铝合金铸造凝固过程有限元模拟的神经网络优化  

Finite Element Modeling Optimization for Aluminum Alloy Casting Solidification Process Based on Neural Network

作　　者：罗蓬[1] 胡侨丹 夏巨谌[1] 胡国安[1] 杨屹[2] 

机构地区：[1]华中科技大学塑性成形模拟及模具技术国家重点实验室 [2]四川大学制造科学与工程学院,四川成都610065

出　　处：《铸造》2005年第1期73-76,共4页Foundry

基　　金：华中科技大学塑性成形模拟及模具技术国家重点实验室开放基金资助(04-3);国家留学基金资助(21852035)。

摘　　要：应用神经网络优化了铝合金铸造凝固有限元模拟(FEM)过程,使得CPU时间最小化。在优化中,通过粗化FEM网格减少CPU时间。模拟精度降低的代价通过神经网络拟合机制来补偿。拟合机制由自适应学习率 动量因子的误差反向传播改进算法来实现。FEM模拟的控制方程是基于焓变的Fourier导热方程。模拟结果为开模循环周期20s的温度场,几何中心与浇道等铸件特定部位的温度 时间曲线。本工作证实了对凝固FEM技术实施神经网络优化的可行性。This research applied neural network optimal scheming to finite element modeling (FEM) of casting solidification for the purpose of minimization of CPU time. In the optimization, an adaptive learning rate and momentum-based error back-propagation algorithm was applied to develop an idea of neural network fitting by which precision losing of FEM based on coarse mesh was compensated, the FEM was theoretically governed by the enthalpy-based Fourier heat conduction equation, and the results were provided in the form of temperature distribution at the time of die-opening cycle of 20 s, and temperature versus solidifying time relations at gating system and geometric center. The study demonstrates the feasibility of neural network optimization to FEM of casting solidification.

关 键 词：铝合金 铸造 有限元模拟 神经网络优化 

分 类 号：TG27[金属学及工艺—铸造]