News thumbnail
Technology / Wed, 18 Nov 2026 PLOS

Calibration algorithm for hybrid solar-blind ultraviolet detectors for discharge detection of high-voltage electrical equipment

2.2 Design of discharge detection processAccording to Sect 2.1, the discharge signal of high-voltage electrical equipment was obtained. The ultraviolet light and background light from corona discharge pass through the ultraviolet lens and the solar-blind ultraviolet filter and are irradiated onto the ultraviolet imaging microchannel plate. Step 3: Discharge detection based on hybrid solar-blind ultraviolet detectionIntegrating solar-blind technology and ultraviolet technology to obtain multiple characteristic information such as frequency, amplitude, and brightness, comprehensively understanding the characteristics of discharge phenomena, and improving the detection coverage and ability of high-voltage electrical equipment discharge. The discharge detection steps based on hybrid solar-blind ultraviolet detection are as follows:(1). The discharge detection output is achieved by utilizing the phase transition degree of the system:(8)Among them, represents the total driving force amplitude of the chaotic system, which is the discharge detection result based on hybrid solar-blind ultraviolet detection; represents the original amplitude; represents the initial phase angle.

In recent years, electromagnetic sensors and related multi-physics detection technologies have made significant progress in the field of high-voltage equipment and industrial defect diagnosis, providing important theoretical support and methodological references for the calibration algorithm of the hybrid solar-blind ultraviolet detector proposed in this paper. For example, one study proposed a dual-polarized broadband MIMO antenna for micro base stations, which achieves dual-band/single-band reconfiguration through orthogonal butterfly dipoles, parasitic element switching, and vertical stubs, with a gain of 9.4–11.2 dBi in the 1.58–2.77 GHz and 4.71–6.18 GHz bands, an isolation of ≥35.3 dB, and an envelope correlation coefficient of ≤0.002. Its design philosophy of multi-band cooperation and high isolation inspires this paper to strengthen band isolation and signal fusion between the solar-blind channel and the ultraviolet channel in the hybrid detection architecture, thereby suppressing spectral cross-interference. Another study proposed a laser-induced surface acoustic wave sensing method for detecting malaria parasites, utilizing photoacoustic signal excitation combined with surface acoustic wave sensing and FFT spectrum analysis to achieve early detection at a low concentration of 0.5%. Its optical-acoustic joint sensing and spectral feature extraction strategy provides a methodological reference for this paper in distinguishing weak discharge signals from background noise in both the time and frequency domains. Yet another study combined air-coupled ultrasonic transducers with laser-induced ultrasonic guided waves to develop an end-to-end deep learning detection model for rail foot corrosion. Using 8,122 field samples at an optimal frequency of 200 kHz, it achieved a detection accuracy of 99.5% and correctly identified abnormal signals. Its multi-modal sensing and deep learning fusion, along with the abnormal signal discrimination mechanism, inspires this paper to introduce feature classification and chaotic system phase transition determination into the calibration algorithm, thereby enhancing the ability to distinguish real discharges from false events in strong interference environments. Additionally, a study proposed a coil-based electromagnetic induction thermoacoustic system for detecting internal defects in railways. Through finite element simulation, equivalent circuit modeling, and ultrasonic propagation analysis, it successfully detected cracks with a diameter of 8 mm. Its electromagnetic-thermal-acoustic multi-physics coupling modeling and equivalent circuit analysis method provide theoretical tools for signal transmission modeling and interference suppression in the calibration of the solar-blind ultraviolet detector in this paper. The above studies indicate that multi-band cooperation, multi-modal fusion, end-to-end feature learning, and multi-physics coupling modeling are effective approaches to improving detection robustness in complex industrial environments. Inspired by these findings, this paper introduces feature classification, color calibration, and chaotic system determination on the basis of solar-blind ultraviolet detection, expanding the theoretical understanding of discharge detection from the three dimensions of frequency domain, time domain, and chromaticity domain, and providing a new technical path for the engineering application of discharge detection in high-voltage electrical equipment under complex scenarios such as strong background light and multiple interference sources.

