Data-Driven Soft Sensor for Energy Performance Prediction in Multi-stage Machining Operations

Abstract Machining operations represent a major source of industrial energy consumption, especially when components are produced through multi-stage process chains. However, most existing studies focus on individual machining operations rather than the integrated behavior of sequential processes. This study investigates a machining chain consisting of internal turning, surface milling, and drilling performed on C40 carbon steel components. A wide experimental campaign was conducted, with real-time monitoring of electrical current and voltage to determine instantaneous power and cumulative energy consumption. A data-driven framework was developed to predict total specific cutting energy and total machining time using only primary machining parameters. The modeling procedure included rigorous data preprocessing, prevention of data leakage, cross-validation, hyperparameter tuning, and external validation. Ensemble learning models achieved high predictive accuracy, enabling reliable estimation of global energy and time indicators. The proposed approach supports virtual sensing and energy-aware process planning in sustainable manufacturing systems.