of the entire cluster is under development. The model simulates the whole system, including the propellant feeding system, engines, and mechanical system. It will be used to study the effect of different faults on the system and compare different reconfiguration strategies.

Artificial-Intelligence-Based Maintenance Scheduling for Complex Systems with Multiple Dependencies

2022-06-12T11:43:27+00:00

Maintenance planning for complex systems has still been a challenging problem. Firstly, integrating multiple dependency types into maintenance models makes them more realistic, however, more complicated to solve and analyze. Secondly, the number of maintenance decision variables needed to be optimized increases rapidly in the number of components, causing computational expensive for optimization algorithms. To face these issues, this thesis aims to incorporate multiple kinds of dependencies into maintenance models as well as to take advantage of recent advances in artificial intelligence field to effectively optimize maintenance polices for large-scale multi-component systems.

Contribution to the Design and Implementation of a Reflexive Cyber-Physical System: Application to Air Quality Prediction in the Vallees des Gaves

2022-06-15T07:06:19+00:00

This thesis aims to set up a scientific approach to monitor and take preventive actions on the air quality for the actors of a territory not covered by conventional measuring stations. Thus, a Cyber-Physical System (CPS) approach combined with Prognostics Health Management (PHM) methodologies is chosen to move toward a self-monitoring and self-reconfiguration system. To collect data in an inexpensive manner, measurement stations with low-cost sensors (LCS) are developed. LCS have drawbacks and the first part of this thesis is the use of redundancy and a proposed algorithm to increase their hardware and data reliability. A first station is deployed as proof of concept and the region is already receiving real-time data. The next phase is to build forecasting models to help authorities make decisions.

Combining Knowledge and Deep Learning for Prognostics and Health Management

2022-06-17T08:03:03+00:00

In the recent past deep learning approaches have achieved remarkable results in the area of Prognostics and Health Management (PHM). These algorithms rely on large amounts of data, which is often not available, and produce outputs, which are hard to interpret. Before the broad success of deep learning machine faults were often classified using domain expert knowledge based on experience and physical models. In comparison, these approaches only require small amounts of data and produce highly interpretable results. On the downside, however, they struggle to predict unexpected patterns hidden in data. This research aims to combine knowledge and deep learning to increase accuracy, robustness and interpretability of current models.