Adaptive Diagnosis for Transformer With Unknown Faults Based on Antenna-Augmented RFID Sensor and Deep Learning

In engineering practice, transformers always have one or more new mechanical fault types that have not been found yet, which would decrease the diagnosis accuracy. This work introduces an adaptive fault diagnosis approach for transformer mechanical failure in incubation period with unknown fault types. An antenna-augmented radio frequency identification (RFID) sensor is used to acquire vibration signal and position information. The faults’ signals of transformer are generally featured as nonlinear; moreover, it always contains unknown faults, making it hard to extract discriminative features from the obtained signals. Therefore, an adaptive fault diagnosis method consisting of sparse stacked denoising autoencoder (SSDA) and unknown fault cluster is proposed. The SSDA is adopted to extract robust feature and quantum particle swarm optimization (QPSO) is employed to find the optimal the layer node structure (LNS) and learning rate (LR) of the SSDA model. The SSDA model along with a full-connected layer (FCL) can divide the data into known faults and unknown faults. The knowns faults are classified by support vector machine (SVM), and the unknown faults are divided into different categories by using the fault cluster. The experiments validate that the augmented RFID sensor has reliable communication performance within the distance of 17.5 m. Moreover, the proposed method achieves prominent results for transformer fault diagnosis with unknown faults.