A learning unit in learning mode generates a cluster from a cluster analysis of data formed from frequency constituent data and state data, obtained from a sensor unit. An abnormality calculation unit computes, as abnormalities, the minimum values among distances to surfaces of the clusters of the data formed from the frequency constituent data and the state data, obtained when in predictive fault indicator sensing mode. A predictive fault indicator determination unit determines a predictive fault indicator of an X-ray tube by comparing the abnormalities with a predetermined threshold.