Simultaneous tool wear condition and surface roughness prediction under limited samples

In computer numerical control (CNC) machining, tool wear degrades surface roughness and poor surface condition accelerates tool wear. While tool wear and surface roughness are inherently coupled, simultaneous prediction of both factors remains challenging due to reliance on multiple intrusive sensors and extensive data collection requirements. To address this issue, the simultaneous tool wear condition and surface roughness prediction (STWC-SRP) framework under limited samples is proposed. First, the multi-domain features are extracted and selected from the spindle current signals. Next, the framework employs the conditional diffusion-based data generation method to create synthetic samples that preserve the underlying relationships between tool wear states and features of spindle current signals. The synthetic sample selection and surface roughness labeling method identifies the high-fidelity synthetic samples and computes corresponding surface roughness values for the synthetic samples. Finally, leveraging the real and the selected synthetic samples, the multi-task model achieves the simultaneous tool wear and surface roughness prediction, with a mean squared error of 0.0179 for surface roughness prediction, and accurate tool wear classification with no misidentifications across tested scenarios. Experimental results demonstrate that the STWC-SRP framework correctly identifies all tool wear conditions and achieves a Mean Squared Error (MSE) of 0.0179 and a Mean Absolute Error (MAE) of 0.1003 for surface roughness prediction, outperforming conventional approaches and confirming its practical applicability in industrial scenarios under limited samples.