基于神经网络遗传算法的CFRP拉挤工艺优化

Optimization for CFRP pultrusion process based on genetic algorithm-neural network

根据固化动力学和传热学理论,建立了碳纤维增强聚合物基复合材料(CFRP,Carbon Fiber Reinforced Polymer)拉挤成型非稳态温度场与固化动力学数学模型.采用有限元与有限差分相结合的方法,结合ANSYS求解耦合场的间接耦合法,对CFRP拉挤过程非稳态温度场和固化度进行数值模拟.使用特殊设计制作的铝毛细管封装的布拉格光栅光纤(FBG,Fiber Bragg Gratings)传感器,对温度场进行实时检测;并采用索氏萃取实验测定CFRP制品固化度.模拟与实验结果基本吻合.以数值模拟结果为样本建立反向传播神经网络,训练得到固化炉温度与CFRP固化度之间的非线性相关关系.采用神经网络与遗传算法相结合的方法,优化得出拉挤固化炉三段最佳温度值,结果表明神经网络结合遗传算法优化拉挤工艺参数快捷有效.

According to the curing kinetics and heat transfer theory, the models of unsteady temperature field and resin curing for the pultrusion process of carbon fiber reinforced polymer （CFRP） were established. The finite element/finite different method associated with the indirect decoupling method based on ANSYS to simulate the temperature and degree of cure for CFRP pultrusion. The fiber Bragg gratings （FBG） sensor, encapsulated in an aluminum capillary, was utilized to real time monitor the temperature during puhrusion; and the final degree of cure was measured by Sorbitic extraction. The results show that the numerical model is reliable and correct. With the simulative results, the neural network was trained and predicted the degree of cure effectively. On the basis of the trained network, the genetic algorithm was used to optimize the temperature of die. It shows that the optimization method is effective and convenient.