An Integrated Model of kNN and GBDT for Fault Diagnosis of Wheel on Railway Vehicle

Wheelsets are critically important for the safety operation of rail vehicle. The real vibration data of railway vehicle wheel, obtained from a Chinese subway company, is used in this article. Principal component analysis (PCA) is conducted to reduce the dimension of the feature indexes. We propose an integrating algorithm with k-Nearest Neighbours (kNN) and gradient boosting decision tree (GBDT) to deal with the typical imbalance and divergent characteristics and satisfy the high requirement of fault detection accuracy. The results show that the classification accuracy of kNN-GBDT reaches 94.66%, 82.35%. While, the kNN and the SVM miss classified all fault samples into normal condition, and GBDT got an accuracy of 56.82% in fault detection. The entire process of the proposed model finished in about 0.35s. Our kNN-GBDT integrated algorithm satisfies the requirements of real-time performance and accuracy for online fault detection.