(596g) Using Large Language Models to Improve Consistency in Process Hazard Analysis

Process Hazard Analysis (PHA) is a critical method for systematically identifying and mitigating risks in chemical engineering and related industries. However, traditional PHA methods suffer from inherent subjectivity in assigning risk ratings, leading to inconsistencies that result either in unnecessary expenditures or inadequate risk mitigation. Specifically, assigning higher-than-actual risk leads to excessive safeguard recommendations and increased costs, while lower-than-actual risk estimations result in significant carried risk exposure. Literature highlights various attempts to address this challenge, including standardizing severity and frequency criteria (Busby et al., 2016), and adopting fuzzy logic and multi-criteria decision-making frameworks (Marhavilas et al., 2020).

In this research, we introduce a novel approach utilizing Large Language Models (LLMs) embeddings to objectively identify discrepancies and outliers in PHA scenario assessments. Our method systematically flags inconsistencies in scenario risk ratings by embedding textual scenario descriptions into vector spaces, allowing for quantitative comparisons. To validate our approach, we present results using two distinct datasets: a publicly available synthetic benchmark generated via detailed process simulations, and a proprietary real-world dataset from industry. Our preliminary results demonstrate substantial improvements in consistency, enabling clearer identification of genuine high-risk scenarios and more efficient allocation of safety resources.

The implications of this work are significant, providing chemical engineers and process safety practitioners with a robust, scalable method to enhance consistency in risk evaluations, ultimately advancing the state of process safety management.

References:

• Busby, J.S., et al. (2016). Guidelines to improve consistency of PHA consequence severity rankings. Global Congress on Process Safety (GCPS).
• Marhavilas PK, Filippidis M, Koulinas GK, Koulouriotis DE. A HAZOP with MCDM Based Risk-Assessment Approach: Focusing on the Deviations with Economic/Health/Environmental Impacts in a Process Industry. Sustainability. 2020; 12(3):993.https://doi.org/10.3390/su12030993.