{"id":13035,"date":"2024-10-01T10:09:43","date_gmt":"2024-10-01T08:09:43","guid":{"rendered":"https:\/\/www.iese.fraunhofer.de\/blog\/?p=13035"},"modified":"2024-10-01T10:10:16","modified_gmt":"2024-10-01T08:10:16","slug":"safety-engineers-conducting-hara-with-lasar","status":"publish","type":"post","link":"https:\/\/www.iese.fraunhofer.de\/blog\/safety-engineers-conducting-hara-with-lasar\/","title":{"rendered":"LLM-Human Co-Engineering to increase efficiency of Safety Engineering Processes (HARA)"},"content":{"rendered":"

Unlocking Automotive Safety: How LLM-Augmented Tools Are Transforming Hazard and Risk Assessments<\/h2>\n

In the ever-evolving landscape of automotive safety engineering, the Hazard Analysis and Risk Assessment (HARA) process is crucial. This procedure demands extensive engineering expertise to meet the requirements of safety standards like ISO 21448 (SOTIF) and ISO 26262 (Functional Safety). The field of generative AI, especially large language models (LLMs), promises to revolutionize how we approach HARA, particularly in the light of more complex operational design domains, faster release cycles, and the shortage of skilled safety personnel. In this article, we present LASAR (LLM-Augmented Situation Space Analysis for Risk Assessment), an approach that demonstrates the potential of LLM-augmented tools to improve HARA for automotive systems. LASAR integrates human-AI collaboration to enhance HARA, ensuring efficiency without compromising on reliability.<\/p>\n

The Challenge of HARA in Automotive Safety Engineering<\/h2>\n

HARA is a cornerstone of safety engineering, essential for identifying and mitigating risks in automotive systems. The process demands a deep understanding of the system, its environment, and compliance with stringent standards like ISO 26262. Historically, this has required the concerted efforts of experienced safety engineers, making it both time-consuming and resource-intensive. Conducting a HARA in the automotive domain involves defining the assessment’s scope, including identifying the vehicle systems, components, or functionalities that will be analyzed. Next, potential hazards are identified. Once hazards are identified, the risks associated with each are evaluated by assessing their exposure value (likelihood of occurrence), the potential severity of harm, and the controllability of the hazardous event. The next step is to formulate safety goals. These are functional objectives or top-level safety requirements, and they must be suitable for preventing or mitigating hazardous events.<\/p>\n

The Intersection of LLM and Automotive Safety Engineering<\/h2>\n

Generative AI, particularly LLMs like ChatGPT, offers new possibilities for automating complex and creative tasks. Large language models are designed to create context-specific data. LLMs can process multiple inputs, while effectively understanding and integrating various context information. This allows us to combine the inputs into a specific scenario in fine detail. The complexity involved in understanding and synthesizing diverse inputs is a highly creative task that requires significant expertise. Conducting a HARA requires creative evaluation of various scenarios and the risks associated with them. Thus, LLMs have the ability to assist in this task. \u00a0Although language models have not yet reached the reliability required to independently oversee safety-critical domains such as automotive engineering, this study presents a promising hybrid approach. By integrating human-AI collaboration, we aim to significantly improve the Hazard Analysis and Risk Assessment (HARA) process.<\/p>\n

\n

More About LLMs and Gen AI<\/h3>\n