Industry 4.0 marks a new era in industrial markets and digitalization is an international research topic to increase productivity and competitiveness. This can be addressed through increased energy and resource efficiency and improved process autonomy, stability and control system flexibility. The industry partners - Hydro, Elkem, Borregaard and Yara - face similar challenges related to complex, nonlinear processes in harsh environments with largely varying time constants, where only sparse, often manual measurements are available. These issues cannot be fully solved through traditional control methods, and new competence is needed to address them.
ML consists of data-driven, flexible and adaptable methods that show improved performance and ability to recognize patterns and connections beyond that of traditional methods in areas such as image analysis and classification, natural language processing and medical diagnosis. However, several challenges and requirements of the process industries cannot be solved using off-the-shelf solutions. In particular, it is not necessarily straightforward to apply the state-of-the-art results in other application areas to fields such as control systems and process optimization. Traditional ML approaches are not based on the laws of phyisics as traditional control methods. Rather, they are non-transparent, lacking in terms of mathematical analysis and often applied in a black box matter. Furthermore, fully understanding why and when purely data-driven methods succeed and fail is presently an unsolved research challenge.
The intersection between control theory and ML is an emerging research field with huge potential where both domain knowledge and collected data are exploited in a hybrid analytics manner. This project will explore, extend and develop new ML methods for control systems that are able to handle the particular challenges of the process industries. To safely apply these control methods in real-life processes, this project seeks to significantly advance the understanding and formal analysis of what can be guaranteed with respect to stability, robustness and convergence when using ML methods in control systems, and what is required related to system observability and training data.
The project results are expected to increase the competence level, productivity and competitiveness for Norwegian process industry as well as improve safety for process operators and reduce emissions, use of chemicals and waste production. TAPI will be a step towards increased autonomy and digitalization, and towards green competitiveness and zero emissions to land-based process industries.
Read more about AI in SINTEF:
