To address the reduced accuracy of structural condition identification caused by the coupled interference of temperature and mechanical loading in pile-supported wharves,an automatic thermo-mechanical separation approach is developed.The method combines complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and K-means clustering,and constructs a multi-feature vector integrating spectral kurtosis,modal energy,and sample entropy to automatically classify and reconstruct intrinsic mode functions (IMFs).Field data from a representative pile foundation at the Ningbo-Zhoushan Port are used for validation.The results indicate that the separated temperature component correlates highly with the measured temperature (PCC＞0.89),and the mechanical component achieves a signal-to-noise ratio improvement of over 12 dB.Compared with the EEMD-K-means and VMD-K-means models,the proposed method reduces the root mean square error by approximately 12% and improves the accuracy of temperature-trend extraction by 4%.Without relying on reference fibers or finite element simulations,the method stably separates thermal and mechanical responses and accurately identifies berthing events under complex marine conditions.The findings clarify the thermo-mechanical coupling behavior of pile foundations and provide an efficient and scalable signal-processing technique for health monitoring of port structures.