Accurate prediction of the remaining useful life（RUL） of the roller with double toothed roll crusher is an important basis for maintenance personnel to make scientific maintenance decision. In engineering practice, some state information is often observed. In order to solve the problem of the remaining useful life of the roller sleeve is difficult to predict accurately in the non-linear and non-stationary state, a new method of remaining useful life prediction based on wavelet transform and SVR fusion is proposed. Firstly, the correlation analysis is used to select the characteristic value of the vibration signal of the roller. Secondly, wavelet technology is used to eliminate the noise of the collected vibration signals. Then, the relationship between the residual life and the characteristic value is established, and the nonlinear dynamic change of the roll sleeve is described. Finally through the SVR model to predict the remaining useful life of the crusher roller. Prediction results show that the method can effectively solve the remaining useful life prediction of partially observable state, thereby reducing the maintenance cost, and has strong engineering applicability and popularization value.