Performance analysis of Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Simulated Annealing (SA) for surface roughness optimization in aluminum alloy milling

Surface roughness (Ra) is one of the most important response indicators used to evaluate machining quality. Although various metaheuristic optimization algorithms have been applied in machining optimization, there remains limited comparative evidence regarding their relative performance under consistent modeling and computational conditions. Therefore, this study aims to evaluate and compare the performance of three widely applied metaheuristic optimization algorithms, namely Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Simulated Annealing (SA), in minimizing Ra during the milling of aluminum alloys. A predictive model for Ra is first constructed using the Group Method of Data Handling (GMDH). Among the eight functions tested in the GMDH network, the double-variable model provides the highest accuracy with an R² of 0.9990, RMSE of 0.0042, and MAPE of 0.2985 for training, and an R² of 0.9962, RMSE of 0.0083, and MAPE of 0.5724 for validation. This model is subsequently integrated into the optimization stage, where GA, PSO, and SA are implemented under consistent computational conditions to ensure a fair comparison. The optimal Ra values obtained by the algorithms are 0.6448 for GA, 0.6445 for PSO, and 0.6472 for SA. PSO achieves the best overall performance due to its rapid convergence, high stability, and small variance across repeated runs. GA produces competitive results with moderate variability, whereas SA converges more slowly and yields a wider spread of solutions. The findings confirm that PSO is the most effective algorithm among the three for surface roughness optimization in aluminum alloy milling and provide practical guidance for selecting machining parameters.