基于神经网络遗传算法的GFRP拉挤双目标优化

Bi-objective optimization for GFRP pultrusion based on neural network and genetic algorithm

为研究玻璃钢(GFRP)拉挤工艺参数对复合材料性能的影响,优化最佳拉挤工艺参数,建立了拉挤工艺过程数学模型,结合基于有限元/有限差分的间接解耦法进行求解,模拟得到了拉挤过程中GFRP内部的非稳态温度场和固化度变化情况.分别采用布拉格光栅光纤温度传感器和索氏萃取法检测拉挤GFRP内部的温度与固化度,实测温度和固化度均与模拟温度和固化度吻合,验证了数值模拟程序的正确性.以数值模拟结果为样本,建立反向传播神经网络,得到拉挤工艺参数(固化温度、拉挤速度)与GFRP固化度之间的非线性相关关系,再结合遗传算法解决拉挤过程中固化炉温度和拉挤速度双目标优化问题.优化得到的拉挤工艺参数可在保证复合材料固化度达标的情况下,提高拉挤速度降低固化炉温度,优化效果显著.神经网络遗传算法优化方法能有效解决此类具有复杂非线性关系的多目标优化问题.

The relationship between technological parameters of glass fiber reinforced polymer（GFRP） pultrusion and products＇ properties was studied to find out the optimal technological parameters.A mathematic model for GFRP pultrusion was established and solved by the combination of finite element,finite difference and indirect decoupling method.The temperature and degree of cure of GFRP during pultrusion were predicted through simulation.And the temperature test with fiber Bragg grating sensor and the degree of cure test with Sorbitic extraction were completed to verify the simulation.The test results agree well with the simulated results.According to the simulated results,the back propagation neural network was trained to form the relationship between the technological parameters（die temperature,pull speed） and the degree of cure of GFRP.The genetic algorithm associated with the neural network was used to deal with the bi-objective optimization for GFRP pultrusion.The optimized parameters have lower die temperature and higher pull speed while guaranteeing the degree of cure of pultruded products.The optimization method based on neural network and genetic algorithm is effective for solving the multi-objective optimization problems with complex relation.