No description available.

The use of knowledge-based systems to aid in the diagnosis of faults in physical devices has grown considerably since their introduction during the 1970s. The majority of the early knowledge-based systems incorporated shallow knowledge, which sought to define simple cause and effect relationships between a symptom and a fault, that could be encoded as a set of rules. Though such systems enjoyed much success, it was recognised that they suffered from a number of inherent limitations such as inflexibility, inadequate explanation, and difficulties of knowledge elicitation. Many of these limitations can be overcome by developing knowledge-based systems which contain deeper knowledge about the device being diagnosed. Such systems, now generally referred to as model-based systems, have shown much promise, but there has been little evidence to suggest that they have successfully made the transition from the research centre to the workplace. This thesis argues that knowledge-based systems are an appropriate tool for the diagnosis of faults in complex devices, and that both deep and shallow knowledge have their part to play in this process. More specifically this thesis demonstrates how a wide-ranging knowledge-based system for quality assurance, based upon shallow knowledge, can be developed, and implemented. The resultant system, named DIPLOMA, not only diagnoses faults, but additionally provides advice and guidance on the assembly, disassembly, testing, inspection and repair of a highly complex hydro-mechanical device. Additionally it is shown that a highly innovative modelbased system, named MIDAS, can be used to contribute to the provision of diagnostic, explanatory and training facilities for the same hydro-mechanical device. The methods of designing, coding, implementing and evaluating both systems are explored in detail. The successful implementation and evaluation of the DIPLOMA and MIDAS systems has shown that knowledge-based systems are an appropriate tool for the diagnosis of faults in complex hydro-mechanical devices, and that they make a beneficial contribution to the business performance of the host organisation. Furthermore, it has been demonstrated that the most effective and comprehensive knowledge-based approach to fault diagnosis is one which incorporates both deep and shallow knowledge, so that the distinctive advantages of each can be realised in a single application. Finally, the research has provided evidence that the model-based approach to diagnosis is highly flexible, and may, therefore, be an appropriate technique for a wide range of industrial applications. / Science and Engineering Research Council, and Alvey Directorate

<p dir="ltr">The industry's interest in electrified powertrain-equipped vehicles has increased due to environmental and economic reasons. Electrified powertrains, in general, produce lower sound and vibration level than those equipped with internal combustion engines, making noise and vibration (N&V) from other non-engine powertrain components more perceptible. One such N&V type that arouses concern to both vehicle manufacturers and passengers is gear growl, but the signal characteristics of gear growl noise and vibration and the threshold of those characteristics that can be used to determine whether a gear growl requires attention are not yet well understood. This study focuses on developing a method to detect gear-growl based on the N\&V measurements and determining thresholds on various severities of gear-growl using supervised machine learning classification. In general, a machine learning classifier requires sufficient high-quality training data with strong information independence to ensure accurate classification performance. In industrial practices, acquiring high-quality vehicle NVH data is expensive in terms of finance, time, and effort. A physically informed data augmentation method is, thus, proposed to generate realistic powertrain NVH signals based on high-quality measurements which not only provides a larger training data set but also enriches the signal feature variations included in the data set. More specifically, this method extracts physical information such as angular speed, tonal amplitudes distribution, and broadband spectrum shape from the measurement data. Then, it recreates a synthetic signal that mimics the measurement data. The measured and simulated (via data augmentation) are transformed into feature matrix representation so that the N\&V signals can be used in the classification model training process. Features describing