Development of an Autograding System for Weld Bead Surface Quality using Feature Extraction and Mahalanobis-Taguchi System

Autograding systems are becoming more prevalent to address the challenges inherent in teaching and learning assessment. Over the past few decades, technological advancements have increased image processing techniques, including pattern recognition research. The Mahalanobis-Taguchi System (MTS) is a technique for assessing system performance by analyzing multivariate data to make quantitative choices via the development of a multivariate measurement scale. This study intends to develop a grading tool that combines a feature extraction technique with MTS theory, which instructors will utilize at the UNITEN Manufacturing Processes Laboratory to assess the quality of weld bead surface work prepared by UNITEN students. The Mahalanobis Distance (MD) will distinguish between normal and abnormal extracted image patterns from workpieces and transform them into a measurable scale. The samples defined better grading with lower MD. A jig was developed to collect consistent and accurate image data for the image-capturing process. The results showed that out of 10 test samples, 2 samples were classified as normal with a grading range between 75% to 82%. Another sample was classified as gray regions, with grading ranges between 65% and 74%. The remaining 6 samples were classified as abnormal, with a grading range between 40% to 64%. An autograding tool for evaluating welding surface quality utilizing MD scales was established.