A diagnostic system of an intelligent component based on Bayesian accurate inference networks

Abstract

The article discusses the features of building an intelligent diagnostic system for technical systems based on Bayesian networks, which are a new method of probabilistic-statistical modeling based on a combination of graph theory, probability theory and methods of applied statistics. An example of a Bayesian network structure for diagnosing and predicting an accident in unmanned vessels is presented. Special attention is paid to the stages of building a Bayesian network for diagnosing technical systems using statistical data. Particular emphasis is placed on solving the problem of forming the Bayesian network output under conditions of decreasing uncertainty when new information arrives within the Bayesian network. The features of constructing an accurate probabilistic inference using the clustering algorithm are also considered. A number of experiments were carried out to compare the computational complexity of the clustering algorithm, which was used in the process of constructing the exact probabilistic inference in the Bayesian network, and the stochastic sampling algorithm to obtain an approximation. To simulate the structure of an intelligent system for technical diagnostics, it is proposed to divide the Bayesian network into subnets, the graphs of which will have a relatively small number of vertices. As an example, the following classes of diagnostic parameters of a technical system are proposed, for each of which it is further recommended to build a Bayesian network: probability of system uptime, system failure rate, average system (element) operation time to failure, average service life, gamma-percentage resource, average shelf life, etc.