A federated digital twin framework leveraging AI and MLOps for The IoT cloud continuum

The complexity of industrial systems and vast data generated by sensors make manual anomaly prediction challenging. Industry 4.0 enhances manufacturing processes through the integration of technologies such as Cloud Computing (CC), Artificial Intelligence(AI), the Internet of Things (IoT), and cyber-physical systems (CPS). Machine Learning (ML) algorithms are essential for Anomaly Detection (AD), predicting equipment failures, and optimizing system configurations. Digital Twins (DTs) play a crucial role in manufacturing, particularly in smart manufacturing and predictive maintenance. However, challenges remain in terms of data privacy, integration, adaptability, and deploying ML models on resource-constrained edge devices. To address these issues, AI-based automated DT framework, C2E-AIOps, has been developed and implemented for Marposs S.p.A. This framework aims to enhance adaptability and interoperability in industrial DT systems while ensuring privacy through custom Federated Learning (FL) framework. FL allows collaborative model training without sharing raw data, addressing privacy concerns in distributed environments. However, applying FL in industrial settings is complicated by non-IID data, high dimensionality, and hierarchical system structures, all of which can affect model performance. C2E-AIOps automates data management and ML model training at the cloud level, with deployment at the edge. It utilizes the Open Neural Network Exchange (ONNX) standard for model portability and employs Continuous Integration and Continuous Deployment (CI/CD) practices for efficient deployment. The framework's effectiveness has been validated through experimental evaluations. Additionally, a novel FL framework, FedAdapt-CAD, has been proposed for AD in industrial control systems. It features Client-Aware Aggregation (CAA) and Dynamic Model Adaptation (DMA) to enhance model performance and fairness across clients. Evaluation on benchmark datasets demonstrates that FedAdapt-CAD significantly improves detection rates in complex attack scenarios. Overall, C2E-AIOps and FedAdapt-CAD represent advances in automated, scalable, and privacy-preserving industrial solutions that facilitate proactive anomaly management and secure data handling in the Industrial Internet of Things (IIoT).