signal characteristics are studied, extracted, and selected. While the root-mean-square (RMS) of the vibration signal and spectral entropy were sufficient for detecting gear-growl with a test accuracy of 0.9828, the acoustic signal required more features due to background noise, making data linearly inseparable. The minimum Redundancy Maximum Relevance (mRMR) feature scoring method was used to assess the importance of acoustic signal features in classification. The five most important features based on the importance score were the angular acceleration of the driveshaft, the time derivative of RMS, the tone-to-noise ratio (TNR), the time derivative of the spectral spread of the tonal component of the acoustic signal, and the time derivative of the spectral spread of the original acoustic signal (before tonal and broadband separation). A supervised classification model is developed using a support vector machine from the extracted acoustic signal features. Data used in training and testing consists of steady-state vehicle operations of 25, 35, 45, and 55 mph, with two vehicles with two different powertrain specs: axles with 4.56 and 6.14 gear ratios. The dataset includes powertrains with swapped axles (four different configurations). Techniques such as cost weighting, median filter, and hyperparameter tuning are implemented to improve the classification performance where the model classifies if a segment in the signal represents a gear-growl event or no gear-growl event. The average accuracy of test data was 0.918. A multi-class classification model is further implemented to classify different severities based on preliminary subjective listening studies. Data augmentation using signal simulation showed improvement in binary classification applications. In this study, only gear-growl was used as a fault type. Still, data augmentation, feature extraction and selection, and classification methods can be generalized for NVH signal-based fault diagnosis applications. Further listening studies are suggested for improved classification of multi-class classification applications.</p>

With the fast evolution of the Industry 4.0, the increased use of sensors and the rapid development of the Internet of Things (IoT), and the adoption of artificial intelligence methods, smart factories can automate their processes to vastly improve their efficiency and production quality. However, even the most well cared-for machines develop faults eventually. Given that Prognostics and Health Management (PHM) is an indispensable aspect for proper machine performance, Predictive Maintenance (PdM) is an emerging topic within maintenance methodologies whose aim is to predict failure prior to occurrence with the goal of scheduling maintenance only when needed. As data can be collected faster than ever before, deep learning is an effective tool that can leverage big data for data-driven fault diagnosis methodologies. This thesis explores two different fault diagnosis methodologies associated with predictive maintenance: an anomaly detection using an Autoencoder Neural Network, and a fault classifier using a Convolutional Neural Network (CNN). The system under analysis is a 3phase AC induction motor commonly used in industry. Results show great performance and indicate the viability for the implementation of both methods in production applications. / Med den snabba utvecklingen av industri 4.0, den ökade användningen av sensorer och den snabba utvecklingen av Internet of Things samt införandet av metoder för artificiell intelligens kan smarta fabriker automatisera sina processer för att avsevärt förbättra effektiviteten och produktionskvaliteten. Även de mest välskötta maskinerna utvecklar dock fel så småningom. PHM är en oumbärlig aspekt för korrekt maskinunderhåll. PdM är ett nytt ämne inom underhållsmetodik som syftar till att förutsäga fel innan de inträffar, med målet att planera underhållet endast när det behövs. Eftersom data kan samlas in snabbare än någonsin tidigare är djupinlärning ett effektivt verktyg som kan utnyttja stora datamängder för datadrivna metoder för feldiagnostik. I den här uppsatsen undersöks två olika metoder för feldiagnostik i samband med förebyggande underhåll: en anomalidetektion med hjälp av ett neuralt nätverk med autoencoder och en felklassificering med hjälp av ett CNN. Det system som analyseras är en induktionsmotor med 3fas växelström som ofta används inom industrin. Resultaten visar på goda resultat och visar att det är möjligt att genomföra båda metoderna i produktionstillämpningar.

No description available.

