Sound Pattern Based Transformer Fault Diagnosis Algorithm And Application Research

As an important member of power equipment,online monitoring and fault diagnosis of power transformers has become an important issue that needs to be addressed.Acoustic vibration analysis,which has no direct electrical connection with the internal components of the equipment and high sensitivity,has certain advantages for diagnosing mechanical type faults inside transformers and is now widely used in the field of online monitoring of transformers.Meanwhile,with the development in the field of artificial intelligence,the application of advanced algorithms in deep learning to transformer fault diagnosis has unparalleled advantages over traditional methods for mining the high-dimensional attributes of transformer vibration data and related features,and deep learning algorithms can better solve the problem of difficult access to transformer fault data in terms of data generation.In this research context,the research work of this paper is summarized as follows:After analyzing deep learning algorithms and generative adversarial network theory in the field of artificial intelligence,the advanced generative adversarial network-derived algorithms are applied to transformer anomaly diagnosis after improvement,and the diagnostic capabilities of the three derived algorithms are analyzed for DC bias anomaly signals,high harmonic anomaly signals,and noise anomaly signals,respectively,after illustrating the experimental evaluation indexes.The STM32 microcontroller-based transformer intelligent acoustic diagnostic device was developed to realize edge computing of transformer acoustic vibration signals,and with a backend server,a system was formed that can monitor transformer operation and perform fault diagnosis online in real time.The developed intelligent acoustic diagnostic device was used to collect transformer vibration signals in a 110kV substation of a municipality.The porting of the generative adversarial network-derived algorithm to the developed device was realized.The trained generative adversarial network model is ported to the developed diagnostic device,and the signal is diagnosed in real time while data is acquired,and the real time curve and the diagnostic curve are displayed simultaneously through the Dwin screen.The experiment proves that the device after transplanting the algorithm has the ability to detect abnormal acoustic vibration signals.