Improving the Development and Reusability of Industrial AI Through Semantic Models

Despite some of the success of AI, particularly machine learning, in industrial applications such as condition monitoring, quality inspection and asset control, solutions are typically bespoke and not robust in the long term. There is a considerable amount of effort in developing these solutions to deliver accurate results within a very limited scenario. In addition, the operationalisation of these models in the factory floor is a challenge. Developing and maintaining these models requires of data science expert knowledge and the digital skills gap in the manufacturing industry is a major barrier. A step towards the development of AI skills can be facilitated in a Learning Factory environment, provided there is a way for operators to develop an understanding of how manufacturing problems can be addressed with different data science tools. To address this, this paper introduces a semantic based framework as one of the key elements to facilitate the development of Industrial AI solutions. By formalising the way data, manufacturing processes and AI models are described and linked before and after the creation of a solution, it is possible not only to automate model creation, but to enable reusability and management. A preliminary conceptualisation on the use of this framework through a process monitoring scenario is presented. By capturing the semantic relationships, it is possible to support a more automated machine learning pipeline, enable manufacturers to understand how solutions can be created and learn how they can then be reused in future similar scenarios.