Abstract: Aiming at the problem that it is difficult for the rolling bearing life prediction method to accurately identify the first predicting time (FPT) and extract the deep features of the time series, this paper proposes a rolling bearing life prediction method combining high-precision FPT points and the multi-module U-Net-BiLSTM network. After wavelet noise reduction, all frequency components in the power spectrum of the original signals at each moment are accumulated and summed, and the Euclidean distance criterion and the 3σ principle are combined to identify high-precision FPT points; the residual blocks, pooling layers and normalization layers are respectively introduced into the encoder and the decoder to achieve multi-scale feature fusion, thereby upgrading the traditional U-Net network and effectively improving the process ability and prediction speed for time series signals of the model. The experimental results show that this method has higher prediction accuracy and faster prediction speed than the existing three deep learning comparison method