基于遗传算法优化神经网络的镁合金化学机械抛光材料去除模型

Simulation of Material Removal in Magnesium Alloy Chemical Mechanical Polishing by Neural Network and Genetic Algorithm

镁合金化学机械抛光（CMP）的材料去除与其工艺参数具有高度非线性的特点，难以采用精确的数学模型来描述。以遗传算法（GA）优化神经网络（NN）建模为基础，利用正交试验设计获取镁合金CMP材料去除样本数据和测试数据，建立镁合金CMP材料去除模型。该模型以抛光压力、抛光盘转速、抛光液流量和抛光时间为输入参数，以材料去除速率为输出目标。结果表明：GA-NN协同模型能够构建镁合金CMP工艺参数与材料去除速率的基本关系；其拟合度波动范围为93．22％-97．97％，大大高于NN模型的拟合度波动范围71．56％-93．56％，因而具有更优的预测能力，基本满足工程实际的需求。

The relationships among the process parameters and material removal rate in magnesium alloy chemical me- chanical polishing （CMP） process are nonlinear, which leads to the difficulty to establish a comprehensive and accurate model to predict the material removal rate. A model for material removal rate in magnesium alloy CMP process based on neural network （NN） optimized by genetic algorithm （GA） was proposed, in which the typical data were obtained from the orthogonal experiments of magnesium alloy CMP.The model was applied to investigate the influences of input parameters involving the polishing pressure, the polishing head speed, the slurry flowing rate and polishing time on the material remov- al rate.The performance of the model was verified by simulation and the experimental data. It is shown that the data of GA-NN forecast are in good agreement with the experimental data, and the fitting degree of GA-NN model is in the range of 93.22% - 97.97%, which shows a much better stable ability than that of NN model in the range of 71.56% - 93.56%.The results show that the GA-NN model can be used to predict the material removal rate in magnesium alloy CMP process, it can meet with the industrial requirements.