Constraint feedback in solving incomplete models: a case study in sheet metal bending

We deal with the problem of generating process plans in a real-life engineering domain, the bending of sheet metal parts. In this domain where geometry plays a key role, traditional engineering knowledge consists of fragmentary, sometimes contradictory pieces of advice and the solutions are evaluated according to multiple criteria. We apply a planning method that extends constraint programming. The solution process is based on the communication of a general-purpose constraint solver and a domain-specific geometric expert. The geometric module works on an exact spatial representation of the workpiece, machine and tools and generates and successively refines constraints that the solutions must satisfy. The other module solves the dynamically evolving constraint models by combining techniques of constraint propagation, branch-and-bound search and multi-criteria optimization.