Assessment of the quality of decisions worked out by an artificial neural network (DIAG) which diagnoses a technical object

Abstract

The issues of the examination and determination of the ambiguity of the identification of the states of a technical object in a diagnostic system with an artificial neural network are presented in this article. It is an important problem in the operation of each inference (decision) system, in which various types of decisions are worked out, particularly including ones for diagnostic systems. For this purpose, a diagnostic system including its elements, in which an artificial neural network is used, is characterized and described. The structure together with the algorithm of a diagnostic neural network is presented. A diagram was drawn up, and the circulation of information in the diagnostic system was described in the perspective of errors brought into decision information by the individual elements of this system. A formula for the general error during working out of decisions in the system is put forward. It was also indicated that a number of factors including interferences and influence of the environment, errors during the measurement of the values of the properties of diagnostic signals, errors at the determination of the ranges of possible (permissible and limiting) changes of the properties of diagnostic signals defined in the inference rules and the accuracy of the drawing up of diagnostic inference rules have all a direct impact on the ambiguity level of the identification of the states of a technical object. In the present article, it is also the possibilities that are put forward for the determination and optimization of the ambiguity of the identification of the states of a technical object. For this purpose, the function that determines the quality of the identification and non-identification of the object’s state was defined in this article. A practical assessment method of the ambiguity level of the identification of the states of a technical object in the examined diagnostic system is additionally presented in an example with the use of a radar system.