This thesis project investigates approaches for malfunction prediction using unsupervised, self-organized models, with an orientation on bus fleets. Certain bus malfunctions are not predictable with conventional methods and preventive replacements are too costly and time consuming. Malfunctions that could result in interruption of service or on degradation of safety  are of high priority to predict.The settings of the desired application define the following constraints: definition of a model by an expert is not desirable as it is not a scalable solution, ambient conditions or usage schedule must not affect the prediction, data communication between the systems is limited so data must be compressed with relevant to the problem features. In this work, definition of normal or faulty operation is not handled by an expert, but using the Wisdom of the crowd idea and Consensus Self-organized models for fault detection (COSMO), or by the system's past state by monitoring an autoencoder's reconstruction error. In COSMO each system constructs a model describing its condition and then all distances between models are estimated to find the Most Central Pattern (MCP), which is considered the normal state of the system. The measure of deviation is the tendency of a system's model to be farther from the MCP after a sequence of observations, expressed as a probability that the deviation is incidental.  Factors that apply to the total of systems, such as the weather conditions are thus minimized.The algorithms approach the problem from the scopes of: linear and non linear relations between signals, distribution of values of a single signal, spectrum information of a single signal. This is achieved by constructing relevant models of each observed system (bus). The performance of the implemented algorithms is investigated using ROC curves and real bus fleet data, targeting at predicting a set of malfunctions of the air pressure system.More tests are performed using artificial data with injected malfunctions, to evaluate the performance of the methods. By applying the method on artificial data, the ability of different methods to detect different malfunctions is exhibited.

Recent refinements in Industry 4.0 and Machine Learning demonstrate the positive effects of using deep learning models for intelligent maintenance. The primary benefit of Deep Learning (DL) is its capability to extract attributes and make fast, accurate, and automated predictions without supervision. However, DL requires high computational power, significant data preprocessing, and vast amounts of data to make accurate predictions for intelligent maintenance. Given the considerable obstacles, meta-learning has been developed as a novel way to overcome these challenges. As a learning technique, meta-learning aims to quickly acquire knowledge of new tasks using theminimal available data by learning through meta-knowledge. There has been less research in the area of using meta-learning for Predictive Maintenance (PdM) and we considered it necessary to conduct this review to understand the applicability of meta-learning’s capabilities and functions to PdM since the outcomes of this technique seem to be rather promising. The review started with the development of a methodology and four research questions: (1) What is the taxonomy of meta-learning for PdM?, (2) What are the current state-of-the-art methodologies? (3) Which datasets are available for meta-learning in PdM?, and (4) What are the open issues, challenges, and opportunities of meta-learning in PdM?. To answer the first and second questions, a new taxonomy was proposed and meta-learnings role in predictive maintenance was identified from selected 55 papers. To answer the third question, we determined which types of datasets and their characteristics exist for this domain. Finally, the challenges, open issues, and opportunities of meta-learning in predictive maintenance were examined to answer the final question. The results of the research questions provided suggestions for future research topics.

Ce travail de thèse concerne le développement d'une méthode de pronostic de défaillance des systèmes de production en série. Une méthode de génération d'un indice de santé brut à partir d'un tenseur de données, appelée Méthode des Points Significatifs a été développée puis validée sur un exemple d'illustration. L'indice généré est ensuite traité par une nouvelle méthode appelée méthode des percentiles, qui permet de générer des profils monotones à partir d'un indice de santé brut. Les profils générés sont ensuite modélisés par un processus Gamma, et la fonction de densité de probabilité agrégée introduite dans ce travail a permis d'estimer le temps de vie résiduel (Remaining Useful Life (RUL)) dans un intervalle de confiance qui assure une marge de sécurité à l'utilisateur industriel. La méthode proposée est appliquée avec succès sur des données expérimentales issues des équipements de production industrielle. / This thesis develops a fault prognosis approach for Discrete Manufacturing Processes. A method of raw health index extraction from a data tensor, called Significant Points was developped and validated on an illustrative example. The generated index is later processed by a new method, called Percentile Method, which allows to generate the monotonic profiles from the raw health index. These profiles are then modelled by a Gamma process, and the aggregate probability density function introduced in this work allowed to estimate the Remaining Useful Life (RUL) in a confidence interval that ensures a safety margin for industrial users. The proposed method is applied successfully on the experimental data of industrial production machines.