Fault Diagnosis of Single Yaw Damper Utilizing Hierarchical Multi-class Classifier

10.3966/199115992017102805009

Daochao Tang；Na Qin；Weidong Jin；Peizhen Xu

Bayesian error estimator ； fault diagnosis ； hierarchical classifier ； yaw damper

電腦學刊

28卷5期（2017 / 10 / 01）

94 - 104

英文

In this paper, we propose a hierarchical multi-class classification approach which is optimized for single yaw damper fault problems. This novel approach associates with an original process of fault detection and localization which is arranged into support vector machine (SVM) with binary tree architecture. In fault detection, the developed method based on concatenated One-Against-One SVMs can significantly reduce the miss rate. Then, the hierarchical structure is built via iteratively partitioning the car bogie structure during fault location. This algorithm is very appealing as it takes advantage of the decision tree architecture and of SVM. Furthermore, the selection of error penalty factor C affects the precision of SVM due to its ability to avoid over fitting. In this paper, the Bayesian error estimator (BEE) which describes the error in a Bayesian framework is applied to obtain the optimal value of C . The effectiveness of this approach is illustrated experimentally on CRH3 EMU vehicle system.