基于小波包能量神经网络的滚动轴承故障诊断方法

FAULT DIAGNOSIS OF ROLLER BEARING BASED ON WAVELET PACKET ENERGY NEURAL NETWORK

针对单一的信号处理诊断方法难以实现滚动轴承故障准确诊断的局限性,提出一种基于小波包能量神经网络相融合的滚动轴承诊断方法。搭建MPS-ICP滚动轴承振动信号的数据采集平台,利用小波包变换对滚动轴承内环、外环及滚动体的故障信号进行去噪、分解与重构,有效提取不同故障下各频段能量的故障特征。将提取的能量故障特征分别输入至建立的BP、RBF和Elman神经网络的诊断系统中,实验分析表明,三种神经网络都能较好的诊断电机滚动轴承的故障类型,且与实际滚动轴承的故障类型较吻合,但就诊断误差和时间综合而言,BP神经网络诊断系统更适合电机滚动轴承故障的检测。

The combination of fault diagnosis methods are presented based on the wavelet packet energy neural network for rolling bearing, as there are limitations which are difficult to achieve accurate diagnosis of rolling bearing fault using a single signal processing diagnosis method. Vibration signal data acquisition platform of the rolling bearing was built by MPS-ICP. Rolling bearing fault signals of inner ring, outer ring and rolling element were done denoising, decomposition and reconstruction using wavelet packet transform, thus fault features of the different band energy were effectively extracted. Fault energy features extracted were inputted to the diagnosis system established of BP, RBF and Elman neural network, the experimental analysis shows that three types of neural networks can well diagnose the type of motor roller bearing fault, and the results are the same as the actual fault type of rolling bearing, but as far as the comprehensive diagnostic error and time are concerned, BP neural network diagnostic system is better for the fault detection of motor rolling bearing.