Global optimization of a dryer by using neural networks and genetic algorithms

For many optimum design problems, the objective function is the result of a complex numerical code and may not be differentiable and explicit. The first aim is to propose a way of solving such complexity on an example problem. A novel and global strategy involving artificial neural networks and a genetic algorithm is presented ann validated for an industrial convective dryer. To begin with, a method to represent a drying model using artificial neural networks is defined. This method is tested and the results are compared with those obtained with classic numerical methods. This approach drastically reduces simulation times and maintains good accuracy and generalization properties. Second the associated optimal design problem is considered. This optimization appears as a difficult combinatorial problem with a complex objective function that involves different economical criteria. The second aim is to present the methodology to solve this problem using genetic algorithms. Final results illustrate the efficiency of this global approach.