Kîfan Deteksyonê ya Bêdixanê ya Daxwazên Derveyî yên Mîchanêkên Vakûmî yên Derveyî Bazirganî yên Li Sernavkirina Wêneya Ultraviolet

Vakûm dîvanekên derveyî (wê di navbera vê nivîsê de wek dîvanekan biceribîn) bi aborên wan wekî mezinahiya kuçik, qalang, şopandina va war-destpêk, karkeşîna hilî, sese ya wêne, dergeha vekehatî ya çavdane, dema şopandina kuçik û rawestina ewle, serperesta da ku pêk hatine. Di dema ku kirîna atmosferî din her destpêkir, di şertên hewlê neyên wekî têkê mezin, şîrîn, peyda bike, an daştan, îhtimala ku dergeha partial discharge (PD) li ser rûya dîvaneka post bikin. Ev dikarin heta flashoveran bibin, dema xebitandina dîvanekan kurtbike û salîbûna weraz û stabîl a systema elektrîk biguherîne.

Di vê nivîsê de, ZW32 - 12 outdoor-pole-mounted high-voltage vakûm dîvaneka (li mirayî wek HV ZW32 - 12 dîvaneka) wekî misal biceribîne, ku testên di şertên hewlê neyên berdara dikin. Prosesa şopandinê li ser rûya dîvaneka post e ZW32 - 12 bi UV imagerê hate tespit kirin, yana raqamê PD hatine mînagirin. Di dema ku imajên UV yên tê gotin, parametreyên xusiyên were tê gotin tu dibêjin cihên imajên wan. Derasê, raqamê PD bi rêbazê least-square support vector machine hatine hesab kirin, ku heta kalibrasyonê imajên UV yên bide. Ev teknîka deteksyonê neyên girîng ên ji bo PD dîvanekan ne.

Dîvaneka ZW32 - 12 yekê ya sê fasla, 50Hz, 12kV AC li derveyî ye. Ev bi tevahî dike ku ampera roja, ampera zafî û ampera kevkî şopande. Sazijeya wê di Şekil 1 de were nîşan kirin.

Bi tevahî ku imajên UV yên şopandinê li ser rûya dîvaneka post tê gotin û raqamê PD mînagirin, sistemê ya testi şopandinê li ser rûya dîvaneka post hate çêkirin, wê di Şekil 2 de were nîşan kirin. Di Şekil 2 de, T voltage regulatorê, B step-up transformer, R₁ resistorê ya piştgirî û C₂ coupling capacitor, ku bi tevahî li ser mînagirina PD tê gotin. 

Transformerê bi karîkerdina YDWT - 10kVA/100kV model, wê di Şekil 3 - a de were nîşan kirin. Ev bi tevahî li ser jînê ya hêviya lêgera bûyer dikin.

OFIL Superb UV imager bi tevahî li ser imajên UV yên şopandinê li ser rûya dîvaneka post tê gotin, wê di Şekil 3 - b de were nîşan kirin. Nermoka testa post insulator e ZW32 - 12 dîvaneka, ku sahsera têriyê dişe, wê di Şekil 3 - c de were nîşan kirin. Nermoka li ser chambera hewlê neyê ya destnîşan hatine tê gotin, ku hûn dikarin bi tevahî li ser reyata humîdiyê kontrol bikin.

Di vê sistemê de, metoda current pulse bi tevahî li ser mînagirina raqamê PD tê gotin. Console voltage regulator û transformer dikare bi tevahî li ser gerêza dîtginî bikin. Derêra, signala PD bi tevahî li ser PD detector JFD - 3 bi tevahî li ser coupling capacitor û detecting impedance tê gotin.

Bi tevahî li ser şopandinê ya humîdiyê, reyata humîdiyê li ser chambera hewlê neyê ya destnîşan dikare bi tevahî li ser reyata bistîn bikin. Insulators li ser bişeyî 2 saat bi tevahî li ser gerêza dîtginî bikin. Derêra, bişeyî 12kV bi tevahî li ser insulator tê gotin 5 dakîk. Li ser dema ku, imajên UV tê gotin, û raqamê PD mînagirin. Destiya tê gotina UV imager 5 metre, bi angle 0° û gain 110%. Testên bi tevahî li ser har reya humîdiyê, ku di 70% derava 90%, bi tevahî li ser processê ya 5% tê gotin.

 Prosesa imajên UV

UV imager video tê gotin, yana ra li ser frame processing tê gotin tu dibêjin framesên UV yên parastîn bikin. Her imaj frame yekê ya RGB true-color imaj [3] ye. Şopandinê li ser rûya dîvaneka post di imajên UV de wekî pikîn tê gotin. Ji kerema xwe, ev dikare ku her çend ku şopandinê li ser rûya digire, herdemê pikê mezin be. Buna, pre-processing û segmentation ya imaj dikarên li ser malpera imaj û pik part tê gotin.

Ji kerema xwe, komponenta red (R), green (G), û blue (B) di space-a RGB de tikîna color-a red, green, û blue û nebêtina tê gotin, yana ra li ser brightness-a imaj tê gotin. HSL komponenta imaj frame yên tê gotin di Şekil 4 de were nîşan kirin. Di Şekil 4 de, hêsasan H û S neyên dikarin pik ji malpera tê gotin, lê L component dikare [4].

