基于改进神经网络的电机轴承故障的诊断

Fault diagnosis of motor bearing based on improved neural network

通过变分模态分解方法(VMD)可以直接得到轴承故障的特征频率参数,但是该参数需要事先对轴承的物理结构及其转速等信息有一定了解才能判定故障类型,不能直接反映轴承的工作状态。多尺度熵的均偏差值参数可以很好地直接表征轴承的运行状态。该实验在利用VMD分解的基础上,将多尺度熵的均偏差值作为轴承的故障特征参数,通过利用改进粒子群算法优化后的BP神经网络算法,实现了轴承的正常运行、内圈裂纹、外圈裂纹、滚轴故障4种状态的有效诊断。

The characteristic frequency parameter of the bearing fault can be directly obtained by the variational mode decomposition method(VMD), but this parameter needs to know the physical structure of the bearing and its rotational speed in advance to determine the fault type, and cannot directly reflect the working state of the bearing.. The mean deviation parameter of multi-scale entropy can directly characterize the running state of the bearing. Based on the decomposition of VMD, the experiment uses the mean deviation of multi-scale entropy as the fault characteristic parameter of the bearing. By using the BP neural network algorithm optimized by the improved particle swarm optimization algorithm, the normal operation of the bearing and the inner ring crack are realized. Outer ring crack, effective diagnosis of four states of roller failure.