7 Prediction of Bearing Remaining Useful Life Based on LSTM Network
89
(a) Bearing 3_1 RUL prediction results based
on SVR
(b) Bearing 1_5 RUL prediction results based
on SVR
(c) Bearing 3_1 RUL prediction results based
on ANN
(d) Bearing 1_5 RUL prediction results based
on ANN
Fig. 7.7 Bearing RUL prediction results based on the SVR model and ANN
Table 7.3 The MSE and MAE between the predicted values and the observation values
Method
MSE
MAE
Bearing3_1
Bearing1_5
Bearing3_1
Bearing1_5
LSTM
3.54
3.07
1.52
1.48
SVC
10.98
9.06
2.39
2.54
ANN
4.61
3.88
2.07
1.52
denoising effect than VMD model. The bearing RUL prediction was completed by
LSTM network. SVR and ANN were used to predict the RUL of the bearing, and
then comparing the prediction RUL based on the three methods with the actual RUL,
the prediction results based on LSTM had a smaller the MSE between prediction
values and observation values. The method solved bearing RUL offline prediction, in
the future, it can be combined with real-time monitoring system to solve the problem
of bearing life online prediction.
References
1. Talllan, T.E.: Weibull Distribution of Rolling Contact Fatigue Life and Deviations. ASLE Trans.
5, 183–196 (1962)
2. Zhang, F.S., Liu, T.T., Liu, J.T., Cui, F.K.: Research on bearing life prediction based on three
parameters Weibull distribution. Adv. Mater. Res. 136, 162–166 (2010)
3. Yin, F.L., Wang, Y.S., Zhang. C.H„ Zhang, X.P.: An improved parameter estimation method
for three-parameter weibull distribution in the life analysis of rolling bearing. Adv. Mater. Res.