Addressing the challenge of diagnosing temperature drift faults in hydro-generator brush temperature sensors caused by long-term operation, a fault diagnosis method based on DBSCAN density clustering and multi-sensor collaborative verification is proposed.By comparing the misjudgment rates, computational efficiency, algorithm parameters, and robustness of K-means, hierarchical clustering, isolation forest, and DBSCAN algorithms in a three-dimensional feature space (rotational speed, current density, temperature), the applicability of DBSCAN for processing 3D data was verified (misjudgment rate below 18.7%, computational efficiency reaching 0.41 s/thousand points).Wavelet transform was applied to denoise historical operational data. Combined with a dynamic threshold early-warning mechanism and collaborative verification from adjacent sensors, this achieved precise diagnosis of gradual sensor offset faults.Under simulated conditions of 0.5°C/h temperature drift and ±10°C noise interference, the mean difference between fault clusters reached 12.23°C, exceeding the threshold to trigger warnings. Diagnostic accuracy was confirmed through shutdown calibration.This method overcomes the insufficient sensitivity of traditional approaches to gradual faults. Field tests demonstrate its suitability for real-time monitoring under complex hydro-generator operating conditions, providing a potentially viable technical solution for intelligent condition monitoring of industrial equipment.