<p>Continuous monitoring and preventive maintenance based on the prognostics and health management (PHM) technology for engines can ensure the flight safety of aircrafts. Wherein, the remaining useful life (RUL) serves as a critical metric to assess the equipment reliability and determine the maintenance schedule. The increase in data volume makes the data-driven model have richer sample sets, and is helpful to enhance the data fitting ability and prediction accuracy. However, the following issues still exist in the data-driven methods of remaining useful life prediction: (i) Single-dimensional and single-scale approaches are insufficient for comprehensive feature extraction. (ii) Extracting features solely from the time-domain overlooks certain degradation information. Hence, a RUL prediction method, which integrates the time–frequency and multi-scale features along the sequence and feature dimensions, respectively, is proposed. The frequency-domain multi-layer perceptrons for time series (FreTS) is utilized to concurrently extract the frequency-domain features along the sequence and feature dimensions. The atrous convolutions with different rates and the selective fusion strategy are used to extract the multi-scale shallow local feature information, and the encoder is adopted to further capture the deep global degradation information within the entire time step. Experiments on the RUL prediction of aircraft engines are conducted based on the C-MAPSS dataset, and the prediction accuracy is studied. The results indicate that the proposed model effectively extracts and fuses the multi-dimensional, multi-scale, and time–frequency features, and has the higher accuracy in the RUL prediction compared to all the latest methods.</p>