Aiming at the problem of the nonlinearity, nonstationarity and background environment complexity of mining equipment under heavy load conditions, it is difficult to extract pure fault signal characteristics, research is proposed on remaining life prediction of shaft based on dualtree complex wavelet and time series. Firstly, the doubletree complex wavelet transform (DT-CWT) was carried out to obtain the transform coefficients, and the modulus coefficients were obtained, and the modulus coefficients with large regularity were extracted and subjected to nonlinear time series processing. Then, the noise signal was added to strengthen the regularity of the noise. Then, the signal source was projected onto different faces to separate the pollution signal source, and the coefficients of the characteristic signal source were extracted, and the double tree complex wavelet reconstruction was carried out in the plural form. Easy ignored fault characteristics of the signal were obtained to get a more complete pure fault signal. Finally, the remaining life prediction of the shaft, then the destructive experimental data of the shaft are compared with the Back Propagation (BP) neural network using the purified characteristic signal, which proves that the method is suitable for the prediction of the life of the crusher shaft and the satisfactory effect is achieved.