Deterioration Trend Prediction of Spalling Damage for Heavy-Haul Railway Based on PCA and TCN-Attention

This paper proposes a new prediction method for damages such as spalling and wear of heavy haul railways caused and deteriorated easily by the continuous rail-wheel contact during the high intensity and heavy load transportation. Based on the combined model of PCA and TCN-Attention, it is capable of accurately predicting the deterioration trend of spalling damage for heavy haul railways. Firstly, a feature set with high dimension is constructed with the features of time domain and frequency domain extracted from the original vibration signal of rail damage. Secondly, PCA is used to reduce the dimension of the high-dimensional feature set of rail damages. The first principal component is used as the health indicator of rail damages. Further, the TCN-Attention model is adopted to predict the deterioration trend of the rail spalling damage. Finally, the feasibility of the proposed method is verified by using the vibration data of the life cycle of a rail track, from normal working to damaged to failure, collected from a railway depot. The results of the proposed method are also compared with those of prediction models constructed by TCN, LSTM and GRU networks. The comparison shows that the proposed method has better performance and higher prediction accuracy. The proposed method can more accurately predict the deterioration trend of the rail spalling damage.