Figure 2
The diagram illustrates a closed-loop manufacturing digital twin architecture. It features a physical manufacturing asset, such as a laser-based additive manufacturing system, connected to sensors that stream data. This data feeds into a physics-informed neural network (PINN) surrogate, which reconstructs full-field thermal states while enforcing governing physics. A genetic algorithm operates at a meta-optimization layer to recalibrate normalized loss weights based on physics-residual performance. The outputs of the PINN support monitoring, anomaly detection, and control feedback within the digital twin loop.Conceptual closed-loop manufacturing digital twin architecture integrating a GA-optimized physics-informed neural network (PINN) surrogate. Sensor measurements from the physical manufacturing asset (e.g. laser-based additive manufacturing system) are streamed to the PINN, which reconstructs full-field thermal states while enforcing governing physics. A genetic algorithm operates at a meta-optimization layer to recalibrate normalized loss weights (λd, λp, λbc) based on physics-residual performance. The resulting predictions support monitoring, anomaly detection, and control feedback within the digital twin loop