Purpose: Dry machining of Ti–6Al–4V presents significant challenges due to its low thermal conductivity, high strength, and strong adhesion at the tool–workpiece interface. These characteristics limit the accuracy of empirical surface-roughness models. This study aims to present and validate an AI-Optimized, Physics-Based Feature Engineering framework capable of predicting surface roughness with high fidelity. Methods: A Central Composite Design with 19 experiments was conducted by varying cutting speed, feed rate, and depth of cut. Surface roughness and cutting forces were measured, and physics-derived features such as specific cutting energy, strain rate, and shear stress were engineered. Multiple Machine Learning models, including boosting ensembles, neural networks, and support vector regression, were trained and evaluated through repeated cross-validation and benchmark comparisons. Results: The proposed Physics-Based framework demonstrated superior performance over classical methods. Specifically, the optimized XGBoost model achieved the highest predictive accuracy ( $$\varvec{R^2 = 0.98}$$ ), reducing the Mean Absolute Error (MAE) to 0.018 $$\mu$$ m compared to 0.23 $$\mu$$ m for Response Surface Methodology. The Neural Network also showed robust generalization ( $$\varvec{R^2 \approx 0.92}$$ ). The framework effectively captured complex thermomechanical behaviors characteristic of Ti–6Al–4V machining, delivering micrometric-precision predictions. Conclusion: The proposed framework provides a reliable tool for optimizing surface quality in dry machining of Ti–6Al–4V. By enabling precise prediction of roughness, it supports the selection of optimal cutting parameters for aerospace and biomedical components.

Fused filament fabrication (FFF) is a widely adopted additive manufacturing technology whose process performance emerges from complex and often non-intuitive interactions among printing parameters, thermal conditions, material deposition mechanisms, and resulting structural properties. Despite its growing adoption in industrial and engineering contexts, identifying robust parameter combinations that simultaneously balance productivity, cost, surface quality, and mechanical performance remains a challenging task, particularly for fiber-reinforced polymer composites. In this study, a data-driven and explainable framework is proposed for the design and optimization of the PET-CF15 FFF process by integrating machine learning, Bayesian hyperparameter optimization, explainable artificial intelligence, and multi-objective metaheuristic optimization. More than twenty machine learning model–hyperparameter configurations are systematically evaluated to construct surrogate models capable of capturing both deterministic and stochastic manufacturing responses. Model interpretability is explicitly addressed through Shapley value analysis, enabling physical insight into the role of key printing parameters. The trained surrogates are embedded into a multi-objective optimization framework based on the Multi-objective Lichtenberg Algorithm, allowing the systematic exploration of competing and synergistic process objectives without additional experimental effort. Notably, the results reveal that, depending on the explored design space, objectives commonly assumed to be conflicting may converge toward compatible optima, highlighting the existence of stable operating regimes that are not evident from empirical or single-objective analyses. By unifying predictive accuracy, explainability, and multi-objective decision-making, the proposed approach provides a robust computational tool to support informed and efficient FFF process design for high-performance composite materials.

Data preprocessing is a pivotal stage in Machine Learning (ML) workflows, directly influencing both predictive performance and computational efficiency. Although widely acknowledged, the quantitative impact of preprocessing choices, particularly dimensionality reduction strategies, remains insufficiently explored in classification tasks. This study compares three data preparation approaches: raw data, Principal Component Analysis (PCA), and Principal Component Factor Analysis (PCFA). While PCA and Factor Analysis (FA) are well-established techniques, their structured integration as PCFA has received limited attention. By evaluating these alternatives, this study quantifies how dimensionality reduction affects predictive capability, result stability, and processing time across diverse learning scenarios. A factorial Design of Experiments (DOE) framework is employed, enabling the assessment of model performance under multiple attribute configurations and their interaction effects. Ten distinct ML algorithms are applied to ten heterogeneous datasets from different domains, with model training performed using cross-validation and hyperparameter optimization via Grid Search and Optuna. The findings show that PCA and PCFA consistently achieve competitive predictive performance while substantially reducing computational cost compared to raw data. Furthermore, data structure and attribute interactions strongly influence outcomes, indicating that the relationship between preprocessing strategies and model behavior is scenario-dependent. We conclude that combining dimensionality reduction, structured feature design, and optimized training strategies is essential for balancing predictive accuracy and computational cost, a trade-off increasingly critical in modern ML applications.

This study proposes a structured decision-support framework for machining operations involving multiple and conflicting performance objectives. The approach integrates multi-objective optimization using the Varimax Rotated Factor-Normal Boundary Intersection method with a post-Pareto decision stage designed to identify the most appropriate solution among Pareto-optimal alternatives. Principal Component Analysis and Factor Analysis, combined with Varimax rotation, are employed to reorganize eight correlated response variables into three independent and interpretable performance dimensions representing Reliability, Economic Behavior, and Quality. This transformation enhances the stability of the search space and improves the robustness of the optimization process. In the post-optimization stage, the Pareto-optimal solutions are evaluated using the Shannon’s Entropy/Global Percentage Error, allowing a systematic and multi-criteria assessment of the solution set. A sensitivity analysis based on the variation of the weighting parameter indicates that the decision-making process is stable and robust with respect to preference changes. In addition, a comparative analysis was conducted with well-established multi-objective optimization methods, including the Non-dominated Sorting Genetic Algorithm II, the Multi-Objective Evolutionary Algorithm based on Decomposition, the Weighted Sum Method, and the Multi-Objective Lichtenberg Algorithm, using the Hypervolume, Inverted Generational Distance and Spacing metrics. The results show that, while traditional methods exhibit strong capability in exploring the Pareto front, the proposed framework provides more uniformly distributed solutions, along with a structured mechanism for selecting the final solution, in addition to lower computational time and cost. A case study on the hard turning of AISI H13 steel using PCBN 7025W inserts confirms the applicability of the proposed framework, optimizing cutting speed, feed rate, and depth of cut with respect to tool life, economic performance, and surface integrity. The proposed method identifies balanced parameter configurations while reducing reliance on subjective decision-making, thereby supporting decision-making in multi-objective contexts within manufacturing environments.

Science & Business Connection: Anais do Congresso Científico Tecnológico (978-65-272-1985-9) - Extração De Características Geométricas Irregulares Por Modelagem Computacional No Contexto Multiobjetivo