Deep Learning Networks (DLN) is a relatively new artificial intelligence algorithm that gained popularity quickly due to its unprecedented performance. One of the key elements for this success is DL’s ability to extract a high-level of information from large amounts of raw data. This ability comes at the cost of high computational and memory requirements for the training process. Estimation algorithms such as the Extended Kalman Filter (EKF) and the Smooth Variable Structure Filter (SVSF) are used in literature to train small Neural Networks. However, they have failed to scale well with deep networks due to their excessive requirements for computation and memory size. In this thesis the concept of using EKF and SVSF for DLN training is revisited. A New family of filters that are efficient in memory and computational requirements are proposed and their performance is evaluated against the state-of-the-art algorithms. The new filters show competitive performance to existing algorithms and do not require fine tuning. These new findings change the scientific community’s perception that estimation theory methods such as EKF and SVSF are not practical for their application to large networks. A second contribution from this research is the application of DLN to Fault Detection and Diagnosis. The findings indicate that DL can analyze complex sound and vibration signals in testing of automotive starters to successfully detect and diagnose faults with 97.6% success rates. This proves that DLN can automate end-of-line testing of starters and replace operators who manually listen to sound signals to detect any deviation. Use of DLN in end-of-line testing could lead to significant economic benefits in manufacturing operations. In addition to starters, another application considered is the use of DLN in monitoring of the State-Of-Charge (SOC) of batteries in electric cars. The use of DLN for improving the SOC prediction accuracy is discussed. / Thesis / Doctor of Science (PhD) / There are two main ideas discussed in this thesis, both are related to Deep Learning (DL). The first investigates the use of estimation theory in DL network training. Training DL networks is challenging as it requires large amounts of data and it is computationally demanding. The thesis discusses the use of estimation theory for training of DL networks and its utility in information extraction. The thesis also presents the application of DL networks in an end-of-line Fault Detection and Diagnosis system for complex automotive components. Failure of appropriately testing automotive components can lead to shipping faulty components that can harm a manufacturer’s reputation as well as potentially jeopardizing safety. In this thesis, DL is used to detect and analyze complex fault patterns of automotive starters, complemented by sound and vibration measurements.

Unmanned Underwater Vehicles (UUVs) are widely used in commercial, scientific, and military missions for various purposes. What makes this technology challenging is the increasing mission duration and unknown environment. It is necessary to embed fault-tolerant control paradigms into UUVs to increase the reliability of the vehicles and enable them to execute and finalize complex missions. Specifically, fault-tolerant control (FTC) comprises fault detection, identification, and control reconfiguration for fault compensation. Literature review shows that there have been no systematic methods for fault-tolerant control of UUVs in earlier investigations. This study presents a hierarchical methodology of fault detection, identification and compensation (HFDIC) that integrates these functions systematically in different levels. The method uses adaptive finite-impulse-response (FIR) modeling and analysis in its first level to detect failure occurrences. Specifically, it incorporates a FIR filter for on-line adaptive modeling, and a least-mean-squares (LMS) algorithm to minimize the output error between the monitored system and the filter in the modeling process. By analyzing the resulting adaptive filter coefficients, we extract the information on the fault occurrence. The HFDIC also includes a two-stage design of parallel Kalman filters in levels two and three for fault identification using the multiple-model adaptive estimation (MMAE). The algorithm activates latter levels only when the failure is detected, and can return back to the monitoring loop in case of false failures. On the basis of MMAE, we use multiple sliding-mode controllers and reconfigure the control law with a probability-weighted average of all the elemental control signals, in order to compensate for the fault. We validate the HFDIC on the steering and diving subsystems of Naval Postgraduate School (NPS) UUVs for various simulated actuator and/or sensor failures, and test the hierarchical fault detection and identification (HFDI) with realistic data from at-sea experiment of the Florida Atlantic University (FAU) Autonomous Underwater Vehicles (AUVs). For both occasions, we model actuator and sensor failures as additive parameter changes in the observation matrix and the output equation, respectively. Simulation results demonstrate the ability of the HFDIC to detect failures in real time, identify failures accurately with a low computational overhead, and compensate actuator and sensor failures with control reconfiguration. In particular, verification of HFDI with FAU data confirms the performance of the fault detection and identification methodology, and provides important information on the vehicle performance. / Ph. D.

