基于振动信号与LSTM网络的船舶齿轮箱剩余寿命精准预测研究OA

To address the issues of excessive maintenance and high failure risk caused by the traditional"periodic maintenance"model for marine gearboxes,a remaining useful life(RUL)prediction method based on vibration signals and LSTM networks is proposed.This paper selects the MB270 type gearboxes of three ocean-going cargo ships of a certain enterprise as the research objects.Vibration signals are collected using piezoelectric acceleration sensors.After preprocessing with the"3σ criterion cleaning+db4 wavelet denoising",24-dimensional multi-domain features are extracted.A prediction model is constructed,and adaptive correction for operating conditions and Bagging integration optimization are introduced.The instance verification shows that the model's RUL prediction MAPE for mid-term worn gearboxes is only 3.3%,with an R2 of 0.968,and the RMSE(35.2 hours)is significantly better than the empirical formula method(128.7 hours)and SVM(72.3 hours).This research can promote the maintenance model from"cycle-driven"to"state-driven",reduce spare parts inventory costs and fault downtime,and provide technical support for ship machinery support.