A Practical Approach to Evolutionary Algorithm based Automated Mechanism Design

Abstract

The complexity of current mechanisms continues to increase and their users keep on demanding even more. Current design methods are not sufficient to keep on fulfilling those requirements. Automated design methods have shown to be a viable solution to keep up with the demands and research has started to make this possible for mechanisms design. But there is a big difference in the predicted performance before and the performance after manufacturing. To minimize the effects of production errors while minimizing its computational costs we introduced a novel robust optimization method. In the genetic programming algorithm we add uniform noise to the production dimensions. Every generation the fitness of a mechanism is calculated with this noise and that fitness is added to a memory bank of that mechanism until that memory is full. The average of the fitness values in the memory bank is used as the fitness value to rank the mechanisms of a population. By changing the maximum bounds of the noise and the maximum amount of values possible in the memory-bank we can influence the robustness of the final mechanisms and computational costs of the algorithm. The optimal settings to get the most robust design with the least computation cost was by keeping the noise between ± 2 times the ISO-2768-m norm and limiting the maximum amount of fitness calculations in the memory-bank to 8. Further decrease of the computational cost was achieved by directly calculating the incidence matrix from the incidence string instead of the iterative method. We also improved the accuracy of the predicted performance by introducing a better weight and moment of inertia calculation based on more accurate descriptions of the shape of the mechanism. The combination of these adjustments resulted in automated mechanism design method that generates robust mechanisms for a wide range of problems within an acceptable time span.