A non-invasive, on-line method for detection of mechanical (rotor, bearings eccentricity) and stator winding faults in a 3-phase induction motors from observation of motor line current supply input. The main aim is to avoid the consequence of unexpected failure of critical equipment which results in extended process shutdown, costly machinery repair, and health and safety problems. This thesis looks into the possibility of utilizing machine learning techniques in the field of condition monitoring of electromechanical systems. Induction motors are chosen as an example for such application. Electrical motors play a vital role in our everyday life. Induction motors are kept in operation through monitoring its condition in a continuous manner in order to minimise their off times. The author proposes a model free sensor-less monitoring system, where the only monitored signal is the input to the induction motor. The thesis considers different methods available in literature for condition monitoring of induction motors and adopts a simple solution that is based on monitoring of the motor current. The method proposed use the feature extraction and Support Vector Machines (SVM) to set the limits for healthy and faulty data based on the statistical methods. After an extensive overview of the related literature and studies, the motor which is the virtual sensor in the drive system is analysed by considering its construction and principle of operation. The mathematical model of the motor is used for analysing the system. This is followed by laboratory testing of healthy motors and comparing their output signals with those of the same motors after being intentionally failed, concluding with the development of a full monitoring system. Finally, a monitoring system is proposed that can detect the presence of a fault in the monitored machine and diagnose the fault type and severity / Ministry of Higher Education, Libya; Switchgear & Instruments Ltd.

The utilization of driverless forklifts necessitates stringent safety measures to prevent any harm to human or material involved in their operation. This thesis addresses the critical need for collision detection algorithms for driverless forklifts, particularly in scenarios where traditional sensors are obstructed during loading and unloading processes. Instead of relying on external sensors, this research focuses on utilizing the internal sensors already present in the forklift. Signals from the forklift were collected during various driving scenarios in a controlled lab environment. Five different algorithms were developed and evaluated, providing detailed insights into their strengths and limitations. These algorithms employ a range of techniques, including physical modeling, regression modeling, residual analysis, and machine learning classification. All five algorithms demonstrate notable accuracy and reliability in collision detection. The research contributes to the advancement of collision detection technology in industrial environments, offering practical insights for safer and more productive material handling operations.

L'objectif de cette thèse est de développer une nouvelle méthodologie pour la détection active de défaillances, basée sur approche multimodèle et robuste des fautes. Ce travail prolonge des recherches effectuées dans le projet Metalau de l'Inria. L'apport essentiel de cette thèse est la prise en compte de modèles évoluant en boucle fermée. On utilise une approche multi-modèle pour modéliser le modèle en fonctionnement normal et le modèle défaillant. Les avantages potentiels de l'utilisation d'un feedback dynamique linéaire et ses propriétés de robustesse sont analysés dans la construction de signaux de détection auxiliaires. On compare les résultats obtenus avec ceux du cas boucle ouverte. La formulation du problème de détection active dans le cas d'un modèle en boucle fermée est nouvelle et repose sur la prise en considération de la norme du signal de détection auxiliaire comme critère d'optimisation. On considère aussi des fonctions coût plus générales, telles celles qui sont utilisées pour mesurer la performance de feedbacks dans des problèmes de la théorie de la commande linéaire robuste. La solution complète repose sur la résolution de plusieurs problèmes d'optimisation non standards / The aim is to develop a novel theory of robust active failure detection based on multi-model formulation of faults. The original method was already proposed by the Metalau group of INRIA. We have continued to work on the extension of this approach to more general cases. The focus is on the effects of feedback on the previous approach. The multi-model approach is still used to model the normal and the failed systems; however the possible advantages of using linear dynamic feedback in the construction of the auxiliary signal for robust fault detection is considered and the results are compared to the previously developed open-loop setup. An original formulation of the active fault detection problem using feedback is developed. The norm of the auxiliary signal is considered as a possible cost criterion. Also, we have considered a more general cost function that has already been used for measuring the performance of feedback configurations in Linear Control Theory. We have given a complete solution to this problem. In order to find a complete solution, several mathematical problems are solved

<p>Rad je posvećen problemu rane i brze detekcije i izolacije neželjenih stanja dinamičkih sistema, sa posebnim naglaskom na rano otkrivanje različitih nepravilnosti u radu i kvarova industrijskih procesa.</p> / <p>The thesys is dedicated to the problem of early and swift detection and isolation of unwanted working regimes of dynamical systems, with particular emphasis on the early detection of various irregularities and failures of industrial processes.</p>

