Application of an improved Levenberg-Marquardt back propagation neural network to gear fault level identification

Chip fault is one of the most frequently occurring damage modes in gears. Identifying different chip levels, especially for incipient chip is a challenge work in gear fault analysis. In order to classify the different gear chip levels automatically and accurately, this paper developed a fast and accurate method. In this method, features which are specially designed for gear damage detection are extracted based on a revised time synchronous averaging algorithm to character the gear conditions. Then, a modified Levenberg-Marquardt training back propagation neural network is utilized to identify the gear chip levels. In this modified neural network, damping factor and dynamic momentum are optimized simultaneously. Fisher iris data which is the machine learning public data is used to validate the performance of the improved neural network. Gear chip fault experiments were conducted and vibration signals were captured under different loads and motor speeds. Finally, the proposed methods are applied to identify the gear chip levels. The classification results of public data and gear chip fault experiment data both demonstrated that the improved neural network gets a better performance in accuracy and speed compared to the neural networks which are trained by El-Alfy's and Norgaard's Levenberg-Marquardt algorithm. Therefore, the proposed method is more suitable for on-line condition monitoring and fault diagnosis.