Download PDFOpen PDF in browser

Machine Learning-based Classifier in the Analysis of Nuclear Power-Specific Requirements

EasyChair Preprint 5297

4 pagesDate: April 6, 2021

Abstract

Typical nuclear power plant projects include the sheer volume of descriptive and non-harmonized requirements, which have to be managed and fulfilled. These requirements are typically hard to interpret and humans' very limited ability to concentrate on a specific task together with a large number of requirements usually cause main errors in the analysis of the requirements. By utilizing artificial intelligence in the analysis of nuclear power plant requirements, designers' decision-making in classification and allocation of requirements could be facilitated and thus, errors reduced.

Fortum developed a machine learning-based requirements classifier utilizing state-of-the-art natural language processing (NLP) and integrated it with a requirements management system. The classifier categorizes nuclear power industry-specific requirements into pre-defined categories. The categories have been determined based on processes and design disciplines of Fortum's nuclear engineering departments that are responsible for fulfilling requirements.

The very promising results include predetermined requirement classes, manually gathered and labeled data, comparison of three models and their classification accuracies. In addition to these results, the established classifier was integrated with the requirements management system. Future development suggestions include focusing on atomizing (i.e., splitting up) long, especially multiclass requirements, combining similar ones and checking requirements syntax based on suggestions generated by an AI-model. Furthermore, new and practical requirement classes and hierarchies are suggested to be developed while also improving current datasets both quantitatively and qualitatively.

Keyphrases: Artificial Intelligence, Classifier, Natural Language Processing, Requirements Classification, Requirements Engineering, Requirements Management System, Systems Engineering, machine learning, nuclear power, requirements

BibTeX entry
BibTeX does not have the right entry for preprints. This is a hack for producing the correct reference:
@booklet{EasyChair:5297,
  author    = {Tapani Raunio and Ilpo Suominen and Santeri Myllynen and Rasmus Karell},
  title     = {Machine Learning-based Classifier in the Analysis of Nuclear Power-Specific Requirements},
  howpublished = {EasyChair Preprint 5297},
  year      = {EasyChair, 2021}}
Download PDFOpen PDF in browser