基于Bi-LSTM和自注意力机制的旋转机械故障诊断方法研究

Research on the fault diagnosis method of rotating machinery based on Bi-LSTM and self-attention mechanism

旋转机械因其特殊的功能要求,通常工作在恶劣的环境中,振动信号易受外界干扰。基于传统信号处理方法的故障诊断技术越来越不能满足故障诊断精度的需要,因此,利用大数据和人工智能技术进行旋转机械故障诊断成为目前的主要研究方向之一。针对以上问题,提出一种基于双向长短时记忆网络(Bi-LSTM)和自注意力机制的旋转机械故障诊断方法。首先,利用转子实验台模拟旋转机械的各种运行状态,采集旋转机械在各种运行状态下的振动信号;然后,将振动信号输入Bi-LSTM网络,自注意力机制将Bi-LSTM各时间步的输出进行加权求和,获得振动信号的深层特征表示;最后,通过全连接层和Softmax层输出旋转机械各种运行状态的预测概率。实验结果表明:本文提出的方法能够有效地实现旋转机械的故障诊断,与其他方法相比,模型的训练稳定性、收敛速度和故障识别准确率均得到提高。

Due to its special functional requirements,rotating machinery usually works in harsh environments and vibration signals are susceptible to external interference.Fault diagnosis techniques based on traditional signal processing methods are increasingly unable to meet the needs of fault diagnosis accuracy.Therefore,the use of big data and artificial intelligence technology for rotating machinery fault diagnosis has become one of the main research directions at present.To address the above problems,this paper proposes a fault diagnosis method for rotating machinery based on bi-directional long and short term memory network(Bi-LSTM) and self-attentive mechanism.Firstly,the vibration signals of the rotating machine are collected under various operating conditions by simulating the various operating conditions of the rotating machine with a rotor test bench.Then the vibration signals are input into the Bi-LSTM network,and the selfattentiveness mechanism weights and sums the outputs of each time step of the Bi-LSTM to obtain the deep feature representation of the vibration signals.Finally,the prediction probabilities of various operating states of rotating machines are output through the fully connected layer and Softmax layer.The experimental results show that the proposed method can effectively implement the fault diagnosis of rotating machines,and the training stability,convergence speed and fault recognition accuracy of the model are improved compared with other methods.