Beyond ALOHA- quickly predict accidental release of toxic chemicals using machine learning

Abstract

Accidental releases of toxic chemicals pose a significant threat to both human safety and the environment. Simulating and preventing chemical leaks is a critical aspect of environmental and process safety. This study utilizes the Areal Location of Hazardous Atmospheres (ALOHA) model to simulate chlorine gas leakage, focusing on four key parameters: wind speed, ambient temperature, gas pressure, and reactor hole diameter. In this study, we carried out over two thousand simulations by changing four specific parameters. The common ALOHA software is well-known for consequence modelling. Typically, the initial run takes around 30–40 min, while subsequent runs can be completed in about 2 min. This delay can be quite a challenge when used in industrial settings. To address this issue, we looked into using machine learning (ML) as a better alternative to traditional consequence modelling methods. Our goal was to cut down the estimation time to just 15 s for each simulation especially initial run. We trained the ML model using 80% of the simulation data. The leftover 20% was used for testing. The results, shown in a series of performance curves, indicate that our model has been effectively trained. It shows high accuracy in predicting chlorine levels across different conditions. In summary, these findings imply that ML models hold significant potential as a more efficient means for conducting in-depth consequence modelling in industrial environments. This study opens future research opportunities in replacing ALOHA with machine learning based models for very quick prediction of accidental release of toxic chemicals.