融合工件几何特征的变工况切削力预测方法

Cutting force prediction under the variable machining condition incorporating workpiece geometric features

在机械加工中,加工工件的几何特征变化会导致切削力统计特征发生变化,使得传统数据驱动的切削力预测模型精度变低,同时不同的加工工况使得采集的切削力建模数据存在明显的数据分布差异,导致切削力预测模型泛化能力出现显著退化。针对上述问题,提出了一种融合工件几何特征的变工况切削力预测方法。首先,进行数据预处理,包括工件几何特征、工况信息编码处理,对切削力信号去除趋势项,对切削力统计特征剔除异常值;其次,考虑工件几何特征及工况变化导致的数据分布差异,将数据集分为源域数据和目标域数据,并将源域数据和目标域数据按照规则划分为训练集和测试集,基于迁移学习构建融合工件几何特征的变工况切削力预测模型;最后,从不同数据量、单一加工几何特征、变工况、不同算法等方面进行了实验验证。实验结果表明,相比传统数据驱动切削力预测模型,该方法更适用于工况以及工件几何特征变化情况下的切削力预测,同时在数据样本较少的情况下保持较高的预测精度,泛化性能更强,具备更好的实用性。

In a machining process,changes of workpiece geometric features can lead to variation in the statistical characteristics of cutting forces,causing significant degradation in the accuracy ability of traditional data-driven cutting force prediction models.and different processing conditions make the collected cutting force modeling data have obvious data distribution differences,causing significant degradation in the generalization ability of traditional data-driven cutting force prediction models.To address these problems,this paper proposes a cutting force prediction method incorporating workpiece geometric features under variable machining conditions.First,data preprocessing is carried out,including the workpiece geometric features and working condition information coding processing,cutting force signal removing trend items,and cutting force statistical features removing outliers.By considering the workpiece geometric features and working condition changes caused by the data distribution differences,the data set is divided into source domain data and target domain data;then the source domain data and the target domain data are divided into training sets and test sets according to the rules,and a variable cutting force prediction model incorporating geometric features of the workpiece is constructed based on transfer learning.Finally experimental verification is carried out from different data quantities,single processing geometry characteristics,variable working conditions,and different algorithms.Experimental results show that the method is more suitable for predicting cutting forces under changing working conditions and workpiece geometrical characteristics than the traditional data driven cutting force prediction model,while maintaining a higher prediction accuracy with fewer data samples and a better generalization performance,so that it can provide a better practicality.