Dissertationen zum Thema „Data-driven maintenance“

Modern Heavy Duty Vehicles consist of a multitude of components and operate in various conditions. As there is value in goods transported, there is an incentive to avoid unplanned breakdowns. For this, condition based maintenance can be applied.\newline This thesis presents a study comparing the applicability of the data-driven Consensus SelfOrganizing Models (COSMO) method and the model-driven patent series introduced by Fogelstrom, applied on the air processing system for leakage detection on Scania Heavy Duty Vehicles. The comparison of the two methods is done using the Area Under Curve value given by the Receiver Operating Characteristics curves for features in order to reach a verdict.\newline For this purpose, three criteria were investigated. First, the effects of the hyper-parameters were explored to conclude a necessary vehicle fleet size and time period required for COSMO to function. The second experiment regarded whether environmental factors impact the predictability of the method, and finally the effect on the predictability for the case of nonidentical vehicles was determined.\newline The results indicate that the number of representations ought to be at least 60, rather with a larger set of vehicles in the fleet than with a larger window size, and that the vehicles should be close to identical on a component level and be in use in comparable ambient conditions.\newline In cases where the vehicle fleet is heterogeneous, a physical model of each system is preferable as this produces more stable results compared to the COSMO method.<br>Moderna tunga fordon består av ett stort antal komponenter och används i många olika miljöer. Då värdet för tunga fordon ofta består i hur mycket gods som transporteras uppstår ett incitament till att förebygga oplanerade stopp. Detta görs med fördel med hjälp av tillståndsbaserat underhåll. Denna avhandling undersöker användbarheten av den data-drivna metoden Consensus SelfOrganizing Models (COSMO) kontra en modellbaserad patentserie för att upptäcka läckage på luftsystem i tunga fordon. Metoderna ställs mot varandra med hjälp av Area Under Curve-värdet som kommer från Receiver Operating Characteristics-kurvor från beskrivande signaler. Detta gjordes genom att utvärdera tre kriterier. Dels hur hyperparametrar influerar COSMOmetoden för att avgöra en rimlig storlek på fordonsflottan, dels huruvida omgivningsförhållanden påverkar resultatet och slutligen till vilken grad metoden påverkas av att fordonsflottan inte är identisk. Slutsatsen är att COSMO-metoden med fördel kan användas sålänge antalet representationer överstiger 60 och att fordonen inom flottan är likvärdiga och har använts inom liknande omgivningsförhållanden. Om fordonsflottan är heterogen så föredras en fysisk modell av systemet då detta ger ett mer stabilt resultat jämfört med COSMO-metoden.

The goal of this PhD thesis is to study, design and develop data analysis methods for naval applications. Data analysis is improving our ways to understand complex phenomena by profitably taking advantage of the information laying behind a collection of data. In fact, by adopting algorithms coming from the world of statistics and machine learning it is possible to extract valuable information, without requiring specific domain knowledge of the system generating the data. The application of such methods to marine contexts opens new research scenarios, since typical naval problems can now be solved with higher accuracy rates with respect to more classical techniques, based on the physical equations governing the naval system. During this study, some major naval problems have been addressed adopting state-of-the-art and novel data analysis techniques: condition-based maintenance, consisting in assets monitoring, maintenance planning, and real-time anomaly detection; energy and consumption monitoring, in order to reduce vessel consumption and gas emissions; system safety for maneuvering control and collision avoidance; components design, in order to detect possible defects at design stage. A review of the state-of-the-art of data analysis and machine learning techniques together with the preliminary results of the application of such methods to the aforementioned problems show a growing interest in these research topics and that effective data-driven solutions can be applied to the naval context. Moreover, for some applications, data-driven models have been used in conjunction with domain-dependent methods, modelling physical phenomena, in order to exploit both mechanistic knowledge of the system and available measurements. These hybrid methods are proved to provide more accurate and interpretable results with respect to both the pure physical or data-driven approaches taken singularly, thus showing that in the naval context it is possible to offer new valuable methodologies by either providing novel statistical methods or improving the state-of-the-art ones.

Recent technological advances in the automotive industry have made vehicularsystems increasingly complex in terms of both hardware and software. As thecomplexity of the systems increase, so does the complexity of efficient monitoringof these system. With increasing computational power the field of diagnosticsis becoming evermore focused on software solutions for detecting and classifyinganomalies in the supervised systems. Model-based methods utilize knowledgeabout the physical system to device nominal models of the system to detect deviations,while data-driven methods uses historical data to come to conclusionsabout the present state of the system in question. This study proposes a combinedmodel-based and data-driven diagnostic framework for fault classification,severity estimation and novelty detection. An algorithm is presented which uses a system model to generate a candidate setof residuals for the system. A subset of the residuals are then selected for eachfault using L1-regularized logistic regression. The time series training data fromthe selected residuals is labelled with fault and severity. It is then compressedusing a Gaussian parametric representation, and data from different fault modesare modelled using 1-class support vector machines. The classification of datais performed by utilizing the support vector machine description of the data inthe residual space, and the fault severity is estimated as a convex optimizationproblem of minimizing the Kullback-Leibler divergence (kld) between the newdata and training data of different fault modes and severities. The algorithm is tested with data collected from a commercial Volvo car enginein an engine test cell and the results are presented in this report. Initial testsindicate the potential of the kld for fault severity estimation and that noveltydetection performance is closely tied to the residual selection process.

