Abstract: The traditional fixed frequency range detection method is difficult to effectively capture fault characteristics due to redundant frequency band division, which weakens the sensitivity, speed, and positioning accuracy of detection. In response to the above issues, this study proposes an intelligent detection method for interturn short circuit faults in synchronous generator rotor windings based on wavelet packet entropy.Firstly, wavelet packet decomposition technology is used to decompose the interturn short circuit signal into different frequency bands,and morphological filtering is used to effectively reduce noise and improve signal quality. Then, the entropy theory is introduced to calculate the sample entropy, multiscale entropy, and wavelet packet energy spectrum relative entropy of each frequency band signal, screen out the frequency bands with significant entropy changes, effectively eliminate redundant frequency bands, accurately extract fault features, and significantly enhance the sensitivity of detection. Combining convolutional neural networks to classify extracted features and achieve intelligent fault detection.Experimental verification shows that compared to traditional methods, this method exhibits significant advantages in improving fault detection sensitivity, accelerating detection processes, and ensuring accurate fault localization. It provides an efficient and reliable solution for synchronous generator rotor winding fault diagnosis, effectively overcoming the limitations of traditional methods.

    Key words: wavelet packet decomposition; entropy calculation; synchronous generator; interturn short circuit; fault detection