Milling surface roughness monitoring using real-time tool wear data
Runqiong Wang,Qinghua Song,Yezhen Peng,Zhanqiang Liu,Haifeng Ma,Zhaojun Liu,Xun Xu
DOI: https://doi.org/10.1016/j.ijmecsci.2024.109821
IF: 7.3
2024-11-24
International Journal of Mechanical Sciences
Abstract:The random disturbances in feature distribution, resulting from the weak rigidity and difficult-to-machine nature of workpieces, significantly impede the intelligent monitoring of aerospace thin-walled parts machining. To address this issue, a surface roughness monitoring model integrating tool state information is proposed. Unlike existing data-driven approaches that typically focus on model-centric strategies, the proposed method constructs a feature performance-guided distributed recognition model using edge-cloud collaboration. It integrates real-time tool condition data into surface quality monitoring through quantitative evaluation of features and collaborative configuration of edge-cloud systems. Firstly, a performance-guided filtering method is proposed to filter redundant cutting data across multiple dimensions, thus avoiding the influence of dynamic disturbances on the distribution of input feature vectors. Secondly, the extracted global-local features are adaptively fused by a feature-weighted stacking function that automatically assigns weight parameters to the features according to their performance. Subsequently, an observation model for detecting tool condition is developed using a parallel heterogeneous deep network and convolutional block attention module, which integrates real-time tool wear information into surface quality recognition through a lightweight system model. Based on this, a distributed architecture that includes service, observation, and system models is constructed by collaborative training and deploying recognition models across edge and cloud systems. A model agile configuration strategy is introduced for varying working conditions. Experimental validation reveals that the feature performance-guided distributed architecture identifies both the tool wear state and the surface roughness of the workpiece in the machining of thin-walled parts. The identification accuracy cumulatively improved by 25.2% after removing the disturbance effects. Compared with the existing state-of-the-art surface roughness monitoring methods, the proposed method has higher prediction accuracy and stronger generalization ability. It lays a theoretical foundation for the surface quality monitoring of aeronautical components.
engineering, mechanical,mechanics