By Jeff Tuzik
The railroad is a system. The holistic approach is the optimal approach. This is a maxim often repeated at the Wheel Rail Interaction (WRI) conference, and at railroads around the world. It’s a simple concept but it belies great complexity. At WRI 2017, a number of experts from various departments came together to discuss the railroad system not from the top down, but from the component level up. Every component from wheels, to ties, to individual cut-spikes exists in a hierarchy – a pyramid, in which a weakness at any level can affect the whole.
Consider a derailment, a system failure of the highest order. A recent derailment investigation on Norfolk Southern (NS) illustrates one of the ways in which a seemingly low-level component failure can quickly escalate.
The Spike Component
The first two questions a derailment investigation has to answer are: “which was the first car to derail,” and “what was the point of derailment (POD)” said Brad Kerchof, Director of Research and Tests at Norfolk Southern.
In 2014, 21 loaded tank cars derailed on an 8.3-degree curve in Vandergrift, PA. Initial analysis ruled out both train handling and track condition as culprits; the train was moving at 32 mph on a 30 mph curve when the derailment initiated. The curve had recently been re-railed and surfaced, and was considered to be in excellent condition, Kerchof said. None of the broken rail sections showed evidence of internal defects or end-batter. On the mechanical side, only one defect was found: a broken stub sill on the 11th-derailed car in the consist. Combined with the apparent absence of train-handling or track-related issues, the investigators initially concluded that this (11th-to-derail car) was the first car to derail, and that the broken stub sill was the likely cause.
“There were some issues with this analysis, however,” Kerchof said, “It’s unlikely that a car derailing would also cause 10 cars ahead of it to derail – typically the first car to derail is at the head of the line.” Further analysis of the broken stub sill on the 11th-derailed car showed the break to be the result of high rotational stress that is consistent with coupled cars rolling over. This led the team to conclude that the broken stub sill was a result of the derailment, not the cause.
Luckily for the investigation, the derailment occurred in front of a steel plant with a number of security cameras that surveyed the area of the derailment. In looking through the footage, the investigators were able to determine the moment of derailment and the true first car to derail. The video also indicated that the derailment was caused by wheel climb or wheel drop-in, Kerchof said. The next step was to examine the rail fragments to determine where wheel marks (evidence of wheel climb or drop-in) first appeared, and thus determine the POD.
Evidence at the POD further narrowed-down the list of potential causes. The presence of intact elastic fasteners ruled out rail rollover, Kerchof said. And the lack of flange marks on the rail head ruled out wheel climb. The investigators then located the POD site’s corresponding ties, and cross-sectioned them for analysis. This is where the culprit lay; they found numerous broken spikes, including multiple broken spikes on top of older broken spikes – indicating a long-term, systemic issue.
The investigation officially determined that the 1st car to derail pushed the high rail out of gauge as a result of broken spikes, allowing the low-side wheels to drop inside the low rail.
Incidents like these led NS to take a closer look at broken spikes and their causes and to draw a number of conclusions based on their experience:
- Broken spikes are found on the high rail on both gauge and field sides, and in both rail and anchor spike positions.
- Broken spikes are found in solid ties, including new ties.
- Broken spikes appear in curves with standard 8×18 tie plates, but appear more frequently in curves with Victor tie plates.
- Spikes typically break 1 to 1.5 in below the tie surface, making them difficult to detect.
- Tie plate movement can be minimal until a cluster of ties with broken spikes develops.
- Broken spikes appear on curves greater than 6 degrees and timetable speeds < 35 mph.
- Broken spikes tend to be associated with non-uniform alignment in the high rail.
- Longitudinal force, such as imparted by heavy braking and high tractive effort, plays a significant role in the development of broken spikes.
Kerchof also noted that the increase in occurrence of broken spikes on NS over the past three years coincides with the use of AC locomotives. NS has also taken steps to prevent the occurrence of broken spikes in susceptible locations; adding anchors to the high rail, and installing “bridge” Victor plates every third tie have proven to be very effective, he said.
