Stray magnetic field data of a synchronous generator with a faulty rotor winding – data report

Synchronous generators (SGs) are electrical machines commonly used in hydroelectric and thermoelectric power plants and are responsible for a large part of power generation worldwide and specially in Brazil, where 58.9% of the electricity is produced in hydro power plants 1. Unscheduled stoppages, whether caused by faults or periodic maintenance, can reduce the reliability of the electric system and cause significant economic and societal damage. On the other hand, preventive maintenance is not only costly, but it can also be inconvenient in times of hydro energy shortages that can become more common with longer drought seasons, potentially caused by climate change. In this context, employing predictive maintenance techniques to detect faults at an early stage is crucial.State-of-the-art fault detection uses measurements of vibration, electric current and magnetic field signals. Thus, maintenance stoppages can be spaced out and carried out at more convenient times. Fault detection based on monitoring the mechanical vibration of synchronous generators is a commercially mature technique, already founded on technical standards 2 3, and is the basis for diagnostics carried out by a wide range of commercial equipment 4 5. In addition, other variables, such as temperature and magnetic flux in the air gap, are used to complement the diagnosis 4 6. However, these sensors are intrusive, and the difficulty, and the cost involved in installing them is one of the main disadvantages of these techniques. It generally involves dismantling parts of the machine and welding or drilling holes into generator structures to attach sensors.In this context, the development of non-intrusive fault detection techniques is essential. It is the case of the stray magnetic field monitoring around the generator. This method consists of measuring the magnetic field external to the generator using induction sensors (search coils sensors) and determining its frequency spectrum by applying a Fourier Transform. This characterizes a kind of synchronous generator magnetic signature 7. The analysis of the magnetic field frequency spectrum has shown the appearance of several spectral components that are smaller than the electric fundamental frequency. Further analyses pointed out that these components are multiples of the mechanical rotational frequency, that can be defined as the ratio between the electric fundamental frequency and the number of pole pairs 8910 11. These components appear due to magnetic asymmetries among the poles of real machines 89 11. Their amplitudes are intrinsically correlated with both generator operation point and the condition of the magnetic circuit. The occurrence of faults modifies this magnetic circuit, changing the patterns that characterize the generator magnetic signature. These amplitude patterns, hence, can be used to characterize the synchronous generator condition 8. Therefore, tracking these patterns changes can be used to detect the occurrence of incipient faults in the generator 12 13.However, progress in detecting incipient faults through the external magnetic field is hindered by the lack of public data concerning faulty synchronous generators. Research employing machine learning and artificial intelligence techniques is almost exclusively centered on induction machines 1415 16 17. Therefore, this data report aims to help fill this gap by presenting an open dataset 18 comprising stray magnetic field data of a synchronous generator with faulty rotor winding. Data concerning the healthy machine is also provided. The files associated with this dataset are licensed under a Creative Commons Attribution 4.0 International license. This data report is divided into three sections. Section one presented an introduction and how this work is positioned in its field. Section two describes the data collection procedure. It details the generator under study, the instrumentation for measuring the magnetic field external to the generator. Furthermore, the same section portrays the performed tests from which the data were gathered and how the faults can be imposed on the machine. Additionally, section two also presents the dataset structure. Finally, section three presents some insights into the data, showing how this type of non-intrusive monitoring can be used to detect faults in synchronous generators.This section describes the data collection process, detailing the workbench used and the generator studied. It is also presented in this section the instrumentation employed to measure the stray field of the 8-pole salient generator. The tests performed to collect the data are also detailed. Finally, the dataset structure is presented.The synchronous generator under study is an 8-pole salient synchronous machine, which is part of a workbench also comprising a direct current motor, as the primary machine, and another 2-pole cylindrical synchronous machine. Figure 1 illustrates the workbench used in this work and Table 1 presents the description of every item shown in the figure. At the same time, Table 2 shows the nominal characteristics of the machines that compose the workbench. This workbench was manufactured by the Brazilian company Equacional Elétrica e Mecânica LTDA. This workbench is specially designed to allow the controlled imposition of the following faults:• Short-circuit in the field (rotor) winding (simulating 20%, 50%, and 100% of pole coil turns removed) • Short circuit in the armature (stator) winding (simulating 20%, 48.6% and 100% of pole coil turns removed) • Open circuit in the damping cage • Short circuit between core laminated sheets • Static eccentricity • Mechanical unbalance.The data described in this