Digital twin driven smart factories: real time physics based co-simulation using edge a.i. and federated learning

Area of Science:

• Smart Manufacturing
• Industry 4.0
• Digital Twin Technology

Background:

• Conventional Digital Twin (DT) approaches often rely on cloud solutions, leading to high latency, scalability issues, and data privacy concerns.
• Real-time, physics-enabled simulation for rapid decision-making in factories is hindered by computational speed and network limitations.

Purpose of the Study:

• To develop a novel method combining Digital Twin, Edge AI, and Federated Learning (FL) for efficient and secure co-simulation in smart factories.
• To address the challenges of latency, scalability, and data privacy associated with cloud-centric DT architectures.

Main Methods:

• Implementing a framework that leverages Edge AI and FL for real-time data processing and inference on edge devices near the factory floor.
• Utilizing Federated Learning (FL) to enable collaborative learning among edge devices while keeping sensitive production data localized.
• Developing an FMI-based engine for real-time co-simulation of physical and virtual events to test system behavior.

Main Results:

• Achieved up to a 35% reduction in latency compared to cloud-only architectures.
• Reduced cloud usage by 28% and increased throughput by 13.2%.
• Demonstrated enhanced scalability and industrial applicability of the proposed framework.

Conclusions:

• The integration of Digital Twin, Edge AI, and FL offers a quick, efficient, and secure solution for co-simulation in smart factories.
• Moving computation to the edge significantly improves performance and data privacy.
• Future enhancements could include blockchain integration, automated learning updates, and cross-plant knowledge transfer.