Knowledge-driven Artificial Intelligence for Predictive Analytics in Steelmaking
Project Overview
Industry 4.0 represents a paradigm shift in manufacturing that leverages technologies such as Internet of Things (IoT), Artificial Intelligence (AI), big data and many others to create smart, automated and interconnected systems. Steelmaking is one example of a domain undergoing transformation.
Steelmaking constitutes the many processes that are used to create the strong alloy, steel. During these processes, there are vast amounts of heterogeneous data generated and processed daily, where information is distributed and stored across different data silos, systems, and formats, making it challenging to interconnect these systems. Additionally, these processes are heavily knowledge driven, i.e., they typically require a human in the loop to make and justify decisions based on their expertise and knowledge. Traditional statistical methods often fall short in capturing domain and expert knowledge and require data to be fully integrated in order to obtain parameters, which can be a strenuous task as it will require a systems expert - someone who understands the system, as well as a technical expert to do the physical integration.
Meanwhile, Knowledge-driven AI, often harnessed using ontologies, has emerged as a powerful paradigm to overcome these shortcomings.
Ontologies serve as structured, semantic models that capture domain-specific knowledge, including concepts, relationships, and rules. They act as a shared vocabulary, where information is captured and represented in a format that is both human interpretable and machine-understandable and can be stored in a knowledge graph structure. By organizing information in a meaningful and coherent manner, ontologies can comprehend complex information through semantics and meta-data without physical data integration, enabling a deeper understanding of the underlying domain and greater semantic interoperability. Additionally, they provide reasoning and inference mechanics to deduce new knowledge from existing data.
This research aims to use ontologies for predictive analytics purposes in the domain of steelmaking. We aim to focus on one aspect of steelmaking, Cold Rolling, and develop our own ontologies to capture the domain and expert knowledge of the cold rolling processes. Using this framework, aim to exploit ontological reasoning and inference to support operators using semantic decision making. Finally, we aim to investigate the possibilities of developing hybrid models that combine ontologies with traditional statistical methods.
