基于迁移学习的小样本齿轮箱故障诊断方法

实际工程场景中齿轮箱受工况、环境等因素影响,数据难以满足特征分布相同、训练数据充足等条件,如何在变工况情况下对齿轮故障进行诊断是故障诊断领域一大难点。为此,提出了一种结合Logistic混沌麻雀搜索优化算法(LSSA)与深度置信网络(DBN)的智能故障诊断方法,即LSSADBN。首先,将时域振动信号进行快速傅里叶变换(FFT)转换为频域信号作为训练数据集,运用Logistic混沌映射对SSA种群进行初始化,采用LSSA方法对训练数据集进行DBN结构寻优;使用最优结构DBN对源域训练集进行预训练,并加入少量目标域样本用于反向权重调优,最终实现在小样本情况下对目标域齿轮箱健康状况的准确识别。实验对比结果证明,LSSADBN方法在模型调优阶段具有更快的收敛速度,且针对不同的目标域进行迁移时都具备较高的准确率,LSSADBN方法的研究对小样本情况下的齿轮箱故障诊断具有一定的应用价值。

Fault Diagnosis Method of Small Sample Variable Load Gearbox Based on Transfer Learning

In the actual engineering scenario,the gearbox is affected by working conditions,environment and other factors,and the data is difficult to meet the conditions of the same feature distribution and sufficient training data.How to diagnose the gear fault under variable working conditions is a major difficulty in the field of fault diagnosis.Therefore,an intelligent fault diagnosis method combining logistic sparrow search algorithm (LSSA) and deep belief network (DBN),namely LSSADBN,is proposed.Firstly,the time domain vibration signal is transformed into frequency domain signal by fast Fourier transform (FFT) as the training data set,the SSA population is initialized by Logistic chaotic mapping,and the DBN structure of the training data set is optimized by LSSA method.The optimal structure DBN is used to pre-train the source domain training set,and a small number of target domain samples are added for reverse weight optimization,so as to accurately identify the gearbox health status in the target domain in the case of small samples.The experimental results show that LSSADBN method has faster convergence speed in the model optimization stage,and has high accuracy when migrating for different target domains.The research of LSSADBN method has a certain application value for gearbox fault diagnosis in the case of small samples.