An efficient algorithm for non-convex engineering optimisation problems

Design optimisation of practical structures often entails hundreds or even thousands of design variables. This results in the fact that design space becomes non-convex. Consequently, most of the optimisers get stuck in local minima or become computationally very expensive. This work presents some examples of non-convex structural optimisation problems of truss and beam structures with up to two thousand design variables. The structures are optimised with the TRLP algorithm which includes Sequential Linear Programming (SLP) and Trust Region Method (TRM) concepts. The combination of SLP and TRM increases the design freedom. In fact, the TRLP algorithm proved itself to be more efficient than other SLP and SQP algorithms in solving non-convex design problems. Besides, the present code resulted also insensitive to changing the algorithm input parameters even in a random fashion.