Optimization of low-power femtosecond laser trepan drilling by machine learning and a high-throughput multi-objective genetic algorithm

The geometric quality (taper), recast layer, and processing efficiency of micro-holes are important issues in femtosecond laser trepan drilling (LTD). Although one-step drilling based on low-power femtosecond LTD may be an ideal drilling method, its disadvantages such as processing time and taper quality still need to be improved. To address these issues, machine learning was successfully applied to establish an accurate predictive model for the femtosecond LTD process. Based on the machine learning model, the femtosecond LTD results for a given parameter set were quickly and accurately obtained, avoiding a large number of complex experiments and characterization requirements. Subsequently, through the combination of our established optimal machine learning predictive model and a high-throughput genetic algorithm, optimized solutions were quickly and successfully designed for a wide range of process spaces. Finally, the reliability of the optimized process was verified by experiments. The combination of machine learning and a high-throughput optimization algorithm provided an efficient and low-cost solution for the optimization of complex laser processing technology.

Automated summary

Machine learning was successfully applied to establish an accurate predictive model for femtosecond laser trepan drilling process, which provided quick and accurate results, improved efficiency, and avoided complex experiments. By combining machine learning with a high-throughput optimization algorithm, optimized solutions were designed for a wide range of process spaces, and the reliability of the optimized process was verified through experiments.