Fault Diagnosis of Complex Systems Based on Modular Knowledge Base and Information Compression

A fault diagnosis method for complex dynamic processes and systems is proposed in the paper. It uses a special modular knowledge base, which is a collection of modules for some typical faulty and normal conditions of the system. Each module is considered as a kind of compressed information model, which keeps in a compact form the most representative characteristics of the original data, collected for the concrete system condition. Here a modified version of the neural-gas learning algorithm for creation of all compressed information models is proposed in the paper, where a preliminary assumed number of neurons is used. The collected data from the current operation of the system are also transformed into a respective compressed information model by the same learning algorithm. The proposed fault diagnosis method is an evaluation procedure for the similarity degree between the compressed information model for the current operation and all the modules form the knowledge base. Here three different measures of similarity are used, with the “center-of-gravity distance ” showing the best results. Real experimental data taken from different operations of a hydraulic excavator are used in the paper to analyse and prove the applicability of the proposed fault diagnosis method