Optimisation of neural network hyperparameters for predictive modelling in laser micromachining of biopolymers

Abstract Laser micromachining has emerged as a precise and efficient technique for modifying biopolymer surfaces, with polylactic acid (PLA) due to its biodegradability and applications in sustainable electronics. This study systematically investigates the influence of key hyperparameters epochs, learning rate, and neural architecture on the predictive accuracy of a neural network model for machining outcomes. A predictive framework is developed to optimise machining width and the heat-affected zone by identifying the most effective activation functions, the optimal number of neurons in the hidden layers, and suitable learning rates. Furthermore, the computational time, a critical factor in practical implementation, is analysed with respect to batch size variations to assess its impact on model convergence and predictive performance. The findings offer significant insights into the integration of deep learning for refining laser micromachining processes, providing a structured approach to enhance precision while minimising thermal damage in polymer-based microfabrication.