Object recognition within smart manufacturing

Process automation has become a norm within industry with cheap and easily accessible automation technology becoming a standard option available to manufacturing firms. However, without the implementation of flexibility in manufacturing by the employment of intelligent systems, the technology will be limited in application. The development of smart sensing technologies has allowed for modularity and versatility to become familiar terms on a manufacturing floor, notwithstanding, it is still widely recognised that a human employee is the most valuable and flexible asset a company may have. Automation falls short in terms of flexibility due to its lack of independence during operations with high levels of variance, such as varying target position from cycle to cycle. Processes with high levels of variance disallow employment at a satisfactory level of standard or more traditional automation methods due to the lack of ability of current systems to deal with the unexpected. This paper aims to examine the technology used, and that can be potentially used in processes with high levels of variance, specifically, vision systems used in collaboration with an algorithmic comparison to compare an obtained image to an image or 3D model of the target for target recognition/object identification. While there have been copious experiments to employ imaging technology for object identification, some systems do currently occupy the factory floors of manufacturing facilities for recognition. Many of these systems run on RFID, barcode or fiducial marker. These technologies, while operational, require a pre-emptive effort to be made to ensure all products or objects to be recognised have an identifying marker attached before recognition is possible. This substantially limits the flexibility of manufacturing as the versatility of a process line to adapt to different products needs to be a possibility without the necessity of rebranding or retagging each object or product, causing a decelerative rate of production. The first section in this paper identifies the most commonly used methods of object recognition and the necessary modules required for each different algorithmic architecture. The need for particular architectures, depending on object illumination, shape, texture etc., will be addressed in the second section, as well as the differences in the construction of these architectures by use of different combinations of modules. Finally, the third section aims to address the best industrial practice and the opportunities currently being offered by research. It is the aim of the authors to determine where there are gaps to be filled between industry and research. This paper will identify areas of research that need to be examined in order to close the gap between theory and practice in research and industry respectively, to allow industry 4.0 to become a reality across the board for manufacturing.