"Machine Learning and Design of Experiments for Optimizing Laser-Engraved Micro Fresnel Lens Mould"

Abstract This research examines the application of Laser Engraving to produce micro Fresnel Lenses on aluminum plates, a novel application of this non-conventional machining method. The research explores the effects of the scan speed, laser power with number of cycles on the roundness deviation using a L9 orthogonal array. Multiple analytical methods, including the Taguchi method, Random Forest Algorithm with sensitivity analysis, are employed to optimize process and predict the outcomes. In this study, a thorough analysis of the fabrication of a micro Fresnel lens on Aluminum plate (10 mm × 10 mm × 2 mm) using fiber laser of wavelength 1064 nm is presented. The study finds that laser power has most significant effect on the roundness deviation, followed by the number of the cycles and scan speed. Scan Speed ranges from 500 to 700 mm s −1 , the Power ranges from 25 to 35 Watts, and the Number of Cycles ranges from 100 to 200. Optimal conditions are identified as 700 mm/s scan speed, 25 W power, and 100 cycles. Microscopic analysis confirms roundness deviation under these conditions. Comparisons between analytical approaches and experimental results reveal that both the Taguchi method and Random Forest Algorithm align closely with experimental outcomes, with the Random Forest Algorithm showing slightly higher accuracy (6.18 percentage points closer to experimental results). This research addresses a gap in comparative studies evaluating traditional statistical methods against modern machine learning algorithms for process optimization in laser machining. It combines knowledge from optics, materials science, and laser machining, utilizing advanced methods and technologies that have only recently become accessible. The findings provide valuable insights for future applications of micro Fresnel lenses on aluminum plates and contribute to the understanding of laser engraving processes for precision optical components. Between the Random Forest Algorithm and the Taguchi method, Random Forest Algorithm fits more closely to the experimental result. Random Forest Algorithm prediction is closer to experimental result by about 6.18 percentage points compared to the Taguchi method prediction.