基于神经网络的双金属复合弯管铸造过程的数值仿真

Numerical Simulation for Casting Process of Double-metal Composite Bend Pipe Based on Artificial Neural Network

给出了双金属复合管的铸造工艺,研究了人工神经网络技术在铸造数值仿真优化中的应用,人工神经网络采用基于自适应学习率-动量项的误差反向传播梯度下降算法。用热电偶对双金属复合管铸造温度场进行了实测,并以温度场实测数据为样本,仿真了双金属复合管充型凝固过程的温度分布。通过实测数据与仿真数据的比较,神经网络优化处理后仿真的最大相对误差为2.1%。铸造过程的仿真为双金属复合管的设计和工艺制订提供了理论依据。

A casting technology of double-metal composite bend pipe is presented, optimization of numerical simulation in casting process is investigated based on artificial neural network, a gradient-descendent algorithm of error back-propagation with adaptive learning rate and momentum is applied. The solidifying temperature of double-metal composite bend pipe is tested by thermo-couples, the data of specimens obtained by the testing results of temperature field is trained, the temperature distribution during filling and solidification is simulated; By contrasting the data of simulation with those of testing, the maximum relative error of simulation is 2.1%, and the theory basis is presented for updating design and technology.