Detection and identification of DC corona discharges by using advanced techniques

Abstract

Corona discharges are an undesirable electrical phenomenon frequently seen in high-voltage systems. These discharges occur in high-voltage overhead lines, hardware, or sharp points on the metal parts of high-voltage devices. Corona discharges cause power loss, television and radio signal interference, ozone formation, aging of insulation materials, acoustic noise, and light emissions. Detecting and preventing corona discharges is significant for a sustainable and reliable power system and power delivery. The symptoms of corona discharge (electrical pulses, ultraviolet and visible lights, electromagnetic signals, acoustic sounds, and chemical reactions) should be well analyzed to diagnose corona discharges. Many sensors and methods have been developed in the literature to detect corona discharges using these symptoms. However, these symptoms may differ depending on the corona discharge modes. Corona discharges manifest in several modes. These modes vary depending on the type of voltage applied, the voltage level, the electrode shape, the gap spacing, and ambient conditions such as temperature, humidity, and pressure. Each mode has its unique characteristics. Fundamental corona modes under positive DC voltage are burst, streamer, glow, and pre-breakdown streamer. A thin light layer is formed on the electrode surface in the burst corona discharge. In the streamer corona, a streamer forms a channel from the anode electrode to the ground electrode. In this mode, electrical pulses and high amounts of acoustic noise occur. In the positive DC glow coronas, also known as Hermstein's Glow, a light layer spreads over the electrode surface rather than a channel towards the ground electrode. In this corona mode, no electrical pulses or acoustic noise are produced. A steady current flows between the electrodes. Before the breakdown, a large amplitude of electrical pulses and acoustic noises occur. If these streamers reach the ground electrode, a breakdown occurs. The primary corona forms under negative DC voltage are Trichel, negative pulseless glow, and pre-breakdown streamers. Negative polarity electrical pulses and acoustic noises are produced in Trichel corona discharge. In negative pulseless glow corona discharge, electrical pulses and acoustic noise are not produced. There are electrical impulses and acoustic noises in the pre-breakdown streamer before the gap breakdown. When these streamers reach the ground electrode, a breakdown occurs. Under AC voltage, both negative and positive corona discharges occur. However, as mentioned before, the formation of the forms of this corona depends on the above parameters. In this Ph.D. thesis, corona discharge modes under DC voltage were examined with advanced methods, and their detection with various sensors was compared. This thesis consists of four main studies. In the first study, the light patterns created by corona discharges were examined under positive, negative DC and AC voltages. Moreover, the applicability of sensors used to detect corona discharges was compared. The key purpose of this study is to create a dataset of the light patterns of corona discharges and to analyze the intensity of the corona discharge using these patterns. In the second study, advanced image classification models were used to classify positive and negative DC corona discharge modes. The light forms of these corona modes were recorded via a digital camera. Photos of corona modes found in the literature were also used to increase the diversity of the created data set. The location of the corona discharge in the recorded photos was determined using YOLO (You Only Look Once) version 8, which is an advanced deep-learning algorithm. Once detected, convolutional neural network-based algorithms determined which mode the corona belonged to. In the third study, the characteristics of electrical pulses produced by positive and negative corona discharges were analyzed. These electrical pulses at different voltage levels were recorded via a shunt resistor and high-frequency current transformer. The appropriate resistance value and resistor type were determined for the shunt resistor. Using a BNC terminator as a shunt resistor was proposed, and its suitability was tested. By extracting the properties of the electrical corona pulses, the best features describing the positive and negative pulses were determined using advanced machine learning algorithms. In the last study, a wavelet transform-based approach was proposed to automatically detect positive and negative corona pulses. Scalogram images were obtained from these pulses using continuous wavelet transform algorithms. Factors affecting these images, such as sampling interval, data recording time, data shift, and external environment noise, were examined. These scalogram images were tried to be classified using convolutional neural networks, and a framework was created to increase the generalization capacity of the algorithm. Under DC voltage, the corona discharge has modes with different characteristics. Therefore, high-frequency current transformers and acoustic sensors cannot detect positive DC glow and negative DC glow modes. Digital cameras and corona cameras can detect all modes of the corona. However, digital cameras are cheaper than corona cameras. Therefore, digital cameras are one step ahead. However, integrating data from different sensors will yield higher accuracy for diagnosing DC corona discharges. Thus, relying on a single sensor for corona diagnosis may give misleading information about the presence or severity of corona discharges.