Condition monitoring techniques can be employed to enhance reliability of electric machinery. The stator winding fault is one of the dominant causes for the failure of induction machines. In this work, the condition monitoring of an inverter-fed induction machine using high-frequency signal injection based technique is investigated. Initially, an analytical model of the induction machine with a stator inter-turn fault is developed. Subsequently, the behaviour of the induction machine in the presence of stator inter-turn fault is analyzed using the symmetrical component theory. Because of their use for fault diagnosis purposes, the analytical expressions for the fundamental and high-frequency symmetrical component currents are derived. The high-frequency signal injection is performed by adding a balanced three-phase high-frequency low-magnitude voltage to the fundamental excitation voltage. The resulting high-frequency negative-sequence current component can be used as reliable fault indicator to detect stator inter-turn faults. The effectiveness of the high-frequency negative-sequence current as a fault indicator is compared with the fundamental negative-sequence current, which is one of the traditionally used fault indicators for detecting these faults. The high-frequency signal injection technique proposed in this work is tested experimentally on a prototype machine in a laboratory set-up. The use of the proposed fault indicator is found to be advantageous when compared to the use of the traditional fault indicator for variable-frequency drives. In particular, it is shown that the proposed fault indicator is less dependent from the drive operating conditions than the traditional fault indicator. / Tillståndsövervakning är en teknik som kan användas för att förbättra tillförlitligheten hos elektriska maskiner. För asynkronmaskiner är fel i statorlindningen en av de dominerande orsakerna som leder till problem. I detta arbete undersöks tillståndsövervakning av en omriktarmatad asynkronmotor med hjälp av en högfrekvent signalinjektionsbaserad teknik. Inledningsvis utvecklas en analytisk modell av en asynkronmaskin med korsslutningsfel mellan varven i statorn. Därefter analyseras beteendet hos maskinen med hjälp av teorin för symmetriska komponenter. Analytiska uttryck för både grund- och övertoner härleds för de symmetriska komponenterna. Den högfrekventa signalinjektionen utförs genom att addera en liten högfrekvent trefasspänning till den matningsspänningen. Den resulterande högfrekventa negativa strömkomponenten kan användas som en tillförlitlig indikator för att upptäcka eventuella kortslutningar i statorlindningen. Förmågan som felindikator hos den högfrekventa negativa sekvensströmmen jämförs med den grundläggande negativa strömkomponentens förmåga, vilken är den traditionella indikatorn för att detektera dessa fel. Den högfrekventa signalinjiceringsmetoden som föreslås i detta arbete undersöks experimentellt på en prototypmaskin. Den föreslagna felindikatorn har visat sig vara fördelaktig jämfört med användningen av den traditionella felindikatorn för frekvensomriktare. I synnerhet visas att den föreslagna felindikatorn är mindre beroende av frekvensomriktarens driftsförhållanden än den traditionella felindikatorn.

<p>This thesis considers the problem of fault diagnosis and fault-tolerant control (FTC) for chemical process systems with nonlinear dynamics. The primary objective of fault diagnosis discussed in this work is to identify the failed actuator or sensor by using the information embodied in a process model, as well as input and output data. To this end, an active fault isolation method is first proposed to identify actuator faults and process disturbances by utilizing control action and process nonlinearity. The key idea is to move the process to a region upon fault detection where the effect of each fault can be differentiated from others. The proposed method enables isolation of faults that may not be achievable under nominal operation. This work then investigates the problem of sensor fault isolation by exploiting model-based sensor redundancy through state observer design. Specifically, a high-gain observer is presented and the stability property of the closed-loop system is rigorously established. A method that uses a bank of high-gain observers is then proposed to isolate sensor faults, which explicitly accounts for process nonlinearity, and to continue nominal operation upon fault isolation. In addition to fault diagnosis, this work addresses the problem of handling severe actuator faults using a safe-parking approach and integrating fault diagnosis and safe-parking techniques in a unified fault-handling framework. In particular, several practical issues are considered for the design and implementation of safe-parking techniques, including changes in process dynamics, the network structure of a chemical plant, and actuators frozen at arbitrary positions. The advantage of this approach is that it enables stable process operation under faulty conditions, avoiding the partial or entire shutdown of a chemical plant and resulting economic losses. The efficacy of the proposed fault diagnosis and FTC methods is demonstrated through numerous simulations of chemical process examples.</p> / Doctor of Philosophy (PhD)

A number of non-invasive fault detection and diagnosis (FDD) techniques have been researched and have proven to have worked well in classifying faults in internal combustion engines (ICE) and other mechanical and electrical systems. These techniques are an integral step to creating more robust and accurate methods of determining where or how a fault has or will occur in such systems. These FDD techniques have the potential to not only save time avoiding a tear-down of a costly machine, but could potentially add another layer of safety in detecting and diagnosing a fault much earlier than was possible before. Looking at the previous research methods and the systems they used to acquire this data, it is a natural progression to try and make a system which is able to encapsulate all of these ideologies into one inexpensive module capable of integrating itself into the advanced set of FDD. This thesis follows along with the development of a new wireless sensor that is developed specifically for the use in FDD for ICE and other mechanical systems. A new set of software and firmware is created for the system to be able to be incorporated into previously designed algorithms. After creating and manufacturing the sensor it is put to the test by incorporating it into several Artificial Neural Networks (ANN) and comparing the results to previous experiments done with previous research equipment. Using vibration data acquired from a running engine to train a neural network, the wireless sensor was able to perform equally as well as its expensive counter parts. It proved to have the ability to achieve 100% accuracy in classifying specific engine faults. The performance of three ANN training algorithms, Levenberg-Marquardt (LM), extended Kalman Filter (EKF), and Smooth Variable Structure filter (SVSF), were tested and compared. Adding to the feasibility of a standalone system the wireless sensor was tested in a live environment as a method of instant ICE fault detection. / Thesis / Master of Applied Science (MASc)

[ES] Las pilas de combustibles son dispositivos de un coste elevado y frágiles ante ambientes contaminados o condiciones inadecuadas de operación como: temperaturas extremas o mala gestión del agua producida como residuo de la pila. Para mejorar la fiabilidad de una pila de combustible es necesario diagnosticar de una manera oportuna los fallos y así evitar daños que reduzcan el desempeño del módulo o que lo inhabiliten. Este trabajo busca contribuir al mejoramiento de la fiabilidad de las pilas de combustible de baja temperatura y de esta forma favorecer el uso de hidrógeno en la transición a una energía descarbonizada. Para lograrlo, se realizaron tres actividades principales: modelado de una pila de hidrógeno, ajuste paramétrico del modelo desarrollado y, por último, aplicación de técnicas de diagnóstico de fallos basados en modelos. En el laboratorio de Recursos Energéticos Renovables Distribuidos LabDER de la Universitat Politècnica de València, se estudia el desempeño de sistemas híbridos renovables, incluyendo una línea de hidrógeno, desde la producción, almacenamiento y reconversión en electricidad en una pila de combustible, por tanto, se ha podido validar el modelo. En un primer momento se identificó la necesidad de un modelo que emplee la temperatura como señal de salida y que retroalimente el sistema, y que tuviese en cuenta señales propias del módulo comercial; sin embargo, el uso de la temperatura como señal y la no linealidad de las ecuaciones físicas, químicas, eléctricas y empleadas, generan un modelo altamente complejo. El ajuste paramétrico del modelo se realizó empleando algoritmos de optimización. Tomando como base al algoritmo de Enjambre de Partículas, se desarrolló una nueva propuesta llamada Scout GA, este algoritmo fue utilizado en otras aplicaciones y pruebas de convergencia para verificar su desempeño frente al fenómeno de estancamiento prematuro y logrando mejorar la precisión y velocidad de convergencia de otras propuestas. Como resultado de la validación de este modelo, en una primera simulación usando datos reales de funcionamiento correspondientes a 1500 segundos, el error de simulación fue del 2,21% en la señal de tensión y del 1,97% en la señal de temperatura, obteniendo un error medio del 2,09%. En un segundo conjunto de datos de algo más de 2.500 segundos de funcionamiento, el error de simulación fue del 2,40% y del 1,96% para las señales de tensión y temperatura, respectivamente. Se estima que el error medio de simulación para ambas señales y condiciones de funcionamiento similares es inferior