Discrete nonlinear optimisation by constraint decomposition and designer interaction

A procedure is developed for optimising discrete non-linear problems by decomposition of the constraints and the merit function and rearranging the decomposed stages into sequential levels. Forward and backward techniques are used to overcome any infeasibility encountered in the subsystems and to improve the design merit. The complex gear train design problem investigated by Pomrehn and Papalambros [20] is used for illustrating the proposed strategy. The improvement in the design merit and the reduction in computational effort gained as a result of designer interaction are demonstrated.