A manufacturing quality prediction model based on AdaBoost-LSTM with rough knowledge. (May 2021)

• Record Type: Journal Article
• Authors: Bai, Yun; Xie, Jingjing; Wang, Dongqiang; Zhang, Wanjuan; Li, Chuan
• Is Part Of: Computers & industrial engineering. Volume 155(2021)
• Publication Date: 2021-05
• Subjects: Manufacturing quality -- Prediction -- Rough set -- Long short-term memory -- AdaBoost ensemble learning; Engineering -- Data processing -- Periodicals; Industrial engineering -- Periodicals; 620.00285
• DOI: 10.1016/j.cie.2021.107227
• Languages: English
• ISSNs: 0360-8352

Abstract

Highlights: A reinforce learning framework with rough knowledge is proposed for manufacturing quality prediction. RS theory is utilized for importance assessment of process parameters. LSTM is utilized as the regression tool for manufacturing quality prediction. AdaBoost algorithm is utilized for model reinforcement of LSTM. Abstract: Manufacturing quality prediction is one of the significant concerns in modern enterprise production management, which provides data support for reliability assessment and parameter optimization, thus improving the intelligent management level of enterprises and helping achieve high-quality products at lower costs. In this paper, an ensemble learning framework using rough knowledge is proposed for manufacturing quality prediction. The proposed model consists of three elements: (1) significant parameters in different production stages are selected based on attribute reduction and decision rule extraction of rough set theory (RS), (2) long short-term memory network (LSTM) is utilized for building the relationship between the significant parameters and manufacturing quality, and (3) the learning performance of the LSTM is reinforced by AdaBoost approach. To estimate the effectiveness of the proposed model, a competition dataset about manufacturing quality control is applied and six models are investigated. The comparison experiments show that the proposed model overwhelms all the comparison models in terms of root-mean-square error, threshold statistics and residuals analysis. In addition, the proposed model has statistically significant difference from all the comparative models. It is recommended from this work that the ensemble learning technique integrating the rough knowledge synchronously improves the sensitivity and regression capacity of the model.