A Multi-Objective Fitness Function for Sequencing Robotic Assembly Operations with Deformable Objects Using a Genetic Algorithm with Constraint Satisfaction

Constructing and optimizing an operation sequence is a major concern in production. However, designing high quality sequences is difficult and assembly sequence planning (ASP) is a NP-hard problem. Due to inherent process uncertainties production processes which include robotic manipulation of deformable objects, ASP can be even more complex than when only rigid objects are manipulated. Genetic algorithms (GA) are a commonly used heuristics for ASP. GA is suitable for ASP of robotic operations with deformable objects, which typically require addressing multiple objectives and which have complex production constraints. However, not all constraint may be explicitly known during the sequence design time. The current work examined ASP with different levels of known production process constraints. Integrating an arc consistency algorithm (AC3) for constraint satisfaction in the initial population generation process was implemented and compared to random population generation. Integrating process duration and the longest common sub-sequence (LCS) index (to sequences of similar products) in the fitness function was implemented and different integration methods were compared. Results show that the effects of using AC3 for initial population generation depend on chromosome length and are not related to the rate of addressed constraints. For long chromosomes AC3 based generation is considerably faster than random generation. The impact of adding LCS to the fitness function depends on the rate of constraints addressed and is not related to chromosome length. When not all the production constrains are addressed LCS increases the number of feasible solutions obtained and is not related to chromosome length.