Edge cloud as an enabler for distributed AI in industrial IoT applications: The experience of the iotwins project

Emerging Industrial Internet of Things (IIoT) applications are pushing the academic and industrial research towards novel solutions for, on the one hand, frameworks to facilitate the rapid and cost-effective exploitation of general-purpose machine learning mechanisms and tools, and, on the other hand, hw/sw infrastructures capable of guaranteeing the desired and challenging quality of service indicators in industrial scenarios, e.g., latency and reliability. We claim that these directions can be effectively and efficiently addressed through the adoption of innovative quality-aware edge cloud computing platforms for the design, implementation, and runtime support of distributed AI solutions that execute on both global cloud resources and edge nodes in industrial plant premises. In particular, the paper presents the first experiences that we are doing within the framework of the H2020 Innovation Action IoTwins, for the implementation and optimization of distributed hybrid twins in the IIoT application domains of predictive maintenance and manufacturing optimization. IoTwins exploits distributed hybrid twins, partly executing at edge cloud nodes in industrial plant localities, to perform process/fault predictions and manufacturing line reconfigurations under time constraints, also by enabling some forms of sovereignty on industrial monitoring data. In addition, the paper overviews our original taxonomy of the stateof-the-art research literature about distributed AI for decentralized learning, with specific focus on federated settings and on emerging trends for the IIoT domain.