基于小生境遗传神经网络的材料力学性能预测

摘要

对工业材料测试,建立优化系统,关于建立材料力学性能与组成、工艺等相关的预测模型,可以减少试验次数、提高效率、实现工艺的优化.提出利用遗传算法优化的神经网络建立材料性能影响因子到力学性能的非线性映射.在遗传算法中采用基于淘汰相似结构机制的小生境技术,使预定距离之内仅存一个优良个体,维护了群体的多样性,从而提高全局搜索能力.以麦杆增强复合材料为例进行仿真研究,建立其力学性能预测的小生境遗传神经网络模型,利用模型优化注塑成型的工艺参数进行仿真.结果表明所建模型具有较好的学习和泛化能力,对于优化成型工艺参数具有可行性,在材料性能研究领域有着较好的应用和推广价值.%Predicting model which refers to mechanical properties with material composition and techniques can be founded to reduce test times, increase the efficiency and realize the optimization of the process. This paper proposes to apply the artificial neural network optimized by genetic algorithm to set up the nonlinear mapping from influence factors of material mechanical properties to mechanical properties. Niche technique based on crowding mechanism is used in genetic algorithm to promote global search capability. Within the pre-specified distance there will be only one highly fit individual. Not only the population has excellent diversity, but also individuals are dispersed throughout the whole constraint space. Taking the wheat straw-reinforced composite for instance, the prediction neural network based on niche genetic algorithm has been built. Besides, the model is used to optimize process parameters of injection molding and find the range of best parameters. The simulation result shows that the founded network model has preferable learning and generalization capabilities, which performs effectively in predicting mechanical properties.Therefore it is feasible to optimize process parameters and the technology is worthy to be applied and spread in the research of material performance.