Prediction of Hydrogen Storage Vessel Explosion with Blast Wall using Machine Learning

We present a prediction of hydrogen storage vessel explosion using machine learning. A neural network model is trained to learn the waveform of a hydrogen explosion with a blast wall and used to predict the waveform with arbitrary blast wall position. The initial training revealed one of the features, blast wall distance, greatly affects the blast waveform. To emphasize the effect of that feature, feature multiplication is used instead of normalization and this enabled the training to learn the waveform correctly. The trained model can predict the blast waveform with an arbitrarily located blast wall. The result can be used in structural analysis and this will help to construct the blast wall in the hydrogen refuelling station. This research uses the CFD (Computational Fluid Dynamics) simulation by Pukyong University, South Korea, as training data.