Real-Time Automated Visual Inspection applied to High Speed Quality control
Dr. Luis Miguel Jiménez García

The overall objective of this thesis is the study and development of new algorithms for image processing for high speed and high resolution real-time detection of defects in production lines. This thesis is presented with a series of publications according to the internal UMH regulations. This report is made up of an overall summary of the results, a discussion and final conclusions. A full copy of the content of the presented papers can be found in the addendum.

In order to segment and classify defects with the high precision and speed imposed by production systems, it is necessary to integrate, in one hand, algorithms adapted to the product to be inspected, as well as the development of specific techniques for real-time implementation using parallel computing architectures. The proposed objectives present advances in several key aspects in the implementation of Computer Vision techniques for automated inspection. These advances include both, new image processing algorithms, and also a prototype implementation, that provides innovative solutions for production lines.

Within this overall objective of the thesis, focused on automated visual inspection systems, the specific objectives are organized in two main lines:

• Automated inspection in aluminum lamination mills: An industrial inspection system that allows real-time inspection with high accuracy on we-based production is presented. Detected defects are caused by the eccentricity of rollers in lamination process (steel, aluminum, etc.). A theoretical study of the problem and a novel algorithm for detection and classification by generating an interference pattern using coherent light, is proposed. The work is complemented by the practical results of the implementation of an industrial prototype for automated inspection in aluminum lamination mills.

• Automated inspection for flexible materials with high-resolution superimposed printings: An automated visual real-time inspection system for printed products is presented. The main contribution relies on the automatic inspection of products with multiple high-resolution superimposed printings, with the peculiarity that the reference system associated with each of the prints is not coincident, and can vary between two consecutive products. This feature increases the complexity of the inspection problem avoiding the use of classical template matching techniques. The proposed solution is implemented in a prototype for inspection of bill sheets, introducing novel morphological algorithms using subpixel techniques, as well as a parallel architecture and an specific scheduling method to meet the resolution and speed requirements.