Case Studies on Predictability in University Chemistry Experiment Accidents

Abstract

In this study, we investigated prediction of fire accidents of university’s chemical experiments based on properties or data of chemicals used. By referencing hazardous compounds and their properties in the Fire Service Act in Japan, potentially dangerous operations were picked up from the textbook of General Chemistry Laboratory for undergraduate students in Department of Chemistry, Faculty of Science, Tokyo University of Science. Moreover, previous examples of fire accidents associated with such hazardous compounds and experimental operations were also searched from some databases. Comparing both facts, we concluded that most of fire accidents are predictable, and some difficult cases (i) required common sense for chemical experiments (implicit knowledge with experience) and (ii) are state-dependent properties of the hazardous compounds, e.g. very reactive nanoparticles. The results will also suggest possibility of AI-aided prediction of fire accidents in the future, the range of data required to be learnt, and remaining technical problems.