A Genetic Algorithm Approach to Multiple-Response Optimization

An artificial neural network approach to multiple-response optimization

“…Myers andMontgomery [15] state that this approach is effective when only a few control factors are involved. The second category discussed by Ortiz et al [17] consists of approaches which can be used to formulate the problem in the form of a constrained optimization problem. Kim et al [10] refer to this category as priority-based approaches.…”

“…Ortiz et al [17] classify the existing methods into three basic categories. The performance of each method depends on the size and complexity of the problem.…”

“…For quantitative responses, the use of individual desirability functions proposed initially by Derringer and Suich [5] and modified later by Ortiz et al [17] is recommended. Derringer and Suich [5] define the individual desirability as:…”

“…However, for differentiable objective functions, one can consider gradient-based methods, such as generalized reduced gradient (GRG) [4]. Heuristic search procedures, such as genetic algorithms (GA), simulated annealing (SA), or tabu search (TS), could be applied to highly nonlinear, complex, and badly behaved surfaces [17].…”

“…It utilizes neural networks to estimate the relation between control factors as inputs and responses as outputs at the first phase, unconstrained desirability functions combined with penalty functions (developed by Ortiz et al [17]) at the second phase, and, finally, a GA as a potent optimization tool at the optimization phase. This procedure enables us to take advantage of neural networks' capabilities in function approximation, the potential of desirability functions in weighting individual responses, and also the aptitude of GAs in optimizing highly nonlinear, complex, and badly behaved functions.…”