Federated Learning-Enabled Digital Twin for Smart Additive Manufacturing Industry

This work introduces a novel architecture of federated learning (FL)-enabled digital twin (DT) for the smart additive manufacturing industry, especially 3D printing. The proposed architecture tackles the previous limitation of the centralized approach that requires a large number of communication costs by efficiently updating the fault detection model on each server with distributed learning methods. A CNN-based model is also proposed to efficiently learn sensory data from a 3D printer for a fast and reliable fault detection model. To provide a robust system in intelligent manufacturing, a DT platform is also designed for seamless monitoring and control purposes. The proposed DT platform is able to initiate, monitor, and terminate the 3D printing process of physical assets via a virtual environment. Based on the simulation results, the FL process demonstrates that the proposed CNN-based model is superior to other DL models with 8% accuracy enlargement while maintaining the low training period. Furthermore, experimental work is conducted to evaluate the proposed architecture with real-world devices. Finally, the findings indicate that the overall latency given by the proposed system is relatively low, with an average of 1026.16 ms from the physical 3D printer to the DT platform.