The spike is not a flashy or expensive or high-tech component. But it’s a crucial component. Keeping track of and monitoring easy-to-overlook components like spikes and fasteners is a big part of an industry-wide push for the development of increasingly granular and comprehensive asset health management systems.
The Wheel Component
One of the shortfalls of traditional inspection techniques is their subjective nature, making results difficult to quantify. A detailed, objective record is critical to monitoring component health. To that end, the BNSF has recently begun using a relatively new kind of vision system to monitor wheel condition with great accuracy and breadth.
Since the introduction of laser-based rail profile measurement, various vision-based measurement systems have found their way into the railroad environment in a number of applications. These range in function from wheel and rail profiling to tie and fastener evaluation. A typical wayside inspection suite may consist of five or more different vision systems, each with different functions: coupler inspection, undercarriage inspection, wheel profile, brake inspection, and truck inspection, for example. But the list doesn’t stop there. BNSF has recently begun to use and evaluate the Beena Vision full wheel inspection system that also includes a wheel surface scanning module that uses several cameras arrayed in series to take 1,500 – 3,000 measurements of each wheel that passes the inspection site. These images are then stitched together to create a map of the wheel surface. Every wheel that passes through the site is measured via the following parameters:
- Wheel Profile
- Wheel Diameter
- Wheel Equivalent Conicity
- Wheel Surface Defects
- Wheel Plate Inspection
- Broken Wheel Sections
- Externally Visible Cracks
- Wheel Hunting
- Angle of Attack
- Back to Back
- Wheel Surface Temperature
Data from other measurement devices such as wheel temperature and impact loads can also be combined with this data. “The idea is to create an as-comprehensive-as-possible dataset for analysis and condition-based monitoring,” said Kambiz Nayebi, Director of Advanced Engineering at Beena Vision.
The aim of the wheel surface scanning component is to fill in some of the gaps in traditional autonomous wheel inspection systems such as wheel profile systems and wheel impact load detectors (WILDs). Wheel profile systems, while invaluable, typically take measurements at one or a few points on the wheel, with the general assumption that wheel wear is uniform, Nayebi said. As a result, these systems can’t detect surface defects like shells, cracks and spalls. Similarly, in order for a WILD to detect a defect, the defect must occur on the contact patch of the wheel. If the defect occurs elsewhere, it might as well be invisible to the system.
Prior to their arrival in North America, wheel surface scanning systems have seen use on transit lines in the European Union and on freight lines in Australia. Their successful implementation has allowed operators to eliminate manual wheel inspection from the regular maintenance routine, Nayebi said. With proper calibration, these systems can autonomously detect and flag surface defects such as:
- Fatigue Cracks
- Out of Round
- Built‐up tread
- Broken and Separated Sections
- Externally Visible Cracks
- Shattered Rim
- Wear Variation along the wheel
- Spread Rim
- Vertical Split Rim
BNSF is currently working with Beena Vision to provide samples of their own wheels with characteristic defects, levels of wear, and profile variances for validation and further calibration of the wheel scan system. These calibrations can be time consuming, as the system is not a simple plug-and-play solution. “Development of an algorithm for the detection of a specific defect type can take several months to a year,” Nayebi said, “and requires close collaboration.”
The ability to autonomously collect detailed and objective data is an important part of building a big-picture of asset health. And the ability to drill down into the condition of an individual component, especially one as critical as wheels, helps to ensure that the details, or the data granularity, isn’t lost in the big picture. But it isn’t only new technologies that are pushing overall asset management into the future; it’s also, perhaps even more so, the integration of the data and how it’s used.
The Truck Component
A recent case study on Brazil’s Vale EFC (Sao Luis Line) took a deep dive into truck performance data over a seven-year-period to explore the relationship between truck performance and track/operating conditions.
