Transformer Fault Diagnosis Using Deep Boltzmann Machine and Multiclass Relevance Vector Machines

Transformers are crucial to electric power systems, yet it can be challenging to accurately diagnose their problems. The research presents a novel method for transformer defect diagnostics that combines ensemble machine learning (EML) with an Internet of Things (IoT) based monitoring model. First, the data measurement subsystem measures the transformer vibration signals, and data reception transmits those signals to remote server. At that time, an EML made up of Multiclass Relevance Vector Machines (MRVMs) and Deep Boltzmann Machines (DBM) functions are suggested. A unique combination strategy and conventional monitoring techniques like vibro-acoustic frequency and short-circuit impedance are used to guarantee the EML's efficiency. The suggested EML readily addresses the insufficient feature extraction information from separate deep learning methods, and the multiclass RVM classification technique clearly outperforms other widely used classifiers. The less power, long-distance watching, and high reliable utilization based on IoT is advantageous for long-term collection and transmission of signals. The study of the results makes it evident that the suggested DBM-MRVM outperformed other models, achieving a greater accuracy of 99.38%.