Il controllo neurale della postura umana è stato investigato a partire da un punto di vista fisico. Il paradigma di controllo intermittente è stato usato allo scopo di capire il peso di quest'ultimo nella generazione delle traiettorie del centro di pressione. Un primo contributo di questo lavoro rigurda quindi l'analisi del centro di pressione generato dal suddetto modello biomeccanico attraverso l'extended detrended fluctuation analysis, recentemente proposta in letteratura. Le proprietà di correlazione a lungo termine e disomogeneità sono risultate strettamente legate al guadagno derivativo del modello di controllo intermittente e anche al grado di intermittenza. Il paradigma di controllo è stato poi esteso verso un sistema biomeccanico più complesso, cioè un pendolo inverso doppio link con controllo intermittente alla caviglia. I contributi più significativi hanno riguardato la modellazione matematica del centro di pressione per una struttura multi-link e la verifica della sua plausibilità fisiologica. Si è poi preso in considerazione il caso della postura perturbata, integrando aspetti cinematici, dinamici e relativi all'attività muscolare. A tal fine, si è utilizzato sia un approccio fisico che basato su dati per l'identificazione dei modelli a struttura variabile Si sono prese in considerazione differenti condizioni di sperimentali, e in tutti i casi l'approccio utilizzato ha garantito un adeguato grado di interpretabilità riguardo il ruolo del sistema nervoso centrale nella regolazione del postura eretta in condizioni perturbate. La seconda parte della tesi ha riguardato la caratterizzazione del controllo motorio attraverso il segnale elettromiografico di superfice. Il primo contributo ha riguardato l'identifcazione dell'onset muscolare in condizioni di basso rapporto segnale rumore, sfruttando operatori energetico di tipo Teager-Kaiser al fine del precondizionamento del segnale mioelettrico. La versioe estesa di questo tipo di operatori è risultata particolarmente utile al miglioramento delle performance di numerosi algoritmi di detection. Si è poi proseguito con l'utilizzo di tali segnali al fine della classificazione dei gesti dell'arto superiore. In particoalre si è prerso in considerazione il problema della motion intention detection dei principali movimenti della spalla , utilizzando sia descrittori del segnale elettromiografico nel dominio del tempo e della frequenza. Quest'ultimo aspetto risulta essere un elemento di novità nel contesto scientifico in quanto si sono considerati il riconoscimento l'intezioni di movimento di otto gesti della spalla con particolare attenzione al ruolo dei descrittori del segnale per la classificazione. Infine, con approcci simili, si è preso in considerazione il problema del riconoscimento della scrittura manuale a partire dal dato elettromiografico. Tale aspetto risulta poco investigato sotto la prospettiva della pattern recognition mioelettrica, ma la sua valenza è data dalla crescente richiesta di interfacce uomo-macchina per compiti riabilitativi che coinvolgono una componente cognitiva significativa, Inoltre, vista la tendenza ad investigare il ruolo del polso per il prelievo del segnale elettromiografico al fine della realizzazione delle suddette interfacce, si è analizzato l'utilizzo dei segnali elettromiografici del polso rispetto a quelli dell'avambraccio al fine di predirre le cifre scritte dall'utente, noto che l'avambraccio risulta essere la zona di prelievo più comunemente utilizzata.<br>The biomechanics and the neural control of the human stance was investigated starting from a physical point of view. In particular the intermittent motor control paradigm was investigated with the aim of understanding how such paradigm mirrors in the center of pressure (COP) trajectories. A first contribution given in this work of thesis regards the analysis of COP generated from intermittent controlled inverted pendulum through the extended detrended fluctuation analysis, which was recently introduced in the literature. It has been found that the long-term correlation and inohmogeniety properties of the COP time series strictly depend on the derivative gain term of the intermittent controller and on the degree of intermittency of the control action. Thus, , another contribution provided in this work of thesis regards the use of a more complex biomechanical model of the stance, e.g. a double-link inverted pendulum intermittently controlled at the ankle. In terms of novelty, it deserves to be pointed out the results regarding the mechanical modeling of the COP for a multi-link structure, and the assessment of its physiological plausibility. . On the other hand, when the perturbed posture motor task was taken into account, there was the need to enlarge the perspective, integrating kinematic, dynamic and muscle activity data. The idea of employing different sources of information to develop models of the CNS represents an important element that was investigated using tools related to hybrid system identification theory. Subjects underwent to impulsive support base translations in three different conditions: considering eyes open, closed, and performing mental counting. Although such data were essentially analyzed through a data-driven approach, the identified models guaranteed physical interpretations of the role played by the CNS in the three different conditions. The second main core of this thesis regards the characterization of the motor control using the surface electromyographic (sEMG) signals. A first contribution given in this work regarded the muscle onset detection considering low SNR scenarios. In this framework, energy operators such as the Teager-Kaiser energy operator (TKEO) and its extended version (ETKEO) were investigated as signal preconditioning steps before the application of state of the art onset detection algorithms. The latter have been significantly boosted when ETKEO was used with respect to TKEO. The use of extended energy operators for the sEMG signal preprocessing constitutes a novel element in this field that can be also further investigated in future studies. From the sEMG muscle, one can also predict which movement the subject is going to perform. This aspect can be enclosed in the motion intention detection (MID) field. In this thesis a MID problem was investigated by taking into account two important aspects: as first the study was centered on the shoulder joint movements. Secondly, the MID problem was faced under a pattern recognition perspective. This allowed to verify whether methodologies encountered in the myoelectric hand gesture recognition can be transferred in the affine field of MID In contrast to what reported in the literature, where MID problems generally consider only few movements, in this work of thesis up to eight shoulder movements have been investigated. Myoelectric pattern recognition architectures were also used in the assessment of the ten hand-written digits. Despite the handwriting can be considered a hand movement that involves fine muscular control actions, it has not been consistently investigated in the field of sEMG based hand gesture recognition. Further, since the literature supports the change from forearm to wrist in order to acquire EMG data for hand gesture recognition, it was investigated whether such exchange can be performed when a challenging classification task, as the handwriting recognition has to be performed.

