Hybrid Machine learning and temporal-spatial fusion decision for real-time monitoring of drilling stage in ultrafast laser drilling

Excessive ablation and low stability significantly limit the wider application of ultrafast laser drilling. To address this challenge, a real-time monitoring and control system for the drilling process was previously developed, effectively enhancing processing stability and preventing excessive workpiece damage. Given the challenges in accurately modeling the relationship between monitoring signals and drilling stages through physical models, machine learning methods have gained popularity. However, using machine learning models directly for process control can lead to substantial decision deviations due to inherent binary decision-making limitations. In this study, a framework for real-time monitoring and decision-making of drilling stages in ultrafast laser drilling is proposed. Unlike conventional methods that rely solely on machine learning for decision-making, we introduce a novel approach, called temporal-spatial fusion decision (TSFD), that integrates the evolution characteristics of the drilling process with sequential three-way decision (S3WD) concepts. By applying the TSFD method to reevaluate machine learning-based decisions, the accuracy in identifying breakthrough times is significantly improved. Experimental results show that TSFD reduces mean decision deviation from 68.06 to 11.01 compared to direct machine learning-based decisions, demonstrating the enhanced decision reliability and stability. The framework and TSFD method proposed in this study enable real-time monitoring and decision-making during ultrafast laser drilling with minimal datasets and higher reliability, offering a solution for stable, high-quality laser material processing.