Burr size minimization using a surrogate artificial neural network (ANN) assisted multi-objective genetic algorithm (MOGA) in micromilling hardened AISI HI3

Abstract The micromilling process still grapples with burr formation, which deteriorates product quality and raises production costs. The use of machine learning models and heuristic optimization algorithms for multi-objective burr control in micromilling remains limited, yet AI models are vital for Industry 4.0’s industrial revolution. This study introduces a multi-objective optimization method that reduces burr sizes relative to other process parameters. A full factorial series of experiments is performed on hardened AISI H13 (50 ± 1 HRC), measuring top burr height, workpiece’s average surface roughness, and mean roughness depth. Sensor signals related to force components and workpiece vibration are collected, processed, and used for model development. Artificial neural network (ANN) models, based on the response parameters, are then trained, tested, and validated before their weights and biases are extracted to create explicit ANN-based mathematical models, hereafter called surrogate ANN models. The models for burr sizes achieve prediction accuracies of 93% and 92% for up and down milling burr sizes, respectively, on validation data. This underscores their robustness. All models are later integrated to develop a multi-objective surrogate ANN, which functions as an objective function in genetic algorithm optimization (hereafter, MOGA). The surrogate ANN-MOGA reduces burr sizes by an average of 35%, and identifies optimal cutting parameters of f z = 4.05 μm/rev and a p = 26.86 μm. These findings are crucial for the successful micromilling of AISI H13 with TiAlN-coated carbide tools in an industrial setting. A comparative analysis with Response Surface Methodology-based multi-objective genetic algorithm (RSM-MOGA) optimization reveals that surrogate ANN-MOGA performs better.