Li ser Şekil 4 - c, L componenta pik parta mezin tê gotin. Buna, threshold segmentation yekê ya rêzikar dikin tu dibêjin pik part tê gotin. Key li ser hilbijartina L-component threshold. Li vir, Otsu's thresholding method bi tevahî li ser L-component threshold tê gotin [5]. Li vir, Matlab coding bi tevahî li ser Otsu's method tê gotin, optimal L-component threshold 216, û resulta segmentation di Şekil 5 - c de were nîşan kirin. Malpera neyên tê gotin, lê pik parta UV tê gotin.

Li ser Şekil 5 - c, ser pirsa pik parta UV, noise points yên zevî tê gotin. Ji kerema xwe, mathematical morphology operations bi tevahî li ser noise points tê gotin [6]. Li vir, mathematical morphology processing, resulta di Şekil 5 - d de were nîşan kirin. Noise points neyên tê gotin, lê pik parta tê gotin. Hûn define number of pixels in the spot part as the "facula area" of this UV image.

Li vir facula area bi tevahî li ser framesên UV yên parastîn tê gotin, hûn dikarin facula area curve tê gotin. Facula area curve at 85% humidity is shown in Fig. 6. As indicated by Fig. 6, the facula area fluctuates within a small range, with a large-sized spot emerging occasionally. Therefore, three parameters are defined to characterize the discharge intensity: the mean facula area, the intermittent facula area, and the repetition times of intermittent facula respectively [7]. We select 100 consecutive frames following the occurrence of partial discharge as the objects of study. The mean facula area is the average of the areas of 100 frames' faculae. The intermittent facula area is the average of the areas of faculae that are larger than the mean facula area, while the repetition times of intermittent facula is the number of faculae with an area larger than the mean facula area. According to Fig. 6, the mean facula area is 665 pixels. The intermittent facula area is 902 pixels. The repetition times of intermittent facula is 32.

Once the three characteristic parameters are calculated and the partial discharge (PD) quantity is measured synchronously, we attempt to determine the PD quantity using these three UV image parameters through the least-square support vector machine method.

Ninety samples of UV videos are selected. For each frame of these samples, three UV image parameters are calculated, and the corresponding partial discharge (PD) quantity is recorded by the JFD3 PD detector. The input arguments for the vector machine are chosen as the mean facula area, the intermittent facula area, the repetition times of intermittent facula, and the relative humidity. The output argument is the PD quantity. The Radial Basis Function (RBF) is selected as the kernel function. After normalization, 80 samples are utilized for training. Both the kernel parameters and the punishment parameters of the vector machine are set to default values. The training result is depicted in Fig. 7.

As shown in Fig. 7, for most of the training samples, the error compared with the measured PD quantity is relatively small. However, for some samples, the error exceeds 20%. The Mean Square Error (MSE) is calculated as follows:

To minimize the Mean Square Error (MSE) of the regression result and enhance the accuracy of the vector machine, a genetic algorithm (GA) is employed to optimize the kernel parameters and punishment parameters. [8 - 9]

The termination generation is set to 100, and the population size is set to 20. The optimization process is illustrated in Fig. 8. As shown in Fig. 8, after 30 generations of evolution, the MSE decreases from 0.07 to 0.01, indicating that the genetic algorithm has reached its optimal point. [10] The optimized kernel and punishment parameters are 0.2861 and 82.65 respectively.

After optimizing the parameters using the genetic algorithm (GA), the same 80 samples are retrained, and the regression result is presented in Fig. 9. As can be seen from Fig. 9, nearly all the samples exhibit a very small error when compared to the measured partial discharge (PD) quantity. The Mean Square Error (MSE) is now 10, which is significantly smaller than the value of 80 before the parameter optimization. Therefore, it is evident that optimizing the GA parameters can effectively reduce the MSE of the regression result and enhance the accuracy of the vector machine.

The final 10 samples are employed to conduct a test on the model. The regression results are presented in Table 1. It can be clearly observed that the error between the regression results and the actual partial discharge (PD) quantity is less than 6.1%. This finding indicates that the trained model demonstrates excellent generalization ability.

UV imaging technology is utilized to detect the surface discharge of outdoor vacuum breaker post insulators. The relationship between the facula area in UV images and the partial discharge quantity is explored through the least-square support vector machine method, offering a novel approach for insulation fault diagnosis of outdoor vacuum circuit breakers based on ultraviolet imaging.

A. Image Processing and Parameter Definition

After performing L-component threshold segmentation and mathematical morphology operations on UV images, the spot part of the UV image is extracted, enabling the calculation of the facula area. Three parameters are defined to quantify the discharge intensity: the mean facula area, the intermittent facula area, and the repetition times of intermittent faculae.

B. Data Acquisition and Analysis

Once UV videos are captured and the partial discharge (PD) quantity is measured synchronously, the relative humidity and the three UV image feature parameters are used as input variables. Through regression analysis via the least-square support vector machine, along with kernel parameter optimization using a genetic algorithm (GA), the PD quantity can be accurately determined.

C. Diagnostic Accuracy and Significance

By conducting regression analysis to establish the relationship between the insulator surface discharge quantity and its UV image facula area, it is found that the PD quantity diagnosed solely from UV images has an error of less than 6% compared to the measured PD quantity. This level of accuracy meets the requirements of practical applications and provides a new non-invasive method for diagnosing external insulation faults in outdoor vacuum circuit breakers based on ultraviolet imaging.

This research was funded by the National Natural Science Foundation of China and the State Key Laboratory of Electrical Insulation and Power Equipment. The authors would like to express their sincere gratitude to all those who provided support for this project.