| description abstract | In nuclear power plant (NPP) projects, requirements engineering manages the sheer volume of requirements, typically characterized by descriptive and nonharmonized requirements. Large projects may have tens of thousands to hundreds of thousands of requirements to be managed and fulfilled. Two main issues impede requirements analysis: tortuous requirements to be interpreted; and humans' very limited ability to concentrate on a specific task. It has therefore been recognized that artificial intelligence (AI) algorithms have the potential to support designers' decision making in classifying and allocating NPP requirements into predefined classes. This paper presents our work on developing an AI-based requirements classifier utilizing natural language processing (NLP) and supervised machine-learning (ML). In addition, the paper presents the integration of the classifier with the requirements management system. The focus is on the classification of nuclear power industry-specific requirements utilizing deep-learning-based NLP. Three classifiers are compared, and the corresponding results are presented. The results include predetermined requirement classes, manually gathered and classified data, a comparison of three models and their classification accuracies, microservice system architecture, and integration of the established classifier with the requirements management system. As the performance of the requirements classifier and related system has been successfully demonstrated, future AI-specific development and studies are suggested to focus on atomizing multiclass requirements, combining similar requirements into one, checking requirements syntax, and utilizing unsupervised learning for clustering. Furthermore, new and advantageous requirement classes and hierarchies are suggested for development while improving current datasets both quantitatively and qualitatively. | |
