Fortum’s R&D project developing artificial intelligence-based models for requirements engineering has developed a tool that allocates requirements into predetermined classes and significantly speeds up these processes. Utilising AI along with our extensive nuclear expertise has allowed us to facilitate more efficient requirements classification and thus will help us better support our nuclear engineering tasks.
The project to create an artificial intelligence (AI) based requirements classifier for requirements engineering tasks of our Nuclear Engineering Department was initiated in 2019. Based on the results of Santeri Myllynen’s master’s thesis, the development project is aiming to respond to the challenges of requirements engineering in a safety-critical industry. The challenges in this area include, for example, the vast amount of requirements to be manually classified and the ability of an individual to make exact decisions when faced with a multitude of options. This kind of classification activity is extremely ineffective as a baseline and further hindered by the limits of human attention span. Errors in requirements specifications due to erroneous interpretations and allocations of requirements may cause massive extra costs because of the need for redesigns or even the reproduction of components. Using AI in requirements classification is expected to support designers in their decision-making process when classifying a vast amount of requirements because computers can constantly retain the same efficiency with tremendous processing power.
Combining an AI solution with 40 years of Loviisa Nuclear Power Plant operating experience
In the first phase of the project, we utilised open-source classifiers and our requirement datasets to train requirements classifiers and compare them with each other. The classifier is utilised at the beginning of the requirements engineering workflow after collecting the initial set of requirements. The AI model takes a requirement as an input and outputs the requirement with the relevant requirement classes representing design disciplines and processes. Fortum’s nuclear engineering knowledge is included in the training datasets in which we have classified requirements based on our know-how acquired both during the 40-year lifetime of Loviisa Nuclear Power Plant as well as our external projects.
The project group has involved numerous experts with a range of different competencies – from artificial intelligence and systems engineering, particularly machine learning, to natural language processing, requirements engineering, and configuration management. This group was further supported by experts from every design discipline of Fortum’s Nuclear Engineering department, especially when forming the training, test, and validation datasets for the classifiers.
So far, the project has already resulted in a classifier integrated with a requirements management system used for writing engineering documents, eliciting, allocating and elaborating requirements, and tracking them through the whole lifecycle of a nuclear power facility. Additionally, the project has resulted in multiple scientific papers. Fortunately, this fruitful and inspiring project continues researching and developing new features, such as atomising long requirements consisting of several actual requirement entries and combining similar requirements into one. We are looking forward to better facilitating our internal and external requirements engineering tasks.
The project was led by Mr. Santeri Myllynen until February 2021 and has since been headed by Mr. Tapani Raunio.
Senior Engineer, Nuclear Safety
tapani [dot] e [dot] raunio [at] fortum [dot] com
- In the first phase, we utilised open-source classifiers and our requirement datasets to train different requirements classifiers and compared them with each other.
- We integrated the best-performing classifier with our requirements management system Polarion ALM®, which we use to write engineering documents and allocate and elaborate requirements to suitable disciplines and architecture levels. This also creates requirement links to enable the traceability of requirements.
- The classifier is utilised in the beginning of the requirements engineering workflow after collecting the initial set of requirements, such as stakeholder needs and requirements, technical requirements, and design process requirements.
- The AI model takes a requirement as an input, classifies it according to the domain knowledge gained from a training dataset and outputs the same requirement associated with relevant requirement classes representing design disciplines and processes.
Myllynen, S., 2019, “Utilization of Artificial Intelligence in the Analysis of Nuclear Power Plant Requirements,” Master’s thesis, Aalto University, Espoo, Finland.
IYNC 2020 Publication
Myllynen, S., and Jitta, A., 2020, “Utilization of Artificial Intelligence in the Analysis of Nuclear Power Plant Requirements,” IYNC2020 Conference Proceedings, Sydney, Australia, March 8-13, pp. 154–157.
ASME Publication 2021
Myllynen, S., Suominen, I., Raunio, T., Karell, R., Lahtinen, J., 2021, “Developing and Implementing AI-Based Classifier for Requirements Engineering, ASME. ASME J of Nuclear Rad Sci.
Raunio, T., Suominen, I., Myllynen, S., Karell, R., 2021, “Machine Learning-based Classifier in the Analysis of Nuclear Power-Specific Requirements,” Automaatiopäivät24 Conference Proceedings, Online, April 13-14.