In this article, we address the problem of fault reconstruction in delayed systems by introducing a time-shifted sliding mode observer (SMO). While time-varying delays of arbitrary duration are considered in the measured output signal, the actuator fault is parametrized as a weighted sum of known regressor functions with unknown coefficients. The prediction scheme utilizes the variation of constants formula to obtain the present time estimate of the unmeasured state. The fault is also identified at present time by means of the continuous-time Least Squares approaches. Ideal sliding mode can be guaranteed in theory, even in the presence of such adverse delays, since there is no chattering in the output estimation error of the SMO. An application to petroleum engineering with numerical simulations is presented to show the effectiveness of the proposed method.