Sulfur hexafluoride (SF 6 ) gas has excellent insulation performance and arc extinguishing ability [ 15 ], and is currently the most commonly used insulation medium in GIS. Its high electronegativity can effectively adsorb free electrons and suppress discharge development; In the research of discharge detection, the use of SF 6 can simulate the operating environment of real GIS equipment, ensuring the consistency between experimental results and engineering reality. Therefore, when using the defect simulation system in Fig 1 to simulate discharge, 0.4MPa SF 6 gas is filled and discharge detection methods such as ultra-high frequency and ultrasonic waves are used to statistically analyze the discharge amount, number of discharge pulses, and number of partial discharges. Fig 2 shows that the stepwise boosting method is used to gradually increase the test voltage in the pressurization mode. For every 5 kV/min increase, a high-speed oscilloscope is used to synchronously record the corresponding relationships and amplitude changes of the ultra-high frequency, ultrasound, and high-frequency current signals at different time bases. When the discharge of the model is detected, the pressure gradient changes to 1 kV/min, and observe whether there is a corresponding relationship between the partial discharge signal and the grounding current in time in the oscilloscope. Maintain a pressurization gradient of 1 kV/min, gradually increase the applied voltage from the initial value to the stable discharge of the model, record the voltage value, maintain the voltage, observe the discharge situation of the model, and when the model is in a continuous discharge state, gradually reduce the applied voltage according to the above gradient to the initial value, no longer generating discharge. Repeat the above pressurization process at unequal intervals to achieve high-voltage electrical equipment discharge simulation.

Electrical equipment defects exhibit different discharge phenomena at each cycle as the operating voltage changes [ 14 ]. If the applied voltage is effectively controlled to place the model terminal voltage at the discharge voltage point, and the spatial composite field strength of the partial discharge model approaches or slightly exceeds the material’s electrical resistance in a random distribution pattern, the partial discharge model will excite discharge with an indefinite time interval. Based on the principle of partial discharge, GIS is used as high-voltage electrical equipment. Through the defect simulation system in Fig 1 , the specific discharge model in the high-voltage electrical equipment is pressurized using the stepped pressurization mode in Fig 2 to simulate discharge of different defect types.

2.2 Design of discharge detection process

According to Sect 2.1, the discharge signal of high-voltage electrical equipment was obtained. However, due to the diversity and complexity of these signals in terms of frequency, amplitude, duration, etc., there may be different types of discharge phenomena that require effective identification and classification. Therefore, design the discharge detection process from the following steps:

Step 1: Analysis of frequency characteristics of discharge signals based on solar-blind technology

Partial discharge is usually manifested as a high-frequency signal in high-voltage electrical equipment. The solar-blind technology is sensitive to high-frequency signals and can effectively detect and analyze the frequency characteristics of discharge signals. By analyzing the frequency changes and trends of signals, potential partial discharge phenomena can be identified. Therefore, this paper chooses the silicon photodetector S2281 as the reference detector. When the response photocurrent is greater than 1.239 × 10-9 A in the 340nm wavelength band, its linear error is within 0.69%, which can be used as a standard detector with linear response in the ultraviolet wavelength band. Due to the selection of the solar-blind phototube R6800U-11 as the detector to be tested, with a spectral response of 160–350nm, a filter with a central wavelength of 340nm and a bandwidth of 10nm was placed at the photosensitive end of the standard detector. The response of the detector to be tested relative to the standard detector is:

(1)

Among them, represents the current value of the light response of the detector to be tested; represents the dark background value of the light response of the detector to be tested; represents the current value of the light response of the standard detector; represents the dark background value of the light response of the standard detector.

In actual measurements, due to the performance differences between various detectors, there may be certain differences in response nonlinearity, which inevitably leads to the calculated relative responsivity value obtained from the measurement not being a fixed value. Therefore, when the relative responsivity value does not change with the radiation intensity of the light source, the linearity of the test detector and the standard detector will be completely consistent. Therefore, the responsivity value obtained from formula (1) can be used as the reference responsivity, and the nonlinear correction factor for the -th measurement point is:

(2)

Among them, represents the current value measured by the detector at the -th measurement point; represents the dark background value measured by the detector at the -th measurement point; represents the current value measured by the standard detector; is the dark background value measured by the standard detector.

