Evolutionary computation in machine vision for manufacturing and logistics industry

Abstract

In the last decade, machine vision technology has improved productivity and quality control processes. Since China continues to develop an infrastructure for Hi-Tech products and the electronics industries while Hong Kong keeps developing the logistics industry, automation solutions are increasingly important in these areas. Thus, there is a growing trend to employ machine vision systems to perform tedious and demanding visual inspection tasks which are traditionally performed by human experts or highly trained operators. In typical cases, industrial visual inspection system is designed to inspect only known objects at fixed positions. These inspection systems have been considered as mature tools, however, application boundaries continue to move outward. There is a growing research curiosity in the development of new vision systems because more and more emerging vision-based problems are identified when the boundaries keep expending. Researchers and developers now consider machine vision as a discipline to develop advanced algorithms and faster hardware in order to further broaden the boundaries. Among the different directions in this research and development trend, evolutionary computation techniques are gaining relevance as there are numerous applications in different areas where they can be applied. Although a number of successful examples of such applications can be found in the literature, they are usually specific to one single problem and reported in a scattered manner. This research is an attempt to propose a schema for incorporating the proposed chaos-based particle swarm optimization (PSO) algorithms in machine vision expert systems. It can be considered as a generic approach to solve other vision-based problems such as inspection, monitoring, guidance and navigation problems in different areas. Benchmarking methods and exploratory case study approaches are adopted as the research method. Although the attempt is mainly research-oriented, particular care has been given to test the feasibility of the proposed approaches, in the thesis, at a level which makes them accessible to researchers and industrial practitioners. Since the expected target of this research includes industrial practitioners and researchers in machine vision who may not be familiar with evolutionary computation techniques, the main focus on applications of the proposed evolutionary approaches are described in detail, and the approaches are tested carefully and compared with other approaches. The chaos-based PSO algorithms that are proposed are also simulated and benchmarked carefully before being applied to solve the formulated vision-based problems. The proposed algorithms outperform other algorithms in both the benchmark tests, in template matching and in circle detection problems. The tests and simulations have revealed the efficiency and feasibility of the proposed algorithms in solving numeric optimization problems as well as vision-based problems. To validate the feasibility of the proposed approaches, two case studies concerned with printed circuit board manufacturing and pick-and-pack processes have been conducted. Furthermore, the design of the proposed system, which encompasses statistical control charts, information handling and digital watermarking, has been described. The suggested solutions have been considered as effective means to address the problems in the case studies. The aim of this research is to explore how evolutionary approaches can be used to solve vision based problems, as well as tackling problems in the manufacturing and logistics industries. The deliverables of this research not only provide the design of the proposed system which will support efficient quality decision making and improve operational performance, but also open the door to incorporating evolutionary computation techniques in future machine vision systems that can be applied to different areas.