Development of a method for generating System-specific failure cases using artificial intelligence based on information from Abstract System Models

Abstract

The aviation industry is progressing rapidly to provide climate-friendly air transport and meet its sustainability targets through electrification and digitalization. These advancements lead to increasing system complexity and integration across multiple domains. To manage these challenges, standards such as SAE ARP4754B provide clear guidelines for using a structured system engineering approach to develop complex aviation systems. Model-Based Systems Engineering (MBSE) with SysML has been the common practice to ensure consistency and traceability across different system levels. It further provides a holistic representation of abstract system architectures and functionalities. One of the key challenges in aviation system development is ensuring system safety, which is governed by SAE ARP4761A. This standard defines analysis methods that iteratively build the required abstract system. To improve workflow efficiency, Model-Based Safety Assessment (MBSA) is recommended for seamless data exchange. While MBSA connects model-based frameworks for system engineering and safety analysis, it remains a highly manual process and involves redundant system data across multiple tools. This thesis proposes a methodology to automate the human-intensive safety assessment process. It employs the current AI technology, specifically large language models (LLMs), to use their contextual data generation capabilities. A novel framework is developed with standard architectural viewpoints and an integrated safety perspective considering model-based approaches to accommodate safety artifacts in compliance with industry standards. The defined framework is further integrated with a localized LLM to generate system-specific failure cases from the abstract model. A defined workflow is derived to implement the developed methodology for the creation of failure conditions for a function using Functional Hazard Analysis (FHA) for aircraft systems. Further, the findings imply the possibility of the usage of AI in current aviation workflows, which can be enhanced to be incorporated into regular aviation system development and safety assessment processes.