Intelligent fault diagnosis for rail transit switch machine based on adaptive feature selection and improved LightGBM. (June 2023)

• Record Type: Journal Article
• Authors: Lao, Zhenpeng; He, Deqiang; Wei, Zexian; Shang, Hui; Jin, Zhenzhen; Miao, Jian; Ren, Chonghui
• Is Part Of: Engineering failure analysis. Volume 148(2023)
• Publication Date: 2023-06
• Subjects: Switch machine -- Fault diagnosis -- Features selection -- Loss function -- LightGBM ensemble learning; System failures (Engineering) -- Periodicals; Fracture mechanics -- Periodicals; Reliability (Engineering) -- Periodicals; Pannes -- Périodiques; Rupture, Mécanique de la -- Périodiques; Fiabilité -- Périodiques; Fracture mechanics; Reliability (Engineering); System failures (Engineering); Periodicals; Electronic journals; 620.112
• DOI: 10.1016/j.engfailanal.2023.107219
• Languages: English
• ISSNs: 1350-6307

Abstract

Highlights: An improved LightGBM fault diagnosis method is proposed for turnout switch machine. Multiscale permutation entropy is applied to extract fault feature information. An adaptive features selection method is proposed to select the optimal feature. An improved Focal Loss function is proposed to enhance the identification ability. Abstract: The turnout switch machine is the critical equipment of the signal system, which has a significant influence on the efficiency and safety of train operation. However, most fault diagnosis technologies of the switch machine are difficult to distinguish samples with similar categories, which leads to the low diagnostic accuracy. Thus, a fault diagnosis method based on improved LightGBM is proposed to deal with the above problems. Time domain features and multi-scale permutation entropy are extracted to capture the weak fault. Moreover, an adaptive feature selection (AFS) method is presented to reduce redundant features. Especially an improved Focal Loss (IFL) function is established, which improves the ability to distinguish samples of similar features in a multi-classification model. The three-phase action current from the switch machine is utilized to testify to the proposed method and compare it with other methods. The experimental results show that the diagnosis accuracies of this method in the normal-reverse and reverse-normal conversion process reach 98.47 % and 96.09 %, respectively, which is well-suitable for practical application.