Quality prediction modeling for multistage manufacturing based on classification and association rule mining

Abstract

Abstract. For manufacturing enterprises, product quality is a key factor to assess production capability and increase their core competence. To reduce external failure cost, many research and methodology have been introduced in order to improve process yield rate, such as TQC/TQM, Shewhart Cycle Deming's 14 Points, etc. Nowadays, impressive progress has been made in process monitoring and industrial data analysis because of the Industry 4.0 trend. Industries start to utilize quality control (QC) methodology to lower inspection overhead and internal failure cost. Currently, the focus of QC is mostly in the inspection of single workstation and final product, however, for multistage manufacturing, many factors (like equipment, operators, parameters, etc.) can have cumulative and interactive effects to the final quality. When failure occurs, it is difficult to resume the original settings for cause analysis. To address these problems, this research proposes a combination of principal components analysis (PCA) with classification and association rule mining algorithms to extract features representing relationship of multiple workstations, predict final product quality, and analyze the root-cause of product defect. The method is demonstrated on a semiconductor data set.

Machine Learning Methods for Quality Prediction in Production

“…The trained model could then be used to predict the quality of units passing through a production line in real-time and contain suspected units before they escape the factory. Applications of machine learning in studies regarding quality in various manufacturing operations had yielded significant success in semiconductor manufacturing [4,41] and additive manufacturing [12].…”

“…However, there have been relatively fewer applications that aim to predict the compliance quality of finished goods from data pertaining to a sequence of operations, on a production line. The research by Kerdprasop et al could predict quality defects from multistage semiconductor manufacturing with Cohen's Kappa of 0.57 [3], while the work by Kao et al and Arif et al in the same context could predict defects with Cohen's Kappa of 0.98 [4] and 0.04 [5], respectively. However, these papers did not use classification algorithms like neural networks or bagged and boosted ensembles.…”

Methodological Assessment of Data Suitability for Defect Prediction

“…Lieber et al (2013) describe a methodical framework based on data mining for predicting the physical quality of intermediate products in interlinked manufacturing processes in the context of a rolling mill case study. Other approaches to defect prediction also do not include specific modules for they systematic assessment of the available database (Arif, Suryana and Hussin, 2013;Kao et al, 2017;Wuest, Irgens and Thoben, 2013;Schmitt and Deuse, 2018).…”

A generative neural network model for the quality prediction of work in progress products

“…Chou et al, [31] successfully combined genetic algorithm with SVMs, to predict the wafer quality. Arif et al [32], suggested a mixture PCA method with an Iterative Dichotomiser algorithm (ID3), for multi-stage quality prediction, while Kao et al, [33] adopted classification and rule mining approach for solving a similar problem. Diao et al [34], introduced another analytical tool called an improved dominant factor (DFs), which is based on an improved principle component analysis (iPCA), for dynamic quality control.…”