Maintenance management is seen as a “necessary evil”, rather than a profit contributing resource that could intensify competitive advantage for the organisation. With the world facing the fourth industrial revolution, a radical increase in the reshaping of companies and competition within asset intensive industries is being observed. Organisations in these industries are being forced to rethink traditional ways of working and gearing the workforce with higher and more diversified competency profiles. This suggests that the traditional way of executing maintenance management, being predominantly reactive with the lack of data driven decision making, is certainly inadequate for a sustainable competitive advantage. An improved way of managing maintenance should be through developing and applying dynamic capabilities within the maintenance domain of the organisation. This research draws on theories of dynamic capabilities (DC), decision making performance (DMP), business process performance (BPP) and firm performance (Fper), in the context of data driven decision making in organisations heavily reliant on good maintenance management practices. The aim of this study was to explore and understand the relationships between these constructs, for insight into further improvement and development of a competitive advantage. The findings presented a statistically significant relationship between DC and Fper, DC and BPP, DC and DMP, but most importantly, a multiple full indirect mediation role was observed, which provides insights for both business and for further studies in academia.<br>Mini Dissertation (MBA)--University of Pretoria, 2020.<br>pt2021<br>Gordon Institute of Business Science (GIBS)<br>MBA<br>Unrestricted

Applied Research in System Analysis (ARiSA) is a company specialized in the development of the customer-specific quality models and applied research work. In order to improve the quality of the projects and to reduce maintenance costs, ARiSA developed Quality Monitor (QM) – a web application for quality analysis. QM application has been originally developed as a basic program to enable customers to evaluate the quality of the sources. Therefore, the business logic of the application was simplified and certain limitations were imposed on it, which in its turn leads to a number of issues related to user experience, performance and architecture design. These aspects are important for both application as a product, and for its future promotion. Moreover, this is important for customers, as end users. Main application issues, which were added to the maintenance list are: manual data upload, insufficient server resources to handle long-running and resource consuming operations, no background processing and status reporting, simplistic presentation of analysis results and known usability issues, weak integration between analysis back-ends and front-end. ­­­­­­­­­­­In order to address known issues and to make improvements of the existing limitations, a maintenance phase of QM application is initiated. First of all, it is intended to stabilize current version and improve user experience. It also needed for refactoring and implementation of more efficient data uploads processing in the background. In addition, extended functionality of QM would fulfill customer needs and transform application from the project into a product. Extended functionality includes: automated data upload from different build processes, new data visualizations, and improvement of the current functionality according to customer comments. Maintenance phase of QM application has been successfully completed and master thesis goals are met. Current version is more stable and more responsive from user experience perspective. Data processing is more efficient, and now it is implemented as background analysis with automatic data import. User interface has been updated with visualizations for client-side interaction and progress reporting. The solution has been evaluated and tested in close cooperation with QM application customers. This thesis describes requirements analysis, technology stack with choice rationale and implementation to show maintenance results.

Rolling element bearings are a vital part in many rotating machinery, including vehicles. A defective bearing can be a symptom of other problems in the machinery and is due to a high failure rate. Early detection of bearing defects can therefore help to prevent malfunction which ultimately could lead to a total collapse. The thesis is done in collaboration with Scania that wants a better understanding of how external sensors such as accelerometers, can be used for condition monitoring in their gearboxes. Defective bearings creates vibrations with specific frequencies, known as Bearing Characteristic Frequencies, BCF [23]. A key component in the proposed method is based on identification and extraction of these frequencies from vibration signals from accelerometers mounted near the monitored bearing. Three solutions are proposed for automatic bearing fault detection. Two are based on data-driven classification using a set of machine learning methods called Support Vector Machines and one method using only the computed characteristic frequencies from the considered bearing faults. Two types of features are developed as inputs to the data-driven classifiers. One is based on the extracted amplitudes of the BCF and the other on statistical properties from Intrinsic Mode Functions generated by an improved Empirical Mode Decomposition algorithm. In order to enhance the diagnostic information in the vibration signals two pre-processing steps are proposed. Separation of the bearing signal from masking noise are done with the Cepstral Editing Procedure, which removes discrete frequencies from the raw vibration signal. Enhancement of the bearing signal is achieved by band pass filtering and amplitude demodulation. The frequency band is produced by the band selection algorithms Kurtogram and Autogram. The proposed methods are evaluated on two large public data sets considering bearing fault classification using accelerometer data, and a smaller data set collected from a Scania gearbox. The produced features achieved significant separation on the public and collected data. Manual detection of the induced defect on the outer race on the bearing from the gearbox was achieved. Due to the small amount of training data the automatic solutions were only tested on the public data sets. Isolation performance of correct bearing and fault mode among multiplebearings were investigated. One of the best trade offs achieved was 76.39 % fault detection rate with 8.33 % false alarm rate. Another was 54.86 % fault detection rate with 0 % false alarm rate.

