In response to the strong background noise of rotating machinery bearings and the difficulties in feature extraction and low time-frequency resolution of traditional time-domain and time-frequency analysis methods, this paper proposes a fault diagnosis method that combines adaptive multiple synchronous compression transform (AM-SCT) with twin networks. AM-SCT improves the clarity and noise resistance of time-frequency representation through multi-scale decomposition, synchronous compression, and local optimization strategies; Twin networks utilize a dual branch structure to learn sample similarity and enhance their ability to identify multiple types of faults. The proposed method was validated by using the bearing dataset from Case Western Reserve University. Through comparative experimental analysis, the proposed method achieved an average accuracy of 99.1 % in strong noise environments, verifying its feasibility and effectiveness in bearing fault diagnosis.