A Master of Science thesis in Electrical Engineering by Mustafa Harbaji entitled, "Classification of Common Partial Discharge Types in Oil-Paper Insulation Using Acoustic Signals," submitted in February 2014. Thesis advisor is Dr. Ayman Hassan El-Hag and thesis co-advisor Dr. Khaled Bashir Shaban. Available are both soft and hard copies of the thesis.
Power transformers' age is highly related to the strength status of its insulation system. Oil-paper insulation system is the main type of insulation for most of the power and distribution transformers in the network. However, such insulation system is subjected to electrical, mechanical and chemical stresses that deteriorate its strength. One of the main sources of such stresses is partial discharge (PD) phenomena. PD within the insulation system can happen at any point from different defect sources like sharp points or voids in the insulation. Knowing the source or type of PD activities provides vital information for maintenance scheduling because PD types have different levels of severity. Since PD activities consume only few mA, they cannot be detected by the typical protective relays. However, PD activities emit energy in different forms that make it possible to detect them using different methods. In practice for power transformers, dissolved gas in oil analysis (DGA), ultra high frequency (UHF) pulses, and acoustic emission (AE) methods are the most common methods used to detect PD. Among these methods, AE has several advantages such as being the most cost effective and easiest to install while the transformer is energized. However, measuring AE in the field can be affected by several measurement challenges. The main contribution of this research is to identify the source of PD under different AE measurement conditions. Four common types of PDs are considered for the classification problem; surface discharge, PD from a sharp point to ground plane, PD from semi parallel plates, and PD from an air void in the insulation. The collected AE signals are processed using pattern recognition techniques to identify their corresponding PD types. For feature extraction and reduction, principle component analysis (PCA) is utilized, whereas k-nearest-neighborhood (KNN) is applied for classification. The measurement conditions include having aged insulation material (oil/paper), a tank size of 1x1x0.5 m dimensions, and high surrounding noise level. In addition, the influence of other practical conditions on the recognition rate is studied including PD location, sensor location, oil temperature, and having a barrier in the line-of-sight between the PD source and the AE sensor. A recognition rate of 94% is achieved while classifying the different PD types measured at the same conditions. In addition, it has been found that PD source location, oil temperature, and barrier insertion have a significant impact on the recognition rate. However, by including AE samples at different conditions in the training process, a recognition rate of around 90% for all cases is achieved.