The conventional method for reliability evaluation struggles to precisely assess and forecast the reliability of vast, complex engineering systems due to the scarcity of experimental data collected from a limited number of trials in the engineering field. Consequently, the general method is not suitable for such systems. This paper proposes an improved approach, introducing the return-to-zero weight coefficient into the traditional fault tree analysis method. This allows for the characterization of recurrence probability and influence degree of particular fault types, as well as the prediction of system reliability for future operations. In this paper, a wind turbine system is used as an example, and the method is simply verified by using the example analysis. The reliability prediction of few experiments system can be enhanced by this method, and its precision will grow with the system's operation, based on the principle of the method.