Do Algorithms Dream of Electric Requirements? Leveraging AI-Based Approaches for Automated Allocation and Classification of Requirements in Railway Engineering

Abstract

In recent years, Artificial Intelligence has experienced an extraordinary growth. Systems Engineering is a discipline where the implementation of AI can be challenging, but that could immensely benefit from its capabilities. This paper presents one of the many implementations that AI can have within the Systems Engineering field. AI has been leveraged to create an algorithm that allows for the automatic identification and classification of requirements within a specific engineering sector: large railway projects. While text classification algorithms are well established, the key to a successful implementation of a requirements classification algorithm lays on the effective structurization of the data, as well as the high quality of the training datasets. This paper describes how an AI-based requirements classification algorithm has been planned and trained to effectively classify requirements in future documents based on systems and subsystems from a System Breakdown Structure (SBS), as well as to predict the adequate method of verification for both the Design and Testing and Commissioning stages of a railway project.

Finally, the paper showcases how the use of this AI-based requirements classifier does not only lower the probability of human error, but also reduces ∼75% human workload per project. Additionally, overall ∼30% cost savings to organizations are expected in a 10-year period in the task of classifying requirements with respect to manual classification performed by subject matter experts.