La dégradation est un phénomène inévitable qui affecte les composants et les systèmes d'ingénierie, et qui peut entraîner leurs défaillances avec des conséquences potentiellement catastrophiques selon l'application. La motivation de cette Thèse est d'essayer de modéliser, d'analyser et de prédire les défaillances par des méthodes pronostiques qui peuvent permettre une gestion prédictive de la maintenance des actifs. Cela permettrait aux décideurs d'améliorer la planification de la maintenance, augmentant ainsi la disponibilité et la sûreté du système en minimisant les arrêts imprévus. Dans cet objectif, la recherche au cours de la thèse a été consacrée à l'adaptation et à l'utilisation d'approches basées sur des modèles et d'approches pilotées par les données pour traiter les processus de dégradation qui peuvent conduire à différents modes de défaillance dans les composants industriels, en utilisant différentes sources d'informations et de données pour effectuer des prédictions sur l'évolution de la dégradation et estimer la durée de vie utile restante (RUL).Les travaux de thèse ont porté sur deux applications pronostiques spécifiques: les pronostics basés sur des modèles pour la prédiction de la croissance des fissures par fatigue et les pronostics pilotées par les données pour les prédictions à pas multiples des données de séries chronologiques des composants des Centrales Nucléaires.Les pronostics basé sur des modèles compter sur le choix des modèles adoptés de Physics-of-Failure (PoF). Cependant, chaque modèle de dégradation ne convient qu'à certains processus de dégradation dans certaines conditions de fonctionnement, qui souvent ne sont pas connues avec précision. Pour généraliser, des ensembles de multiples modèles de dégradation ont été intégrés dans la méthode pronostique basée sur les modèles afin de tirer profit des différentes précisions des modèles spécifiques aux différentes dégradations et conditions. Les principales contributions des approches pronostiques proposées basées sur l'ensemble des modèles sont l'intégration d'approches de filtrage, y compris le filtrage Bayésien récursif et le Particle Filtering (PF), et de nouvelles stratégies d'ensemble pondérées tenant compte des précisions des modèles individuels dans l'ensemble aux étapes de prédiction précédentes. Les méthodes proposées ont été validées par des études de cas de croissance par fissures de fatigue simulées dans des conditions de fonctionnement variables dans le temps.Quant à la prédictions à pas multiples, elle reste une tâche difficile pour le Prognostics and Health Management (PHM) car l'incertitude de prédiction a tendance à augmenter avec l'horizon temporel de la prédiction. La grande incertitude de prédiction a limité le développement de pronostics à pas multiples dans les applications. Pour résoudre le problème, de nouveaux modèles de prédiction à pas multiples basés sur la Long Short-Term Memory (LSTM), un réseau de neurones profond développé pour traiter les dépendances à long terme dans les données de séries chronologiques, ont été développés dans cette Thèse. Pour des applications pratiques réalistes, les méthodes proposées abordent également les problèmes supplémentaires de détection d'anomalie, d'optimisation automatique des hyper-paramètres et de quantification de l'incertitude de prédiction. Des études de cas pratiques ont été envisagées, concernant les données de séries chronologiques collectées auprès des Générateurs de Vapeur et de Pompes de Refroidissement de Réacteurs de Centrales Nucléaires<br>Degradation is an unavoidable phenomenon that affects engineering components and systems, and which may lead to their failures with potentially catastrophic consequences depending on the application. The motivation of this Thesis is trying to model, analyze and predict failures with prognostic methods that can enable a predictive management of asset maintenance. This would allow decision makers to improve maintenance planning, thus increasing system availability and safety by minimizing unexpected shutdowns. To this aim, research during the Thesis has been devoted to the tailoring and use of both model-based and data-driven approaches to treat the degradation processes that can lead to different failure modes in industrial components, making use of different information and data sources for performing predictions on the degradation evolution and estimating the Remaining Useful Life (RUL).The Ph.D. work has addressed two specific prognostic applications: model-based prognostics for fatigue crack growth prediction and data-driven prognostics for multi-step ahead predictions of time series data of Nuclear Power Plant (NPP) components.Model-based prognostics relies on the choice of the adopted Physics-of-Failure (PoF) models. However, each degradation model is appropriate only to certain degradation process under certain operating conditions, which are often not precisely known. To generalize this, ensembles of multiple degradation models have been embedded in the model-based prognostic method in order to take advantage of the different accuracies of the models specific to different degradations and conditions. The main contributions of the proposed ensemble of models-based prognostic approaches are the integration of filtering approaches, including recursive Bayesian filtering and Particle Filtering (PF), and novel weighted ensemble strategies considering the accuracies of the individual models in the ensemble at the previous time steps of prediction. The proposed methods have been validated by case studies of fatigue crack growth simulated with time-varying operating conditions.As for multi-step ahead prediction, it remains a difficult task of Prognostics and Health Management (PHM) because prediction uncertainty tends to increase with the time horizon of the prediction. Large prediction uncertainty has limited the development of multi-step ahead prognostics in applications. To address the problem, novel multi-step ahead prediction models based on Long Short- Term Memory (LSTM), a deep neural network developed for dealing with the long-term dependencies in the time series data have been developed in this Thesis. For realistic practical applications, the proposed methods also address the additional issues of anomaly detection, automatic hyperparameter optimization and prediction uncertainty quantification. Practical case studies have been considered, concerning time series data collected from Steam Generators (SGs) and Reactor Coolant Pumps (RCPs) of NPPs

