Robust Parameter Design of Dynamic Systems With Mixed Response Types: A Learning‐Based Optimization Method

ABSTRACT

Managers and industrialists constantly strive to improve quality to remain competitive and increase profitability. One of the quality engineering tools is the multi‐response surface methodology, which simultaneously considers multiple responses with different optimization directions, including larger‐the‐better, smaller‐the‐better, and nominal‐the‐best. These problems are mainly categorized as nonlinear programming models and become more complex when the quality characteristics are supposed to meet targets at multiple time points. The present research aims to propose a model and an efficient solution method for optimizing multi‐response problems with time‐oriented responses based on robust parameter design, accounting for all three types of quality characteristics and response correlations. In the proposed approach, the Taguchi multivariate loss function has been extended to aggregate the multiple objective functions, and artificial neural networks have been applied for computational ease. Finally, the proposed model was solved using genetic algorithm, simulated annealing, and classic optimization algorithms. The results indicate that simultaneous optimization of multiple responses, along with consideration of the time factor, provides more information and yields more realistic results. Moreover, implementing artificial neural networks coupled with metaheuristics facilitates the global optimization procedure.