Railway fastener clips detection algorithm and algorithm automation
Presented by Shan Cao
Advisor:
Dr. Yuanchang Xie, Assistant Professor
Civil & Environ. Eng.
University of Massachusetts Lowell
A critical aspect of the transit state of good repair is the inspection of rail transit infrastructure. The current rail transit inspections are primarily based on visual observation, which is time-consuming and labor-intensive. Also, they cannot effectively identify subsurface hazards. For instance, ultrasound is usually used for rail inspection twice a year, and an inspection vehicle is used to check the rail geometry every few months. For all other structures and facilities (e.g., ties and fastening systems), they are inspected manually on a weekly or monthly basis, which requires considerable time and efforts. This practice is typical for most transit agencies in the United States.
This project developed an integrated system to automatically collect and georeference surface and/or subsurface data for rail, concrete ties, fastening systems, and ballast. It consisted of Ground Penetrating Radar (GPR), 3D laser, Geographic Information Systems (GIS), encoder, accelerometer, and Global Positioning System (GPS).
Modern fasteners are elastic fastenings which allow for dampening of the vibrations. The rails are clamped on to the sleepers using the fastening system, which is illustrated in the schematic of this figure.
Fastener working
Failure one
Advanced algorithms and software tools were developed to interpret the data and to identify rail infrastructure surface and subsurface defects and safety hazards, such as broken ties (e.g., cracks in ties), missing fasteners, fouled ballast, and wide rail gauge.
Rail transit agencies in the U.S. rely heavily on visual observation for track inspections. This manual method is time-consuming, costly, and unsafe and cannot effectively identify subsurface safety hazards. With the aging rail infrastructure, this proposed system is expected to substantially benefit the rail transit industry by improving track inspection efficiency, accuracy, and the safety of both the rail transit systems and track workers.
Each major component (e.g., GPR, laser, GPS, and accelerometer) of the system was developed/tested first in the lab to ensure that they worked properly. For testing the performance of the GPR subsystem, a wooden box was constructed in the lab and filled with soil, sand, and clean and fouled ballast. The results suggested that the developed GPR subsystem was able to accurately identify the locations of fouled ballast and subsurface pipelines. The team also conducted lab tests to evaluate the accuracy of the laser subsystem and found that its horizontal accuracy was less than 0.3 mm and its vertical accuracy was less 0.5 mm.
alla
The integrated system was then used to collect surface and surface data from Metro St. Louis and the Massachusetts Bay Transportation Authority (MBTA) during the summer of 2013. The system was designed to be easily mounted on a high-rail vehicle and performed reliably during the field tests.
A laser software tool for measuring rail gauge, identifying rail positions, cross-ties and fasteners, and detecting cracks in concrete cross-ties. A 3D template matching algorithm was developed for detecting missing fasteners;
Image captured in real railroad(this figure is provide by Professor Yuanchang Xie of Umass Lowell)
These are the fasteners that need to be detected out from the image
As we could see from this figure, the rail fastener clips are put in a straight line, so we could cut the nearby area out
newdataL1=data;
newdataL1(:,75:2040)=0;
newdataL2=data;
newdataL2(:,1:209)=0;
newdataL2(:,331:2040)=0;
newdataR1=data;
newdataR1(:,1:1939)=0;
newdataR2=data;
newdataR2(:,1:1689)=0;
newdataR2(:,1816:2040)=0;
as shown in the figure, the index of the fastener clips is between 80 and 130, the others should all be noise which could be eliminated
newdataL2(newdataL2>130)=0;
newdataL2(newdataL2<80)=0;
w=[20,20];
newdataL2 = minfilt2(newdataL2,w);%灰度腐蚀。
w=[50,50];
newdataL2 = maxfilt2(newdataL2,w);%灰度膨胀。
newdataL2(newdataL2~=0)=255;
The fasteners collected by different camera were aligned by the algorithms
The fasteners were detected by the
algorithm, if any fasteners is missing, the GPS data will be reported to the Railway maintenance company.
The end