Machine learning approach for laser-drilled Ni-alloy plates fabrication technology
摘要
The complex relationship between laser drilling parameters and hole properties complicates parameter selection and requires advanced optimisation methods. In this study, a fully connected artificial neural network (ANN) with three hidden layers was used to model the relationships between laser drilling parameters and the properties of holes produced in a 0.3 mm thick Haynes 242 alloy plate. The laser parameters varied in the following ranges: power (P) from 40 to 100%, frequency (ν) from 4 to 50 kHz, pulse duration (ti) from 16 to 350 ns, and laser delay time (tΣ) from 2 to 20 ms. Hole properties, including inlet (din) and outlet (dout) diameters, standard deviation of the inlet diameter (SDin), and eccentricity (εin), were determined using scanning electron microscopy image processing. For each laser setting, 9 holes were drilled with entrance diameters ranging from 15 to 60 μm. The trained ANN accurately predicted the hole properties based on the given laser drilling parameters, with deviations of no more than 2–3%. In addition, the ANN was combined with a genetic algorithm (GA) to solve the inverse problem of determining optimal laser parameters for a target din with minimum SDin and maximum εin. For a target din of 27 μm, the optimal parameters were P = 46%, ν = 49 kHz, tΣ = 19 ms and ti = 348 ns. The resulting sample showed deviations of 3.38% in din, 8.99% in εin and 7.34% in SDin, confirming the effectiveness of the proposed approach. By improving machining quality, minimising defects and increasing precision, this methodology can serve as a robust framework for optimising laser drilling processes in various materials, including steels and composites.