朱欢欢,葛爱丽,迟玉伦,张梦梦,李厚佳.基于梯度提升决策树的硬车表面白层预测方法[J].表面技术,2023,52(2):328-342.
ZHU Huan-huan,GE Ai-li,CHI Yu-lun,ZHANG Meng-meng,LI Hou-jia.Surface White Layer Prediction Method of Hard Turning Based on Gradient Boosting Decision Tree[J].Surface Technology,2023,52(2):328-342 |
| 基于梯度提升决策树的硬车表面白层预测方法 |
| Surface White Layer Prediction Method of Hard Turning Based on Gradient Boosting Decision Tree |
| |
| DOI：10.16490/j.cnki.issn.1001-3660.2023.02.031 |
| 中文关键词: 白层 混淆矩阵 硬态切削 智能预测 梯度提升决策树 |
| 英文关键词:white layer confusion matrix hard cutting intelligent prediction gradient boosting decision tree |
| 基金项目: |
|
|
| Author | Institution |
| ZHU Huan-huan | Higher Vocational and Technical College, Shanghai University of Engineering Science, Shanghai 200437, China |
| GE Ai-li | University of Shanghai for Science and Technology, Shanghai 200093, China |
| CHI Yu-lun | University of Shanghai for Science and Technology, Shanghai 200093, China |
| ZHANG Meng-meng | Higher Vocational and Technical College, Shanghai University of Engineering Science, Shanghai 200437, China |
| LI Hou-jia | Higher Vocational and Technical College, Shanghai University of Engineering Science, Shanghai 200437, China |
|
| 摘要点击次数: |
| 全文下载次数: |
| 中文摘要: |
| 目的 实现硬态车削过程中每个产品零件白层现象的实时在线检测，提高产品生产加工效率和加工质量，提出一种基于梯度提升决策树的硬态车削加工工件表面白层预测方法。方法 首先，利用功率传感器、声发射传感器和振动传感器采集硬态车削过程中的动态切削信号数据，并对上述各种传感器信号数据进行特征提取;然后，结合特征重要性分析和梯度提升决策树建立硬态车削加工表面白层预测模型;最后，基于混淆矩阵提出一套评估梯度提升决策树模型预测性能的评价方法。结果 与功率、振动信号等特征相比，声发射信号特征能够进一步提升模型的白层预测性能。实验结果表明，该方法的预测准确率达到90%，F1为92%，Auc为89%，与SVM、XGBoost分类方法所得结果相比，该方法能更准确有效地实现硬态车削加工工件表面白层现象的在线预测。结论 该方法基于智能传感技术和梯度决策树模型对硬车过程中产生的白层现象进行了有效预测识别，对实现硬车过程白层现象的在线智能预测具有重要意义。 |
| 英文摘要: |
| In the process of hard turning, the workpiece surface is easy to cause white layer, which affects the surface quality of the workpiece. The current methods of detecting white layer not only affects the processing efficiency, but also fail meet the requirements on full detection of product parts. The work aims to propose a prediction method of white layer on workpiece surface in hard turning based on gradient lifting decision to realize the real-time on-line detection of white layer phenomenon for each product in hard turning process, and improve the production efficiency and processing quality of products. The method mainly includes signal data acquisition, feature extraction and analysis, prediction model construction and prediction result analysis. Firstly, the normal (without white layer) and abnormal (with white layer) signal data in the process of hard turning were collected by power sensor, acoustic emission sensor and vibration sensor. Secondly, the time-domain parameters and wavelet packet energy parameters were extracted as the characteristic parameters to identify the white layer, and the influence degree of the above characteristic parameters was sorted by the feature importance analysis method, then the principal feature quantity related to the white layer was extracted by the PCA method as the input sample in the gradient boosting decision tree model. Thirdly, the grid search method was used to optimize the model parameters, reducing the dependence of the model on the amount of data and improving the accuracy of the model in predicting the white layer. Finally, based on the confusion matrix, a set of evaluation methods was proposed to ensure the prediction performance of the model. In order to verify the effectiveness of the above method, a hard turning experiment was carried out on the actual bearing products in the factory. The machining processing signal data was collected such as acoustic emission signal, three-way vibration signal and power signal for on-line prediction and identification of hard turning process; then the microstructure test specimens were prepared for the machined parts to explore the relationship between the white layer and the above sensor signals. The experimental results showed that compared with the characteristics of power and vibration signal, the characteristics of acoustic emission signal were more sensitive to the white layer. The model with acoustic emission signal characteristics greatly improved the accuracy, precision, recall, F1 value and Auc value. At the same time, compared with other model algorithms (SVM algorithm and XGBoost algorithm), the prediction accuracy based on the gradient boosting decision tree model can reach 90%, F1 score can reach 92% and Auc value can reach 89%, which can better reflect the white layer phenomenon. Based on various sensor technologies, combined with gradient boosting decision tree, this method can effectively predict and identify the white layer phenomenon in hard turning. This can not only realize the on-line intelligent prediction of the white layer phenomenon on the workpiece surface in hard turning, but also is of great significance to ensure the processing quality and improve the processing efficiency of products. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
渝公网安备 50010702501715号