Smart defect identification for manufacturing applications

Abstract

Abstract Quality control procedures are fundamental to any manufacturing process to ensure that the product conforms to a defined set of requirements. To meet the ever-growing demand for high-quality products and address the disadvantages of manual quality control procedures, the use of intelligent visual inspection systems is gaining importance for deployment in production lines. Many works imbibing image processing techniques, machine learning, and neural network models have been proposed to perform defect detection and segmentation focused on specific domains of defects. However, defects in manufacturing manifest in varied forms and attributes which add to the woes of developing one-shot detection methodologies, while it is also expensive to generate a dataset of images capturing the variety to train a one-shot machine-learning model. This paper presents a framework consisting of three mind-maps to capture the essence of defect detection. The first proposes a classification of defects in manufacturing based on visual attributes. The second aims to identify the relevant image processing methodologies, such as thresholding, Fourier analysis, line detection, neural networks, etc. The third mapping is to relate the class of defects with the specific image processing methodologies. Taken together, the mind-maps provide the basis for the development or adaptation of defect detection approaches for specific use cases. This paper also proposes an empirical recommendation formula based on three image metrics, namely, entropy, universal Quality Index (UQI) and Rosenberger's to judge the performance of a method over a given class of images. This paper showcases the implementation of a Smart Defect Segmentation Toolbox assimilating methodologies like Wavelet Analysis, Morphological Component Analysis (MCA), Basic Line Detector (BLD), and presents case studies to support the working of the recommendation formula.