A novel approach applied to fault diagnosis for micro-defects on piston throat. (March 2021)

• Record Type: Journal Article
• Authors: Chen, Zhongxin; Zhao, Feng; Zhou, Jun; Huang, Panling; Song, Wenping
• Is Part Of: Measurement. Volume 173(2021)
• Publication Date: 2021-03
• Subjects: Piston throat 1.D -- SVM 1.D -- Active learning 1.D -- Micro-defects detecting 2.B; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2020.108508
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: The E-SVM-AL algorithm is proposed in this paper. The proposed method for preprocessing unlabeled samples. The proposed model to determine the number of labeling samples. The system based on E-SVM-AL is used in real applications. Abstract: The identification of different types of micro-defects on piston throat is of great significance to improve the efficiency of piston production and the safe operation of the engine. The two challenges to the micro-defects classification are to reduce the impact of imbalanced labeled samples on training, and to spend less cost of experts for labeling. To overcome these challenges, SMOTE and a new selecting strategy applied to the active learning on SVM algorithm (E-SVM-AL) is proposed. The strategy includes two portions: (1) a method for preprocessing unlabeled samples, which can be combined with uncertainty-based sampling criterion to improve labeling efficiency; and (2) a model to determine the number of labeling samples with experts, which is based on the current hyperplane, and can obtain an additional stopping criterion. Experiments using the system based on E-SVM-AL algorithm show the efficiency and accuracy in the classification of micro-defects on different batches of pistons.