APPLICATION OF PATTERN RECOGNITION THEORY FOR DIAGNOSING THE TECHNICAL CONDITION OF A DIESEL ENGINE

Abstract

The theoretical foundations of applying pattern recognition systems to the technical diagnostics of diesel engines are considered. Approaches to the formation of a diagnostic feature space, assessment of feature informativeness, and classification of technical conditions under complete and incomplete a priori information are systematized. The main types of recognition systems and the specific features of their application in diesel engine condition assessment are characterized. The mathematical foundations of parametric estimation using the maximum likelihood method and the Bayesian approach are presented. Relationships for univariate and multivariate normal probability density functions, estimates of the mean vector and covariance matrix, as well as the use of the Mahalanobis distance for evaluating the distance of a feature vector from the class center are given. It is shown that the effectiveness of technical condition recognition is determined not by the number of monitored parameters, but by their informativeness, discriminating ability, and the possibility of reliable measurement under real operating conditions. Using fuel pump plunger pairs as an example, the selection of the most informative diagnostic feature and the elimination of parameters that do not provide clear separation between serviceable and unserviceable conditions are substantiated. For diesel engine diagnostics, fault classes are formed for the fuel supply, lubrication and cooling systems, as well as for the crank mechanism and valve timing mechanism. The Bayesian decision rule, conditional risk, and the criterion for minimizing the probability of misclassification are considered. The proposed approach can be used as a methodological basis for developing automated systems for the technical diagnostics of diesel engines. Its practical application makes it possible to formalize the procedure for assigning the current engine condition to a predefined class and to improve the validity of diagnostic decisions based on a set of measured features.