Fuel cell technology is the fourth generation technology after hydropower, thermal power and nuclear power. Once the solid oxide fuel cell system fails, if it can't be found in time, the initial glitch may slowly evolve and spread to the subsequent components. Therefore, fault diagnosis is a prerequisite to ensure its stability. In order to diagnose fuel leakage fault of solid oxide fuel cell system, a decision tree is proposed to diagnose the fuel leakage fault of solid oxide fuel cell. Compared with other machine learning methods, it can be clearly observed that the decision tree method can effectively identify the severity of faults. This method can be extended to the fault diagnosis of air leakage.