Surface treatments are often used in the manufacturing industry to change the surface of a product, including its related properties and functions. The occurrence of degradation and corrosion in surface treatment processes can lead to critical breakdowns over time. Critical breakdowns may impair the properties of the products and shorten their service life, which causes increased lead times or additional costs in the form of rework or scrapping.  Prevention of critical breakdowns due to machine component failure requires a carefully selected maintenance policy. Predictive maintenance is used to anticipate equipment failures to allow for maintenance scheduling before component failure. Developing predictive maintenance policies for surface treatment processes is problematic due to the vast number of attributes to consider in modern surface treatment processes. The emergence of smart sensors and big data has led companies to pursue predictive maintenance. A company that strives for predictive maintenance of its surface treatment processes is Seco Tools in Fagersta. The purpose of this master's thesis has been to investigate the occurrence of critical breakdowns and failures in the machine components of the chemical vapor deposition and post-treatment wet blasting processes by mapping the interaction between its respective process variables and their impact on critical breakdowns. The work has been conducted as a Six Sigma project utilizing the problem-solving methodology DMAIC.  Critical breakdowns were investigated combining principal component analysis (PCA), computational fluid dynamics (CFD), and statistical process control (SPC) to create an understanding of the failures in both processes. For both processes, two predictive solutions were created: one short-term solution utilizing existing dashboards and one long-term solution utilizing a PCA model and an Orthogonal Partial Least Squares (OPLS) regression model for batch statistical process control (BSPC). The short-term solutions were verified and implemented during the master's thesis at Seco Tools. Recommendations were given for future implementation of the long-term solutions. In this thesis, insights are shared regarding the applicability of OPLS and Partial Least Squares (PLS) regression models for batch monitoring of the CVD process. We also demonstrate that the prediction of a certain critical breakdown, clogging of the aluminum generator in the CVD process, can be accomplished through the use of SPC. For the wet blasting process, a PCA methodology is suggested to be effective for visualizing breakdowns.

Electrical contacts provide a well-proven solution to switching various loads in a wide variety of applications, such as power distribution, control applications, automotive and telecommunications. However, electrical contacts are known for limited reliability due to degradation effects upon the switching contacts due to arcing and fretting. Essentially, the life of the device may be determined by the limited life of the contacts. Failure to trip, spurious tripping and contact welding can, in critical applications such as control systems for avionics and nuclear power application, cause significant costs due to downtime, as well as safety implications. Prognostics provides a way to assess the remaining useful life (RUL) of a component based on its current state of health and its anticipated future usage and operating conditions. In this thesis, the effects of contact wear on a set of electromagnetic relays used in an avionic power controller is examined, and how contact resistance combined with a prognostic approach, can be used to ascertain the RUL of the device. Two methodologies are presented, firstly a Physics based Model (PbM) of the degradation using the predicted material loss due to arc damage. Secondly a computationally efficient technique using posterior degradation data to form a state space model in real time via a Sliding Window Recursive Least Squares (SWRLS) algorithm. Health monitoring using the presented techniques can provide knowledge of impending failure in high reliability applications where the risks associated with loss-of-functionality are too high to endure. The future states of the systems has been estimated based on a Particle and Kalman-filter projection of the models via a Bayesian framework. Performance of the prognostication health management algorithm during the contacts life has been quantified using performance evaluation metrics. Model predictions have been correlated with experimental data. Prognostic metrics including Prognostic Horizon (PH), alpha-Lamda (α-λ), and Relative Accuracy have been used to assess the performance of the damage proxies and a comparison of the two models made.

This thesis presents a novel hybrid prognostic methodology, integrating physics-based and data-driven prognostic models, to enhance the prognostic accuracy, robustness, and applicability. The presented prognostic methodology integrates the short-term predictions of a physics-based model with the longer term projection of a similarity-based data-driven model, to obtain remaining useful life estimations. The hybrid prognostic methodology has been applied on specific components of two different engineering systems, one which represents accelerated, and the other a nominal degradation process. Clogged filter and fatigue crack propagation failure cases are selected as case studies. An experimental rig has been developed to investigate the accelerated clogging phenomena whereas the publicly available Virkler fatigue crack propagation dataset is chosen after an extensive literature search and dataset analysis. The filter clogging experimental rig is designed to obtain reproducible filter clogging data under different operational profiles. This data is thought to be a good benchmark dataset for prognostic models. The performance of the presented methodology has been evaluated by comparing remaining useful life estimations obtained from both hybrid and individual prognostic models. This comparison has been based on the most recent prognostic evaluation metrics. The results show that the presented methodology improves accuracy, robustness and applicability. The work contained herein is therefore expected to contribute to scientific knowledge as well as industrial technology development.

