Decision-level fusion detection method of hydrogen leakage in hydrogen supply system of fuel cell truck

Abstract

The hydrogen supply system is the high-pressure section of fuel cell system, consisting of many valves, pipelines and joints. A hydrogen leak may occur due to the reduced sealing performance or crash accidents. Therefore, a quick and accurate hydrogen leakage detection method is essential to support hydrogen safety. The traditional multisensor-based hydrogen leakage detection method directly classifies the hydrogen leakage fault according to the hydrogen concentration. However, using hydrogen concentrations at local points to describe leak fault severity has significant limitations. A novel hydrogen leakage detection method based on Stacking ensemble machine learning is proposed to classify hydrogen leak rate. The hydrogen leakage simulations are performed to form the database containing a series of hydrogen leakage scenarios. The sensor placement is optimized and the features extracted from hydrogen concentration time series are selected by applying ReliefF algorithm. The model-based hydrogen leakage detection method is analyzed and fused with proposed hydrogen leakage detection method based on Stacking ensemble machine learning by applying decision-level fusion algorithm. Compared with traditional multisensor-based hydrogen leakage detection method, the detection time of proposed multisensor-based hydrogen leakage detection method is shorter under nearly 80% of all scenarios. The classification accuracy of the proposed hydrogen leakage detection method based on Stacking ensemble machine learning and model-based hydrogen leakage detection method is 92.1% and 93.5%, respectively. The probability of class is increased and classification accuracy can be improved through D-S evidence theory.