A hybrid multi-class imbalanced learning method for predicting the quality level of diesel engines. (January 2022)

• Record Type: Journal Article
• Authors: Qin, Wei; Zhuang, Zilong; Guo, Liangxun; Sun, Yanning
• Is Part Of: Journal of manufacturing systems. Volume 62(2022)
• Page Start: 846
• Page End: 856
• Publication Date: 2022-01
• Subjects: Quality level prediction -- Multi-class imbalanced learning -- Dynamic oversampling -- Feature selection; Manufacturing processes -- Periodicals; Production engineering -- Data processing -- Periodicals; Robots, Industrial -- Periodicals; Production, Technique de la -- Informatique -- Périodiques; Robots industriels -- Périodiques; Electronic journals; 670.42
• DOI: 10.1016/j.jmsy.2021.03.014
• Languages: English
• ISSNs: 0278-6125

Abstract

Highlights: Develop a hybrid imbalanced learning method for predicting the quality level of diesel engines. PSO algorithm is applied to adaptively select the optimum feature subset. DySBoost is utilized to address the multi-class imbalanced classification problem. Experimental results demonstrate the superiority of the proposed PSO-DySBoost. Abstract: Establishing an effective quality level prediction of diesel engines is of great significance for controlling the production quality through subsequent process improvements and reducing manufacturing costs. Data-driven methods have become the most promising methods due to their self-learning ability and independence of complex manufacturing processes. However, the dataset collected from a diesel engine assembly line is characterized by high dimensionality and multiclass imbalance. To realize the high-accuracy quality level prediction of diesel engines, this study develops a novel hybrid multi-class imbalanced learning method that combines particle swarm optimization (PSO) based wrapper method and dynamic oversampling approach with adaptive boosting (DysBoost), where PSO based wrapper method is applied to adaptively select the optimum feature subset and DySBoost is utilized to address the multi-class imbalanced classification problem. Computational experiments are conducted on three standard imbalanced datasets and one diesel engine assembly dataset, and the experimental results demonstrate the effectiveness of the proposed method in terms of accuracy and efficiency.