Les approches de Pronostic et gestion de la santé des systèmes (Prognostics and Health Management : PHM) guidées par les données sont fortement dépendantes de la disponibilité et de la qualité d’historiques de défaillances, une exigence souvent difficile à satisfaire dans les systèmes de surveillance en conditions industrielles. Cette divergence crée un écart significatif entre les méthodologies de PHM et leur application pratique sur des systèmes réels. L’accent prédominant mis sur les ensembles de données unimodales dans les travaux de recherche en PHM basée sur les données met en lumière le potentiel des données multimodales pour combler cet écart.Cette thèse explore l’intégration des données multimodales afin d’améliorer les méthodes et les algorithmes de PHM appliqués aux machines industrielles. Elle aborde de manière systématique des défis cruciaux tels que l’absence de données, les données bruitées, les données clairsemées et irrégulières, le déséquilibre des classes et la rareté des données de fonctionnement jusqu’à la défaillance. Elle propose des méthodologies innovantes qui intègrent plusieurs modalités de données et exploitent l’expertise spécifique au domaine pour créer des modèles prédictifs robustes.Les contributions principales de la thèse se déclinent comme suit :1. Apprentissage basé sur l’attention intermodale : une nouvelle méthode d’apprentissage multimodal conçue pour atténuer les limites posées par les données manquantes et bruitées. Elle permet d’intégrer des informations provenant de multiples modalités, améliorant ainsi la précision et la robustesse des modèles prédictifs.2. Méthodologie de diagnostic multimodal assisté par les connaissances d’experts : cette méthodologie combine l’expertise du domaine avec l’apprentissage multimodal pour permettre des diagnostics complets, améliorant ainsi la détection et la classification des défauts dans les machines industrielles.3. Approche de pronostic basée sur des graphes : cette approche innovante construit des trajectoires de fonctionnement jusqu’à la défaillance à partir de données incomplètes en utilisant des techniques basées sur les graphes, offrant une avancée significative dans le domaine du pronostic de défaillances.Ces méthodologies ont été rigoureusement validées sur des données de simulation ainsi que sur des données industrielles provenant d’hydro-générateurs, démontrant des améliorations significatives des algorithmes de PHM et de maintenance prédictive. Les résultats soulignent le potentiel des données multimodales pour améliorer considérablement la fiabilité et l’efficacité des modèles de PHM.Cette thèse apporte un cadre complet pour exploiter diverses sources de données et l’expertise du domaine, promettant de transformer les stratégies de maintenance et de réduire les coûts opérationnels dans diverses industries. Les résultats ouvrent la voie à des recherches futures et à des applications pratiques, positionnant l’intégration des données multimodales comme une avancée essentielle dans le domaine du PHM<br>Prognostics and Health Management (PHM) with data-driven techniques is heavily dependent upon the availability of extensive and high-quality datasets, a requirement often challenging to fulfill in industrial condition monitoring environments. This discrepancy creates a significant gap between state-of-the-art PHM methodologies and their practical application in real-world scenarios. The prevailing focus in data-driven PHM research on unimodal datasets highlights the potential of multimodal data to bridge this gap.This thesis explores the integration of multimodal data to advance PHM models for industrial machines. It systematically addresses pivotal challenges such as data missingness and noise, sparse and irregular datasets, class imbalance, and the scarcity of run-to-failure data. The research develops innovative methodologies that incorporate multiple data modalities and harness domain-specific expertise to create robust predictive models.The primary contributions of this research include:1. Cross-modal attention-based learning: A new multimodal learning method is designed to mitigate the limitations posed by missing and noisy data. It allows integrating information across multiple modalities, thereby enhancing the accuracy and robustness of predictive models.2. Expert-knowledge-assisted multimodal diagnostics methodology: This methodology combines domain expertise with multimodal learning to enable comprehensive diagnostics, thereby improving fault detection and classification in industrial machinery.3. Graph-based prognostics approach: This innovative approach constructs run-to-failure trajectories from incomplete data using graph-based techniques, offering a significant advancement in failure prognostics.These methodologies were rigorously validated using both simulation and industrial dataset of hydrogenerators, demonstrating significant improvements in PHM and predictive maintenance capabilities. The results underscore the potential of multimodal data to significantly enhance the reliability and efficiency of PHM methods and algorithms.This thesis proposes a comprehensive framework for leveraging diverse data sources and domain expertise, promising to transform maintenance strategies and reducing operational costs across various industries. The findings pave the way for future research and practical implementations, positioning multimodal data integration as a pivotal advancement in the field of PHM

Hlavní cílem dizertace je poskytnutí uceleného pohledu na problematiku technické prognostiky, která nachází uplatnění v tzv. prediktivní údržbě založené na trvalém monitorování zařízení a odhadu úrovně degradace systému či jeho zbývající životnosti a to zejména v oblasti komplexních zařízení a strojů. V současnosti je technická diagnostika poměrně dobře zmapovaná a reálně nasazená na rozdíl od technické prognostiky, která je stále rozvíjejícím se oborem, který ovšem postrádá větší množství reálných aplikaci a navíc ne všechny metody jsou dostatečně přesné a aplikovatelné pro embedded systémy. Dizertační práce přináší přehled základních metod použitelných pro účely predikce zbývající užitné životnosti, jsou zde popsány metriky pomocí, kterých je možné jednotlivé přístupy porovnávat ať už z pohledu přesnosti, ale také i z pohledu výpočetní náročnosti. Jedno z dizertačních jader tvoří doporučení a postup pro výběr vhodné prognostické metody s ohledem na prognostická kritéria. Dalším dizertačním jádrem je představení tzv. částicového filtrovaní (particle filtering) vhodné pro model-based prognostiku s ověřením jejich implementace a porovnáním. Hlavní dizertační jádro reprezentuje případovou studii pro velmi aktuální téma prognostiky Li-Ion baterii s ohledem na trvalé monitorování. Případová studie demonstruje proces prognostiky založené na modelu a srovnává možné přístupy jednak pro odhad doby před vybitím baterie, ale také sleduje možné vlivy na degradaci baterie. Součástí práce je základní ověření modelu Li-Ion baterie a návrh prognostického procesu.

