基于EEMD和GA-LSTM算法的行星齿轮故障诊断方法

Fault diagnosis method of planetary gear based on EEMD and GA-LSTM algorithms

在强烈的背景噪声下,如何较好地提取行星齿轮的微弱故障特征,这是行星齿轮故障诊断领域需要解决的难题。针对行星齿轮振动信号的非线性和非平稳性,为了提高其故障诊断的准确性,提出了一种利用遗传算法优化的长短时记忆网络(GA-LSTM)和集合经验模式分解(EEMD)的行星齿轮故障诊断方法。首先,通过试验采集了4种行星齿轮故障类型的振动信号,并采用EEMD方法将行星齿轮的原始振动信号分解为6个本征模态函数(IMF)分量,将其作为特征分量以便进一步处理;然后,使用遗传算法(GA)对LSTM网络的超参数进行了优化,以提高故障类型识别的准确性;最后,将特征分量输入到已经训练好的GA-LSTM模型中,将其网络模型作为最终分类器,对行星齿轮进行了故障诊断识别,并通过对比未经优化的网络,以及在原始信号中人为地加入噪声模拟的实际工程信号,验证了基于EEMD和GA-LSTM算法的有效性和泛化性。研究结果表明:训练后的网络实现了不到2%的损失率,具有良好的稳定性,GA-LSTM方法故障分类精度达到了94.17%;与未经优化的网络相比,GA-LSTM模型的验证精度高于LTSM,该网络模型在所有分量上都表现出更好的时序性能,在识别添加了噪声的工程信号时,也能保持较高的故障诊断精度,从而表明其在行星齿轮故障诊断中的优越性。该研究在提高机械传动设备故障诊断能力方面有一定的理论参考价值。

How to effectively extract weak fault features of planetary gears under strong background noise is a difficult problem that needs to be solved in the field of planetary gear fault diagnosis.For the nonlinear and non-stationary vibration signals of planetary gears,in order to improve the accuracy of fault diagnosis,a planetary gear fault diagnosis method optimized by genetic algorithm-optimized long-short-term memory network(GA-LSTM)and ensemble empirical mode decomposition(EEMD)was proposed.First,the vibration signals of four types of planetary gear faults were collected in the experiment,and the original vibration signal of the planetary gear was decomposed into six intrinsic mode function(IMF)components by using EEMD method,which was used as the feature components for further processing.Then,the hyperparameters of the LSTM network were optimized using the genetic algorithm(GA)to improve the accuracy of fault type identification.Finally,the feature components were inputted into the trained GA-LSTM model,the network model was used as the final classifier to diagnose and identify the faults of the planetary gears.By comparing the unoptimized network and artificially adding noise to the original signal to simulate the actual engineering signal,the validity and effectiveness of the method based on EEMD and GA-LSTM algorithms were verified effectiveness.The research results show that the trained network achieves a loss rate of less than 2%,and has good stability.The fault classification accuracy of the GA-LSTM method reaches 94.17%.Comparing with the non-optimized network,the verification accuracy of the GA-LSTM model is found to be higher than that of the LTSM,which shows better timing performance on all components;even when identifying engineering signals with added noise,high fault diagnosis accuracy can also be maintained,which shows its superiority in planetary gear fault diagnosis.This study has certain theoretical guidance and reference value in improving the fault diagnosis ability of mechanical transmission equipment.