Data-driven diagnosis method of high-pressure hydrogen leakage based on actual driving conditions and probabilistic neutral network

Fuel cell vehicles rely on high-pressure hydrogen storage to ensure sufficient driving range. However, the potential for high-pressure hydrogen leakage in the supply system poses a significant safety concern, highlighting the importance of timely leakage monitoring. Traditional monitoring methods often use hydrogen concentration sensors, limited to stationary vehicles in confined spaces and reliant on sensor layout settings. Additionally, other data-driven methods utilizing internal fuel cell parameters face challenges in accurately obtaining data, especially for humid, complex, and time-varying fuel cells. Therefore, using easily obtainable parameters like hydrogen pressure inside the storage tank for fault diagnosis, is more practical. Nevertheless, the variation of gas pressure inside the hydrogen storage tank is not only related to hydrogen leakage but also influenced by the normal running of the vehicle and environmental temperature factors particularly in winter and summer. In this study, a comprehensive heat transfer model for hydrogen storage tank is established to simulate such summer and winter driving scenarios to obtain original data. Subsequently, linear discriminant analysis and probabilistic neural network are adopted for preprocessing data and pattern recognition respectively. The results show that when vehicles are driving at a speed of 10 m/s and a temperature of 293.15 K, the diagnostic accuracy can reach 97% for hydrogen leakage faults with a mass flow rate of more than 0.04 g/s. Compared to diagnostic methods that do not consider temperature factors, the accuracy of diagnostic methods that consider temperature factors has also increased by 23%.