A remaining useful life prediction method based on time–frequency images of the mechanical vibration signals

•A multi-scale CNN prognostic model combined with attention mechanism is proposed.•The quadratic degradation function is used to establish an effective health index.•The multi-scale CNN model is used to extract the deep features of the data.•The attention mechanism is used to catch the more important features out from MSCNN.•The proposed CBAM-MSCNN model achieves accurate estimation of bearing degradation state.