流体机械

2026, v.54;No.649(07) 102-109

[打印本页] [关闭]
本期目录(Current Issue) | 过刊浏览(Archive) | 高级检索(Advanced Search)

基于AWTD和LSTM的循环泵口环磨损趋势预测研究
Research on the wear trend prediction of circulating pump mouth ring based on AWTD and LSTM

魏道红,于雅静,林彬,刘瀚文,于涵
Wei Daohong,Yu Yajing,Lin Bin,Liu Hanwen,Yu Han

摘要(Abstract):

因受流体噪声与多源机械干扰的叠加影响,循环泵口环磨损状态的早期识别与趋势预测面临挑战。针对这一问题,提出一种基于自适应小波阈值降噪(AWTD)和长短期记忆神经网络(LSTM)的循环泵口环磨损趋势预测方法。通过搭建循环泵试验测试平台,分别采集口环间隙为0.2,0.5,1.0 mm三种不同磨损程度的振动信号,并利用连续小波变换(CWT)进行多尺度时频特征分解;利用基于信号能量分布的自适应阈值函数,对含噪振动信号实施降噪处理,降噪后信号的信噪比R_(SN)分别提升0.32dB,18.34dB和15.30dB,有效凸显磨损特征成分,进一步构建LSTM预测模型。结果表明,模型的均方根误差E_(RMS)低至0.021 039,平均绝对误差E_(MA)为0.016 598,说明模型具有较高的预测精度,进一步验证该方法在循环泵口环磨损早期趋势预测中的有效性。
Due to the combined influence of fluid noise and multiple mechanical disturbances,the early identification and trend prediction of wear conditions of circulating pump mouth ring face challenges.To address this issue,a method for predicting the wear trend of circulating pump was proposed based on adaptive wavelet threshold denoising(AWTD) and long short-term memory(LSTM).By establishing circulating pump experimental test platform,the vibration signals were collected under three different wear levels(mouth ring clearance of 0.2,0.5,1.0 mm),and continuous wavelet transforms(CWT) was used to perform multi-scale time/frequency feature decomposition.By using an adaptive threshold function based on signal energy distribution,noise reduction processing was carried out on the noisy vibration signals.After noise reduction,the signal-to-noise ratio R_(SN) of the signals increased by 0.32dB,18.34dB and 15.30dB respectively,effectively highlighting the wear feature components,and further constructing an LSTM prediction model.The results show that the root mean square error E_(RMS) of the model is as low as 0.021 039,and the mean absolute error E_(MA) is 0.016 598,indicating that the model has high prediction accuracy and verifies the effectiveness of this method in early trend prediction of wear in circulating pump mouth ring.

关键词(KeyWords): 循环泵;口环磨损;自适应小波阈值降噪(AWTD);长短期记忆神经网络(LSTM);趋势预测
circulating pump;wear of mouth ring;adaptive wavelet threshold denoising(AWTD);long short-term memory(LSTM);trend prediction

Abstract:

Keywords:

基金项目(Foundation): 国家自然科学基金青年科学基金项目(52409113)

作者(Author): 魏道红,于雅静,林彬,刘瀚文,于涵
Wei Daohong,Yu Yajing,Lin Bin,Liu Hanwen,Yu Han

参考文献(References):

扩展功能
本文信息
服务与反馈
本文关键词相关文章
本文作者相关文章
中国知网
分享