In 2009, Vale EFC, working with Wayside Inspection Devices Inc., began collecting detailed truck performance on: angle of attack, tracking position, inter-axle misalignment, tracking error, lateral shift, and hunting. In September of 2009, shortly after data collection began, about 10% of the empty traffic was hunting and 0% of the loaded traffic was hunting (in the 0 – 14mm low severity bracket), said Paul Bladon, Vice President of Business Development at Wayside Inspection Devices Inc. But by 2015, up to 60% of all loaded traffic was hunting (in the same severity bracket). Clearly something was awry.
Overlaying other data, particularly major track-related events, helped the Vale/Wayside Inspection Devices team break down their study into more manageable units.
- In 2009, just prior to the installation of the “TBOGI” truck performance detector system, the rail was replaced with a lower-quality rail in short (25m) sections.
- In 2011, a large number of new trucks began to enter the fleet.
- In 2012, significant manual track maintenance was performed, including fixation-renewal, tie-leveling, tamping and manual grinding.
- In 2013, a large-scale rail intervention replaced the lower-quality 25m rail with higher-quality 250m rail.
- In 2015, there was another major rail replacement, ballast renewal, and bogie maintenance.
- In the period from 2009 to 2015, the volume of traffic on the network doubled.
In terms of truck performance, the 2009 – 2011 phase was fairly standard for a heavy-haul environment, said Bladon. There was some low-severity hunting, typically in empty cars, but nothing to alarm the team. The degradation trends for specific bogies – the increase in their peak-to-peak hunting severity – during this period was also considered normal. During this period, hunting trucks degraded from mild-hunting to requiring intervention within 14 months.
The 2012 – 2013 phase saw an increase in both unloaded and loaded truck hunting; loaded truck hunting was much more dramatic. However, the early-2013 rail replacement program targeting the previously-laid 25m rail, immediately eliminated the spike in loaded car hunting, leading the team to conclude that the uptick was largely due to poor and rapidly degrading rail conditions.
In 2014 – 2015, both loaded and unloaded hunting increased dramatically. During this period, Vale also increased the average loaded train speed from 30 kph to 40 kph. The speed increase clearly coincided with the increased hunting and rate of hunting degradation, Bladon said. Vale undertook a major track overhaul at this point and implemented an aggressive bogie maintenance regime particularly focused on various bogie defects. The positive effect on hunting was immediate.
Based on their observations over the 2009 – 2015 period, the Vale/WID team organized bogies with a propensity to hunt into three classes: gradual-onset-while-empty, gradual-onset-while-loaded, and sudden-onset-while-loaded. Based on their years of collected data, several patterns emerged.
- A bogie that develops gradual-onset hunting will do so in either loaded or unloaded condition, but not both.
- Sudden-onset hunting develops in the loaded condition only.
- Gradual-onset hunting typically reaches higher peak-to-peak severity than sudden-onset hunting, which tends to plateau.
Additional measurements taken during the 2016 – 2017 period attest to the efficacy of the remedial actions Vale undertook in late 2015. The combination of major rail, tie and ballast renewal, and the modified bogie maintenance regime returned loaded and unloaded hunting trends back to “normal” baseline parameters and have held steady since then.
This kind of case study reaffirms the importance not just of precise data, but of data collected over long periods of time. Historical data on any given component allows patterns to emerge that could otherwise easily be missed.
The Sharing Component
Collecting truck or wheel data over time on a captive-fleet system like Vale EFC isn’t easy to replicate in North America given the interchange operating environment. Since freight cars go offline, individual properties have a hard time building useful databases on cars, trucks, and wheels that pass through their territory.
Railinc, through the creation of centralized databases of various components, aims to make the kind of asset management that is possible on captive fleet railroads more accessible to North American railroads, car owners and operators. Due to North American interchange practices, many issues that affect wheel/rail interaction can’t be easily addressed locally. Over the past several years, Railinc, working with the AAR, has sought to identify common problems that interchange practices make difficult to address. One such area of focus is reducing mechanical service interruptions.
Beginning in 2013, Railinc began development work on a methodology for reducing service interruptions based on a few key directives: bad actor identification, component serialization and equipment failure analysis.
