This paper investigates the problems of fault detection and diagnosis in a class of discrete-time switched systems with unknown state delay against component faults. The component fault, also called the multiplicative fault or loss of actuator effectiveness, is a common type of operation fault for switched systems, which has been paid more attention. In this paper, the learning observers (LOs) technology are proposed to reconstruct both the system state and component faults. The proposed LOs technique is also extended to the system with unknown time delay and external disturbance. The sufficient conditions for the convergence of the output of LOs and the uniformly ultimately boundedness of the reconstructed errors are proved by the switched Lyapunov function and H ∞ technique. The proposed fault reconstruction strategy has certain tolerance for fault types, which is effective for both slow and fast time-varying faults by adjusting the iteration step size. In the simulation parts, three kinds of fault signals are taken into consideration to verify the effectiveness and robustness of the proposed method.