Osmotic Cloud-Edge Intelligence for IoT-Based Cyber-Physical Systems

Area of Science:

• Cyber-Physical Systems
• Artificial Intelligence
• Edge Computing

Background:

• AI in Cyber-Physical Systems leverages massive IoT data for accurate machine learning inference.
• Edge Intelligence executes inference locally using cloud-trained models, but lacks flexible training migration.
• Current Edge nodes struggle with dynamic training task migration between edge and cloud environments.

Purpose of the Study:

• To propose a novel Cloud-Edge AI microservice architecture based on Osmotic Computing principles.
• To enable flexible and dynamic training and inference capabilities across distributed edge and cloud resources.
• To facilitate the opportunistic exploitation of computational resources for optimal AI prediction accuracy.

Main Methods:

• Developed a containerized microservice architecture for AI training and inference.
• Encapsulated architectural modules for direct mapping with Commercial-Off-The-Shelf (COTS) components.
• Implemented a prototype using commodity hardware and open-source technologies.

Main Results:

• A functional prototype demonstrating the proposed Cloud-Edge AI architecture was realized.
• Experiments were conducted within a small-scale intelligent manufacturing case study.
• The results validated the feasibility and significant benefits of the proposed architecture.

Conclusions:

• The Cloud-Edge AI microservice architecture effectively supports dynamic training and inference.
• The approach enhances AI model accuracy by opportunistically utilizing edge and cloud resources.
• The architecture's modularity and COTS compatibility promote practical implementation in intelligent systems.