Fault diagnosis for rail turnout system based on improved introspective CBR method

Turnout system has great impact on the safety of China high-speed railway, thus the fault diagnosis process of the turnout becomes an urgent and challenging task. Aiming at the diagnosis issue, an improved CBR approach is introduced. First, a compound distance metric including a dynamic time warping metric and an improved Euclidean distance metric with feature extraction is introduced to give a more accurate retrieval solution on the data sequences, then for the extracted features, a dynamic feature weight allocation method based on introspective learning is adopted to increase the diagnosis performance of turnout. The evaluation of the method was carried out with a real turnout data set, the comparison experimental results with other commonly used methods show that the proposed method achieves the best diagnosis performance, proves the effectiveness of improved CBR method and the proposed method could enhance the diagnosis performance, and guarantee the safety of high-speed rail way.