AN INTELLIGENT APPROACH TO WELDING ROBOT SELECTION

In a shipyard where multiple stationary and mobile workcells are emplo yed in the fabrication of components of complex sub-assemblies, effici ent operation requires an intelligent method of scheduling jobs and se lecting workcells based on optimum throughput and cost. The achievemen t of this global solution requires the successful organization of reso urce availability, process requirements, and process constraints. The Off-line Planner (OLP) of the Programmable Automated Weld System (PAWS ) is capable of advanced modeling of weld processes and environments a s well as the generation of complete weld procedures. These capabiliti es involve the integration of advanced Computer Aided Design (CAD), pa th planning, and obstacle detection and avoidance techniques as well a s the synthesis of complex design and process information. These exist ing capabilities provide the basis of the functionality required for t he successful implementation of an intelligent weld robot selector and material flow planner. Current efforts are focused on robot selection via the dynamic routing of components to the appropriate work cells. It is proposed that this problem is a variant of the ''Traveling Sales man Problem'' (TSP) that has been proven to belong to a larger set of optimization problems termed nondeterministic polynomial complete (NP complete). In this paper, a heuristic approach utilizing recurrent neu ral networks is explored as a rapid means of producing a near optimal, if not optimal, weld robot selection.