15 Dic Indústia Ferroviaria
Automated predictive maintenance in the rail industry has come to stay. It’s not in the future, but right now, as you read, that such activities are becoming a remarkable breakthrough when it comes to parts replacement (even critical parts), wheels CNC lathe refurbishing, bogie maintenance and overhauling.
Automated predictive maintenance is achieved through the deployment of reliable data gathering systems (sensors, machine vision systems, smart embedded systems, etc…) both in-car and on-track, data storage and classification, and decentralized prognosis software. In addition, artificial intelligence adoption is quickly increasing to complete the contribution in the rail industry digital revolution.
By monitoring and combining data regarding each component of a train, the prognosis system is able to raise warning and alarm signs as they wear or must be replaced. These systems are revolutionizing many aspects of traditional maintenance based solely on timely replacement.
Automated visual inspection and measurement
One of the systems playing a significant role in reliable data gathering are 3D machine vision systems. They are used for accurate measurement of several aspects of wheels, brake pads, nuts and bolts presence, panel integrity, underframe integrity, pantograph integrity, etc… By using 3D laser scanner data, not only conventional images are obtained, but also reliable volumetric and dimensional measurements.
Visual data acquisition is applied both for rolling stock inspection and measurement, as well as infrastructure inspection and measurement.
Parameters of wheels include flange width, flange height, wheel diameter, back-to-back distance, tread rollover, false flange, steps in flange profile, flange profile radius, flange back excess material and QR factor.
Measuring these parameters is achieved through the use of specialized, dedicated contact gauges or by deploying non-contact laser scanners along a section of track.
Figure 1 shows a sample of a 3D point cloud of a wheel, gathered from a single 3D scanner mounted on the side of a track. Measurement takes place at max speed of 10KPH in this current model of scanner, although active work is undergoing to obtain reliable measurements at speeds of 100KPH+.
Figure 2 shows how readings from 4 different cameras are overlapped in order to maximize the amount of data and to take advantage of redundancy for minimizing occlusions and maximize measurement reliability. Overlapping readings from 4 different scanners allow for very accurate measurements, better than ±0,1mm for all the above mentioned parameters except for wheels diameter which is currently better than ±0,2mm.
Figure 1: Point cloud of a wheel from a single scanner
Figure 2: Wheel scan composition from 4 different scanners
Figure 3: Wheel profile
Figure 3 shows a raw wheel profile as a combination of two scanners (inner track and outer track). The raw data is just a superposition of data from different scanners. These individual data are merged in order to obtain a unique dataset per profile, on which automatic profile dimensions are measured and reported in real time.
Figure 4: Wheel profile dimensions as defined according to standards
Figure 4 shows how different wheel profile dimensions are defined according to standards. Measurements on wheel profiles are taken following these directions after the obtention of the unified profile data.
Figure 5: Back-to-Back measure using all wheel scanners
Back-to-Back measurements (Figure 5) are performed using all data from scanners at both sides of the track. Just like with profile measurements, data from the 8 scanners are merged in order to obtain a unique data set from which the Back-to-Back distance is estimated.
Bogies and general check
The whole train can be scanned using multiple scanners from which data is merged into a single dataset. A large number of checks are performed for presence / absence, open / closed, locked / released, etc… on items such as nuts and bolts, panels, lockers, lights, bobby pins, pantograph carbon strips, pantograph structures, roof items, etc…
Figure 6 shows a side view of a bogie merged from two scanners.
Figure 6: Side scan of a bogie
Figure 7 shows the side scan of a whole car where, in addition to the bogie, other elements are clearly visible, identifiable and locatable (since all 3D coordinates are recorded). Covers, panels, etc… are shown both in 2D texture and 3D point cloud.
Figure 7: Side scan of a whole car
Brake Pad measurement
Figure 8 shows a close-up view of a brake disk with one of its brake pads and caliper subsystem. All brake pads can be monitored for wear keeping track of their individual size log along time. As happens with wheels, this size evolution for each of the brake pads allows for predictive maintenance tasks to be trustfully fulfilled.
Figure 8: Brake disk, brake pads and caliper
Defective tracks is one of the most common causes of train accidents, inducing derailments and derived consequences.
Rail state monitoring on the head, web and foot sections of the rail, as well as switchblades, bolts, bolt holes, welds, gauge, crosslevel, warp, alignment, crossties, ballast, rail breaks, rail joints, etc… is of paramount importance for:
- Early detection and resulting replacement
- Wear monitoring
- Track condition (bolts, gauge, ballast, etc…)
- Predictive maintenance of all items involved
By systematically 3D and 2D (see Figures 9 and 10) scanning the track (rails, sleepers, switchblades, etc…), applying vision-based gauging, measurement and deep learning technologies, OPSIS is able to deliver track monitoring systems for assembling on both special vehicles and train cars, even in service.
Figure 9: 3D Scan of a rail with bolts and ballast (height coloring)
Currently full scanning and detection is possible at speeds of up to 180 KPH with an inter-profile distance of 5mm. This resolution, combined with the high accuracy inherent in our technology, allows for the detection of even very small cracks on the rail, crossties, track fixings, etc…
Higher speed systems at the same resolution is currently under development, allowing for even lower inter-profile distance, achieving sub-millimeter scans.
Figure 11: Close-up view showing profiles
Figure 11 shows a close-up view of how profiles are arranged in a section of the point cloud that’s being scanned along the track.
By combining 3D machine vision with other sensors, global positioning and highlight of the defects can be shown on a control panel, allowing for replacement of items to be held promptly, for both predictive maintenance and emergency actions.