To solve the problem of accurate forecasting of remaining-useful-life, an on-line fault prediction method based
on hiding semi Markov model(HsMM) is proposed in this paper. The fuzzy ??-means(FCM) clustering algorithm is used to
cluster the input-output data of system. The variables with continuous value are transformed to discrete value variables. Then,
the structure of HsMM is redesigned, and the modeling process of HsMM is simplified. The fault detection is accomplished
by the HsMM model which is established with data of normal system. At the same time, the remaining-useful-life can be
forecasted by the time duration of the hidden states. In addition, the approximate time when the fault has established is
estimated. Therefore, the fault prediction is also accomplished. By using the continuous stirred tank reactor(CSTR) as the
simulation model, the proposed method can forecast the remaining-useful-life of failure system accurately when the data of
system has noise.