What tools detect rail defects and rolling contact fatigue?

Which signature points to which tool — and what the finding commits you to

The distinction that matters commercially: only the last two rows are consumable decisions on their own. The first three change what the maintenance plan is, and a grinding pass done before the internal condition is known can turn a manageable defect into an urgent one.

The 0.2 mm threshold that links detection to grinding

Repeated wheel loads on the rail surface form a rolling contact fatigue layer through what the literature calls the ratchet effect: within that layer, hardness and brittleness rise sharply and the microstructure develops higher dislocation density with microcrack initiation. Those microcracks then propagate laterally and longitudinally, which is how delamination and eventually rail breakage begin.

Crack development splits into nucleation, initiation and propagation — and nucleation plus initiation consume by far the largest share of the life. That is the whole basis of the preventive grinding strategy: grind while crack propagation depth is still no more than about 0.2 mm, because that is the point at which a normal cut removes the crack array before it becomes a defect. Once the crack has propagated past that, the same wheel removes the surface but leaves the crack root in place.

The intervals are then set by traffic, not by the calendar — for example, on Chinese high-speed line, one grinding pass per 30–50 million tonnes of traffic and no more than two years apart; on conventional line, one pass per 100 million tonnes on straight and large-radius track, tightening to 30–50 million tonnes below 1,200 m radius. Preventive passes are shallower than corrective ones, which is precisely why they extend rail life and cost less than repairing the damage after the fact.

Case and data — what a defect detection programme has to stay inside

Detection is only useful if the corrective pass itself leaves acceptable metal behind. The standard for that is measurable: surface roughness Ra ≤ 10 μm, no blue burning, flatness within 0.3 mm under a 1 m straightedge, and grinding depth held to ≤ 0.5 mm with a finished profile blending out at better than 1‰ so no step is left at the end of the cut.

Get the parameters wrong and detection's work is undone: an excessively hard, brittle white layer of about 870 HV0.3 forms against roughly 340 HV0.3 for the parent metal — a 155% increase — and that layer fractures in service and drives new crack initiation. In the same test series, surface roughness moved from 6.6 to 8.4 μm as grinding speed rose from 1,000 to 3,000 rpm, while white-layer thickness grew from about 45 μm to 320 μm. Speed and depth are defect-prevention parameters, not productivity knobs.

Where detection fed a Molaton specification, the records are explicit. On the Liuzhou network a GMC-96X removed 18.72 pass-kilometres in a two-hour possession with 20–30 mm of wheel consumption against 43.5–59 mm for imported wheels. On the Hewu high-speed line, verified comparison gave 4.28 versus 3.27 pass-kilometres per millimetre, a best wheel of 214.22 pass-kilometres and no continuous blue burning; independent testing at the China Academy of Railway Sciences measured wear 1.4–2.6× better than specification, and a G1b 260×90×153 mm 50 m/s wheel survived 4,775 rpm for 30 s without breaking, with unbalance at 21 g. On the Loram DM01 machine tested in 2026, inspection of the finished surface returned no blue burning and roughness of Ra 1.05–9.0 μm, mostly 2–4 μm. Correcting turnout profiles and verifying them cut car-body lateral acceleration from about 0.20 m/s² to 0.04–0.08 m/s². RailwayCare has manufactured these wheels since 2004 — the first dedicated producer in China — to the standard JB/T 11431 the company drafted.