At a remote siding in single-track territory, a crew brings train 498 to a halt. A wayside hot bearing detector has sensed a wheel bearing on a flatcar loaded with lumber that’s getting hotter, and has now exceeded tolerances. The car cannot safely continue as part of the train. As the crew begins the work of removing it, a grain train traveling the opposite direction is waiting at a siding. An intermodal train following No. 498 eases to a stop to wait out the operation.
In this scenario, three trains are delayed by an hour. The customer’s lumber shipment, needed to keep a distant lumberyard stocked, will sit for at least a day. And a mechanical truck and crew have to be deployed from the nearest terminal to go to the location to make the fix.
There’s a better way. CP has patented it, and other railroads are paying attention.
“We can predict, using an acoustic detector, when the bearing is going to fail,” says Kyle Mulligan, CP’s Assistant Chief Mechanical Engineer, who holds a Ph.D. in predictive analytics. “The model is able to see three months in advance.”
Railroaders sometimes refer to cars in the early stages of bearing failure as growlers due to the whining sound they make while in transit. Such cars are not yet at risk of suffering an overheated bearing, but some will go on to develop that condition. As early as 2003, CP’s mechanical engineers were looking at this, and began using acoustic detectors to identify impending failures. Loaded cars flagged by acoustic detectors could proceed to destination, be unloaded, than forwarded empty to an appropriate car shop for repairs.
This plan had the potential to dramatically reduce service failures. Problem was, the detectors sensed too many impending failures. Bearing replacements shot up, taxing the capabilities of the mechanical department. If the system was working properly, the number of bearing replacements should have stayed the same; the only difference should have been the ability to identify failure earlier in the process. So we returned to relying on hot bearing detectors.
“We can predict, using an acoustic detector, when a bearing is going to fail”
The reason we’ve been able to make it work now, Kyle says, is that his team has created an algorithm to analyze acoustic detector results to more precisely identify which bearings will fail, and when. “Around 92 percent of them are valid repairs,” he says.
The algorithm came through careful study of acoustic detector data. Kyle, along with CP engineer in training Solange de Blois, analyzed the precise sonic patterns that emanated from bearings as they passed acoustic detectors. Then, they watched those same bearings as they passed over infrared detectors in subsequent days, weeks and months. Through this process, they were able to identify which sonic signatures were precursors to bearing failure and which ones weren’t. Turns out, the solution wasn’t in developing a flashy new technology, but in gaining a better understanding of data we were already collecting.
Prior to joining CP, Kyle studied the failure of bearings on gas turbines for the National Research Council of Canada. This research, he says, “was directly applicable to the railroad industry.”
Because acoustic detectors can predict failure so far in advance, there’s little need to saturate CP’s network with them. Currently, CP employs four acoustic detectors at strategic points on our system.
Meanwhile, CP has altered the way we use data from hot bearing detectors. Cars that are showing elevated readings, but are still within tolerances, are now opportunistically sent to repair shops.
Infrared detectors are still necessary as a last resort to catch hot bearings, and are still important in identifying hot wheel treads, but we are relying on them less and less. In the recent past, CP could expect to suffer around 60 service failures per month as a result of hot bearings. Our improved understanding of acoustic detector data and analyzing the trends have reduced that by around 90 percent.
The results of better data analytics don’t stop there. Our existing detectors are now weighing railcars to catch overweight loads, identifying malfunctioning railcar brakes by identifying which wheels are cold, and calling out wheelsets that need to be changed out. All this stems from better understanding the data we were already collecting from wayside detectors.
Our existing detectors are now weighing railcars to catch overweight loads, identifying malfunctioning railcar brakes by identifying which wheels are cold, and calling out wheelsets that need to be changed out.
“We effectively save a lot of money doing this,” Kyle says, “and we still maintain a high standard of safety.”