When selecting the reference responsivity, linear fitting is first performed using the least squares method [16,17], and the obtained slope value is used as the standard to calculate the theoretical response value of the detector to be tested, and its relative error is calculated. Then, use the responsivity at the data point with the smallest relative error as the reference responsivity. Due to the deduction of dark background values in actual calculations, the intercept b can be set to 0 during linear fitting. Therefore, the slope value a is:

(3)

Among them, represents the abscissa of the -th measurement point; represents the vertical coordinate of the -th measurement point; represents the number of measurement points.

According to formula (3), the linear error, i.e., the frequency characteristic analysis of discharge signals based on solar-blind technology, can be obtained as follows:

(4)

According to formula (4), the discharge signal of high-voltage electrical equipment for noise removal was obtained. Among them, is the linear error defined in this paper, which is used to quantify the degree of linear deviation of the response of the test detector (solar-blind photoelectric tube R6800U-11) relative to the standard detector (silicon photoelectric detector S2281). This formula is self defined in this paper. The closer the value is to 0, the more consistent the response linearity of the tested detector is with the standard detector, and the higher the reliability of the frequency characteristic analysis results. In this paper, a silicon photodetector S2281 is selected as the reference detector, and a solar-blind phototube R6800U-11 is chosen as the detector under test. The central frequency band for frequency analysis is selected as 13.0 kHz, and the rationale is as follows: The energy of UHF electromagnetic wave signals generated by partial discharges in high-voltage electrical equipment is mainly concentrated in the 3 kHz–30 MHz frequency band. Among them, 13.0 kHz corresponds to the characteristic repetition frequency peak of discharge pulses. The signal-to-noise ratio at this frequency is higher than that in the low-frequency band (<1 kHz) where power frequency harmonic interference exists, and lower than the radio broadcast interference band above 30 MHz, making it convenient to obtain sufficient time-domain resolution at the oscilloscope sampling rate (7 GHz). Meanwhile, the responsivity of the R6800U-11 in the 160–350 nm spectral range reaches a stable output region at a modulation frequency of 13.0 kHz (response fluctuation <2%), and the linearity error of the S2281 at this frequency (0.69%) is also the best across the entire frequency band. Considering three factors—anti-interference capability, detector response stability, and measurement linearity—13.0 kHz is selected as the characteristic analysis frequency for this study.

Step 2: Brightness detection of high-voltage electrical equipment based on ultraviolet detection technology

The solar-blind technology mainly identifies potential partial discharge phenomena by analyzing the frequency characteristics of signals, but it cannot fully obtain other characteristics such as brightness and heat of high-voltage electrical equipment signals, which limits the ability to comprehensively understand and diagnose discharge phenomena. UV detection technology can capture the ultraviolet light events generated during discharge, providing another perspective and supplementary information on the discharge phenomenon. Therefore, this paper mainly uses a spectral solar-blind UV visible dual spectrum camera as the main equipment, based on UV detection technology to detect the brightness of high-voltage electrical equipment, and provides a more comprehensive understanding of the situation of electricity liberation.

The structure of a spectral solar-blind UV visible dual spectrum camera is shown in Fig 3.

The structure shown in Fig 3 mainly consists of a spectroscopic system, a UV lens, a microchannel plate UV Charge-Coupled Device (CCD), a visible light CCD, and an image processing module. The ultraviolet light and background light from corona discharge pass through the ultraviolet lens and the solar-blind ultraviolet filter and are irradiated onto the ultraviolet imaging microchannel plate. The excited photoelectrons are multiplied by photoelectric amplification and then bombarded onto the rear screen, producing visible light images; Then, through light cone coupling, it is irradiated onto the CCD image plane to form the ultraviolet light path of the dual spectral camera; The other part of the light passes through two reflections of reflector 1 and reflector 2 and enters the visible light lens. The solar-blind dual spectral camera achieves real-time acquisition and display of ultraviolet signals through the collaborative work of the above process.

