Artificial Intelligence-Driven Monitoring of Surface Polishing by Laser Remelting: Process Stability and Regime Classification
Abstract
Laser remelting (LRM) is a novel advanced multifunctional surface engineering technology with rapidity, versatility, accuracy and precision traits. LRM uses a high-powered laser to melt a superficial thin layer of material which redistributes through thermophysical phenomena within the molten pool for later resolidification. Surface polishing (SP-LRM) uses constant laser parameters to reallocate and redistribute molten micro-peak material into micro-valleys to create a flat surface. Surface structuring LRM (SS-LRM) uses variation in laser parameters affecting the process’ flow dynamics allowing for material redistribution control to create functional micro-structures. Being a novel and micro-process, there are numerous inputs that affect the process conditions of LRM that are not fully investigated resulting in optimization of LRM remains a heuristic trial-and-error process. Artificial intelligence (AI) is used for online monitoring and analysis of the process as a preliminary step toward AI-driven self optimizing LRM. The LRM process consists of two extreme regimes, shallow-LRM (S-LRM) and deep-LRM (D-LRM), and a novel transitional regime intermediate-LRM (I-LRM) that differ in process conditions and formation of the resultant surface. Statistical digital twins were developed for phenomenological regime understanding for effective AI usage. K-means clustering was deployed for near infrared LRM thermography stability classification. LRM processes with one optimal cluster based on Silhouette scoring was thermographically stable and processes with > 1 optimal clusters were unstable. Furthermore, Jaccard index is an image statistic that best represents k-means clustering criterion. Thermographic images where Jaccard index was < 0.8 were correlated to surface formation instabilities whereas images with high Jaccard index (> 0.8) correlate to surface formation stability. These results prove a strong correlation between thermography and surface topology exists and AI can be incorporated into LRM for online monitoring, regime recognition and stability formation self-optimization.