A systematic review on machine learning methods for root cause analysis towards zero-defect manufacturing

Abstract

The identification of defect causes plays a key role in smart manufacturing as it can reduce production risks, minimize the effects of unexpected downtimes, and optimize the production process. This paper implements a literature review protocol and reports the latest advances in Root Cause Analysis (RCA) toward Zero-Defect Manufacturing (ZDM). The most recent works are reported to demonstrate the use of machine learning methodologies for root cause analysis in the manufacturing domain. The popularity of these technologies is then summarized and presented in the form of visualizing graphs. This enables us to identify the most popular and prominent methods used in modern industry. Although artificial intelligence gains more and more attraction in smart manufacturing, machine learning methods for root cause analysis seem to be under-explored. The literature survey revealed that only limited reviews are available in the field of RCA towards zero-defect manufacturing using AI and machine learning; thus, it attempts to fill this gap. This work also presents a set of open challenges to determine future developments.

Applying Machine Learning to Construct a Printed Circuit Board Gold Finger Defect Detection System

“…(5) Visual Defect Detection using CNN: Jha et al's [7] survey of current approaches in visual defect detection focused on CNNs and pixel-level segmentation techniques. (6) Root Cause Analysis: Papageorgiou [8] provided a systematic review of ML methods for root cause analysis, utilizing techniques such as Principal Component Analysis and Dynamic Time Warping.…”

“…The learning process is shown in Figure 4, in which the red frame is the preset anchor frame, and the green box is the real feature box. For coordinate translation, the solid red box is first translated to the dotted red box using Equations ( 6) and (7), and then the dotted red box is scaled to the green box according to Equations ( 7) and (8), in which (G x , G y ), G w , G h are the center coordinates, width, and height of the real feature box, respectively, and (c x , c y ) is the upper-left coordinate of the grid.…”