al 2,5%. Buscando mejorar la fiabilidad de la pila, se realizó el trabajo de diagnóstico de fallos, este partió de la simulación de fallos, mediante la modificación de algunas señales de entrada del modelo, los fallos se caracterizaron mediante el tratamiento estadístico de 12 residuos, obteniendo firmas de fallos, que, en su conjunto, formaron una matriz de fallos. Luego, un algoritmo de diagnóstico propuesto permitió identificar y aislar 14 fallos. permitiendo concluir que, el modelo predice eficazmente los fallos de las pilas PEMFC y podría extrapolarse a otras pilas de combustible. / [CA] Les piles de combustibles són dispositius d'un cost elevat i fràgils davant ambients contaminats o condicions inadequades d'operació com: temperatures extremes o dolenta gestió de l'aigua produïda com a residu de la pila. Per a millorar la fiabilitat d'una pila de combustible és necessari diagnosticar d'una manera oportuna les fallades i així evitar danys que reduïsquen l'acompliment del mòdul o que l'inhabiliten. Este treball busca contribuir al millorament de la fiabilitat de les piles de combustible de baixa temperatura i d'esta manera afavorir l'ús d'hidrogen en la transició a una energia *descarbonizada. Per a aconseguir-ho, es van realitzar tres activitats principals: modelatge d'una pila d'hidrogen, ajust paramètric del model desenvolupat i, finalment, aplicació de tècniques de diagnòstic de fallades basades en models. En el laboratori de Recursos Energètics Renovables Distribuïts *LabDER de la Universitat Politècnica de València, s'estudia l'acompliment de sistemes híbrids renovables, incloent-hi una línia d'hidrogen, des de la producció, emmagatzematge i reconversió en electricitat en una pila de combustible, per tant, s'ha pogut validar el model. En un primer moment es va identificar la necessitat d'un model que empre la temperatura com a senyal d'eixida i que retroalimente el sistema, i que tinguera en compte senyals propis del mòdul comercial, no obstant això, l'ús de la temperatura i la no linealitat de les equacions físiques, químiques, elèctriques i tèrmiques empleades, deriven en un model altament complex. L'ajust paramètric del model de pila de combustible es va realitzar emprant algorismes d'optimització. Prenent com a base a l'algorisme d'Eixam de Partícules, es va desenvolupar una nova proposta anomenada Scout GA, aquest algorisme va ser utilitzat en altres aplicacions i proves de convergència per a verificar el seu acompliment enfront del fenomen d'estancament prematur i aconseguint millorar la precisió i velocitat de convergència d'altres propostes. La simulació i identificació del model té un cost computacional entre 7 i 20 ms per iteració, on es van aconseguir errors de simulació menors al 2.5% Com a resultat de la validació d'aquest model, en una primera simulació usant dades reals de funcionament corresponents a 1500 segons, l'error de simulació va ser del 2,21% en el senyal de tensió, del 1,97% en el senyal de temperatura i un error mitjà del 2,09%. En un segon conjunt de dades d'una mica més de 2.500 segons de funcionament, l'error de simulació va ser del 2,40% i del 1,96% per als senyals de tensió i temperatura, respectivament. S'estima que l'error mitjà de simulació per a tots dos senyals i condicions de funcionament similars és inferior al 2,5%. Buscant millorar la fiabilitat de la pila, es va fer el treball de diagnòstic de fallades, aquest va partir de la simulació de fallades, mitjançant la modificació d'alguns senyals d'entrada del model, les fallades es van caracteritzar mitjançant el tractament estadístic de 12 residus, obtenint signatures de fallades, que en el seu conjunt, van formar una matriu de fallades. després un algorisme de diagnòstic proposat, va permetre identificar i aïllar 14 fallades. Permetent concloure que, el model prediu eficaçment les fallades de les piles PEMFC i podria extrapolar-se a altres piles de combustible. / [EN] Fuel cells are high-cost devices that are fragile in contaminated environments or in inadequate operating conditions, such as extreme temperatures or poor water management, produced as battery waste. To improve the reliability of a fuel cell, it is necessary to diagnose failures promptly and thus avoid damage that reduces the module's performance or disables it. This work seeks to contribute to improving the reliability of low-temperature fuel cells and thus promote the use of hydrogen in the transition to decarbonized energy. To achieve this, three main activities were carried out: modeling a hydrogen fuel cell, parametric adjustment of the developed model, and application of model-based fault diagnosis techniques. In the LabDER Distributed Renewable Energy Resources laboratory of the Polytechnic University of Valencia, the performance of renewable hybrid systems is studied, including a hydrogen line, from production, storage, and reconversion into electricity in a fuel cell, therefore, has been able to validate the