Physics-guided adaptive asymmetric wavelet and compact time-frequency representation for single-source cross-condition fault diagnosis of rotating machinery

Cross-condition fault diagnosis from a single labeled operating condition is challenging because speed and load variations alter fault frequencies, resonance responses, and vibration distributions, while scarce fault samples encourage condition-dependent representations. This paper proposes a physics-guided framework that integrates a finite-window adaptive asymmetric quadrature wavelet, complementary angular-envelope and order-spectrum representations, and a fault-order-guided training objective. The wavelet front end learns its bilateral scale, decay-to-rise asymmetry, and carrier band. The two representation streams propagate complementary task information to the shared wavelet parameters, promoting stronger temporal localization, frequency concentration, and reduced joint spread in the learned filter bank. The physics-informed objective further enhances demodulated components near nominal candidate fault orders and promotes a concentrated order-spectrum structure. The three mechanisms are integrated into a single-source cross-condition diagnostic framework. Experiments on five bearing and gearbox datasets yield an equally weighted five-dataset mean accuracy of 91.16%. Paired ablations show that the complete dual-stream model exceeds the order-spectrum-only and angular-envelope-only variants by 2.844 and 15.220 percentage points, respectively. Representation analyses show increased candidate-order concentration and spectral compactness, while all 80 dataset-level channel observations of the complete model exhibit reduced joint spread, with a mean reduction of 5.91%. These results establish a physically structured and computationally efficient approach to single-source cross-condition diagnosis. The implementation is available at: https://github.com/EVANGELION-NO1/Physics-guided-Adaptive-Asymmetric-Wavelet-and-Time-frequency-Resolution-Balanced-Representation .