In condition based maintenance (CBM), maintenance activities are scheduled based on the predicted equipment failure times, and the predictions are performed based on conditon monitoirng data, such as vibration and acoustic data. The reported health condition prediction methods can be roughly classified into model-based, and data-driven, and integrated methods. Our research mainly focuses on CBM optimization using data driven methods, such as proportional hazards model (PHM) and artificial neural network (ANN), which don't require equipment physical models. In CBM optimization using PHM, the accuracy of parameter estimation for PHM greatly affects the effectiveness of the optimal maintenance policy. Directly using collected condition monitoring data may iv introduce noise into the CBM optimization, and thus the optimal maintenance policy obtained based on this model may not be really optimal. Therefore, a data processing method, where the actual measurements are fitted first using the Generalized Weibull-FR function, is proposed to remove the external noise before fitting it into the PHM. Effective CBM optimization methods utilizing ANN prediction information are currently not available due to two key challenges: (1) ANN prediction models typically only give a single remaining life prediction value, and it is hard to quantify the uncertainty associated with the predicted value; (2) simulation methods are generally used for evaluating the cost of the CBM policies, while more accurate and efficient numerical methods are not available. Therefore, we propose an ANN based CBM optimization approach and a numerical cost evaluation method to address those key challenges. It is observed that the prediction accuracy often improves with the increase of the age of the component. Therefore, we develop a method to quantify the remaining life prediction uncertainty considering the prediction accuracy improvements by modeling the relationship between the mean value as well as standard deviation of prediction error and the life percentage. An effective CBM optimization approach is also proposed to optimize the maintenance schedule. The proposed approaches are demonstrated using some simulated degradation data sets as well as some real-world vibration monitoring data set. They contribute to the general knowledge of CBM, and have the potential to greatly benefit various industries.

碩士<br>國立中央大學<br>企業管理學系<br>106<br>Utilizing preventive maintenance can reduce machine’s shutdown and improve the equipment efficiency. Traditional preventive maintainance methods focus on maintaining single component. The research, however, strives to maintain a group of components to further reduce the maintenance time. Components are clustered into group according to the their distributions of lifespans. Clusters that save the most maintenance costs are recommended to managers for maintenance scheduling. The methodology is applied to an auto component company for experiments. The results show that OEE is improved from 81% to 84%.