Ce mémoire de thèse présente une étude fondamentale enrichie par des contributions qui sont articulées autour de la modélisation de processus ainsi qu'un diagnostic de défauts en utilisant l'analyse en composantes principales (ACP). Dans l'objectif d'un choix optimal du modèle ACP, une étude comparative de quelques critères connus dans la littérature nous a permis de conclure que le problème rencontré est souvent lié à une ignorance des variables indépendantes et quasi-indépendantes. Dans ce cadre, nous avons réalisé deux démonstrations mettant en évidence les limitations de deux critères en particulier la variance non reconstruite (VNR). En s'appuyant sur le principe d'une telle variance, nous avons proposé trois nouveaux critères. Parmi eux, deux ont été considérés comme étant empiriques car seule l'expérience permettra de prouver leur efficacité. Le troisième critère noté VNRVI représente un remède à la limitation du critère VNR. Une étude de sa consistance théorique a permis d'établir les conditions garantissant l'optimalité de son choix. Les résultats de simulation ont validé une telle théorie en prouvant ainsi que le critère VNRVI étant plus efficace que ceux étudiés dans cette thèse. Dans le cadre d'un diagnostic de défauts par ACP, l'approche de reconstruction des indices de détection ainsi que celle des contributions ont été utilisées. A travers une étude de généralisation, nous avons étendu le concept d'isolabilité de défauts par reconstruction à tout indice quadratique. / This thesis presents a fundamental study enhanced by some contributions that are focused on process modelling and fault diagnosis using principal components analysis (PCA). In order to find an optimal PCA model, we have concluded through a comparative study of some popular criteria that the problem is often related to an ignorance of the independent and quasi-independent variables. In this framework, we have performed two demonstrations highlighting the limitations of two selection criteria in particular the unreconstructed variance (VNR). Based on the principle of VNR approach, we have proposed three new criteria, among them two methods were considered as empirical criteria because only the experience will prove their effectiveness. However the third one which is noted VNRVI represents a cure for the limitation of the classical VNR criterion. Thus, the conditions that ensure an optimal selection were derived according to a theoretical consistency study of the VNRVI approach. The simulation results have successfully validated the VNRVI criterion by proving that is more effective than the other studied criteria in the present thesis. The reconstruction and contributions approaches were used for fault diagnosis using PCA. According to a unified study, we have extended the fault isolability concept based on the reconstruction method to any detection index which has a quadratic form. Such generalization has allowed us to develop a theoretical fault isolability analysis based on the reconstruction of the combined index versus those of SPE and T2 indices. This analysis has highlighted the advantage of using the combined index for fault isolation.