Due to the different focal lengths and sizes of ultraviolet and visible light lenses, the obtained image fields of view do not coincide. Therefore, this paper utilizes the mathematical relationship between UV and visible optical systems to perform coordinate point transformation and solve image fusion calibration for non central fields of view. Select the origin (0,0) as the center pixel of the UV and optical systems, and transform according to formula (5). After calibrating the discharge signal coordinate of the high-voltage electrical equipment in the field of view to remove noise, obtain:

(5)

Among them, represents the coordinate point of after field calibration; represents the focal length of the visible light lens; represents pixels in the ultraviolet region; represents 1/2 ultraviolet lens field of view; represents the size of pixels in visible light systems.

Based on the calibrated coordinates, perform brightness detection image fusion. Assuming that the visible light image is the background image and the ultraviolet image is the foreground image, the ultraviolet imaging system first converts the collected video signal into a digital signal stream and inputs it into the image processing section; Perform affine transformation on the digital signal, and jointly process it with visible light video signal to obtain a video signal with display; Finally, display the signal. The specific process of brightness detection image fusion algorithm is shown in Fig 4.

To ensure smooth integration, first use the aforementioned method to adjust the size of the two field of view ranges to be consistent; During the fusion process, the pixels in the image are calculated one by one. If the grayscale value of the pixels in the UV image is 0, then the grayscale value of the fused image is the grayscale value of the visible light image; If the grayscale value is not 0, the fused image grayscale value is adjusted to (255,0,0), which is displayed in red.

Step 3: Discharge detection based on hybrid solar-blind ultraviolet detection

Integrating solar-blind technology and ultraviolet technology to obtain multiple characteristic information such as frequency, amplitude, and brightness, comprehensively understanding the characteristics of discharge phenomena, and improving the detection coverage and ability of high-voltage electrical equipment discharge. The discharge detection steps based on hybrid solar-blind ultraviolet detection are as follows:

(1). Signal similarity calculation

The discharge process of high-voltage electrical equipment consists of two parts: electronic discharge and ion discharge [18]. Under normal circumstances, the duration of ion discharge is greater than that of electron discharge, and the Gaussian current pulse signal generated during this process has characteristics such as rapid oscillation, short formation time, and low amplitude. Therefore, selecting appropriate complex wavelets for wave analysis of the phase frequency characteristics of discharge signals.

Using correlation coefficients, compare the discharge signal with the wavelet function to obtain the similarity between the two. The comparison result is:

(6)

Among them, represents similarity; represents the mean of the discharge signal; represents the set of wavelet functions; represents the mean of the wavelet function. The larger the calculation result of formula (6), the higher the degree of similarity between the two.

(2). Discharge detection output based on chaotic system

Fit the obtained to a linear function using formula (7):

(7)

Among them, represents the amplitude; represents the absolute frequency difference; represents the time; represents the phase difference. It should be noted that although is a cosine function in mathematical form, its amplitude, absolute frequency difference, and phase difference are obtained by linear least squares fitting of the envelope peak sequence of the measured discharge signal. Firstly, extract the peak amplitude and corresponding time points of each discharge pulse, then fit these discrete points into a linear function to estimate the amplitude variation trend, and finally substitute the linear fitting results into cosine form to fully describe the signal oscillation characteristics. Therefore, linear fitting refers to parameter estimation methods, rather than itself being a linear function. This expression is used to describe the oscillation waveform of a discharge signal processed by a chaotic system. Its physical meaning is that the amplitude reflects the discharge intensity, the frequency difference reflects the offset of the signal frequency relative to the driving frequency of the chaotic system, and the phase difference reflects the time alignment relationship between the discharge start time and the system reference phase.

Put formula (7) as the driving force of the chaotic system into the chaotic system, and directly use the degree of phase transition of the system to complete the detection of discharge signals. When the phase trajectory of the chaotic system is in a large period critical state, the discharge signal generated by high-voltage electrical equipment during the discharge process is introduced into the chaotic system to obtain the signal sensitivity and noise immunity in the original driving force of the system. The discharge detection output is achieved by utilizing the phase transition degree of the system:

(8)

Among them, represents the total driving force amplitude of the chaotic system, which is the discharge detection result based on hybrid solar-blind ultraviolet detection; represents the original amplitude; represents the initial phase angle.

© All Rights Reserved.