Remaining Useful Life Assessment of Slewing Bearing Based on Spatial-Temporal Sequence

Slewing bearing is one of key components in the large size machinery and its remaining useful life (RUL) prediction is required to schedule a future action to avoid catastrophic events, extend life cycles, etc. The vibration-based method has been widely used in the RUL prediction. However, the spurious fluctuation usually exists in the vibration signal when the machines are operated under complex conditions. In order to enhance performance of RUL prediction model, two kinds of new health indicators are constructed by the spatial-temporal (ST) information firstly. One is the temporal indicators, which are derived by using the smoothing mean values of positive and negative vibration signal. Another is the spatial indicator, which is defined by fusing the multi-features extracted from the balance position information of vibration signal. During this process, a new data processing method proposed in this paper improves the quality of the vibration data and increases the number of samples. And then, the RUL prediction model is presented by combing the ST indicators and long-short-term memory network (LSTM) to establish the relationship between the ST indicators and the RUL of slewing bearings and overcome the sparsity of data. Moreover, in order to accelerate the adjustment of ST-LSTM model, a fine-tuning ST-LSTM model is further proposed by incorporating the generative adversarial networks (GAN) into the ST-LSTM. Experimental results verify that the proposed RUL prediction model can well estimate the RUL of slewing bearings and its performance is superior to some existing methods.