Dans cette thèse, on s'est intéressé à trois problématiques des réseaux de capteurs sans fil (WSN). Dans un premier temps nous avons analysé l'impact de la chute de tension dans la batterie du nœud sur la puissance du signal en réception. On propose alors une méthode pour compenser l'augmentation apparente de la distance calculée entre les nœuds due à la diminution de l'énergie de la batterie. Pour les nœuds passant par deux états principaux endormi et actif, on propose d'étudier, la relation entre la diminution de la tension de la batterie en fonction du temps passé par un nœud dans l'état actif. Ensuite, on calcule le rapport entre la RSS et la distance entre les nœuds connectés avec des batteries complètement chargées. Après on mesure la RSS en faisant varier la tension de la batterie du nœud émetteur et en gardant le nœud récepteur à une distance constante. Finalement, on propose une relation entre la RSS observée et la tension actuelle de la batterie du nœud émetteur. Cette fonction permet de calculer la valeur corrigée de la RSS qui correspond à la distance réelle entre les nœuds connectés. Ainsi l'efficacité des méthodes de la localisation basée sur la RSS se trouvent améliorées. Dans la deuxième partie de cette thèse on propose une méthode d'estimation des positions des nœuds dans un WSN. Dans l'algorithme de localisation proposé, on utilise des nœuds ancres comme des points de référence. On a utilisé une approche heuristique pour trouver la topologie relative avec l'aide de la matrice de distance. Le but de la matrice de distance est d'indiquer s'il existe une connexion entre une paire de nœuds donnée et en cas de connectivité, la distance estimée entre ces nœuds. En utilisant les informations de connectivité entre les nœuds et leurs distances, on obtient la topologie du réseau. La méthode proposée utilise la solution de l'intersection de deux cercles au lieu de la méthode classique de triangulation, où un système quadratique de trois équations avec deux variables est utilisé ce qui rend la complexité de calcul augmentée. Lorsque deux nœuds connectés ont un autre nœud en commun, puis en utilisant les informations de distances entre ces nœuds interconnectés, nous pouvons calculer deux positions possibles pour le troisième nœud. La présence ou l'absence d'un lien entre le troisième nœud et un quatrième nœud, permet de trouver la position précise. Ce processus est réitéré jusqu'à ce que toutes les positions des nœuds aient été obtenues. Une fois la topologie relative calculée, il faut trouver la symétrie, l'orientation et la position de cette topologie dans le plan. C'est à ce moment que la connaissance des positions des trois nœuds entre en action. La topologie donne les coordonnées temporaires des nœuds. En ayant une comparaison de certaines caractéristiques entre les coordonnées temporaires et les coordonnées exactes, on trouve d'abord la symétrie de la topologie relative qui correspondrait à la topologie originale. En d'autres termes on vérifie si oui ou non la topologie relative est une image miroir de la topologie originale. Des opérateurs géométriques sont alors utilisés pour corriger la topologie relative par rapport à la topologie réelle. Ainsi, on localise tous les nœuds dans un WSN en utilisant exactement trois ancres. Dans la dernière partie de cette thèse, on propose une méthode pour la détection de défauts dans un WSN. Il y a toujours une possibilité qu'un capteur d'un nœud ne donne pas toujours des mesures précises. On utilise des systèmes récurrents et non récurrents pour la modélisation et on prend comme variable d'entrée, en plus des variables du nœud en question, les informations des capteurs voisins. La différence entre la valeur estimée et celle mesurée est utilisée pour déterminer la possibilité de défaillance d'un nœud / In this thesis three themes related to wireless sensor networks (WSNs) are covered. The first one concerns the power loss in a node signal due to voltage droop in the battery of the node. In the first part of the thesis a method is proposed to compensate for the apparent increase in the calculated distance between the related nodes due to decrease in the energy of the signal sending node battery. A function is proposed whose arguments are the apparently observed RSS and the current voltage of the emitter node battery. The return of the function is the corrected RSS that corresponds to the actual distance amongst the connected nodes. Hence increasing the efficiency of the RSS based localization methods in WSNs. In the second part of the thesis a position estimation method for localization of nodes in a WSN is proposed. In the proposed localization algorithm anchor nodes are used as landmark points. The localization method proposed here does not require any constraint on the placement of the anchors; rather any three randomly chosen nodes can serve as anchors. A heuristic approach is used to find the relative topology with the help of distance matrix. The purpose of the distance matrix is to indicate whether or not a pair of nodes has a connection between them and in case of connectivity it gives the estimated distance between the nodes. By using the information of connectivity between the nodes and their respective distances the topology of the nodes is calculated. This method is heuristic because it uses the point solution from the intersection of two circles instead of conventional triangulation method, where a system of three quadratic equations in two variables is used whereby the computational complexity of the position estimation method is increased. When two connected nodes have another node in common, then by using the information of distances between these interconnected nodes, two possible positions are calculated for the third node. The presence or absence of a connection between the third node and a fourth node helps in finding the accurate possibility out of the two. This process is iterated till all the nodes have been relatively placed. Once the relative topology has been calculated, we need to find the exact symmetry, orientation, and position of this topology in the plane. It is at this moment the knowledge of three nodes positions comes into action. From the relative topology we know the temporary coordinates of the nodes. By having a comparison of certain characteristics between the temporary coordinates and the exact coordinates; first the symmetry of relative topology is obtained that would correspond to the original topology. In other words it tells whether or not the relative topology is a mirror image of the original topology. Some geometrical operators are used to correct the topology position and orientation. Thus, all the nodes in the WSN are localized using exactly three anchors. The last part of the thesis focuses on the detection of faults in a WSN. There is always a possibility that a sensor of a node is not giving accurate measurements all of the time. Therefore, it is necessary to find if a node has developed a faulty sensor. With the precise information about the sensor health, one can determine the extent of reliance on its sensor measurement. To equip a node with multiple sensors is not an economical solution. Thus the sensor measurements of a node are modeled with the help of the fuzzy inference system (FIS). For each node, both recurrent and non-recurrent systems are used to model its sensor measurement. An FIS for a particular node is trained with input variables as the actual sensor measurements of the neighbor nodes and with output variable as the real sensor measurements of that node. The difference between the FIS approximated value and the actual measurement of the sensor is used as an indication for whether or not to declare a node as faulty

