Received on October 08, 2014

The article analyzes effectiveness of the diagnostic systems of small dimensions (2—3 faults) with a small number of features (1—3). The systems are based on a fuzzy inference. Research is conducted using a stochastic model of an object, which implies generation of random characteristic values with given expectation (MO) and values of the standard deviation (SD) for the respective types of faults (pathologies). Herewith, various options are generated, membership functions, the types of distributions generated by the signs, their MO and SD are explored. A stochastic model of the object allows us to measure the performance of the diagnostic system with small volumes of training samples. It is shown that in many cases the diagnostic system of full dimension does not provide an unambiguous and quality solution to the problem of diagnostics. It is advisable to start construction of a multi-level diagnostic system using fuzzy logic with a mutual comparison of the sets of attribute values for the relevant problems in order to determine the feasibility of construction of the first, low-level diagnostics. At this level an obvious fault can be determined, or groups of faults can be selected. At the next level, using the additional features, we can diagnose specific faults in the pre-selected lower level group by the methods of fuzzy logic. In the developed system the fuzzy logic diagnostic method is most efficient when used as a part of the terms of triangular membership functions. With the increasing number of faults, and depending on the location of the sets of attribute values it makes sense to increase the number of the terms. In this case, it would be expedient to ensure that each term corresponds to at least one fault.

Keywords: stochastic models, fuzzy logic, multilevel system

Acknowledgements: This work was supported by the Russian Foundation for Basic Research, projects no. 13-01-00082а.

For citation:
Kolosov O. S., Anisimov D. N., Khripkov D. V. Research of the Multilevel Fuzzy Diagnostic Systems with the Use of Stochastic Models, Mekhatronika, avtomatizatsiya, upravlenie, 2015, vol. 16, no. 4, pp. 254—261.
DOI: 10.17587/mau.16.254-261

Corresponding author:
Kolosov Oleg S., Professor, Dr. Su. Tech, National Research University "Moscow Power Engineering Institute", 111250, Moscow, Russian Federation, e-mail: KolosovOS@mpei.ru

To the contents