Abstract The deployment of autonomous artificial intelligence (AI) within complex physical systems remains a grand challenge. While intelligent manufacturing aims to address this by infusing industrial processes with real-time adaptation, the prevailing approach often involves applying machine learning (ML) to isolated, well-defined sub-problems. This component-wise strategy, however, proves insufficient for processes like Wire Arc Additive Manufacturing (WAAM), where geometric fidelity is governed by deeply coupled, multi-scale thermo-fluid dynamics that defy simple predictive modelling.