Modelisations et outils d’aide au diagnostic de defauts de machines synchrones et a reluctance variable

Abstract

The « Condition monitoring » of machines has become an art that can diagnose and precisely potential defects to act quickly before the « break »! Monitoring and diagnosis of electric machines represent a scientific and economic challenge motivated by the goals of dependability and continuity of service of electric drives. Electromechanical converters (engines, generators, actuators…) occupy an increasingly important in industrial equipment, especially with the new requirements for electric traction or decentralized energy production with new structures machines. In addition, these devices can play a critical role in the process which generates severe constraints in terms of dependability and availability rates, hence the need for robust diagnosis increased and thus the development of diagnostic tools more efficient. Effective diagnosis paves the way for a fault tolerant control, and should therefore increase the robustness of the industrial process. Many approaches and diagnostic procedures are developed for fault detection and diagnosis by different research communities’ automatic, productics and artificial intelligence. The methods differ in the type of a priori knowledge about the process that they require. So they can be classified generally as model-based methods, based on knowledge and methods based on historical data. Methods based models consider a structural model of the behavior of the process based on fundamental physical principles. These models can be quantitative type, expressed as mathematical or qualitative equations, expressed for example in the form of logical relations. Methods based knowledge exploits the skills, reasoning and expert knowledge about the process to turn them into rules, so as to solve specific problems. Finally, methods for database search to discover information, as typical examples and trends in the measurements from the sensors and actuators, which can identify the behavior of the process. These methods include, among others, learning methods and classification (or recognition). The most recent studies have been devoted to the electrical monitoring of induction machines in particular the inspection of the stator current. In recent years, research in the field of electric motor drives for critical industrial applications such as automotive, aerospace, robotics, nuclear power plants or decentralized production are focused on research level of the drive motor and various topologies. The concept of the fault tolerant device and the development of devices dependability are often required to improve the availability of systems integrating this type of machine, minimize maintenance costs and ensure more effectively the security of goods and people in direct or indirect relationship with the application. That is why we seek to apply proven methods by asynchronous machine (ASM) for other machine structures, types, synchronous reluctance machines (SynRM) and variable reluctance machines (VRM) by their growing presence in the areas of fault tolerance. Therefore, all new results can be of significant interest to all researchers working in the area of fault tolerance. The work and research developed within the LGEC (Electrical Engineering Laboratory Constantine) in collaboration with Electrical Machines and Drives Department Technical University of Cluj-Napoca, Romania are in this part of the Modeling & support tools to diagnose faults synchronous machines & a variable reluctance. The research topics cover aspects: digital finite element modeling, using the FLUX-2D 7.6 software and MATLAB-Simulink with the flowFLUX-2D 10.4 and in order to improve the operation of the first type of machine studied and modeled we proposed new converter model based on the principle of the separation of the phases then understanding and analysis of different stator and / or rotor faults, research and development of monitoring tools, diagnosis and fault monitoring driver assistance and human-system interaction based on the optimal time-frequency representation, called "dependent class signal (DCS)" whose plane ambiguity is smoothed by a kernel designed to achieve maximum separation between the defect class and healthy class machine. The separation of classes is performed by Fisher contrast, based on compactness and reparability of classes. The assignment criteria or classification of a new signal is based on several intelligent methods by classification algorithms in order to automate the process of diagnosis: the hidden Markov model (HMM ), combined with Neural Networks (RN ), K-means (KPP) and the Mahalanobis (MAH ) or Ecludienne (ED) distance. Different decision rules are compared in the presence or absence of defects and rejected observations are analyzed to determine the possible emergence of a new mode of operation. The Kalman filter approach is used for monitoring of evolution developed and allows prediction modes included in the training set and determine the future state of these modes. In this context, the thesis consists of four chapters: The first presents an overview on the supervision and the different approaches for the detection and diagnosis of faults developed by different research communities, including the signal approach and the system approach which we have supported our work. The second chapter provides a finite element design of the fault tolerant switched reluctance machine consists in modifying their windings. Splitting phase’s independent coils is the method most widely used for machinery fault tolerance. It is necessary to compensate for the absence of a phase fault or coil has the least possible changes in the torque characteristic. The power converter of the machine must also be designed to be fault tolerant. Using the programmed intelligence converter must be able to reconfigure its control and power of the machine according to the severity of defects, to continue operating of the machine The third chapter explains the failures which may form on a whole ‘converter - variable reluctance machine’ and the occurrence of each of these defects. This chapter is divided into two main parts. The first part describes the different sources (electrical, mechanical,) failures that can occur to the machine variable reluctance. As for the second part, it presents different failures that can undergo a power converter. Finally analyzing of the different measured signals such as current flow, and electromagnetic torque through Fourier transform (FFT). Chapter 4 is dedicated to the development of our diagnostic system for variable reluctance machines. We describe in this section, the means used to obtain the states, transitions and events associated with these transitions. We show also how this system can be used by the operator for the purposes of supervision. The selection algorithms parameters vector form used by the decision-making system are implemented and presented (learning phase). Failures correspond to a short circuit, open circuit, for various levels of load supply by voltage inverter. We conclude this paper with a chapter five dedicated to the experimental results obtained during the application of our diagnostic classification; a description of the experimental and different modes studied (healthy, faulty) bench is presented. Classification of the new observations with the implementation of the proposed methods in combination with the experimental data of the asynchronous machine and variable reluctance machine proves the effectiveness of these classification methods independently of the type of fault and the type of machine.