Leakage source localisation employing 3D-CFD simulations and gated recurrent units

Abstract

The risk of explosion due to gas leakages and its human, environmental and economic losses in accident scenarios constitute serious safety hazards in industries. Fast location of leak sources enables quick corrective maintenance, avoiding the most hazardous cases. In the present study, gated recurrent units were developed to identify CH4 leaks in a chemical process module. The training and test databases were obtained through 3D-CFD simulations for four leaks and a non-leakage scenario. The inputs utilised were the concentration profiles at eleven sensors for four leak sources, four wind speeds, and eight wind directions using different temporal lengths. Additionally, noise was added to the database to assess the performance in more realistic cases. The findings indicated better performance with higher values of input time-steps, and accuracy over 93.9% for unseen data, indicating good generalisation of the models and their potential of predicting the leaks applying easily acquired inputs.