Partial discharge pattern classification of angled point-oil-pressboard degradation A. Abubakar Mas’ud, B. G. Stewart, S. G. McMeekin and A Nesbitt School of Engineering and Built Environment, Glasgow Caledonian University, G4 0BA, UK. Email: abdullahi.masud@gmail.com Abstract- This paper compares single network (SNN) and ensemble neural network (ENN) capabilities to recognize and distinguish surface discharges between two point- interface-pressboard arrangements with point angles of 10 0 and 45 0 . The training fingerprints for both the SNN and ENN comprise statistical parameters from the measurement of the surface discharge patterns captured over a period of 15 hours. The results shows that there is minimal statistical variability for surface discharges from a 45 0 point-interface-pressboard angles in comparison to that of 10 0 , which shows different behavior over a similar degradation period. In comparison to the widely applied SNN, the ENN also consistently provides improved recognition of PD patterns while the SNN actually shows improved discrimination potential between the two point- oil-pressboard degradation angle geometries. Keywords- surface discharges; single neural network; ensemble neural network. I. INTRODUCTION Over the years, several fault phenomena are known to occur within high voltage (HV) transformer oil insulation systems. Among them, surface discharge along an oil-pressboard interface is one of the most widely investigated [4, 6, 9]. If left unattended and subjected to long stressing periods, the discharges can degradate the oil/cellulose material and may lead to total failure of the equipment. One technique utilized by researchers to study surface discharge on an oil-pressboard interface is the point-earth configuration [4, 9]. Persistent surface discharge on the pressboard surface leads to surface tracking which appear as a treeing pattern emanating from the needle point position. According to the literature [9], surface tracking is caused by the drying process of the pressboard and the carbonization of the cellulose and oil. Recently, surface discharge for an oil-pressboard interface over different applied voltages and degradation levels has been investigated and classified by artificial intelligence systems [4]. However, no work has been carried out to investigate the influence of the angular positioning of the needle interface pressboard and how this may affect the surface discharge patterns and the ensuing degradation. Further as a consequence, the capability of a recognition system to identify angled point-oil-pressboard degradations has not been investigated. To recognize partial discharge (PD), statistical fingerprints from φ-q-n (phase-amplitude-number) patterns have been widely applied for recognition feature extraction, i.e. to choose the training and testing parameters for the recognition systems [1]. Recently, the ensemble neural network (ENN) has been regarded as a potential technique to improve discriminating statistical variations in the surface discharge φ-q-n patterns emanating from prolonged oil-pressboard degradation [4]. An ENN is a model technique for training a number of single neural network (SNN) models and combining their predictions [8]. As a further investigation, this paper therefore applies the ENN and SNN to classify and discriminate different angular positions of a needle point on pressboard surface. The novelty of the work presented in this paper is as follows: Firstly examining and comparing the PD patterns captured for two different point-pressboard-interface angles (i.e. 10 0 and 45 0 ) during an extended time period and correlating the discharges with tracking damage on the pressboard surface. Secondly to determine the ENN performance capabilities in discriminating surface discharges from the two angle geometries over similar degradation intervals and compare it with SNN performance. II. EXPERIMENTAL TEST SYSTEM The surface discharge fault was simulated by an experimental test arrangement as shown in Fig. 1. A pressboard, conditioned to 4.2% moisture content was immersed in a container filled with Castrol insulating oil [5]. The moisture content in the pressboard was determined by the weighing technique in accordance with the literature [8]. Fig. 1: Experimental test system for capturing surface discharges in oil. The point needle was of length 30mm with a 10μ m tip radius. For the 2 point-interface-pressboard angles (Ф o ) of 10 0 and 45 0 , measurement of surface discharges were made by a PD detection system, in accordance to the IEC 60270 standard [7]. The system allows the capture of apparent charge and also 2013 Annual Report Conference on Electrical Insulation and Dielectric Phenomena 978-1-4799-2597-1/13/$31.00 ©2013 IEEE 1217