This paper deals with the problem of robust longitudinal speed estimation for rail transportation system under wheel-adhesion related faults that corrupt speed measurements and lead to poor closed-loop behavior. Two observers are used: the first one is a sliding mode filtering differentiator which gives an estimate of the adhesion force in finite-time. The second one is a continuous-discrete high-gain observer to estimate the longitudinal train speed from aperiodic measurements of the position beacons. It is shown that the speed estimation error is practically stable in spite of variable adhesion, measurement noise and aperiodically sampled position data. Simulations are provided to show the benefits of the proposed approach.