基于卷积双向长短期记忆网络与混沌理论的滚动轴承故障诊断

Fault diagnosis of rolling bearing based on CCNN-BiLSTMN method

针对传统滚动轴承故障诊断方法在大噪声与变载荷环境下诊断困难的问题。基于混沌理论,通过卷积神经网络(CNN)与双向长短期记忆网络(BiLSTM)提出CCNN(Chaotic CNN)-BiLSTM智能故障诊断方法。采用相空间重构法将一维时间序列转化为二维混沌序列,学习并提取混沌序列中有效非线性信息,并输入Softmax层中完成分类。结果表明,较之现有方法,所提CCNN-BiLSTM方法在变载荷和大噪声(信噪比为-8 dB)环境下的准确率分别至少高出3.76%与5.21%,表明该方法具有良好的鲁棒性和泛化性能。

Here,aiming at the problem of traditional rolling bearing fault diagnosis methods being difficult to diagnose faults under environments of large noise and variable load,based on chaos theory,combing convolution neural network(CNN)and bidirectional long-short term memory network(BiLSTMN),the chaotic CNN-BiLSTMN(CCNN-BiLSTMN)intelligent fault diagnosis method was proposed.The phase space reconstruction method was used to transform one-dimensional time series into two-dimensional chaotic sequences,effective nonlinear information in the chaotic sequences was learned,extracted,and input into layers of the software Softmax to complete classification.The results showed that compared with the existing methods,the accuracy of the proposed CCNN-BILSTMN method can at least increase by 3.76%and 5.21%under environments of variable load and large noise with signal-to-noise ratio of-8 dB,respectively;CCNN-BILSTMN method has good robustness and generalization performance.