Transformer Fault Diagnosis Using Self-Powered RFID Sensor and Deep Learning Approach

This paper introduces a novel fault diagnosis approach for transformer based on self-powered radio-frequency identification (RFID) sensor and deep learning technique. The exploited RFID sensor tag with functionalities of signal collection, data storage, and wireless transmission employs surrounding electromagnetic field as power source. A customized power management circuit, including ac–dc converter, supercapacitor, and its corresponding charging circuit, is presented to guarantee constant dc power for the sensor tag. The measured vibration signal contains miscellaneous noises and is characterized as nonlinearity and nonstationarity, so it is difficult to extract robust and useful features by using traditional feature extraction approaches. As one of the deep learning techniques, stacked denoising autoencoder (SDA) shows satisfactory performance in learning robust features from complex signal. Hence, in this paper, SDA approach is employed to learn robust and discriminative features from measured signals. The experimental results show that the presented power supply can generate 2.5-V dc voltage, which is the rated operating voltage for the rest of the sensor tag. The developed sensor tag can achieve a reliable communication distance of 17.3 m in the test environment. Furthermore, the SDA approach shows satisfactory performance in learning robust and discriminative features. Experimental results indicate that the presented approach is effective and time-saving in terms of fault diagnosis for transformer winding and core.