New Detection Method for Gear Faults Based on Kernel Independent Component Analysis and BP Neural Network

Abstract

Gearboxes are widely used in various kinds of applications. The normal operation of the gears contributes important roles on the machine performance. Due to harsh environment the rolling bearings are prone to failures. Hence, it is essential to detect the gear faults. However, the vibration signals of the gearbox are often contaminated, leading to deterioration of the fault diagnosis performance. To address this issue, a new approach is proposed based on the kernel independent component analysis (KICA) and BP neural network (BPNN). The KICA was used to extract sensitive signals to eliminate noise signals. Then a BPNN was adopted to detect the gear fault. To improve the fault identification, the Genetic Algorithm (GA) was adopted to optimize the BP parameters. Experiment tests using the gearbox fault simulator have been implemented. The test results show that the noise signals have been eliminated by the KICA and the GA-BPNN can detect the gear fault accurately. Moreover, through comparison with other existing methods, the proposed KICA-GA-BPNN produced the best detection rate of 93.7%.