Multi-objects Optimization Design Of Extrusion-die System Using Genetic Algorithms

The design of extrusion-die is a very complicated process which needs compositive analysis and repeat operation. The design with modeling, optimizing and flexibility is developing forward integrating, intelligent, auto-matization. Many numerical optimization methods have been developed and used for design optimization of extrusion-die system. Most of these available extrusion-die system analysis programs make use of gradients to search feasible design parameters to achieve optimal objective functions. These methods are reasonably effective for well-behaved objective functions, because the gradient of the function helps to guide the direction of the search. However, when the continuity and existence of derivatives of objective function are not present, gradient methods lack robustness and may become trapped in local optima. These problems with the application of numerical optimization are difficult to over come. The development of faster computers has allowed implement of more robust and efficient optimization methods. Genetic algorithms are one of these robust methods. Genetic algorithms are classified as guided random search techniques. They use objective function information instead of derivatives.In this dissertation, we make a full improving research for the disadvantages of a binary encoding simple genetic algorithm(SGA),such as premature convergence, lower seeking efficiency and difficultly selecting parameters, etc. Based on this, a float-encoding efficient non-gradient optimization method which need not to take too much cost of differential equation solution ----- multi-population parallel genetic algorithm based on real (RPGA) have been developed and presented for design optimization of extrusion-die system. The RPGA has multiple, independent sub-populations, each sub-population evolves using a steady-state genetic algorithm, but each generation some individuals migrate from one population to another. The migration algorithm is deterministic stepping-stone, each sub-population migrates a fixed number of its best individuals to its neighbor. Through De Jong functions test, it is demonstrated that the RPGA is more efficient and robust than a standard GA. In order to solve the multi-objective optimization of the extrusion-die system, a new multi-objective optimization technique based on the RPGA and complex shape method has developed. In this approach, sub-populations of the next generation are reproduced from the current population according to each of the objectives, separately. Using RPGA operators, the selection method is repeated for each individual objective to fill up a portion of the mating pool, then the mating pool is shuffled and the other operators (crossover and mutation) are performed, and the new population is divided into sub-populations randomly. Combined with complex shape method, by the operation of reflection, constriction, constricting to the best points and turn operation, the satisfaction of decision-maker is reflected. The synthesized gene helps to accelerate the optimal speed, and diversify the mating pool population. It is proved though some examples and numerical simulation that this method is feasible and superior. According to the theoretical results above, this paper puts forward for design optimization of extrusion-die system, which integrates existing engineering numerical analysis(Rigid-viscoplastic FEM), intelligent technology(BP Neural Network) and genetic algorithms(RPGA). First, a series of experiential parameter are selected to investigate the effect of process variables, numerical analysis construct the learning samples for training the Artificial Neural Network. Then, the fitness values are obtained on basis of a BP neural network, the inter-action of RPGA and Artificial Neural Network is utilized to obtain the optimal variables corresponding to the optimal target. The results are verified in a further step, where a complete extrusion-die system is optimized. Significant improvements in the simulated product quality and the total calculation time saving have...