Multi-scale quality monitoring for laser additive remanufacturing: Example with surface continuity and hardness monitoring in ceramic reinforced metal matrix composite

• An innovative multi-scale quality monitoring method with a single sensor for comprehensive quality description of LARM-prepared PRMMCs. • A CRM-F model with lightweight single-path shared backbone and multi-task branch is proposed to predict comprehensive quality in LARM. • The mesoscopic surface continuity and macroscopic hardness is first predicted simultaneously for MMCs fabricated by LARM. Ceramic particle reinforced metal matrix composites (PRMMCs) fabricated by laser additive remanufacturing (LARM) provides a new route for the in-situ restore and surface enhancement of high value-added parts. However, the absence of an effective quality description method for LARMed PRMMCs hindered their application due to the complex multi-scale defects. This work presents promising results aim to develop an innovative multi-scale quality monitoring method for comprehensive quality description of LARM-prepared PRMMCs utilizing a single sensor. An innovative classification-regression multi-task framework with fully depth-separable convolutional backbone (CRM-F model) was introduced to simultaneously predict mesoscopic surface continuity and macroscopic hardness. The results show that the prediction results of the CRM-F model exceed 95 % accuracy on the surface continuity branch, and the prediction errors on the hardness regression prediction branch are stabilized within 5 %. Meanwhile, the CRM-F model with concise single-path architecture demonstrates excellent inference efficiency, and its inference time of 8.10 ms/frame provides an irreplaceable advantage for real-time monitoring on micro-industrial computers with Inter-CPUs. A prototype human–machine interfacing system developed on the basis of the CRM-F model lays a solid foundation for subsequent closed-loop control and digital twins in LARMed PRMMCs.