Electron cascades from electrical discharge produce secondary emissions from atmospheric plasma in the ultraviolet band. For a single point of discharge, these emissions exhibit a stereotypical discharge morphology, with latent information about the discharge location. Morphological processing can uncover the location and therefore have diagnostic utility.
© 2016 Optical Society of America
In electrical power transmission, great care is taken to avoid sharp protrusions on structures that are on or near high-voltage conductors, to avoid a local concentration of the electric field that exceeds the threshold for coronal discharge. Electron avalanches from high-potential points produce, via impact ionization and subsequent recombination of atmospheric plasma, an ultraviolet (UV) photon spray. While most coronas are benign, some are indicative of a severe degradation that requires immediate attention; location, diagnosis, and disposition of coronas are necessary components of transmission line inspection. However, this method is applied infrequently due to the cost and difficulty of image capture (commonly in remote areas) and due to the cost of human inspection of UV-band imagery. More frequent inspection of power transmission infrastructure may be possible by low-cost aerial (UAV) image capture and automated image analysis.
The dielectric response to a high field is a complex, multiregime phenomenon , particularly when the field arises from an alternating (e.g., 60 Hz) source: cathodal and anodal coronas initiate, propagate and extinguish in each positive and negative half-cycle of alternation . This paper describes morphological processing that automates UV-band corona analysis in the static discharge regime  in which the dielectric is air (at standard temperature and pressure) and in which the field gradient is insufficient, for a fixed source-to-sink distance, to produce spark-over or conductive shorting, as these are the conditions for which corona is difficult to locate and analyze. Coronal ionization/recombination sites decrease monotonically with distance from the initiation point; in a planar projection into the viewport of a UV imager, the photon distribution appears as a solid ball of emission in the nucleus and a speckled halo in the periphery .
Because the peripheral emission sites appear as blobs with an area much smaller than the center site, a simple erode-AND operation can extract the center . However, in electrical inspection, visible-band imagery is needed to determine the corona cause, and the context of the corona is necessary. For example, adjacency to a high-voltage conductor and surface conditions in the area surrounding the corona center are of critical diagnostic value; methods that remove information about the corona radius are of limited utility. In this paper, we report two additional methods to determine the corona center, and a method to determine the entire extent of the corona emission image.
Center determination. The first center-finding method of this study uses temporal averaging to accomplish, via the time domain, an analog to spatial erosion method of . We found that, for common camera settings, averaging frames over 1 s of 30 fps video are sufficient to robustly lower the average intensity of peripheral emission sites, so that a blob test based on area and circularity followed by a persistence threshold removes all sites except the center and uniquely identifies the center coordinates:
If the UV imager viewport is not stable, this method suffers from distortions.
The second center-finding method counts the number of maxima of image intensity after Gaussian blurring:
Extent determination. Corona extent varies with the gain of the UV imager multichannel plate [4,6]. To diagnostically determine a discharge location, human inspectors commonly (1) start with a relatively high gain to find discharge figures, (2) lower the gain to remove stray ambient emissions, and, if a characteristic radial morphology is recognized, (3) center the camera view on the coronal nucleus, and (4) lower the gain further until only the center is visible and take a snapshot in the visible band with the UV nucleus overlaid. This sequence is based on the broadly reliable assumption that the corona nucleus is coincident with the high field initiation point of the discharge, and effectively pinpoints the center within a diagnostic context.
To derive a computational basis with which to mimic the third and fourth steps of this behavior, we create an intermediate representation of corona morphology by applying a series of difference of box (DoB) filters with increasing size , centered at the corona nucleus (Fig. 1). In this representation, if the DoB is balanced and the UV image is binary, the shape of the DoB score through scale indicates discharge isolation and morphological coherence. The positive extent of the DoB (central green boxes in Fig. 1) is a suitable frame for a detail snapshot in the visible band, while the negative extent (outer red boxes in Fig. 1) frames the contextual snapshot and can be used to prompt an operator to zoom out to capture the full diagnostic context [Fig. 1(c)].
Multiple coronas in a frame are located by successive application of Eq. (2): after a max is found at scale , the DoB step is applied at center , and the frame is modified by zeroing the pixel values within the positive portion of ; a second max is found at scale , thus the DoB step is applied at center , etc., until the integrated intensity of the th nucleus is less than 30% of the value of . For well-framed coronas, we found that the integrated intensity of the spray of the corona is always less than 30% of the nucleus’ integrated intensity and should be ignored.
If the camera is pointed well to one side (only spray is visible), or far from a corona (so that only the nucleus is imaged and the spray is below the camera resolution), this test does not discern between nucleus and spray. As a heuristic, a positive extent of the DoB center less than 1/5 of the camera frame dimension is labeled as “spray” [Fig. 1(f)], prompting the operator to pan or zoom in (increase gain).
NASA Safe Autonomous Systems Operations Program.
We thank the Dominion Electric Transmission Forestry & Line Services team for test imagery and helpful discussions about diagnostics, Fabio Bologna and Andrew Phillips of the Electric Power Research Institute for discussions on phenomenology, Taumi Daniels of NASA for discussions on plasma physics, and Scott Dorsey of NASA for discussions on corona generation.
1. F. W. Peek, Dielectric Phenomena in High Voltage Engineering (McGraw-Hill, 1920).
2. A. E. W. Austen and S. Whitehead, “Discharges in insulation under alternating-current stresses,” J. Inst. Electr. Eng. III 88, 18–22 (1941).
3. E. E. Kunhardt, “Electrical breakdown of gases: the prebreakdown stage,” IEEE Plasma Sci. 8, 130–138 (1980). [CrossRef]
4. K. M. Shong, Y. S. Kim, and S. G. Kim, “Images detection and diagnosis of corona discharge on porcelain insulators at 22.9 kV D/L,” in IEEE International Symposium on Diagnostics for Electric Machines, Power Electronics and Drives (IEEE, 2007), pp. 462–466.
5. B. Hu, L.-X. Ma, S.-J. Yuan, and B. Yang, “New corona ultraviolet detection system and fault location method,” in China International Conference on Electricity Distribution (CICED) (IEEE, 2012), pp. 1–4.
6. Y. Kim and K. Shong, “The characteristics of UV strength according to corona discharge from polymer insulators using a UV sensor and optic lens,” IEEE Trans. Power Del. 26, 1579–1584 (2011). [CrossRef]