STPA Automation Using Natural Language Processing and Machine Learning

• Grant number: 23/03393-5
• Support Opportunities: Regular Research Grants
• Start date: May 01, 2025
• End date: April 30, 2028
• Field of knowledge: Engineering - Aerospace Engineering - Aerospace Systems
• Principal Investigator: Luiz Eduardo Galvão Martins
• Grantee: Luiz Eduardo Galvão Martins
• Host Institution: Instituto de Ciência e Tecnologia (ICT). Universidade Federal de São Paulo (UNIFESP). Campus São José dos Campos. São José dos Campos , SP, Brazil
• Associated researchers: Ana Estela Antunes da Silva ; Fabricio José Pontes ; Marcelo Jose Ruv Lemes ; Niklas Lavesson ; Tony Gorschek
• Associated scholarship(s): 25/16287-4 - Automation of STPA Using Natural Language Processing and Machine Learning: Automatic Generation of Safety Requirements, BP.DR

Abstract

Safety-Critical Systems (SCS) are increasingly present in the daily life of modern societies. The complexity of SCS has increased significantly in recent decades, demanding new approaches to the analysis of accidents, hazards, safety and risks of these systems. Among the new approaches are STAMP (System Theoretic Accident Model and Processes) and STPA (System Theoretic Process Analysis), both proposed by Nancy Leveson. The objective of this project is to promote an automation of the STPA technique, through the development of classifier models that automate the identification of losses, hazards, and constraints (step 1 of the STPA) and unsafe control actions (step 3 of the STPA). The automatic identification of these concepts will occur through the analysis of textual documents, which are typically created during the processes of hazard analysis, risk analysis and specification of SCS requirements. For the development of the classifier models, we intend to develop an NLP pipeline, create a textual documents base that will be used to compose the training and testing datasets, and use tools such as BERT (Bidirectional Encoder Representation from Transformers), which consists of a Natural Language Processing (NLP) model to generate word representations (embeddings) based on a Transformer-like architecture. This project is being proposed in cooperation with the Software Engineering Research Lab. of the Blekinge Institute of Technology (BTH - Sweden), and Embraer (Brazil). Among the expected results, it is intended to provide a set of tools that allow an automation of the STPA, as well as to improve the quality and productivity of the teams responsible for the analysis of risks and hazards in SCS, with a reduction in cost and time in the development process and SCS certification. (AU)