13. Imaging analysis method for detecting various anomalies in overhead contact lines

Overhead contact line fittings are subject to various anomalies including general deformation, localized anomalies such as dropper bar breaks that are traditionally harder to detect, and incorrect bolt configurations. We developed a new machine learning algorithm capable of detecting both localized and combination anomalies via training on digitized data correlations as used in natural language processing (see Figure 1).
Using images of overhead contact line fittings captured outdoors during daylight hours, the tool successfully detects anomalies with an accuracy of 90% or better.
The tool screens for the four most common anomalies in overhead contact line fittings: deformation, bar breaks, misalignment, and loose or missing bolts.

Shadows within wires such as messenger wires that comprise wire strands have traditionally made it difficult to detect changes in color that indicate corrosion.
But now, thanks to machine learning, we can isolate wire strand components within the image, and create color and brightness distributions that can be used as the basis for an overall corrosion score, as shown in Figure 2. As the example in Figure 3 shows, corrosion scores for messenger wires increase and then decrease as the corrosion progresses. Using the corrosion score, it will be possible to track the chronological progression to determine the extent of corrosion in the future.

By combining the tool with images of overhead contact lines captured from the top of an electric inspection car or ordinary commercial car, we can screen the images, detect anomalies sites and assess the extent of corrosion. This is significantly more efficient than the current process of manual inspection for anomalies based on overhead contact line image.