In this paper, the adaptive decentralized fault-tolerant control problem is discussed for a class of uncertain nonlinear large-scale systems. A novel nonlinear-mapping technique is proposed to deal with the constrained output variables, so that the system outputs can start from within the initial output constrained space; the bounded reference signals are used to replace the independent argument of the unknown interconnected functions, then the fuzzy approximators are used to deal with these unknown functions, and the adaptive technique is combined to deal with the errors of replacement and approximation, so that the globally stability of the closed-loop system is achieved; all of the four types of actuator faults can be effectively compensated for on line. Simulation results are provided to show the effectiveness of the control approach.