基于特征迁移的面铣刀磨损监测方法

实时监测刀具磨损状态对保证工件加工质量和确定合理换刀时间至关重要. 数据驱动的多源信号融合预测是解决刀具磨损预测难题的可行方案. 本文中通过时域和频域分析提取了多维信号特征，并结合机器视觉方法处理刀具磨损图像获得的磨损特征，针对涂层面铣刀建立了随机森林磨损预测模型. 对于同类型的刀具和工件材料，使用特征迁移方法解决多工况场景下新刀样本不足问题. 试验结果表明，基于迁移特征建立的磨损预测模型对目标刀具的磨损量预测效果较迁移前显著提升，准确性评价指标R2决定系数从0.37提升到0.96. 基于特征迁移的磨损预测模型为数据驱动模型在刀具磨损预测和实时监测领域的应用提供参考依据.

Real-time tool wear monitoring is critical for ensuring the quality of machined workpieces and determining appropriate tool change intervals. Data-driven multi-source signal fusion prediction represents a feasible solution to the challenging problem of tool wear prediction. In this study, multi-dimensional signal features are extracted through time-domain and frequency-domain analyses, and wear features obtained from processing tool wear images via machine vision methods are incorporated. A random forest-based wear prediction model is developed for coated face milling cutters. For tools and workpiece materials of the same type, a feature transfer method is employed to address the issue of insufficient new tool samples under multi-working-condition scenarios. Experimental results demonstrate that the wear prediction model established based on transfer features achieves significantly improved wear amount prediction performance for the target tool compared to the pre-transfer model. The accuracy evaluation metric, the coefficient of determination R², is increased from 0.37 to 0.96. The feature transfer-based wear prediction model provides a valuable reference for the application of data-driven models in the fields of tool wear prediction and real-time monitoring.

搜集汇总

背景与挑战

背景概述

该数据集研究基于特征迁移的面铣刀磨损监测方法，通过时域和频域分析提取多维信号特征，结合机器视觉方法建立随机森林磨损预测模型。试验结果表明，特征迁移方法显著提升了磨损量预测的准确性，R2决定系数从0.37提升到0.96。