Bearing life prediction based on grey model and LSTM network

  • SHU Tao , ZHANG Yichi , DING Rixian, Yuan Dizhe


Many sophisticated weapons have bearings, but it always has many failures, so we need improve its accuracy to make the maintenance people repair or replace it in time easily, this can enhance the work efficiency of weapons. CV(confidential value) can be quantitative assessment of bearing working condition, according to the known data, calculate the model bearing real CV value; increasing the smooth ratio and accumulative integral optimization of background value improve GM(1,1), and combined with Long-Short Term Memory network (LSTM) for new prediction model to predict the CV value; compare combination model with GM(1,1), improved GM (1,1) and LSTM network in four performance indicators and failure time prediction. Four kinds of indexes' averages of the combined model are all better than the other three single forecasting models; combined model's prediction of failure moment is the closest to the real results. The results prove that grey model and LSTM network combination forecast method is feasible; performance of the combined model is significantly better than the single model.