基于特征融合的齿轮箱故障诊断  

Fault Diagnosis of Gearboxes Based on Feature Fusion

作　　者：周芸 吴胜利[1] 邢文婷 ZHOU Yun;WU Shengli;XING Wenting(School of Traffic and Transportation,Chongqing Jiaotong University,Chongqing 400074,China;School of Management Science and Engineering,Chongqing Technology and Business University,Chongqing 400067,China)

机构地区：[1]重庆交通大学交通运输学院,重庆400074 [2]重庆工商大学管理科学与工程学院,重庆400067

出　　处：《噪声与振动控制》2025年第2期63-69,共7页Noise and Vibration Control

基　　金：国家自然科学基金资助项目(51705052);国家社会科学基金资助项目(23BGL220);重庆市研究生联合培养基地建设资助项目(JDLHPYJD2020028)。

摘　　要：齿轮箱故障的振动信号具有非线性、非平稳性,再加上齿轮箱运行工况复杂的特点,导致传统的信号处理方法难以有效提取齿轮箱故障特征,严重影响传动精度和设备运行安全。基于此,本文研究振动信号无量纲指标与经验模态分解(Empirical Mode Decomposition, EMD)信息熵进行特征融合的方法,并利用随机森林(Random Forest, RF)对不同特征之间的重要性进行比较、排序,有效克服信息冗余,同时将新构建的样本集作为输入,对LSTM(Long Short-Term Memory)神经网络进行训练,实现对齿轮箱不同局部故障的有效识别。并利用东南大学实验数据验证所提方法的有效性,通过与其他方法对比证明本文所提方法具有计算效率高和识别精度高的特点,可为齿轮箱的智能诊断提供新的实践和方法基础。The vibration signal of gearbox faults exhibits the characteristics of nonlinearity,non-stationarity and complexity of gearbox operating conditions,which leads to the difficulty for traditional signal processing methods to effectively extract gearbox fault characteristics and severe impacts on transmission accuracy and equipment operational safety.In view of this,in this paper,a method that combines dimensionless indicators of vibration signals with Empirical Mode Decomposition(EMD)information entropy was proposed.This method utilized the Random Forest(RF)to compare and rank the importance of different features and effectively overcome information redundancy.At the same time,the newly constructed sample set was used as input to train the Long Short-Term Memory(LSTM)neural network,and achieve the effective identification of various local faults in the gearbox.The effectiveness of the proposed method was validated using experimental data from Southeast University.A comparison with other methods demonstrates the computational efficiency and high identification accuracy of the proposed approach.This work contributes a new practical and methodological foundation for the intelligent diagnosis of gearboxes.

关 键 词：故障诊断 齿轮箱 融合特征 无量纲指标 随机森林 LSTM神经网络 

分 类 号：TH132.4[机械工程—机械制造及自动化] TH165.3