Identification of Variables Impacting Cascading Failures in Aerospace Systems: A Natural Language Processing Approach

The emergence of new technological advancements increases the complexity level of systems, such as those in aerospace, resulting in unpredictable behaviors. Systems engineering practices have been extensively used to conceptualize different complex problem domains and reveal the underlying patterns in systems. Tools such as causal loop diagrams (CLDs) are implemented to identify reinforcing and balancing loops within a system, along with quantitative techniques such as stock and flow models. Identifying critical variables in a complex system is challenging, and it prevents the domain expert from developing a systematic approach to the problem definition. To address this challenge, a natural language processing (NLP) approach has been applied to aircraft accident reports to systematically identify variables and assist the domain expert in generating a CLD. The accident report data is analyzed using a large language model (LLM) Llama 2 to identify the cause of accidents and categorize risks in this system. The critical variables associated with the cascading failures in aerospace systems are extracted using topic modeling techniques. The keywords in each topic were assigned to different risk categories based on the coherence analysis. The results generated a CLD that underscores this system’s feedback loops. The findings highlight NLP’s applications in systems engineering methods and its contribution to the risk assessment of aerospace systems.