Resumo:
Turning is one of the main machining processes used in the manufacture of metallic components with high dimensional accuracy and good surface finish. The efficiency of this process directly depends on the condition of the cutting tool, whose wear affects the quality of the workpiece, the cycle time, and production costs. In this context, monitoring and predicting tool wear are essential to ensure productivity and reliability in manufacturing systems. This work aimed to predict tool wear in the turning process of AISI 52100 steel through the analysis of vibration signals using Artificial Intelligence (AI) techniques. The signals were collected by two different sensors — a laser vibrometer and a three-axis accelerometer — allowing precise acquisition of the system’s dynamic responses during machining. From the raw signals, relevant features were extracted in both time and frequency domains, such as amplitude, energy, kurtosis, skewness, and dominant frequency, composing the set of variables used for modeling. Several machine learning models were evaluated, including Artificial Neural Networks (MLP), Support Vector Regression (SVR), Extra Trees, and XGBoost, as well as ensemble learning techniques. The stacking ensemble model achieved the best performance, reaching R² = 0.88 and MSE = 0.00552, surpassing the individual approaches. The variable importance analysis performed using the SHAP (SHapley Additive exPlanations) method revealed that the energy and kurtosis of the vibration signals exert the greatest influence on wear, confirming the physical consistency of the model. In addition, an interactive interface was implemented in Python (Gradio), integrating the predictive model with a Large Language Model (LLM) capable of automatically generating cutting parameter recommendations and natural language interpretations. This integration enhances the explainability of the results and demonstrates the potential of AI applied to intelligent machining, in line with the principles of Industry 4.0 and Explainable Artificial Intelligence (XAI). The results confirm the feasibility of the proposed approach for predictive and assisted monitoring of tool wear, indicating promising paths for the industrial application of intelligent decision-support systems within the context of advanced manufacturing.
