基于条件变分自编码器的熔铸炸药成型缺陷快速模拟和预测

Fast Simulation and Prediction of Molding Defects in Melt-cast Explosives Based on Conditional Variational Autoencoder

为了实现凝固缺陷的快速模拟和预测,提出了一种基于条件变分自编码器(CVAE)的熔铸炸药成型缺陷预测模型;以注液温度、冒口预热温度等工艺参数为条件,通过条件变分自编码器建立工艺参数与熔铸炸药缺陷的条件概率模型;采用多层神经网络和变分推断方法结合进行模型训练,实现了RHT和DNP基熔铸炸药凝固成型缺陷预测。结果表明,成功构建了熔铸炸药凝固过程数值模拟的条件概率分布,实现了基于仿真数据的RHT和DNP基熔铸炸药凝固缺陷预测;与有限元直接数值计算结果比较,CVAE算法计算缺陷位置的准确率可达到99%,计算时间小于2 s;CVAE在熔铸炸药缺陷概率分布建模上具有性能高、泛化性强的特点,能有效实现熔铸炸药成型缺陷的智能预测。

In order to realize the fast simulation and prediction of solidification defects,a molding defect prediction model of melt-cast explosives was proposed based on conditional variational autoencoder(CVAE).Taking the process parameters such as injection temperature,riser preheating temperature and so on as conditions,the conditional probability model of the relationship between the melt-cast explosives defects and the process parameters was established by the CVAE.For RHT and DNP-based melt-cast explosives,the prediction of melt-casting defects was implemented by training the models of the multilayer neural network combined with the variational inference method.The results show that,compared with the results of the direct numerical calculations of finite elements,the prediction accuracy of the CVAE algorithm in calculating the defect location reaches 99%,and the computation time is less than 2 s.The CVAE has an excellent performance in the modeling of the probability distribution of defects in melt-cast explosives with a strong generalization,and the trained models can be used to realize intelligent prediction of molding defects of melt-cast explosives.