谢超林,唐昌平,刘筱,等.基于机器学习的AZ31镁合金轧制板材腐蚀电位预测研究[J].精密成形工程,2024,16(11):75-81.
XIE Chaolin,TANG Changping,LIU Xiao,et al.Prediction of Corrosion Potential of Rolled AZ31 Magnesium Alloy Sheet Based on Machine Learning[J].Journal of Netshape Forming Engineering,2024,16(11):75-81. |
| 基于机器学习的AZ31镁合金轧制板材腐蚀电位预测研究 |
| Prediction of Corrosion Potential of Rolled AZ31 Magnesium Alloy Sheet Based on Machine Learning |
| 投稿时间：2024-08-15 |
| DOI：10.3969/j.issn.1674-6457.2024.11.009 |
| 中文关键词: AZ31镁合金 轧制工艺 机器学习 腐蚀电位 Pearson相关系数 |
| 英文关键词: AZ31 magnesium alloy rolling process machine learning corrosion potential Pearson correlation coefficient |
| 基金项目:国家自然科学基金(52471132，52475356，52475345，U20A20275);重庆市自然科学基金(CSTB2023NSCQ-MSX0886);福建省科技计划杰出青年基金(2024J010031);青海大学盐湖化工大型系列研究设施开放研究项目(2023-DXSSKF-04) |
|
| 摘要点击次数: 143 |
| 全文下载次数: 14 |
| 中文摘要: |
| 目的 为了实现对AZ31镁合金轧制板材耐腐蚀性能的调控。方法 以AZ31镁合金轧制板材的下压量、轧制温度和应变速率3个轧制工艺参数作为输入变量，腐蚀电位作为输出变量，采用支持向量机(SVM)、随机森林(RF)、K近邻(KNN)和反向传播神经网络(BP)4种机器学习模型对AZ31镁合金轧制板材的腐蚀性能进行预测。结果 计算得出，支持向量机(SVM)、随机森林(RF)、K近邻(KNN)和反向传播神经网络(BP)4种机器学习模型的平均绝对误差(MAE)分别为0.013 65、0.012 59、0.010 72和0.015 38;均方误差(MSE)分别为0.000 247、0.000 182、0.000 169和0.000 354;决定系数(R2)分别为0.61、0.71、0.74和0.44;下压量、轧制温度和应变速率与腐蚀电位的Pearson相关系数分别为0.755、0.262和0.015。结论 对比上述4种机器学习模型，K近邻(KNN)模型对AZ31镁合金轧制板材腐蚀电位的预测效果更好;由Pearson相关系数热力图可知，下压量与腐蚀电位呈正相关，这是由于随下压量的增加，晶粒逐渐细化，镁合金的耐腐蚀性能随之增加;轧制温度与腐蚀电位呈正相关，归因于随轧制温度的增加，孪晶数量逐渐减少，孪晶与镁基体形成的微电偶效应减弱，提高了腐蚀性能。 |
| 英文摘要: |
| The work aims to adjust the corrosion resistance of rolled AZ31 magnesium alloy sheet. Rolling reduction, rolling temperature and average strain rate were taken as input variables, and corrosion potential was taken as output variables in the rolled AZ31 magnesium alloy. Support vector machine (SVM), random forest (RF), k-nearest neighbor (KNN) and backpropagation neural network (BP) were used to predict the corrosion resistance of the rolled AZ31 magnesium alloy sheet. The mean absolute error (MAE) of support vector machine (SVM), random forest (RF), k-nearest neighbor (KNN) and backpropagation neural network (BP) were 0.013 65, 0.012 59, 0.010 72 and 0.015 38, respectively. The mean square errors (MSE) of SVM, RF, KNN and BP were 0.000 247, 0.000 182, 0.000 169 and 0.000 354, respectively. The determination coefficient (R2) of SVM, RF, KNN and BP were 0.61, 0.71, 0.74 and 0.44, respectively. The Pearson correlation coefficient of rolling reduction, rolling temperature and average strain rate with corrosion potential were 0.755, 0.262 and 0.015, respectively. Through the comparison of the above four machine learning models, KNN model exhibits the best performance in predicting the corrosion potential of the rolled AZ31 magnesium alloy sheet. According to the thermal map of Pearson correlation coefficient, rolling reduction and corrosion potential are in positive correlation, resulting from that with the rolling reduction increasing, the grain is refined, followed by the enhancement of the corrosion resistance of the magnesium alloy. Rolling temperature and corrosion potential are in positive correlation, attributing to that with the rolling temperature increasing, the number of twins decreases gradually, followed by weakening electric couple effect between the twins and the magnesium matrix accelerates the corrosion, resulting in improvement of the corrosion resistance of magnesium alloy. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
渝公网安备 50010702501719号