In this thesis, the design and development of an industrial prototype for herring roe grading is presented. In the fish processing industry of British Columbia, grading of herring roe is currently done manually. The highly repetitive and laborious activities that are involved in grading of herring roe, and the problem of maintaining a skilled labor force for a seasonal industry that is active for less than 3 months/year, are motivating factors for automation of the grading process. For this purpose, sensor technology for analyzing shape, size, firmness, color, and weight of herring roe skein has been developed in the Industrial Automation Laboratory. An intelligent decision making algorithm has been developed as well, incorporating this sensor technology, employing fuzzy logic to infer a classification measure. Further development of the system requires the integration of both hardware and software in a functional and automated herring roe grading prototype. System design, integration and performance evaluation are also presented in the thesis.

A typical herring roe processing operation in industry is described in the thesis. It will highlight problems in manual grading. It will also serve to identify appropriate technology for an automated grading machine. The conceptual design of each component of a system for automated grading of herring roe is presented. A control system, with appropriate component level and monitoring functions, in a multitasking environment is developed for the overall grading system.

An industrial prototype using a CCD camera and optical object-detection sensor as the key sensing means, along with a variety of mechanical and control devices has been developed by us for herring roe grading. The selection of the conveyor system and the pneumatic valves of the roe ejection system that has been implemented is described. A special illumination system has been designed to generate high-quality images of roe. A control function to execute various tasks in the prototype machine also has been developed and implemented. The present version of the machine is capable of grading herring roe by shape and weight at a rate of 1 skein/second. Two refinements in weight estimation and broken-tail analysis are developed and incorporated in the machine.

Testing of the prototype, both in our laboratory and at an industrial plant, has shown that the accuracy of selecting out Grade 1 Large skeins from a typical, industrial batch is approximately 94%, with our prototype. The experimental results are evaluated, and a possible application for the current version of the grading machine is given. A cost-benefit analysis is also carried out for the present version of the machine, and further improvements are recommended for the automated herring roe grading system.