DEVELOPMENT OF THE CONCEPT OF BUILDING DIAGNOSTIC SYSTEMS OF ROTATING ELECTRICAL MACHINES UNDER THE CONDITIONS OF LIMITED INFORMATIONALITY OF DIAGNOSTIC SIGNS

Abstract

The article examines the peculiarities of the construction of systems for diagnosing rotating electric machines in the real conditions of their operation. It is shown that in the specified modes of operation there is a problem of limited informativeness of input information parameters that can be used to build such systems. At the same time, an additional limiting factor that must be considered when designing and implementing such equipment is the limited possibility of intervention in the design of the electric machine, which is usually limited to the manufacturing plant. As a result of a thorough analysis of the latest research in the direction of the development of diagnostic systems for rotating electric machines, a systematization of the technological parameters of electric machines that are most suitable for use in diagnostic systems was carried out. It is shown that when choosing input parameters of diagnostic systems, it is advisable to consider their informativeness, selectivity, expressiveness and complexity of the acquisition algorithm. At the same time, it is substantiated that the choice of the optimal combination of diagnostic features cannot be considered from the point of view of superposition, since each of them will be characterized by the entropy of selectivity and severity relative to defects of different types. The expediency of choosing the type of input information of diagnostic systems based on the method of evolutionary search is shown. It is demonstrated that the mentioned method allows to more completely cover the search space than, for example, gradient optimization methods, and to obtain a solution close to the optimal one in a relatively short time (a small number of iterations). The concept and typical structural diagram of the system for diagnosing rotating electric machines based on a modified non-standard artificial neural network (ANN) and the structure of the ANN itself, which considers the current mode of operation of the electric machine during diagnosis and is characterized by high adaptability to the object of diagnosis, are proposed. An example of its hardware implementation is given.