Neurocomputing strategies for solving reliability-robust design optimization problems

Purpose - This paper, by taking randomness and uncertainty of structural systems into account aims to implement a combined reliability-based robust design optimization (RRDO) formulation. The random variables to be considered include the cross section dimensions, modulus of elasticity, yield stress, and applied loading. The RRDO problem is to be formulated as a multi-objective optimization problem where the construction cost and the standard deviation of the structural response are the objectives to be minimized. Design/methodology/approach - The solution of the optimization problem is performed with the non-dominant cascade evolutionary algorithm with the weighted Tchebycheff metric, while the probabilistic analysis required is carried out with the Monte Carlo simulation method. Despite the computational advances, the solution of a RRDO problem for real-world structures is extremely computationally demanding and for this reason neurocomputing estimations are implemented. Findings - The obtained estimates with the neural network predictions are shown to be very satisfactory in terms of accuracy for performing this type of computation. Furthermore, the present numerical results manage to achieve a reduction in computational time up to four orders of magnitude, for low probabilities of violation, compared to the conventional procedure making thus feasible the reliability-robust design optimization of realistic structures under probabilistic constraints. Originality/value - The novel parts of the present work include the implementation of neurocomputing strategies in RRDO problems for reducing the computational cost and the comparison of the results given by RRDO and robust design optimization formulations, where the significance of taking into account probabilistic constraints is emphasized.