基于深度学习和贝叶斯优化的压缩机故障诊断  

Compressor Fault Diagnosis Based on Deep Learning and Bayesian Optimization

作　　者：董丽娟 方召 陈会涛[2] DONG Li-juan;FANG Zhao;CHEN Hui-tao(Xuchang Digital Learning Engineering Technology Research Center,He'nan Xuchang 461000,China;School of Mechanical and Power Engineering,He'nan University of technology,He'nan Jiaozuo 454003,China)

机构地区：[1]许昌电气职业学院机电工程系,河南许昌461000 [2]河南理工大学机械与动力工程学院,河南焦作454003

出　　处：《机械设计与制造》2023年第2期45-52,共8页Machinery Design & Manufacture

基　　金：2018年度河南省重点研发与推广专项(182102310793)。

摘　　要：由于往复式压缩机的故障诊断需要复杂而耗时的特征提取过程,并且对超参数优化存在局限性,提出了一种基于深度学习和贝叶斯优化的压缩机故障诊断方法。首先通过时域计算短窗口的预处理方法降低模型复杂性,并且不损失时间相关信息。然后从压缩机振动信号的时间序列表示中迭代训练长短期记忆模型,在每次迭代中限定搜索空间,并利用贝叶斯优化方法对超参数进行优化。通过实验结果显示提出模型的故障识别率达到93%,与其他方法的对比结果证明该方法在性能上有了显著的提高。Because the fault diagnosis of reciprocating compressor needed complex and time consuming feature extraction process,and there were limitations in super parameter optimization,a compressor fault diagnosis method based on deep learning and Bayesian optimization was proposed.Firstly,the time-domain short window pre-processing method was used to reduce the complexity of the model without loss of time-related information.Then,the long-term and short-term memory model was trained iteratively from the time series representation of compressor vibration signal,and the search space was limited in each iteration,and the super parameters were optimized by Bayesian optimization method.The experimental results show that the fault recognition rate of the proposed model reaches 93%.Compared with other methods,the performance of the proposed method is significantly improved.

关 键 词：往复式压缩机 故障诊断 深度学习 贝叶斯 

分 类 号：TH16[机械工程—机械制造及自动化] TH133.33