Predicting Surface Roughness in AISI H13 Steel Turning: A Comparative Analysis of Extreme Learning Machine, Gaussian Process, and Transformer Models

Predicting surface roughness (Ra) in machining processes is critical for ensuring product quality and optimizing manufacturing efficiency. This study aimed to predict Ra during the CNC turning of AISI H13 steel using advanced machine learning models. Real-time feed force (Fz) measurements, combined with input variables such as depth of cut (ap), cutting speed (vc), and feed rate ($f$), were leveraged to develop accurate predictive models. The methodology involved designing experiments with a factorial approach, collecting surface roughness and cutting force data, and training three machine learning models: Extreme Learning Machine (ELM), Gaussian Process Regression with Deep Kernel Learning (GP-DKL), and a Transformer-based model. The dataset was normalized and split using 5-fold cross-validation to ensure robust evaluation, and the models were assessed based on their predictive performance through Mean Squared Error (MSE), Mean Absolute Error (MAE), and R2 metrics. The ELM model demonstrated the highest accuracy, achieving an R2 of 0.980, followed by GP-DKL (R2: 0.932) and the Transformer model (R2: 0.856). These results highlight the potential of machine learning for real-time prediction and opti-mization of machining processes. Future work could incorporate additional real-time variables, such as temperature and vibration, to further enhance predictive accuracy.