A study of thermal hazards assessment from the estimation of physical properties based on neural networks

Abstract

The accidents, caused in chemical industries, can cause a loss of lives and properties and damage to the surrounding environment as well. Therefore, when designing the new chemical process, it is vital to assess the safety of the overall processes. Especially, the risk assessment for the chemical reactor should be evaluated more attentively than other equipments because it is one of the most hazardous equipments. It seems to be useful to predict the amount of heat release of the reactions and the adiabatic temperature rise by reactions as the preliminary screening procedures of reactive chemical hazard evaluation. These data can be measured experimentally or calculated theoretically. Theoretical approaches are effective because they can reduce the cost and the risk caused by the experiments and besides it is not easy to get the experimental data, especially for the new materials or the high molecular substances. In this study, we developed the new method for the prediction of physical properties. This method can be applied to predict 16 physical properties for the chemicals that consist of C, H, N, O, and S. In order to improve the accuracy and applicability of the predictive model, we extended the database in large numbers, modified the existing group contribution methods and then established new method for predicting the physical properties using neural networks. Neural network-based approach could develop nonlinear structure property correlations more accurately and easily in comparison with other conventional approaches. The results from the new estimation method were found to be more reliable, accurate and applicable. Besides, this model could distinguish isomers.