基于数值模拟的定向连续铸造固液界面位置人工神经网络模型的研究

基于人工神经网络原理及数值模拟技术，对定向凝固连续铸造过程中控制参数的选取进行了研究。利用自行设计的上引式定向凝固连铸机，结合数值模拟，提取了引晶速率、熔体温度、结晶器温度、冷却水温度、冷却水流量、冷却距离等控制参数值及相应目标参数值的固液界面位置。通过归一化处理所得数据，采用BP算法训练网络，对定向凝固连铸控制参数与固液界面位置之间的映射关系进行了函数逼近，建立了固液界面位置神经网络模型，依据该模型，可定量预测定向凝固连铸过程的工艺状态，并可为将来的定向凝固连铸神经网络控制提供可行的控制模型。

Artificial Neural Network Model for S/L Interface Position in Directional Solidification Continuous Casting Process

Based on artificial neural network theory and numerical simulation technology, it has been stu died that the selection of control parameters in directional solidification continuous casting process (DSCC process). With combination of the experiments using our self designed upward DSCC device and numerical simulation, control parameters of this device such as the mold temperature, the bath temperature, the cooling water temperature, the pulling speed, the cooling distance and the cooling water flux etc as well as a target parameter of the position of S/L interface are picked up. A series of values of these parameters compose a batch of samples. We used these sample data appropriately treated to train a BP algorithm neural network, which could unlimitedly approach the reality of the mapping relation between those control parameters and the S/L interface position in DSCC process. So a neural network model for the S/L interface position is built. Depending on this model, we could quantitatively predict the process status of directional solidification continuous casting and provide an available control model for future neural control of the DSCC process.