|
| 基于机器学习的热塑性树脂/金属异质结构界面键合特征预测模型研究 |
| Research on Machine Learning Based Prediction Model for Interface Bonding Characteristics of Thermoplastic/Metal Heterostructure |
| Received:May 24, 2024 |
| DOI：10.3969/j.issn.1674-6457.2025.02.024 |
| 中文关键词: 热塑性树脂 金属 异质结构 机器学习 计算材料学 键长 |
| 英文关键词: thermoplastic metal heterostructure machine learning computational material science bond distance |
| 基金项目:山东省重大科技创新工程计划(2021ZDPT02);国家重点研发计划(2022YFB4300102，2019YFE0110700) |
| Author Name | Affiliation | | WANG Lingjun | National Innovation Center of High Speed Train, Shandong Qingdao 266111, China | | SONG Kunlin | National Innovation Center of High Speed Train, Shandong Qingdao 266111, China | | ZHANG Lijiao | CRRC Industrial Institute Co., Ltd., Beijing 100070, China | | ZHU Hongbin | CRRC Industrial Institute Co., Ltd., Beijing 100070, China | | LI Yan | School of Mechanical & Automotive Engineering, South China University of Technology,Guangzhou 510641, China | | WANG Zhenmin | School of Mechanical & Automotive Engineering, South China University of Technology,Guangzhou 510641, China |
|
| Hits: 20 |
| Download times: 2 |
| 中文摘要: |
| 目的 探索一种能精确预测热塑性树脂/金属异质界面键合典型特征的方法，采用计算材料学与机器学习结合的方式构建预测模型。方法 提出了基于机器学习与计算材料学的热塑性树脂/金属异质结构界面键合特征预测新策略。基于热塑性树脂/金属界面焊接冶金特征，建立了界面反应精细原子模型，通过分子动力学模拟构建了异质界面键合典型特征数据集。根据键合原子间物理、化学及结构特征选取输入特征向量，随后分别通过随机森林回归模型、梯度提升回归树模型、LASSO回归模型以及BP神经网络模型对该数据集进行训练，并综合评估预测性能，选出预测结果最优算法模型。最后通过基于皮尔逊相关系数和主成分分析的特征工程，降低特征向量维度，进而有效降低模型计算复杂度。结果 建立了热塑性树脂/金属异质结构界面键长预测模型，其中随机森林回归模型、梯度提升回归树模型、LASSO回归模型以及BP神经网络模型的均方根误差RMSE分别为0.046 6、0.044 6、0.099 9、0.062 7。特征工程使特征向量从20维降到5维，降维后的新特征向量在梯度提升回归树模型中预测的RMSE为0.038 6。结论 梯度提升回归树模型对热塑性树脂/金属异质结构界面键合特征的预测效果最好，特征工程能有效降低模型复杂度，并提升预测性能。 |
| 英文摘要: |
| The work aims to explore a method for accurately predicting the typical characteristics of thermoplastic resin/metal heterojunction bonding, and construct a prediction model by combining computational materials science and machine learning. In this article, a new strategy for predicting the interfacial bonding characteristics of thermoplastic/metal heterostructures based on machine learning and computational materials science was proposed. Based on the metallurgical characteristics of thermoplastic/metal welding interface, a fine atom model for interface reactions was established. Then a dataset of typical characteristics of heterogeneous interface bonding was constructed through molecular dynamics simulation. Input feature vectors were selected based on the physical, chemical, and structural characteristics of bonding atoms. On this basis, the random forest regression model, gradient boosting regression tree model, LASSO regression model, and BP neural network model were constructed to train the dataset. Then the predictive performance was evaluated comprehensively and the optimal algorithm model was selected for the prediction results. Through the feature engineering based on Pearson correlation coefficient and principal component analysis, the dimensionality of feature vectors was reduced, which effectively reduced the computational complexity of the model. Finally, a prediction model for the interfacial bond distance of thermoplastic/metal heterostructures was established. The root mean square errors (RMSE) of the random forest regression model, gradient boosting regression tree model, LASSO regression model, and BP neural network model were 0.046 6, 0.044 6, 0.099 9, and 0.062 7, respectively. Feature engineering reduced the feature vector from 20 dimensions to 5 dimensions, and the RMSE of the new feature vector predicted by the gradient boosting regression tree model after dimensionality reduction was 0.038 6. In conclusion, the gradient boosting regression tree model has the best prediction effect on the typical interface bonding characteristics of thermoplastic/metal heterostructures. Feature engineering can effectively reduce model complexity and improve prediction performance. |
|
View Full Text
View/Add Comment Download reader |
| Close |
|
|
|
渝公网安备 50010702501719号