A hybrid genetic algorithm for multiresponse parameter optimization within desirability function framework

The desirability function method is widely used for simultaneous optimization of several independent or uncorrelated responses. In practical applications, a second-order polynomial is usually employed to represent each response based on RSM. Then a functional form for the overall desirability will be reached. The so-obtained overall desirability function is usually nondifferentiable, highly nonlinear and multimodal for practical problems, especially when there are quite a number of responses and design variables. This paper proposes a hybrid approach, which merges a global search procedure, the genetic algorithm, with a local search procedure, the pattern search method, to tackle this kind of problems. A numerical example from literature is discussed for illustrative purpose. Results reveal that the proposed approach has good convergence characteristics.