Contents lists available at ScienceDirect Electric Power Systems Research journal homepage: www.elsevier.com/locate/epsr Localization of partial discharge in a transformer winding using frequency response assurance criterion and LMS adaptive filter Amir Mohammadirad, A.A. Shayegani Akmal ⁎ , Ramin Vakili School of Electrical and Computer Engineering, University of Tehran, Tehran, Iran ARTICLE INFO Keywords: PD location LMS adaptive filter Transformer winding Least mean square (LMS) Frequency response assurance criterion (FRAC) ABSTRACT Transformers are one of the essential and costly pieces of equipment in power grids. Monitoring and detecting insulation faults in transformers at the shortest time helps prevent catastrophic failures. Partial discharge (PD) is one of the most significant insulation failures in power transformers. Due to the complex structures of the windings, the accurate location of a PD source is very difficult to determine. In this paper, a technique based on a frequency response assurance criterion (FRAC) is proposed to determine the location of a PD source in a transformer winding. The responses of the winding to the proposed Heidler function injected in parallel to all sections of the winding are considered as PD reference signals. Moreover, the winding responses of any arbitrary PD pulses applied along the various sections of the winding, are considered as PD test signals. The maximum FRAC value between the reference signals and test signal specifies the location of the PD source. The simulation case studies clarify the superior performance of the proposed method. A least mean square (LMS) adaptive filter is suggested to minimize background noise effects and increase the accuracy of the method. Finally, the proposed method is validated with a laboratory winding. 1. Introduction Power transformers are very expensive and essential parts of power generation and transmission systems. They have a conspicuous position in power systems, being the vital connection between power stations and points of consumptions [1]. Normal operation of the power trans- formers plays a very imperative role in increasing the security and re- liability of power grids [2]. Therefore, the issue of improving the con- dition assessment tools is at the center of attention. One of the useful condition assessment tools is partial discharge (PD) measuring. PD is one of the major indicators of insulation conditions in power transfor- mers [3]. PD can happen due to aging and degradation activities of the insulation system. PD activities in power transformers, occurring over time, can be a result of thermal and electrical overstressing [4]. Iden- tifying the presence of a PD activity at early steps makes the initial insulation faults detectable, and prevents from the procedure leading to a total breakdown of the insulation [5]. Therefore, identifying, mea- suring and estimating the PD location are three vital actions of the procedure diagnosis and thus, decreasing the likelihood of further de- gradation of the insulation system and a catastrophic failure [4]. Also, estimating PD location at early steps provides an opportunity to make decisions about taking the power transformer out of service for main- tenance or to increase monitoring while operating [6]. Several methods have been employed in previous studies to find the location of PD in power transformers. In general, these methods have been divided into two categories: acoustic methods and electrical methods [7]. A number of acoustic-based methods are suggested for locating the PD source [8–13]. Although the acoustic methods are straightforward, they have low sensitivity and are expensive methods [7]. Among the various kind of methods for locating PD sources, the electrical methods are the most precise and economical ones [14]. Early research on the electrical methods for locating PD in Ref. [15] presumed that the function of a transformer in high frequency resembles a capacitive network. Digital filtering is described in Ref. [16], and it is indicated that the capacitive network model of a transformer is only valid over a limited frequency range. In Refs. [5,6,17–20], the frequency positions of the poles and zeroes in the transfer function of the measured currents have been utilized to find the PD location. In Ref. [21], a time-domain correlation method has been proposed for locating PD in a transformer winding. This method has a limitation in time domain due to the fact that the PD test and reference signals have to be the same in order to locate the PD source. In Ref. [22], hence, a new algorithm has been proposed for overcoming this limitation by converting the signal to the frequency domain. In Ref. [14] a technique based on the Archimedean Copula has been proposed for determining the PD location. However, these https://doi.org/10.1016/j.epsr.2018.07.020 Received 25 November 2017; Received in revised form 11 June 2018; Accepted 16 July 2018 ⁎ Corresponding author. E-mail addresses: amir.mohammadirad@ut.ac.ir (A. Mohammadirad), shayegani@ut.ac.ir (A.A. Shayegani Akmal), vakili.r1992@ut.ac.ir (R. Vakili). Electric Power Systems Research 163 (2018) 461–469 0378-7796/ © 2018 Elsevier B.V. All rights reserved. T