XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE Condition Assessment of Hydrogenerator Stator Bar Insulation using Partial Discharge Measurements Torstein Grav Aakre Department of Electric Power Engeneering Norwegian University of Science and Technology (NTNU) Trondheim, Norway torstein.aakre@ntnu.no Erling Ildstad Department of Electric Power Engeneering Norwegian University of Science and Technology (NTNU) Trondheim, Norway Sverre Hvidsten Department of Electric Power Technology SINTEF Energy Research Trondheim, Norway Abstract—This paper presents results from laboratory measurements of partial discharge (PD) activity in 50 cm long samples cut from the mainwall section of old hydrogenerator stator bars. All stator bars were manufactured in 1976 and samples were taken after 35 years in service from both the low and high voltage sections of the generator, as well as non-energized back-up bars. The PD activity, using a phase resolved (PRPDA) measuring system, was investigated at different test voltages up to 9.6 kV (1.5 U0), frequencies and temperatures in the range 20- 155 ˚C and 0.1-50 Hz, respectively. The service-aged and the unaged reference samples showed a clear difference in voltage frequency dependence. It was, however, not possible to distinguish between service-aged bars from high and low electric stress. The observed frequency and temperature dependences are discussed with respect to theoretical assumptions regarding possible void degradation and surface conductivity. Keywords— Hydropower, partial discharges, variable voltage frequency I. INTRODUCTION Condition assessment of hydrogenerator stator bars, based on detection of changes in partial discharge (PD) activity are performed either as online or offline measurements. During PD- testing of installed generators, the temperature varies with the current load or time after shutdown. It is usually beneficial to perform offline PD tests at low frequency (0.1 Hz) due to the large capacitance in the generator windings, keeping the test equipment small. Although, several standardized test procedures are available [1-3] it is unclear how PD results obtained at 0.1 Hz relate to results measured at 50 Hz, see e.g. [4, 5]. One example of results from trend analysis of online PD monitoring is given by Bélec et al. [6]. In case of large change in PD characteristics, a more detailed examination of the generator revealed the presence of slot discharges. Based upon this, it was decided to rewind the generator. In case of considering and comparing the quality of large number of generators, statistical analysis of PD measurements are used to locate the critical generators suggested for further more detailed investigations [7]. An alternative approach is to use measured phase resolved PD signatures (PRPDA) to build a database of possible PD sources, for example Hudon et al. [8]. The main purpose of this paper is to examine experimentally, how PD features change with applied voltage frequency and temperature and examine if the variable voltage frequency PD technique can be used as a diagnostic tool for stator bars. II. THEORY REGARDING VOID DISCHARGES PDs occur when the voltage across voids exceeds its threshold value and a starting electron is available. This means that the PD inception voltage strongly depends on void shape, size and availability of starting electron. Charge is deposited during a PD, opposing the applied field. The apparent PD charge qa, which can be measured by an external circuit, can approximately be expressed as where b according to the well-known abc equivalent of the test object is the insulation capacitance in series with the void. It is assumed that the test object capacitance, a, is much larger than the void capacitance. ΔU is the voltage change across the void during the PD occurrence. In order to consider possible effects of testing at low frequencies the deposited charge qi is here considered to decay exponentially where τ is a time constant given by where σ is the resulting apparent conductivity and ε the permittivity. A statistical time lag influence the voltage ΔU and is more pronounced at higher frequencies, where a certain delay before initiation gives a PD at a higher voltage. The influence of statistical time lag on frequency behavior is e.g. described in the experimental and numerical work in polycarbonate by Forssén et al. [9]. Here a statistical time lag in the millisecond range was used to describe an increase in PD magnitude with applied voltage frequency. In case of varying the frequency of the applied voltage, three different cases may occur: I. The time constant is much smaller than the voltage half cycle. Deposited charge decay fast. This work is founded by the project "Hydrogenerator Stator Winding Insulation Assessment". The project is supported by The Research Council of Norway (Project No. 255099/E20), and industrial partners. ݍ ୟ =  ⋅ Δ (1) d ݍ ௜ ሺݐሻ d ݐ =− 1  ݍ ௜ (2) = ߝ ߪ (3)