A HAZOP of dust explosion testing and explosibility modelling using artificial neural networks

This work presents artificial neural network (ANN) models to determine explosion severity parameters (e.g., maximum explosion pressure and maximum rate of pressure rise) of given dust samples. ANN-based models for explosibility parameters are presented for carbon black, zinc, urea, and oat grain flour dust samples based on data generated in a 20-L explosion chamber. The optimal hyper-parameters of the models have been explored using the Broyden-Fletcher-Goldfarb-Shanno, stochastic gradient descent, and Adam solvers. A hazard and operability study has also been conducted for each of the following to diagnose issues at different stages in developing the ANN-based dust explosibility models: dust testing, model selection, parameter learning, and evaluation. Using different guidewords, the deviation of numerous factors from their design intent, causes, consequences, and specific safeguards have been provided for enabling optimal performance. This can be helpful in understanding, evaluating, and analyzing dust explosions for safer operation of industrial activities handling combustible dusts.