基于BI-LSTM的小样本滚动轴承故障诊断方法研究

Rolling Bearing Fault Diagnosis Method Based on BI-LSTM under Less Samples Condition

微弱故障特征的有效提取与判别模型的精确性是滚动轴承状态监测和故障诊断的关键。针对长短时记忆网络(LSTM)模型在少样本条件下存在故障诊断准确度较低的问题,提出一种基于双向长短时记忆网络(BI-LSTM)的小样本滚动轴承故障诊断方法:首先采用自适应白噪声的集合经验模态分解(CEEMDAN)与傅里叶变换对信号进行分解变换构成特征矩阵,然后采用BI-LSTM进行特征提取,获取每个样本序列的故障特征,最后采用逻辑回归(LR)将多个故障特征汇总学习。结果表明:所提出的方法在随机的小样本测试集上平均精确度相对传统LSTM模型提高30.8%,可为滚动轴承健康状态监测提供重要算法支撑。

Effective extraction of the weak fault features and the accuracy of the discriminant model are the key of condition monitoring and fault diagnosis of rolling bearings.Aiming at the problem that the traditional long short-term memory(LSTM)model has low accuracy in fault diagnosis under less samples conditions,this paper proposes a rolling bearing fault diagnosis method with less samples based on bidirectional long short-term memory network(BI-LSTM).Firstly,the complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)and Fourier transform are used to decompose and transform the signal to form a feature matrix.Then,BI-LSTM is used for feature extraction to obtain the fault characteristics of each sample sequence.Finally,logistic regression(LR)is used to summarize the fault features.The results show that the average accuracy of the proposed method in the random small samples test set is 30.8%higher than the traditional LSTM model,which can provide an important algorithm support for the monitoring of rolling bearing health.