The goal of this work is to inspect metallic surfaces of engine parts in a Reconfigurable Manufacturing System (RMS) environment using machine vision. We utilize learning for superior adaptation to different parts and varying inspection conditions. Using an appearance-based approach, the inspection system automatically derives the most discriminating features from samples of a specific application. This way, it can be reconfigured to inspect different parts in different conditions without the need for reprogramming. The proposed method of inspection uses the Hierarchical Discriminant Regression (HDR) algorithm for feature extraction and classification, which is a new appearance-based classification method for machine inspection systems. The efficiency of the HDR enables fast classification and provides an adaptive, real-time inspection method (for different parts and inspection conditions). Texture types of different defects, non-defective machined surfaces and metallic casting surfaces have been classified with low error rates. In addition, texture was classified according to its local orientation, and defect boundaries were determined according to local texture contrast. Using experiments and theoretical complexity analysis, it was shown that the method is fast and can be applied in real-time online engine part inspection applications, where new parts are introduced, and conditions vary.
展开▼
