Dynamic threshold extraction method for bearing bush temperature based on different operating conditions of hydro-generator unit

The bearing bush temperature of hydro-generator units is a crucial indicator for assessing the safe operating condition of the unit. A warning system using dynamic thresholds and trend prediction is employed to address excessive bearing bush temperature. This paper proposes a method for determining dynamic thresholds for bearing bush temperature based on operating condition classification. We first categorize different load ranges according to the unit's vibration characteristics, using box plots to construct an anomaly data cleaning model for extracting temperature features. Subsequently, combining the original threshold, prediction correction factors, and trend adjustment factors, the final dynamic threshold is calculated comprehensively. Validation via case studies indicates that the dynamic threshold established by the proposed model exhibits a false alarm rate of merely 1.07%, with an average value of 56.59°C. This is considerably lower than the shutdown threshold of 23°C. These findings highlight the model's high reliability and provide operators with a trustworthy reference for monitoring temperature anomalies.