Study on remaining useful life prediction method of grinding wheel based on CEEMDAN-LSTM-MHSA

CHI Yulun; HUANG Haolun

doi:10.16579/j.issn.1001.9669.2026.06.004

Abstract：[Objective]Aiming at the problem that traditional signal reconstruction relies heavily on prior knowledge and manual experience,which reduces model generality and fails to simultaneously realize the prediction of grinding wheel dressing life and replacement life,the complementary ensemble empirical mode decomposition with adaptive noise(CEEMDAN),multi-head self-attention(MHSA)mechanism and long-short term memory(LSTM)network were adopted to replace the manual signal reconstruction process.A general grinding wheel life prediction method based on CEEMDAN-LSTM-MHSA was proposed.[Methods]Firstly,according to frequency characteristics,CEEMDAN was used to decompose grinding acoustic emission and vibration signals into intrinsic mode function(IMF).Secondly,multiple LSTM networks were applied to extract time-domain features from each IMF.The MHSA mechanism was employed to establish the dependency between influencing factor features and the changes in grinding wheel surface morphology and diameter,and construct health index(HI)curves for grinding wheel dressing and replacement.Then,the isolation forest anomaly detection algorithm and exponential weighted moving average(EWMA)were utilized to improve the monotonicity and smoothness of HI curves.Support vector regression(SVR)was adopted to fit the HI curves,and the remaining useful life(RUL)prediction of the grinding wheel was realized combined with the threshold of HI curves.Finally,continuous grinding and grinding wheel dressing tests were carried out for verification.[Results]The results show that the proposed model can accurately predict the dressing life and replacement life of grinding wheels,and effectively improve the generality of grinding wheel RUL prediction.