提升小波包和改进BP神经网络相融合的新故障诊断算法

A New Fault Diagnosis Algorithm Combining Lifting Wavelet Packet with Improved BP Neural Network

摘要:

针对传统的小波变换和BP神经网络应用于故障诊断中存在自适应性差、效率低等问题,提出一种提升小波包和改进BP神经网络相融合的新故障诊断算法；利用插值细分思想,设计了提升小波包的预测算子和更新算子,结合传统小波包算法和提升模式的原理,完成了提升小波包算法的设计,并将该算法应用于故障信号的消噪和能量特征量的提取；利用遗传算法优化标准BP神经网络的初始权值和阈值,采用L-M算法优化标准BP神经网络的搜索方式；以美国凯斯西储大学提供的滚动轴承实验数据,将新算法应用于实验中,分析结果表明:新故障诊断算法比传统的BP神经网络算法具有收敛速度快、诊断精度高等实效性。

Abstract:

According to the problem of poor adaptability and low efficiency when traditional wavelet transform and BP neural network used for fault diagnosis, a new fault diagnosis algorithm for the fusion of lifting wavelet packet and improved BP neural network is proposed. Takes advantage of interpolating subdivision thinking, the prediction operator and update operator of lifting wavelet packet were designed, combining traditional wavelet packet algorithm and the principle of lifting mode, the lifting wavelet packet algorithm’s design was completed, and the algorithm was applied in extinction noise and energy feature extraction of fault signal. Use GA algorithm to optimize initial weights and thresholds of standard BP neural network algorithm, and use L-M algorithm to optimize the search of the standard BP neural network. Make use of the rolling experimental data provided by Case Western Reserve University, the new algorithm is applied to the experiment, the results show that:the new fault diagnosis algorithm has faster convergence, higher precision diagnostic effectiveness than the traditional BP neural network algorithm.