Data-driven maintenance bears the potential to realize various benefits based on multifaceted data assets generated in increasingly digitized industrial environments. By taking advantage of modern methods and technologies from the field of data science and analytics (DSA), it is possible, for example, to gain a better understanding of complex technical processes and to anticipate impending machine faults and failures at an early stage. However, successful implementation of DSA projects requires multidisciplinary expertise, which can rarely be covered by individual employees or single units within an organization. This expertise covers, for example, a solid understanding of the domain, analytical method and modeling skills, experience in dealing with different source systems and data structures, and the ability to transfer suitable solution approaches into information systems. Against this background, various approaches have emerged in recent years to make the implementation of DSA projects more accessible to broader user groups. These include structured procedure models, systematization and modeling frameworks, domain-specific benchmark studies to illustrate best practices, standardized DSA software solutions, and intelligent assistance systems. The present thesis ties in with previous efforts and provides further contributions for their continuation. More specifically, it aims to create supportive artifacts for the selection, evaluation, and application of data-driven methods in the field of industrial maintenance. For this purpose, the thesis covers four artifacts, which were developed in several publications. These artifacts include (i) a comprehensive systematization framework for the description of central properties of recurring data analysis problems in the field of industrial maintenance, (ii) a text-based assistance system that offers advice regarding the most suitable class of analysis methods based on natural language and domain-specific problem descriptions, (iii) a taxonomic evaluation framework for the systematic assessment of data-driven methods under varying conditions, and (iv) a novel solution approach for the development of prognostic decision models in cases of missing label information. Individual research objectives guide the construction of the artifacts as part of a systematic research design. The findings are presented in a structured manner by summarizing the results of the corresponding publications. Moreover, the connections between the developed artifacts as well as related work are discussed. Subsequently, a critical reflection is offered concerning the generalization and transferability of the achieved results. Thus, the thesis not only provides a contribution based on the proposed artifacts; it also paves the way for future opportunities, for which a detailed research agenda is outlined.:List of Figures List of Tables List of Abbreviations 1 Introduction 1.1 Motivation 1.2 Conceptual Background 1.3 Related Work 1.4 Research Design 1.5 Structure of the Thesis 2 Systematization of the Field 2.1 The Current State of Research 2.2 Systematization Framework 2.3 Exemplary Framework Application 3 Intelligent Assistance System for Automated Method Selection 3.1 Elicitation of Requirements 3.2 Design Principles and Design Features 3.3 Prototypical Instantiation and Evaluation 4 Taxonomic Framework for Method Evaluation 4.1 Survey of Prognostic Solutions 4.2 Taxonomic Evaluation Framework 4.3 Exemplary Framework Application 5 Method Application Under Industrial Conditions 5.1 Conceptualization of a Solution Approach 5.2 Prototypical Implementation and Evaluation 6 Discussion of the Results 6.1 Connections Between Developed Artifacts and Related Work 6.2 Generalization and Transferability of the Results 7 Concluding Remarks Bibliography Appendix I: Implementation Details Appendix II: List of Publications A Publication P1: Focus Area Systematization B Publication P2: Focus Area Method Selection C Publication P3: Focus Area Method Selection D Publication P4: Focus Area Method Evaluation E Publication P5: Focus Area Method Application<br>Datengetriebene Instandhaltung birgt das Potential, aus den in Industrieumgebungen vielfältig anfallenden Datensammlungen unterschiedliche Nutzeneffekte zu erzielen. Unter Verwendung von modernen Methoden und Technologien aus dem Bereich Data Science und Analytics (DSA) ist es beispielsweise möglich, das Verhalten komplexer technischer Prozesse besser nachzuvollziehen oder bevorstehende Maschinenausfälle und Fehler frühzeitig zu erkennen. Eine erfolgreiche Umsetzung von DSA-Projekten erfordert jedoch multidisziplinäres Expertenwissen, welches sich nur selten von einzelnen Personen bzw. Einheiten innerhalb einer Organisation abdecken lässt. Dies umfasst beispielsweise ein fundiertes Domänenverständnis, Kenntnisse über zahlreiche Analysemethoden, Erfahrungen im Umgang mit verschiedenen Quellsystemen und Datenstrukturen sowie die Fähigkeit, geeignete Lösungsansätze in Informationssysteme zu überführen. Vor diesem Hintergrund haben sich in den letzten Jahren verschiedene Ansätze herausgebildet, um die Durchführung von DSA-Projekten für breitere Anwendergruppen zugänglich zu machen. Dazu gehören strukturierte Vorgehensmodelle, Systematisierungs- und Modellierungsframeworks, domänenspezifische Benchmark-Studien zur Veranschaulichung von Best Practices, Standardlösungen für DSA-Software und intelligente Assistenzsysteme. An diese Arbeiten knüpft die vorliegende Dissertation an und liefert weitere Artefakte, um insbesondere die Selektion, Evaluation und Anwendung datengetriebener Methoden im Bereich der industriellen Instandhaltung zu unterstützen. Insgesamt erstreckt sich die Abhandlung auf vier Artefakte, die in einzelnen Publikationen erarbeitet wurden. Dies umfasst (i) ein umfangreiches Systematisierungsframework zur Beschreibung zentraler Ausprägungen wiederkehrender Datenanalyseprobleme im Bereich der industriellen Instandhaltung, (ii) ein textbasiertes Assistenzsystem, welches ausgehend von natürlichsprachlichen und domänenspezifischen Problembeschreibungen eine geeignete Klasse von Analysemethoden vorschlägt, (iii) ein taxonomisches Evaluationsframework zur systematischen Bewertung von datengetriebenen Methoden unter verschiedenen Rahmenbedingungen sowie (iv) einen neuartigen Lösungsansatz zur Entwicklung von prognostischen Entscheidungsmodellen im Fall von eingeschränkter Informationslage. Die Konstruktion der Artefakte wird durch einzelne Forschungsziele im Rahmen eines systematischen Forschungsdesigns angeleitet. Neben der Darstellung der einzelnen Forschungsbeiträge unter Bezugnahme auf die erzielten Ergebnisse der dazugehörigen Publikationen werden auch die Verbindungen zwischen den entwickelten Artefakten beleuchtet und Zusammenhänge zu angrenzenden Arbeiten hergestellt. Zudem erfolgt eine kritische Reflektion der Ergebnisse hinsichtlich ihrer Verallgemeinerung und Übertragung auf andere Rahmenbedingungen. Dadurch liefert die vorliegende Abhandlung nicht nur einen Beitrag anhand der erzeugten Artefakte, sondern ebnet auch den Weg für fortführende Forschungsarbeiten, wofür eine detaillierte Forschungsagenda erarbeitet wird.:List of Figures List of Tables List of Abbreviations 1 Introduction 1.1 Motivation 1.2 Conceptual Background 1.3 Related Work 1.4 Research Design 1.5 Structure of the Thesis 2 Systematization of the Field 2.1 The Current State of Research 2.2 Systematization Framework 2.3 Exemplary Framework Application 3 Intelligent Assistance System for Automated Method Selection 3.1 Elicitation of Requirements 3.2 Design Principles and Design Features 3.3 Prototypical Instantiation and Evaluation 4 Taxonomic Framework for Method Evaluation 4.1 Survey of Prognostic Solutions 4.2 Taxonomic Evaluation Framework 4.3 Exemplary Framework Application 5 Method Application Under Industrial Conditions 5.1 Conceptualization of a Solution Approach 5.2 Prototypical Implementation and Evaluation 6 Discussion of the Results 6.1 Connections Between Developed Artifacts and Related Work 6.2 Generalization and Transferability of the Results 7 Concluding Remarks Bibliography Appendix I: Implementation Details Appendix II: List of Publications A Publication P1: Focus Area Systematization B Publication P2: Focus Area Method Selection C Publication P3: Focus Area Method Selection D Publication P4: Focus Area Method Evaluation E Publication P5: Focus Area Method Application