DEVELOPMENT OF A FRAME-BASED MODEL FOR CUT-OFF DIMENSIONING
Abstract
The cut-off optimization problem has traditionally been solved using mathematical, or computer-simulation algorithms. Owing to the unpredictability of the material, and the diversity of constraints to satisfy, conventional methods fall short of properly handling the problem. A modeling approach that revolves around the frame data-structure is proposed. It authorizes to represent the 'qualitative' nature of the problem; it also allows the embodiment of various forms of knowledge, within a cohesive shell. The knowledge identified is: (i) Qualitative knowledge; to represent the tolerability of instances of defects according to their characteristics. (ii) Topological knowledge; to manage the physical spread of defects. (iii) Production data; to consider the dynamic status of the cutting bill and to adaptively modify the 'importance' of the cuttings. A synergy of knowledge handling techniques is provided through the frame. Qualitative knowledge is managed via a rule-based forward chaining inferencing engine. Topological knowledge is managed via both heuristic methods, and a dynamic program algorithm. Heuristics are invoked to control the evolution of the state of completion of the cutting bill. The system proposed considers both sides of a blank. The selection of a side depends on the 'quality' requirements imposed by the finished part. Ripping is then performed on the most appropriate face of the blank. A value is tagged to each candidate blank to indicate the goodness of the rip. 'Look ahead' is then performed on the yield of the board-residual, in order to establish the impact of selecting a blank. Proper selection is also assessed in terms of the rate of completion of the cutting-bill. A preference value function combining all three objectives is used to select the most appropriate cutting. A heuristic tree search is performed using a breadth-first selection strategy. Local optimums are sought, that are returned by the preference value function. They depend on the dynamic state of the system. The search space is directly proportional to the number of cutting stages. This drops the complexity to an order of 10$\sp{3}$, as opposed to 10$\sp{8}$ when using exhaustive enumeration on a problem of the same size. The yield obtained with the proposed method fluctuates aroung 60%.
Degree
Ph.D.
Subject Area
Industrial engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.