Machine-learning-based inventory models to support life cycle assessment of forged steam-turbine components

Life Cycle Assessment (LCA) of turbomachinery components is highly data-intensive and often constrained by limited manufacturing information. This study develops machine-learning-based surrogate models to estimate key forging inventory parameters for blade carriers, including material requirements and processing times. Following the CRISP-DM framework, a dataset was generated through CAD-based engineering design-space exploration and independently evaluated using LeanCOST. Since multiple configurations originated from the same component geometry, a group-aware validation strategy was adopted to prevent data leakage and assess model generalisation to unseen blade-carrier designs. Different regression algorithms were evaluated and selected according to their predictive performance for each inventory parameter. The models were also applied to additional geometries within the investigated dimensional range and descriptively compared with proprietary analytical approaches. The machine-learning models showed the clearest benefits for manufacturing parameters characterised by more complex geometric relationships, while simpler analytical models remained competitive for some mass-related quantities. From an LCA perspective, model-development effort should be prioritised according to the environmental relevance of each inventory parameter. The proposed approach supports rapid generation of Life Cycle Inventory data during early design, when detailed process information is not yet available.