基于L-M神经网络的齿轮故障诊断

Gear Fault Diagnosis Based on Levenberg-Marquardt Neural Network

齿轮传动工况的复杂性使得其特征参量与故障形式呈非线性映射关系。提出基于Levenberg-Marquardt算法的前向多层神经网络的齿轮故障诊断方法,该方法通过利用二阶导数信息,可以提高收敛速度和增强网络的泛化性能。并以一种齿轮箱故障信号采集实验系统为例,通过MATLAB软件及其神经网络工具建模和仿真研究。结果表明,Levenberg-Marquardt神经网络对齿轮常见故障有良好的识别能力,能稳定、准确地识别各类故障,与标准BP网络相比,收敛速度快且诊断更为准确。

Because of the complexity of gear working condition, there are non-linear relationship between characteristic parameters and fault types. Proposes to apply the feed forward artificial neural network with Levenberg-Marquardt training algorithm, to the problem of gear fault diagnosis. By using second derivative information,the network convergence speed is promoted and the generalization performance is enhanced. Taking a certain gearbox fault signal acquisition experimental system for an example, MATLAB software and its neural network toolbox are used to model and simulate. The experiment result shows that Levenberg-Marquardt neural network has good performance for the common gear fault diagnosis and it can identify various types of faults stably and accurately. Furthermore ,compared with conventional BP neural network, the Levenberg-Marquardt neural network reduces training epochs and promotes prediction accuracy.