<![CDATA[Artificial Neural Networks (ANNs) have gained increasing attention as effective tools for modeling nonlinear and multivariate relationships in complex manufacturing processes, where conventional predictive approaches often exhibit limited accuracy. In this study, an ANN-based predictive framework was developed to estimate surface roughness (Ra) and resultant force (F) in CNC turning of hardened AISI H13 steel. The framework was constructed using an experimental dataset comprising 324 machining records, with cutting speed (vc), feed rate (f), and depth of cut (ap) as input parameters, all normalized using the Min-Max scaling method to ensure stable and efficient model training. To identify the optimal training configuration, eight optimization algorithms: Adam, RMSprop, Nadam, Adagrad, Adadelta, Adamax, FTRL, and Stochastic Gradient Descent (SGD) are systematically evaluated, and Nadam was selected as the most effective optimizer with a learning rate of 0.0001 and a batch size of 16. Two dedicated feed forward ANN models are designed separately for Ra and F prediction and validated using the Leave-One-Out Cross-Validation (LOOCV) technique to enhance generalization and minimize overfitting. The resulting models achieved excellent predictive accuracy for resultant force (R² = 0.9939, MAE = 4.3313 N, RMSE = 7.5955 N) and moderate accuracy for surface roughness (R² = 0.6454, MAE = 0.1440 µm, RMSE = 0.1960 µm). These results demonstrate that the proposed ANN-based framework provides a reliable decision-support tool for process optimization, monitoring, and surface quality control in high-performance machining environments.]]>