work comprise measurement results of tests performed on the 8-pole salient generator operating with short-circuit in the rotor winding.The time-varying magnetic flux passing through the sensor induces a voltage at its terminals. This voltage is proportional to the time derivative of the magnetic field. Thus, the magnetic field external to the machine is estimated based on the voltage measured at the sensor terminals, which can be determined by the following equation.In which □□(□□) represents the voltage induced at sensors terminals, □□ is the number of turns, □□ represents the coil cross section, □□ 0 is the magnetic permeability of air and □□ represents the external magnetic field.Two sensors (CH1 and CH2), positioned 90° apart, are responsible for measuring the tangential component of the magnetic field external to the generator. An example of these sensors and their placement are presented in Figure 2. Their constructive characteristics are detailed in the following table. The devices were calibrated using uniform magnetic fields generated by a Helmholtz coil. They demonstrated a linear response and sensitivity compatible with those calculated from the constructive parameters. Voltage signals coming from the sensors pass through a conditioning stage, where they are amplified and filtered. Those signals pass by an input filter, a pre-amplifier, a variable gain amplifier and an anti-aliasing filter. The input filter frequency response is shown in the Figure 3. In addition, the gain provided by the board can be selected from 11, 110, 1100 or 11000. The circuit provides a differential voltage output of ±5 V, the reference being the same as the 12 V power supply. A National Instruments (NI) board, model USB 6361 BNC, is used to acquire the signals from the conditioning boards. The device is capable of simultaneously acquiring 8 analog channels at a maximum frequency of 125 kHz. The board has a 16 bits A/D converter that supports voltages up to ±10 V. The signals acquired by the acquisition board are transferred via a USB cable to a computer and read out using a LabVIEW Virtual Instrument. More information about the acquisition board can be found at its website.This dataset presents data from tests imposing short-circuit in the field (rotor) winding faults to the 8-pole generator. Figure 4 shows the rotor winding configuration and how the fault can be imposed. The available data correspond to seven operation points (from P15 up to P21) which are detailed in the following table. During these tests, the generator was synchronized with the local 60 Hz distribution network. The tests involved progressively short-circuiting turns on one pole of the field winding. The turns short-circuit was performed by an automatic system, that can switch the terminals J (healthy generator), J1 (20% turns removed), J2 (50% turns removed) and J3 (50% turns removed) without shutting down the generator. Hence, for each fault condition (and every operation point), the voltage at the terminals of both sensors was collected for 10 seconds, at 125 kHz sample rate. Those measures were named voltage vectors.The dataset is available in open access in the Mendeley Data repository under the DOI "10.17632/d75sb25f7m.1", and under a Creative Commons Attribution 4.0 International license 18. At this repository, a compressed rar file can be downloaded. Once extracted, there is a directory called "Data", in which the data (voltage vectors) are stored. Inside this folder, there are seven subfolders, named P15 up to P21, that contain the data of every generator operation point presented in Table 4. Data are stored in comma separated values (CSV) files, with comma (",") as delimiter and point (".") as decimal separator.The data files are named 'VoltageVectorXX.csv,' where 'XX' represents the voltage vector number. Table 5 presents the correspondence between voltage vectors numbers and the generator fault condition for each operating point. The data within the dataset presented here enable several analyses. This paper shows some examples of the rotor winding fault occurrence in both time-domain and frequency-domain. It also presents certain limitations of the dataset, especially regarding data quality issues.The first step into the analysis of the generator's external magnetic field is to observe the voltage issued from the induction sensors, as this voltage reflects the time derivative of the magnetic field multiplied by a constant that depends on the characteristics of the sensor and the magnetic permeability of the air. Because of the sinusoidal nature of the waveform, the time derivative of the magnetic field can also be seen as the magnetic field itself displaced by 90º. Hence, the voltage issued from the sensors can be directly used to analyze the magnetic field.The notebook available within the dataset allows you to choose the file to be analyzed by selecting the generator's operating point and the desired voltage vector. An example can be seen in Figure 5, as it illustrates voltage waveform issued from both sensors corresponding to the voltage vector 48 of the operation point P15. This voltage vector corresponds to fault condition with 50% of the turns removed from one pole of the rotor winding, emulating a short-circuit.The waveform in Figure 5 is characteristic of the magnetic field variation in an eight-pole synchronous generator operating at 60 Hz. The mechanical fundamental frequency, which is 15 Hz for this kind of operation, can be seen by the periodic amplitude reduction observed in both signals every eight half-cycles. This reduction is directly related to the removal of 50% of the turns from a pole of the rotor winding. The pole weakening results in a decrease of the generated magnetic field and, consequently, a reduction of