Алгоритм классификации технического состояния топливного регулятора в пространстве параметров рабочего процесса

Abstract

This article solves the problem of constructing an algorithm for classifying the technical state of the fuel regulator of a turboshaft engine of a helicopter in the parameter space of the regulator's working process and obtaining estimates of the state of the investigated product. In the introduction to this work, the main methods of classifying the technical state are considered, and a brief justification for the choice of the method for classifying the state of the product in the space of measured parameters used to construct the algorithm is considered in this article, is given. The criteria for the presence of a malfunction in the investigated controller are determined and the basic requirements for the classification algorithm are formed. To simplify the problem being solved, many assumptions about the diagnosed defects were made. The article provides descriptions of all components of the classification algorithm. A brief description of the mathematical model of the fuel regulator is given. The operating mode of the regulator for analysis is selected, and the list of state parameters required for diagnostics and the list of diagnostic parameters of the working process is given. The technique of linearization of the mathematical model of the controller and the technique of constructing the matrix of influence coefficients, which is the basic element of the entire algorithm, is described. The probabilistic characteristic of the credibility of the classification algorithm is determined, and the derivation of the formula for its calculation, based on the Bayes theorem, is also given. To assess the quality of the classification by the diagnostic algorithm, a test sample of workflow parameters was formed. The methodology for constructing a test sample is described and its size is determined. After the product condition estimates are obtained according to the test sample data by the classification algorithm, such a quality criterion as the recall is calculated. As a result of assessing the recall, a table was formed with the values of this criterion for each class. The recall of the algorithm on average for all defects was 89 %. The conclusions indicate possible methods for improving the quality of diagnosis by the algorithm.