Electric Machines, the most important consumer of electric energy, are considered as the dominant driving force in different industries where most of the production lines are depended on the proper and unintruppted performance of electric machines. In this regard, condition monitoring is being considered as a strategy to increase the life long and to reduce the production line shutdown’s losts. Condition monitoring includes measuring physical signals and process the gathered raw data in order to obtain interpretable indices in monitoring the operating status of the machine and difrentiating the normal and fault condition of the system. In this dissertation, the feasibility of condition monitoring for electric machine based on the temperature distribution profile on the body surface of the machine is studied. A multi-physics modelling approach including electromagnetic, heat transfer, and fluid flow, is being employed to analyze the flow of the fluid inside the machine and to predict the temperature distribution. The model is then used to predict the body temperature distribution for several working codition. The result of such study has resulted to develop a simplified multi-physics model based on which, the temperature on the surface of the machine can be predict using a less ordered yet accurate model. The simplified model is validated in practice. The validated simplified model is useful to predict the temperature profile in a wide range of working condition of the machine, and create a databased for training an intelligent algorithm in evaluating the working condition of the machine based on the predicted temperature profiles. In this regard, first the generated data are employed to define the minimum possible number of required temperature sensors and the best installing position to reduce the costs and reduce the complexity of the condition monitoring system. In the second step, the database is used to study and extract the efficient indeces as the inputs of the condition monitoring algorithm in order to report the abnormal working condition of the machine and then predict the type, cause, and the severity of the abnormality. In this dissertation a three-stage condition monioting algorithm is developed in order to increase the accuracy and functionality of the algorithm. Keywords: Multi-Physics Modelling, Eletro-Thermal Modelling, Switched Reluctance Machines, Fault Diagnosis, Condition Monitoring, Machine Learning
