To study the robustness of complex networks while encountering random failure or deliberate attack, the cascading failure model based on probability propagation is constructed by considering the characteristics of repeated failures and recovery of nodes. In particular, we consider that the failure probability of node decreases with the increase of number of failures and the failure probability function is constructed accordingly. Then, two kinds of node recovery strategies (i.e., probability R and stage T) are proposed and implemented on ER random network, WS small-world network, nearest-neighbor coupled network and BA scale-free network, respectively, to test the network robustness of cascading failure model. Finally, simulation experiments are carried out and the parameters changes are considered to reveal their effects on robustness during cascading failure process. Additionally, performance tradeoff between edge robustness and node robustness is analyzed as well. The simulation results show that cascading failure scale of four types of network decreases with the increase of recovery rate. Under the stage recovery strategy, the network robustness index of four types of network shows a mutation phenomenon when the parameter T increases to a threshold.