直写成形工艺制备的功能梯度材料零件时变挤出系统建模OA
Modelling of Time-varying Extrusion Systems for Fabrication of FGMs Parts by Direct Ink Writing Processes
高精度的计算流体力学表征模型会带来极高的时间成本,这给具有高频次复杂梯度变化的功能梯度材料零件的表征带来挑战.建立了以贝叶斯正则化神经网络为预测模型的时变挤出系统,首先通过高精度的计算流体动力学仿真模型获取数据集并用于训练神经网络模型,将材料目标比例、料腔中初始比例、双进料口流量总和以及适配的螺杆转速作为输入参数,标记交付延迟时间以及过渡延迟时间作为输出参数,再将训练后贝叶斯正则化神经网络融合经典控制理论对系统描述的方法构建完整的时变挤出系统.最后通过打印功能梯度材料样件验证了所构建的计算流体动力学仿真模型以及时变挤出系统的准确性与适用性.
High-precision CFD models are time-consuming,creating challenges for the frequent gradi-ent variations in FGMs part printing.Therefore,a time-varying extrusion system was established using a Bayesian regularization neural network as the prediction model.High-precision CFD simulation data sets were first obtained to train the neural network model,with input parameters including the target materials ratio,initial ratio in the chamber,total flow rate of the dual feed rate,and the adapted screw speed.The output parameters were labeled as delivery delay time and transition delay time.Then,the trained Bayes-ian regularized neural network was merged with the classical control theory approach to system description to construct the complete time-varying extrusion systems.
王世杰;段国林
河北工业大学机械工程学院,天津,300401河北工业大学机械工程学院,天津,300401
通用工业技术
功能梯度材料零件直写成形工艺计算流体动力学贝叶斯正则化神经网络时变挤出系统
functionally graded materials(FGMs)partdirect ink writing processcomputational fluid dynamics(CFD)Bayesian regularized neural networktime-varying extrusion system
《中国机械工程》 2026 (2)
466-475,10
中央引导地方科技发展资金(216Z804G)
评论