The RNVP-based process monitoring with transforming non-normal data to multivariate normal data. (February 2023)

• Record Type: Journal Article
• Authors: Lee, Chang Ki
• Is Part Of: Engineering applications of artificial intelligence. Volume 118(2023)
• Publication Date: 2023-02
• Subjects: Statistical process monitoring -- Deep learning -- Non-normal data -- Generative model -- Change-of-variable; Engineering -- Data processing -- Periodicals; Artificial intelligence -- Periodicals; Expert systems (Computer science) -- Periodicals; Ingénierie -- Informatique -- Périodiques; Intelligence artificielle -- Périodiques; Systèmes experts (Informatique) -- Périodiques; Artificial intelligence; Engineering -- Data processing; Expert systems (Computer science); Periodicals; 620.00285
• DOI: 10.1016/j.engappai.2022.105623
• Languages: English
• ISSNs: 0952-1976

Abstract

Abstract: In the modern industry, process monitoring with control charts is essential to improve processes by decreasing defects. Control charts, which are based on grounded statistical theory, have been used as an online process monitoring method to detect out-of-control. However, because control charts assume that the process data are normally distributed (a.k.a. the assumption of normality), they do not perform as well as failing to detect out-of-control when assumptions are not held. In practice, most process data violate the assumption of normality, the application of control charts in real manufacturing sites is limited. Therefore, to address this limitation, this study proposes a real-valued non-volume preserving (RNVP)-based control chart. The proposed method first transforms the process data to follow a multivariate normal distribution, and then monitors the transformed data using a control chart. As a result of conducting numerical experiments to evaluate the effectiveness of the proposed method, it was found that the performance of the proposed method was superior to that of the existing control charts in terms of Type II error rate and average run length. Graphical abstract: