基于神经网络的炭/炭复合材料烧蚀性能预测

采用人工神经网络(ANN)对炭/炭复合材料烧蚀性能进行了预测。确定了炭/炭复合材料的密度、 石墨化程度和基体炭类型为其烧蚀性能的关键控制因素, 通过人工神经网络表征了炭/炭复合材料的密度、 石墨化程度与其烧蚀性能之间的关系。在大量实验基础上对神经网络结构与参数发生变化时的网络性能进行了评估。结果表明, 当网络训练集规模、 隐层节点数、 初始学习率与动量项等参数的取值分别为35、 7、 0.5和0.2时网络预测性能达到最佳状态, 在此基础之上对炭/炭复合材料的质量烧蚀率进行了预测与评价。实践证明, 采用人工神经网络对炭/炭复合材料的烧蚀性能进行预测时误差小于11％, 满足工程实践的精度要求。 

Prediction on the ablative performance of carbon/carbon compositesbased on artificial neutral network

The artificial neutral network (ANN) method is applied to the prediction on the ablative performance of carbon/carbon composites. The key control factors for the ablative performance, namely, the density, degree of graphitization and the matrix kind, were selected. Further, a relation between those factors and ablative performance was determined. Through large numbers of experimental data, the structure and the performance of ANN had been evaluated with the variation of training parameters. It can be achieved from the results that there exists an optimal predicting ratio when the training set scale, the hidden unit, initial learning rate and momentum coefficient are 35, 7, 0.5 and 0.2, respectively. Based on the ratio, prediction and evaluation on the mass ablative rate have been made for the ablative performance of carbon/carbon composites. With the application of ANN, the prediction error is within 11%, which can satisfy the precision requirements for practical engineering purposes.