Author(s)

Guoqiang Luan, Ning Li, Bing Bai, Heru Wang

Corresponding Author

Guoqiang Luan

Abstract

The research of mechanical system fault diagnosis has become one of the key enabling technologies of intelligent manufacturing. With the increase of the application fields of rotating machinery, many countries actively carry out research on the theory and application technology of mechanical fault diagnosis. The fault diagnosis technology of mechanical system in this paper is based on PNN(Probabilistic neural network) and GA(genetic algorithm), which realizes the collection, processing and analysis of vibration signals of rotating machinery. Firstly, GA is used to search the global optimal values of PNN weights and thresholds, and then it is input into PNN to replace the randomly generated weights and thresholds. Finally, network training is carried out. The simulation results show that the correct rate of fault state classification and identification of the test sample bearing is 99.06%, and the recognition rate is high, which can meet the needs of engineering application. The GA_PNN model proposed in this paper is superior to BPNN(Back propagation neural network) in reducing the loss function value, which proves the feasibility of GA_PNN model. The feasibility of extracting fault characteristics of bearing vibration signals with different scales by the model is verified.

Keywords

Vibration signal; Fault diagnosis; Probabilistic neural network; Genetic algorithm