基于SABO-VMD-IWT与LSTM的变负载滚动轴承故障诊断  

Fault Diagnosis of Rolling Bearing under Variable Load Based on SABO-VMD-IWT and LSTM

作　　者：郗涛[1] 王龙 王莉静 XI Tao;WANG Long;WANG Lijing(School of Mechanical Engineering,Tiangong University,Tianjin 300387,China;School of Control and Mechanical Engineering,Tianjin Chengjian University,Tianjin 300384,China)

机构地区：[1]天津工业大学机械工程学院,天津300387 [2]天津城建大学控制与机械工程学院,天津300384

出　　处：《组合机床与自动化加工技术》2025年第3期192-198,共7页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：国家科技重大专项项目(2019zx04055-001-014);天津市科委企业科技特派员项目(20YDTPJC00840)。

摘　　要：针对滚动轴承在强干扰噪声、变负载工况下故障特征难以提取和故障识别不准确问题,提出了一种减法平均优化算法(SABO)-变分模态分解(VMD)-小波阈值降噪(IWT)的故障特征提取与长短时记忆神经网络(LSTM)相融合的滚动轴承故障诊断模型。同时,提出了一种新的阈值函数,克服了传统软硬阈值的缺点,提高了降噪的精度。首先,通过SABO-VMD-IWT对信号进行降噪;然后,提取降噪重构后信号的包络谱值作为故障特征向量;最后,在某大学公开数据基础上构建了低、中、高3种变负载、强干扰噪声的数据集,把提取的特征向量输入到LSTM中进行训练,并使用不同负载数据集进行交叉测试。结果表明,在噪声干扰、负载动态变化条件下,此模型的准确率达到97.08%,验证了本模型的有效性。Aiming at the problem that fault features of rolling bearings are difficult to extract and fault recognition is inaccurate under strong interference noise and variable load conditions,a fault diagnosis model of rolling bearings based on subtractive average optimization algorithm(SABO)-variational mode decomposition(VMD)-wavelet threshold denoising(IWT)and long short-term memory neural network(LSTM)was proposed.At the same time,a new threshold function is proposed to overcome the shortcomings of the traditional soft and hard threshold and improve the accuracy of noise reduction.Firstly,the signal was denoised by SABO-VMD-IWT,and then the envelope spectrum value of the denoised and reconstructed signal was extracted as the fault feature vector.Finally,data sets with low,medium and high variable load and strong interference noise were constructed based on the public data of university,and the extracted feature vectors were input into LSTM for training,and different load data sets were used for cross testing.The results show that the accuracy of this model reaches 97.08%under the condition of noise interference and dynamic load change,which verifies the effectiveness of this model.

关 键 词：变负载滚动轴承 减法平均优化算法 变分模态分解 改进阈值 长短时记忆网络 

分 类 号：TH133.33[机械工程—机械制造及自动化] TG659[金属学及工艺—金属切削加工及机床]