Analysis of model prediction results of fan gearbox under abnormal state based on LSTM network

The test set data is input into LSTM model to verify the effectiveness of model early warning. The prediction results and residual variation trend of test set are shown in Figure 1 and Figure 2.

The prediction error in the first half of the model is small and relatively stable. Starting from the 3rd 000 point, the residual error increases gradually and fluctuates in a certain range, and the gearbox bearing may have a potential failure. The width of the sliding window is still 150, and the changes of the mean and standard deviation of the test set residuals are analyzed, as shown in Fig. 3 and Fig. 4.

It can be seen from Figure 3 that the mean value of residual error exceeds the alarm threshold of the mean value from the 3130 sliding window, falls below the alarm line at the 3463 window, and then exceeds the alarm line at the 3531 window, and then above the alarm line. Looking at Figure 4, it is found that the standard deviation exceeds the alarm threshold of the standard deviation at the 3197 sliding window. The actual alarm information of the fan is sent from the 5th sampling point, which shows the bearing failure of the fan gearbox. According to this method, when the mean value and standard deviation exceed the alarm threshold at the same time, the alarm signal will be sent out, so the alarm will be given at the 3197 sliding window, that is, the 3347 sampling point, and the alarm time will be obviously advanced (about 27.3 h in advance), which avoids the further aggravation of the gearbox bearing fault.

spacer