Tailoring MLOps Techniques for Industry 5.0 Needs

It is a very popular era for machine learning (ML) applications, and Industry5.0 aims to have AI as one of its key technologies. Still, only a few ML initiatives make it to a production-grade implementation, mostly due to lacking proper Continuous Integration and Delivery framework and MLOps practices. This is especially true for industrial use cases, where the trust and reliability of ML applications are mission-critical. Most of these applications fail during the final stage of the development lifecycle, i.e. acceptance testing and validation of the ML application, while being integrated into Cyber-Physical System of Systems (CPSoS). This paper explores the key requirements for deploying ML applications in industrial scenarios, emphasizing the critical role of Digital Twins, edge AI, and responsible-explainable AI techniques in ensuring efficient and responsible operations. Building upon previous models, this paper suggests two process models: (i) the Olympics model for MLOps-coupled CPS engineering and (ii) the MLOps engineering toolchain for industrial applications.