A finite element model was developed, using weld techniques established within TWI, which simulates each of the 22 tack welds in sequence and predicts the resulting distortion. Figure 1 shows the typical distortion of the weld assembly that can arise due to different tack weld sequences.
The number of possible weld sequences – 22 factorial or approximately 1021 – is far too large to enable analysis of each one. TWI’s solution therefore was to develop a genetic algorithm (GA) which generates a set of sequences (a ‘generation’), identifies the most successful and uses them to initiate the next generation.
A computer programme was developed which uses the GA to drive the inputs to the FEM and feedback the results. In this way, progressively better sequences are identified and analysed.
The graph in Figure 2 summarises the predicted distortion for almost 3000 simulations (49 generations of the GA). The prediction for the current process is shown as a dashed line.
The results show that there are a small number of weld sequences which cause comparatively large distortions. However the GA routine identifies a number of possible weld sequences that can improve and minimise the distortion compared to the existing manufacturing sequence. As a result, current work is now being experimentally validated as part of the Optipress project.