The initial concept in the bad actor identification program was to identify cars that were involved in multiple undesired emergency stops, said Steve Josey, Director of New Products & Services & Asset Health Strategy at Railinc. Sometimes the cause of an undesired emergency stop is obvious, but often the root cause is not found, he said. Meaning, once back in service, the car may continue to cause problems, often on another property. By tracking undesired emergency stops over time, the repeat offenders can be identified and brought into a shop for closer inspection.
Based on work Norfolk Southern did to identify bad actors on their property, Railinc developed (and is still developing) an expanded program to cover all north American heavy haul operations. By collecting data that spans multiple properties, the trends are clearer and bad actors can be addressed more quickly, ideally before initiating another undesired emergency stop. Based on work done so far, Josey estimated that well over 3,000 line-of-road failures have been prevented. “[It’s] a great example of a problem that’s better solved as an industry, rather than locally,” he said.
Many railroads use temperature detectors (or cold-wheel detectors) to detect bad actors in brake performance as well. This is another type of data that is well-suited to centralized collection and inter-organizational sharing. If an operator records an instance of a cold wheel on their property and can query a database to see that car’s history, there is potential for identifying bad actors earlier or for saving time and money on unnecessary changeouts and inspections, said Tod Schneider, Director of Advanced Freight Car Engineering at Union Pacific. Comprehensive cold-wheel detection data could also lay the groundwork for a more condition-based approach to monitoring brake performance, rather than relying on set intervals or mileages to determine when maintenance is required, he said.
It’s not only wheels and cars that are difficult to keep track of in the interchange shuffle – every component that makes up the car is subject to the same. For this reason, component serialization has also been a major focus in intra- and inter-organization asset management.
Many railroads have developed serialization programs in the past, but for various reasons many of these have been discontinued, Josey said. The primary reason is that while a component may have good visibility on one railroad’s property, once it leaves the property, that visibility is gone. Once again, interchange makes this a difficult problem to solve locally.
One example of the importance of component serialization and tracking is a series of derailments in the early 2000s. These particular derailments turned out to have been caused by defective truck bolsters. In total, 58,000 units were deemed defective and had to be changed out. “There were no barcodes, no serial numbers, no easy way to track down the bolsters. Particularly in the after-market, finding [the bolsters] was hopeless,” said Pat Ameen, Vice President of Industry Relations at Amsted Rail. The ability to locate the defective bolsters might very well have prevented additional derailments in this case.
Similarly, in 2014, there was an early warning issued for a batch of CJ33 wheels due to several broken-plate-caused derailments, Ameen said. However, at this time, Railinc’s component serialization project was underway (though still in its early stages). As a result, there was enough data to significantly narrow the search for the affected wheels. “Without an industry-wide push for serialization and industry-wide collaboration, it would have been another needle-in-a-haystack situation,” Ameen said.
Component serialization and inter-organizational asset management requires tremendous cooperation. Luckily the potential benefits of component serialization are such that car owners, railroads and manufacturers are largely willing and eager to work together to realize those benefits. Railinc’s project is ongoing, but there has been significant progress thus far. About two-thirds of North American heavy-haul wheels are now serialized, Josey said. Large scale serialization projects like these take years to fully realize, but there are already tangible benefits.
As quantity and quality of the data that feeds a centralized hub increases, the accuracy of trending and predictive analysis also improves. Just as there are benefits to sharing data between silos within an organization, there are benefits to sharing information between organizations, particularly in the North American interchange environment.
The industry’s focus on condition monitoring from the component, up to the system level seems to be the result, not only of more mature technologies, but also of more sophisticated maintenance and asset management strategies. And while the proliferation of technologies providing incredibly detailed data seems to suggest an increasingly-complex system, it is only an increasingly-measured system. The experiences presented at WRI 2017 suggest that the key to optimization is not to be overwhelmed by the detail but to embrace it, and use it to better manage the system as a whole.
Jeff Tuzik is Managing Editor of Interface Journal.
All photos courtesy of Jack Lindquist.