the induced voltage during the pole's passage through the sensor region. This reduction can be observed in both sensors. However, it does not occur simultaneously, as the sensors are displaced 90º apart. In an eight-pole generator, 90º physical degrees correspond to 360° electrical degrees, as can be seen in the figure. Additionally, the amplitude variation between the measured signals, despite the sensors having identical constructive characteristics, is due to the different attenuation of the external magnetic field in the specific position of each sensor. Another approach to analyze the magnetic field data is to transform the time-domain data into the frequency-domain. Because of the discrete nature of the measured data, the Fast Fourier Transform (FFT) is the way to do it. With the application of the Fast Fourier Transform (FFT) it is possible to obtain the harmonic components that constitute the signal. The amplitude of each harmonic component depends on multiple factors, such as the SG's operation point, its constructive characteristics and the condition of the magnetic circuit. The ensemble of these harmonic component amplitudes forms the magnetic signature of the generator. Figure 6 presents the FFT of Voltage Vector 48 at P15 operation point. In this figure, it can be highlighted the fundamental mechanical harmonic (fmh), which manifests as 15 Hz in an eight-pole synchronous generator synchronized to the 60 Hz power grid. It is also possible to visualize several other harmonics components that are integer multiples of fmh. Additionally, it can be seen the strong presence of the fundamental electrical frequency (feh), at 60 Hz and its third harmonic, at 180 Hz. For this analysis, a rectangular window was employed, with a spectral resolution of 0.1 Hz, given the total acquisition time of 10 s for each voltage vector. Periodic monitoring of the generator's magnetic signature enables the detection of changes in harmonic component amplitudes, which may indicate incipient faults. Therefore, due to the existence of several voltage vectors, measured at different operation points and different rotor winding conditions, within the dataset, it is possible to build a history with the amplitudes of every harmonic component, that enables the detection of a faulty condition. Figure 7 illustrates the evolution of the fmh amplitude, issued from CH1, for operation point P15, across all 95 voltage vectors. In this figure, it can be seen how the sequential removal (20%, 50% and 100%) of turns from the rotor winding of a pole affects the 15 Hz amplitude of the generator frequency specter. Beyond the fundamental mechanical harmonic, numerous additional harmonics are affected by the fault in diverse ways. The analysis, while not based on optimized parameters, is sufficient to illustrate the potential and utility of the presented data.Consequently, monitoring the historical amplitudes of harmonics, combined with machine learning techniques applied to FFT signal amplitude histories, emerges as a promising methodology for detecting incipient faults in SGs. This approach underscores the critical importance of sharing experimental data derived from external magnetic field measurements in SGs under fault conditions. The magnetic field captured by the sensors is highly dependent on the generator's field current and, consequently, its reactive power. It is therefore expected that the magnitude of the magnetic field will be stronger when the generator operates with high reactive power. Nevertheless, this could be an issue because of the amplification step nature.In this context, signal saturation can be observed in some measurements. One example is the signal saturation observed in the CH1 measurement, for the P21 operating condition, Considering the fault condition of total short-circuit of the turns of one pole of the rotor winding. This example can be verified in Figure 8. However, despite the occurrence of saturation under this specific condition, it can be observed that saturation does not occur for CH2, due to the different attenuation of the external magnetic field in the specific position of each sensor. Hence, the data within this dataset enable a wide range of analyses for all operation points and fault conditions, despite their quality issues at some instances. This work aimed to present a dataset composed by data issued from measurements of stray magnetic field of an eight-salient-pole synchronous generator with faulty rotor winding. Magnetic field data can be used for various applications, such as detecting the generator's speed or identifying incipient faults. The raw sensor data were provided so as not to limit the type of analysis that can be performed. For example, while a frequency analysis via FFT was presented in this paper, it is possible to use techniques involving wavelet transform, combined with spectrograms and neural networks, for the detection and determination of fault severities. Other types of analyses can be based on statistical analysis methods, machine learning, or artificial intelligence. In this context, the dataset contains data regarding not only the healthy condition of the generator, but also three severities of rotor winding short-circuit fault. Hence, this dataset is intended to fill the gap in the availability of open magnetic field data from synchronous generators operating with faults.Electricity Regulatory Agency (ANEEL) and developed under the R&D Program of ENGIE Brasil Energia and Itá Energética S.A., PD-0403-0048/2019 and PD-00403-0057/2023, both entitled "Non-Invasive Equipment for Fault Detection in Synchronous Generators through the External Magnetic Field", and carried out with the support of the Brazilian National Council for Scientific and Technological Development (CNPq).