Compressor surge in aero-engines is a primary cause of catastrophic engine failure, making precise early warning essential for flight safety. Prevailing methods commonly encounter a critical bottleneck when balancing accuracy with model efficiency, which restricts their practical engineering applications. To address this limitation, this paper proposes an aero-engine surge early warning method based on the Discrete Wavelet Transform (DWT) and a lightweight Convolutional Neural Network, termed Multiwave Fusion-based Lightweight Convolutional Neural Network (MF-LWCNN). The proposed method first employs DWT to decompose sensor signals into multiple scales, effectively separating and extracting high-frequency detail features and low-frequency trend features. Subsequently, transposed convolution is introduced to establish a scale-consistency learning mechanism. Combined with pointwise convolution, this mechanism achieves channel-wise feature fusion. This design circumvents the “manual dependency” issue inherent in traditional DWT feature selection. Meanwhile, a residual network is incorporated to enhance the model’s feature representation capability. Finally, a depthwise separable convolutional network is employed to construct the classification module for implementing surge early warning. To validate the efficacy, experiments were conducted on the public Case Western Reserve University bearing dataset and an experimental dataset from an aero-engine compressor test rig. The results demonstrate that MF-LWCNN offers significant advantages in both accuracy and lightweight model design, providing theoretical foundation and a technical reference for the engineering application of surge early warning technology.
Zhou et al. (Sat,) studied this question.