In this paper, we propose a fault detection scheme of examining if a human operator controlling a switched system stays in normal condition. Our attention focuses on an expert who manually operates a system with on-off switches. To keep the system in an acceptable state, we expect that the expert flips switches almost always with impeccable timing when the expert is in good condition. Therefore, dispersion of the timing of flipping on-off switches is a reliable indicator of the condition of the expert. The proposed fault detection scheme is based on on-line statistical hypotheses testing between the dispersion being significantly wider than usual and not so.