Balancing optimization of robotic welding lines: model and case study

Abstract

Robotic welding manufacturing lines are production lines common in automobile industries. During a vehicle's production, the vehicle's metal structure must be welded in a single resistant body. This is made by hundreds of spot-welding points, each of which tie locally two or more metal plates. Efficiently distributing these welding points amongst robots is particularly challenging, taking in account that: not all robots can perform all weld points, robots must move their welding tools between weld points, and robots might interfere with one another if they use the same geometrical space. There are multiple feasible manners to distribute the welding points. However, each of these forms generates different economical results: If a robot performs too many points, it will become a line bottleneck and reduce average throughput. To find the set of operational decisions that yields the best output is the goal of optimization techniques. There are a wide variety of such techniques described in operations research and computer sciences literature: mathematical models, algorithms, heuristics, meta-heuristics, etc. In the industrial context, these techniques were adapted to related line balancing problems. However, these adaptations can only solve the specific variants they were designed to address. While parallels can be drawn between aspects of robotic welding lines and many of such variants, the full combined set of characteristics of the studied lines is not treatable by (or convertible to) any of them. This dissertation develops a framework to optimize such lines, based on mixed-integer linear programing model developed to describe the problem. It also presents a case study to discuss and illustrate possible difficulties and how to overcome them. The presented model was applied to data from the factory's robotic welding lines composed of forty-two robots (divided in thirteen stations), four vehicle models and over seven hundred welding points for each vehicle. The weighted average reduction percentage in cycle time obtained by the model was 17.5%. Model variants, designed to aid further works are presented and discussed.