Intelligent inspection of surface defects in metal castings using machine learning

Abstract

Metal casting is considered to be one of oldest manufacturing processes, and its root has been found back to 5000 years ago. It is mainly driven by different sub-processes including pattern making, mold making, melting and pouring. Metal casting is usually employed to produce metallic components that further used in different industrial sectors including aerospace, automobile, chemical, bio-medical, defense, etc. These industrial casting are required to be conformed with desired mechanical properties and absence of defects. The occurrence of defects is mainly related to geometrical inequalities, and surface as well as sub-surface discontinuities. Various techniques are employed to detect these defects however detection of surface discontinuities is relatively challenging as it requires decent technical skills as well as domain knowledge. In the present work, intelligent inspection for detection of surface discontinuities has been developed. Metal casting images were captured using automated camera, and they were further preprocessed for removing noise in images using Gaussian filter. Different feature extraction algorithms Harris, Otsu, Hough and Canny were used to extract various topographical features including corners, contours, discontinuities as well as edges. Two models, Support Vector Machine (SVM) and K- Nearest Neighbor (KNN) models were trained on topographical features extracted from more than 1400 images Aluminum metal casting. These models were tested on the data set of more than 350 images. Both models performed better in inspection of discontinuities in metal castings however SVM model provided more accurate results. The accuracy level of both models can further be improved by improving environment conditions, resolution of images as well as incorporating intensity invariants object recognition techniques.