基于LSTM和证据理论的引风机轴承故障诊断

Fault Diagnosis of Induced Draft Fan Bearing Based on LSTM and Evidence Theory

引风机作为锅炉的重要辅机,其轴承的异常振动可能会给电厂造成重大损失,但引风机内部结构复杂,难以构建精确的机理模型对其进行故障诊断。针对此问题提出了基于数据驱动的故障诊断方法,此方法通过长短期记忆(LSTM)神经网络的预测能力和证据理论的多源信息融合能力对引风机轴承的状态进行诊断,再利用LSTM神经网络对于去噪数据进行预测,求取预测值与去噪值的均方根误差(RMSE),最后利用改进后的证据理论对不同参数的RMSE融合并进行故障诊断。结果表明:该方法能提前约350 s预测引风机轴承异常振动,准确地诊断出引风机轴承振动的状态,可有效改善数据冲突的问题。

As an important auxiliary equipment of the boiler,the abnormal vibration of the bearing of induced draft fan would cause great losses to the power plant,but the internal structure of induced draft fan is complex,so it is difficult to build an accurate mechanism model for fault diagnosis.To solve this problem,a method of data-driven fault diagnosis was proposed.In this method,the state of induced draft fan bearing was diagnosed through the prediction ability of long-short term memory(LSTM)and the multi-source information fusion ability of evidence theory.Then LSTM neural network was used to predict the denoised data,and the root mean square error(RMSE)between the predicted value and the denoised data was calculated.Finally,the improved evidence theory was used to fuse the RMSE of different parameters and diagnose the fault.Results show that this method can predict the abnormal vibration of the induced draft fan bearing about 350 s in advance,accurately diagnose the vibration state of the induced draft fan bearing,and effectively improve the problem of data conflict.