TSB R23E0029: a 1991 railcar wheel, dirty steel and a shattered rim fatigue failure

On 24 September 2026 the Transportation Safety Board of Canada (TSB) released report R23E0029. It covers the derailment of a Canadian National grain train near Uncas, Alberta, on 5 April 2023. Eighteen cars left the north main track and grain spilled from several of them. No injuries were reported.

The cause was a single wheel. The L4 wheel of car FURX 841817 had a shattered rim failure and lost part of its outboard rim and tread. It then hammered the rail on every revolution. Impact marks repeated about every 2.7 to 3.0 m (9 to 10 ft), matching the circumference of a 914 mm (36 in) wheel. The repeated blows broke the rail in several places until the train derailed.

What the investigation found

The TSB laboratory found beach marks on the fracture, confirming a fatigue crack that initiated and grew in service. Metallography showed manganese sulfide and oxide inclusions with scattered voids. The wheel was made in 1991, before the Association of American Railroads (AAR) introduced steel cleanliness standards for new wheels in 2003.

Test	Result
Current AAR limit (worst field, oxides plus voids)	0.75%
Sample 1	4.29% (5.7 times the limit)
Sample 3	2.74% (3.6 times the limit)
Other 4 samples	Within limits; sulfides within specification in all 6
Hardness (Brinell and Rockwell C)	Within AAR limits

The TSB notes the samples came from deeper than the standard location because of rim wear, so they give context rather than a formal acceptance test. Its finding is that the imperfections acted as stress raisers, subsurface cracks formed and grew to the tread, and the rim failed.

Detection failed at three levels. Visual inspections before the derailment found nothing. Ultrasonic testing can only be done at manufacture or when a wheel is removed for reprofiling. Current practice (RP-631) requires radial scans only, which can miss defects in some orientations. Wayside wheel impact load detectors (WILDs) did see the wheel getting worse, as its peak impacts rose at successive sites. On the day of the accident, a reading of about 496 kN (111.6 kip) triggered a replacement code. Under CN’s policy the wheel was due to be changed at Edmonton, about 31 km (19 miles) further on. It failed first.

The TSB says WILD data is judged one pass at a time against thresholds such as the 90 kip (about 400 kN) trigger in AAR Rule 41. Its exploratory analysis of this and three other broken-rim derailments showed trend-based, multi-metric screening could flag such wheels earlier. Recommendation R26-02 asks Transport Canada and industry to develop it.

Why it matters for design engineers

Material cleanliness is a fatigue design input, not only a procurement line. An inclusion is a notch you cannot see or dimension. Under rolling contact the highest shear stress sits below the surface, which is where these cracks started. Hardness was within limits, so a hardness check alone would not have flagged this wheel.

The monitoring lesson carries over to any fleet. A threshold tells you when a part is bad. A rate of change tells you when it is going bad fast. This wheel was flagged and scheduled, and it still failed on the way to the shop.

If you	Check
Specify steel for a rolling-contact or high-cycle part	Cleanliness limits (inclusion rating, oxide and void content), not only hardness and tensile strength
Assess fatigue at a geometric notch	$K_t$ for the notch, but also the size of the largest likely internal defect
Keep legacy parts in service after rework	Whether they were made to an older, weaker material or inspection standard
Rely on NDT at overhaul only	Whether the crack growth time can be shorter than the inspection interval
Monitor a condition signal against a limit	Trend and rate of change, and how long it takes to get the part to a place where it can be replaced