De par l’augmentation de leur utilisation, la sécurisation du vol des drones devient de plus en plus importante. La commande tolérante aux fautes peut alors contribuer à l’obtention d’un niveau de sécurité acceptable. Le but de cette thèse est de développer une méthode de commande tolérante aux fautes basée sur deux types d’approches : l’approche Automatique qui utilise une représentation de systèmes à l’aide de modèles décrivant des évolutions continues et l’approche Intelligence Artificielle qui se base sur la représentation de systèmes à l’aide de modèles discrets ou logiques. Ainsi la première contribution de cette recherche est le développement d'une méthode générique de commande tolérante aux fautes utilisant les cadres de modélisation discret et continu. L’idée consiste à combiner une modélisation continue permettant d’estimer l’état et les paramètres de fautes et une modélisation discrète permettant de prendre une décision en ligne quant au contrôleur à utiliser. L’estimation continue permet d’avoir plus d’informations sur la faute qu’avec une modélisation discrète, alors que celle-ci prend en compte des probabilités de panne et des techniques d’optimisation qui sont plus adaptées à la tâche de décision. La seconde contribution concerne le développement et la validation d’une méthode permettant de détecter et de diagnostiquer la faute. Pour ses avantages, l’idée a été de développer un filtre de Kalman sensibles aux sauts de panne pour l’estimation de l’état et des paramètres de fautes. Pour la détection et le diagnostic de la panne, l’idée a été d’utiliser les données de l’estimation de façon probabiliste. Une fois la faute détectée et identifiée, le système de commande doit réagir pour pouvoir compenser cette faute. La troisième contribution porte donc sur l’amélioration du suivi de la trajectoire par reconfiguration du système de commande. L’objectif est de combiner les méthodes de commutation et d’adaptation, afin de limiter le nombre de contrôleurs en utilisant des contrôleurs adaptatifs pour les modes dégradés, tout en ayant des contrôleurs faciles à concevoir. Des techniques d’optimisation sont alors utilisées de façon à prendre une décision en ligne quant au choix du contrôleur. Finalement, la méthode développée doit être vérifiée avant de pouvoir être implémentée sur un drone. La dernière contribution est l’évaluation de la capacité de la méthode à suivre une trajectoire d’atterrissage en cas de pannes capteurs ou actionneurs grâce à un modèle de drone. / Major security risks appear with the increase of the number of UAV in the air space. Thus, UAV security is more and more important and Fault Tolerant Control (FTC) methods could support the achievement of acceptable security level. The aims of this research is to develop a FTC method which combines two approaches : Automatic Control approach which is based on model which have a continuous representation of the system and Artificial Intelligence approach which is based on discrete or logical model to represent the system. Thus, the first contribution of this thesis is the development of a generic fault tolerant control method which uses discrete and continuous frameworks. The idea was to combine a continuous framework to estimate the state and fault parameters and a discrete framework to take on line a decision about the controller. The continuous estimation provides more knowledge on the fault whereas a discrete model allows the use of different optimization tools which are more adapted to decision task. The second contribution is the development and the validation of a method for fault detection and diagnosis. For its potential, a Kalman filter is adapted in order to be sensitive to abrupt faults and used for state and fault parameters estimation. These estimates are then used in a probabilistic way to detect and identify the fault. Once the fault is detected, the control system should react to compensate the fault. Thus, the third contribution of this thesis is the improvement of the trajectory tracking by reconfiguration of the control system. The aim is to combine switching and adaptive methods in order to limit the number of controllers by using adaptive controllers for degraded modes while having convenient controllers. Optimization tools are then used to take the decision on the controller to use. Finally, the method has to be validated before being implemented on line. The last contribution is the evaluation of the ability of the method to follow its trajectory despite the apparition of actuator or sensor faults during a landing approach.

[pt] A interpretação de falhas sísmicas é uma tarefa complexa e trabalhosa, que está sujeita à experiência do geólogo. Normalmente ela é auxiliada pela análise de atributos sísmicos, que podem não ser suficientes para uma clara visualização das falhas. Este trabalho realiza uma análise dos métodos atuais que utilizam ACO para o realce de atributos de falha, de forma a entender a contribuição de cada etapa para o resultado. Com base nessa análise, um novo método é proposto, o qual elimina as fraquezas encontradas de forma a buscar uma convergência mais estável e rápida ao resultado. / [en] The interpretation of seismic faults is a complex and labourious task, which is dependent on the experience of the geologist. The interpretation is normally aided by seismic attributes. However, they may not be enough for a clear visualization nor to be used in automatic extraction methods. This dissertation accomplishes an examination of the state of the art ACO algorithms for fault enhancement. This study reveals the importance, contributions and weaknesses of each step of these methods. From there, a new method is proposed, which eliminates some of the problems found, acquiring a more stable and quick convergence of the end result.