Fault diagnosis approach of gearbox based on Support Vector Machine with improved bi-layers quantum genetic optimization

Support vector machine (SVM) is a powerful structural risk minimization principle based pattern classification method which is particularly available for the nonlinear and high dimensional problem with small-size samples. However, poor choice of parameters involving penalization coefficient and kernel parameters decided by kernel function type, dramatically decreases the classification accuracy and generalization ability of SVM. To address the above issue, quantum genetic algorithm (QGA) is introduced into parameters optimization of SVM, and moreover a novel dynamic adjustment mechanism of quantum gate rotation angle based on quantum entropy, Hadamard gate-based mutation strategy and population catastrophe strategy are appended to classical QGA to ameliorate local searching ability of the algorithm, and furthermore an improved bi-layer quantum genetic algorithm (IBQGA) which possess the ability of solving the blindfold preinstall problem of the mutation probability and the timing parameter of catastrophe, is put forward to realize the parameters optimization of SVM. Finally the optimal SVM is utilized for fault diagnosis of gearbox. The effectiveness of the proposed diagnosis approach is validated by a gearbox's fault simulation experiment. Experiment results demonstrate that SVM optimized by the developed IBQGA outperforms classical quantum genetic-optimized SVM and genetic-optimized SVM in converging rate and fault identification rate.