Predicting Railroad Ballast Fouling Conditions Based on Ballast Image

The University of South Carolina is offering licensing opportunities for Predicting Railroad Ballast Fouling Conditions Based on Ballast Image




Evaluating railway ballast fouling conditions are critical to assessing track conditions and arranging proper ballast maintenance. The fouled ballast affects the performance of the track and can accelerate track deterioration. Hence, regular ballast maintenance activities are required, and the specific maintenance activity selection depends on the fouling severity. To quantify the fouling severity is important to schedule proper maintenance activity at the proper timeframes.

Invention Description:

We can predict how many fine particles are in aggregates from a photo of the aggregates. Instead of segmenting each particle out, our method analyzes the aggregates’ overall appearance. If there are fewer fine particles/dust, we should see big aggregates and gaps between them. If there are more fine particles/dust, we should see the picture to have a more consistent appearance as fine particles take all the voids.

Potential Applications:

Approaches using Ground Penetration Radar, Impulse Response, Surface Wave, and SmartRock have been developed to estimate the fouling conditions. Those methods require special sensors or equipment, and well-trained technicians, so they are very expensive. The current image-based ballast fouling prediction approaches are particle segmentation-based methods. Those models identify and segment individual particles to a specific size to quantify the ballast grain size distribution and then use the quantified grain size distribution to calculate ballast fouling conditions. The camera resolution limits their accuracy and particle segmentation algorithm, especially the fine particles (typically finer than 1mm) cannot be captured or segmented. So, they are not accurate and required sophisticated models.

Advantages and Benefits:

Different from the current particle segmentation-based approaches, our method is based on the overall image characteristics instead of individual particles. Depending on the fouling conditions, the combination of ballast particles, fines, and voids will determine the sample appearance which can be quantified by the frequency distributions of the color channels, Red, Green, and Blue, respectively. Images of ballast samples having different fouling conditions are processed through statistical analysis. Our method is not limited by fine particle sizes as the particle-segmentation methods would face. The computational expense is also significantly less than the particle-segmentation methods.