Abstract
This report is an introduction to machine learning, and it focuses on topics that can be seen as challenging on safety critical domains such as nuclear power industry. The requirements for designing, developing and testing a safety-critical system for operation are stricter and more regulated than in many conventional domains because of the high emphasis on correct and predictable system behaviour for the protection of the public. While machine learning techniques and the science behind them is developing in an unparalleled fast pace, there are still many fundamental challenges that need special attention when they are to be used in safety-critical context, where there are no room for errors.
In the report, the three main machine learning paradigms are briefly presented and the basic steps in their use are explained. Then, the use of machine learning components in safety critical systems and industrial domains is discussed, while highlighting different fundamental properties and challenges. Finally, it goes through some of the most interesting machine learning topics that were identified in informal discussions with Finnish nuclear stakeholders as potential current or future directions for research and development.
There is clear motivation in nuclear safety engineering to look for and carefully adopt technologies that can improve safety in a proven and measurable manner. The authors, based on this study and the discussions, recognize the growing interest to utilize artificial intelligence based systems to support the everyday work of engineers over the different lifecycle phases of plant design and operation. Topics, which raised most interest were related to using Natural Language Processing for document and requirements, different predictive maintenance topics and supporting the work of operators in control rooms.
In the report, the three main machine learning paradigms are briefly presented and the basic steps in their use are explained. Then, the use of machine learning components in safety critical systems and industrial domains is discussed, while highlighting different fundamental properties and challenges. Finally, it goes through some of the most interesting machine learning topics that were identified in informal discussions with Finnish nuclear stakeholders as potential current or future directions for research and development.
There is clear motivation in nuclear safety engineering to look for and carefully adopt technologies that can improve safety in a proven and measurable manner. The authors, based on this study and the discussions, recognize the growing interest to utilize artificial intelligence based systems to support the everyday work of engineers over the different lifecycle phases of plant design and operation. Topics, which raised most interest were related to using Natural Language Processing for document and requirements, different predictive maintenance topics and supporting the work of operators in control rooms.
Original language | English |
---|---|
Place of Publication | Tampere |
Publisher | VTT Technical Research Centre of Finland |
Number of pages | 27 |
Publication status | Published - 20 Oct 2020 |
MoE publication type | D4 Published development or research report or study |
Publication series
Series | VTT Research Report |
---|---|
Number | VTT-R-01124-20 |
Keywords
- machine learning
- artificial intelligence
- safety critical industry
- nuclear power