Optimization of process driven dimensional variations in vehicle body using genetic algorithm and finite element analysis

The vehicle body is arguably the biggest contributor to the vehicle dimensional outcome and is the most integral factor shaping the perception of the design and brand image. In car body manufacturing, one of the big challenges in high rate body production lines and a hidden contributor to dimensional quality of the body is designing an optimum weld process that delivers the minimum process driven dimensional variations. The most challenging part of the body production line is the Framing Geometry Setting Station, in which main assemblies of the body are joined to shape the bodyshell. Even in the biggest car manufacturing companies, the body process design is mainly done by process engineers relying on their individual experiences, without proper simulation or CAE analysis to optimize the processes. Through collaboration with Ford Motor Company, in this research we investigated a real industry problem using one of the Ford's products CAD data. From the cycle time assumptions, the station weld capacity was calculated as maximum of 60 welds in Framing Geometry Setting station. This meant all the main assemblies in this station had to be dimensionally set and fixed using only 60 welds. Knowing the product design of this vehicle included 850 welds for Framing process, the question was how to design a weld process of 60 welds out of 850 welds to best set the geometry and to minimize the process driven dimensional variation. Mathematically, there are unlimited solutions for a combination of 60 out of 850, hence unlimited process designs. So, the key challenge was how to design an optimum process with 60 welds that could best set and fix the geometry of the bodyshell. We first assessed the application of Genetic Algorithm in this field and performed the first level optimization through Genetic Algorithm. This resulted in around 99% reduction of non-viable process designs (solutions). Next, we investigated the application of Finite Element Analysis in assessing the process driven dimensional variations. The results of the FE simulation were assessed through evaluating the dimensional variation of 30 critical-to-quality measurement points on the FE models. Root Sum of Square (RSS) of the dimensional variation of the 30 measurement points was defined as an index to evaluate the geometrical quality of each process design. The design with the smallest RSS represented the optimum process design delivering the minimum process driven dimensional variation. Using all the findings from Genetic Algorithm models and the FE assessment, a set of process design guidelines were developed to help designing an optimum process without the need to run Genetic Algorithm or FE models. This is a novel approach that solves a real industry problem.