model. Initially, a fuel cell model that uses temperature as an in/output signal is required. Also, the model must be able to use the reals signals supplied for the commercial module. However, using temperature and an equation set that includes the non-linearity of the physical, chemical, electrical, and thermal equations resulted in a highly complex model. The parametric adjustment of the fuel cell model was performed using optimization algorithms. Based on the Particle Swarm algorithm, a new proposal called Scout GA was developed. This algorithm was used in other applications and convergence tests to verify its performance against the premature stagnation phenomenon and improved the accuracy and speed of convergence of other proposals. The simulation and identification of the model have a computational cost between 7 and 20 ms per iteration, where simulation errors of less than 2.5% were achieved. As a result of the validation of this model, in a first simulation using real operating data corresponding to 1,500 seconds, the simulation error was 2.21% for the voltage signal, 1.97% for the temperature signal, and an average error of 2.09%. In a second data set for slightly more than 2500 seconds of operation, the simulation error was 2.40% and 1.96% for the voltage and temperature signals, respectively. The average simulation error for both signals and similar operating conditions is estimated to be less than 2.5%. To improve the reliability of the stack, the fault diagnosis work was carried out, starting from the simulation of faults by modifying some input signals of the model; the faults were characterized by the statistical treatment of 12 residuals, obtaining fault signatures, which formed a fault matrix. Then, a proposed diagnostic algorithm allowed to identify and isolate 14 faults. Allowing to conclude that the model effectively predicts the PEMFC stack faults and could be extrapolated to other fuel cells. / Ariza Chacón, HE. (2024). Contribuciones al modelado y diagnóstico de fallos en PEMFC para mejorar la fiabilidad en sistemas híbridos renovables [Tesis doctoral]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/203614

Condition Monitoring (CM) is widely used in industry to meet sustainability, safety, and equipment efficiency requirements. Intelligent Fault Diagnosis (IFD) research focuses on automating CM data analysis tasks, to detect and prevent machine faults, and provide decision support. IFD enables trained analysts to focus their efforts on advanced tasks such as fault severity estimation and preventive maintenance optimization, instead of performing routine tasks. Industry datasets are rarely labelled, and IFD models are therefore typically trained on labelled data generated in laboratory environments with artificial or accelerated fault development. In the process industry, fault characteristics are often context-dependent and difficult to predict in sufficient detail due to the heterogeneous environment of machine parts. Furthermore, fault development is non-linear and measurements are subject to varying background noise. Thus, IFD models trained on lab data are not expected to transfer well to process industry environments, and require on-site pre-training or fine-tuning to facilitate accurate and advanced fault diagnosis. While ground truth labels are absent in industrial CM datasets, analysts sometimes write annotations of faults and maintenance work orders that describe the fault characteristics and required actions. These annotations deviate from typical natural language due to the technical language used, characterised by a high frequency of technical terms and abbreviations. Recent advances in natural language processing have enabled simultaneous learning from unlabelled pairs of images and captions through Natural Language Supervision (NLS). In this thesis, opportunities to enable weakly supervised IFD using annotated but otherwise unlabelled CM data are investigated. This thesis proposes novel machine learning methods for joint representation learning for IFD directly on annotated CM data. The main contributions are: (1) the introduction and implementation of technical language supervision to merge advances in natural language processing and, including a literature survey; (2) the creation of a method to improve technical languageprocessing by substituting out-of-vocabulary technical words with natural language descriptions, and to evaluate language model performance without explicit labels or downstream tasks; (3) the creation of a method for small-data language-based fault classification using human-centricvisualisation and clustering. Preliminary results for sensor and cable fault detection show an accuracy of over 90%. These results imply a considerable increase in the value of annotated CM datasets through the implementation of IFD models directly on industry data, e.g. for improving the decision support to avoid unplanned stops. / KnowIT FAST