基于GA和LM组合优化BP神经网络的滚动轴承故障诊断方法∗

摘要

针对传统BP神经网络在滚动轴承故障诊断中存在收敛速度慢且易陷入局部极小等问题，提出一种GA和LM组合优化BP神经网络的故障诊断方法。利用小波包变换对不同故障类型的振动信号进行软阈值消噪处理，然后进行三层小波包分解及重构，并成功提取了8个频带构建的故障能量特征向量。利用GA优化了BP神经网络的隐含层层数及节点数、初始权值和阈值的网络参数，采用LM算法改进网络的搜索空间。以美国凯斯西储大学提供的滚动轴承实验数据进行诊断，结果表明，与GA优化的诊断结果相比，组合优化后的BP神经网络具有更高的诊断效率和精度。%Aim to the problem of slow convergence and easy to fall into local minimum for traditional BP neural network in rolling bearing fault diagnosis, a fault diagnosis method is proposed based on GA and LM combined-optimization BP neural network. Vibration signal of different failure types were used wavelet packet transform to carry out soft threshold de-noising, and then were taken advantage of three wavelet pack-et for decomposition and reconstruction, and were successfully extracted fault energy eigenvectors of eight bands. Use GA to optimize the hidden layers and the number of nodes, the network parameters of initial weights and thresholds for BP neural network, and use LM algorithm to improve the search space of the net-work. Rolling bearing experimental dates that were provided Case Western Reserve University were diag-nosed, the results show that, compared with the diagnostic results before GA optimization, the combined-optimization BP neural network has a higher diagnostic efficiency and accuracy.