Rolling Bearing Fault Diagnosis Based on QGA Optimized DBN-ELM Model

The traditional fault diagnosis model is easy to fall into the local optimum when training, the model generalization ability is poor, and the fault recognition accuracy is easily affected by the quality of artificial feature extraction. Firstly, combine Deep Confidence Network to Extreme Learning Machine(DBN-ELM)model, extract features in bearing data by using 4-layer Restricted Boltzmann Machine (RBM) network, then integrate Extreme Learning Machine (ELM) into Depth Belief Network (DBN), input these features into DBN-ELM model for classification, avoiding global fine-tuning of DBN. Due to difficulties in selecting parameters such as node number and learning rate of hidden layer, adopt Quantum Genetic Algorithm (QGA) for optimization, in which dynamic improvement strategy is used to adjust rotation angle according to evolution, The convergence speed and accuracy of QGA are improved, and the quantum catastrophe operation increases the possibility of finding the optimal solution. Finally, without feature extraction, the bearing data set of Case Western Reserve University is used for experimental research and fault feature analysis, and compared with DBN, PSO-DBN, QGA-DBN and other algorithms. The results show that the cross validation classification accuracy of the DBN-ELM model optimized by QGA is higher.