Discrete parameter-nonlinear constrained optimisation of a gear train using genetic algorithms

This paper investigates the optimal design of a four-stage gear train using genetic algorithms. Five different genetic encoding schemes, which incorporate various heuristic search techniques, are proposed to deal with the most critical constraints of the problem. The fitness criterion used by all genetic algorithms includes a merit function for minimising the size of the gearbox. The results show improvement in the design merit over previous approaches without reliance on the designer's interaction to avoid geometric constraint violations and facilitate the convergence.