Leak localization using distributed sensors and machine learning for hydrogen releases from a fuel cell vehicle in a parking garage

Abstract

The detection and localization of hydrogen releases are of vital importance in large confined spaces to enable safe commercial penetration of hydrogen energy into society. In the present study, a hydrogen leak localization system was developed based on measured concentration data and a machine learning system. A scale parking garage model was used to experimentally model fuel cell vehicle leaks in a confined space with helium used as a surrogate of hydrogen for safety reasons. Twelve gas sensors were placed along the ceiling to measure the helium concentrations. The leak position was changed in each test to gather a range of concentration data. Then, the measured helium concentrations were used as input data to train the machine learning models. Two machine learning algorithms were used to locate the leak position, an artificial neural network (ANN) and the K-DTW algorithm. The helium concentrations recorded by the twelve sensors at a specific time point were used as the input data for the ANN model, while time-series concentration data was used as the input for the K-DTW algorithm. The results show that the ANN model can predict leak locations that were not included in the training datasets and that the K-DTW algorithm can classify the unknown leak location to the given location label. The ANN prediction accuracy was 78.4%. The K-DTW algorithm identified the leak location with 87.5% accuracy. The localization technology developed in the present work can provide safe monitoring of large parking garages and the model accuracy can be improved by feeding more training data to the machine learning models.