A process FMEA for tyre tread separation starts from one failure mode: the tread pulling away from the tyre, the definition used in US Federal Motor Vehicle Safety Standard 139. Its causes are traced to the manufacturing steps that weaken adhesion between the tread, the belts and the casing. Each cause is rated for severity, occurrence and detection on the AIAG-VDA 1–10 scales, and ranked by Action Priority rather than a multiplied score. Two places need the most care: the detection rating, because cameras are a poor fit for internal belt delamination of the kind the NTSB found, and the evidence trail from a failed tyre back to its production records.
What a tread separation FMEA has to list
FMVSS 139 does not treat separation as a single defect. It defines a family of them: bead separation (a breakdown of the bond between components in the bead), ply separation (rubber parting between adjacent plies), sidewall separation, innerliner separation, chunking, cracking and tread separation. Keep these as separate rows in the FMEA, even where they share causes, because the endurance test acceptance criteria name them separately.
The mechanism matters as much as the label. In a fatal crash the NTSB investigated in 2014, the separation began as a delamination between the inner and outer steel belts, grew into a thumbnail-shaped pocket along the shoulder, and the tread came off when a second delamination joined it. The NTSB judged from the uneven wear that the pocket had probably been there for at least hundreds, if not thousands, of miles.
For the effects chain, recall filings give concrete wording. Yokohama’s 2018 recall report describes it in three stages: a small separation may produce vibration, the separation can grow into tread detachment if the tyre stays in service, and detachment can cause unpredictable handling and a crash. NHTSA’s EWR compendium defines the tread area broadly, as everything from the tread rubber through the sub-base, belts and skim coat to the body ply and inner liner under the tread.
Which manufacturing causes of tread separation to rate
Recall reports name specific process failures. Three from the NHTSA files show the range.
In Yokohama’s recall 18T-016, a batch of undertread compound failed lot confirmation testing after mixing and was put on hold. The procedure required good material on either side of the nonconforming section to be scrapped as well, but material from one end of the suspect batch was not fully removed and was extruded into undertread. The failure was in removing the suspect material, not in detecting it.
In recall 17T-018, part of the back stitcher on a building machine loosened during production, which may have weakened the bond between the ply cords and the sidewall near the bead. The reported failure mode was sidewall separation (voids) around the bead during endurance testing.
Kumho’s recall 26T009 attributed possible FMVSS 139 noncompliance to performance degradation over time, depending on storage. Ageing also appears in the NTSB’s account: as the skim and wedge rubber oxidise they lose elasticity and adhesion, and cracks form along the belt edges. Underinflation and ambient heat speed up that ageing. Those two are service conditions, not process causes.
The table below sets out how we would structure the cause rows. Occurrence is left to plant data on purpose: AIAG-VDA guidance is that occurrence ratings are checked against current ppm or Cpk figures.
Cause-and-control table for tread separation
| Cause (process step) | Evidence that it happens | Prevention control | Detection control | Detection anchor on the AIAG-VDA scale |
|---|---|---|---|---|
| Out-of-spec undertread or skim compound reaches extrusion | Recall 18T-016 | Lot confirmation testing after mixing; written boundary-scrap rule for held batches | Hardness testing of tyres around the suspect period; recurrence testing at mixing and extrusion | Sample checks sit near 6–7 (“sample visual / dimensional check”, “sample SPC”) |
| Building machine fault weakens ply-to-sidewall adhesion | Recall 17T-018 | Not stated in the recall file | 3D strip-edge scanning on the building drum; internal endurance testing on samples | 100% automated inspection with feedback anchors at 3 |
| Belt-edge ageing and oxidation (design robustness) | NTSB SIR-15-02 | Compound and belt package design | Aged endurance testing beyond FMVSS 139 | Lab test on samples, not a production detection control |
Severity for tread separation leading to detachment belongs at the top of the scale. The AIAG-VDA scale gives 10 to a hazardous failure with no warning that breaks a safety or regulatory requirement. Some separations do warn (vibration that gets worse), so a team may argue for 9, but under Action Priority a severity of 9 or 10 is High priority whatever occurrence and detection say.
Action Priority, not RPN, and why visual detection gets overrated
The 2019 AIAG-VDA harmonised handbook replaced the Risk Priority Number with Action Priority for automotive FMEAs. RPN multiplies three ordinal scales, which is statistically inappropriate, and different S, O and D combinations can produce the same RPN even when their operational meaning is very different. Action Priority instead uses a lookup table that weights severity most heavily and returns High, Medium or Low. For a tyre maker supplying OEMs under IATF 16949, the AIAG-VDA 1–10 scales are the expected baseline, and collapsing or rewording them is a routine source of audit findings and PPAP rejections.
The detection column needs particular scrutiny. ReliaMag’s FMEA guide is blunt about it: visual inspection that catches “most” defects is not D=1, and realistic field data typically puts most visual inspection processes at D=4 or higher. Good practice is to cross-check any detection rating below 5 against the Control Plan’s reaction plan, and compare severity against the customer impact recorded in warranty data.
One judgement worth stating plainly: a detection rating attaches to a cause and failure mode pair, not to a station. A finished-tyre camera can rate well for a sidewall bulge and still be no control for an internal belt delamination. We would rather see two rows with honest ratings than one row with an average.
Ratings also need maintaining. Recommended calibration practice is to anchor each rating with two or three examples from your own scrap, warranty and PPAP history, publish the anchored tables as a controlled document, and twice a year re-rate five random rows from different FMEAs as a group. After a corrective action the FMEA must be re-rated, or nobody can show the action worked. Field failure data captured through FRACAS replaces estimated occurrence with actual failure rates.
Where inline vision inspection can lift the Detection rating
Inline vision earns a better detection rating when it inspects every unit with feedback, on a defect it can image. On the AIAG-VDA scale, “100% automated inspection with feedback” anchors at 3. That gap is real, but its boundary is just as real: inline vision is usually a poor fit when the main failure mode is internal and out of camera view. Belt delamination is internal, forming between the inner and outer steel belts.
So the honest answer has two parts. Vision lifts detection for surface bulges and dents, and for strip-edge errors it can see on the drum. We would cover the internal interfaces with another control, such as X-ray inspection or sampled endurance testing. Both belong in the Control Plan: the FMEA documents whether each control acts on occurrence or detection, and the Control Plan states error-proofing test frequency, with records kept.
Which early signs of tread separation inline inspection can catch
Not all the warning signs named in recall filings exist when a tyre leaves the plant. Kumho’s 26T009 report lists localised tread wear, vibration, noise and bulging in the tread area. Vibration is a sign that appears in service as the separation grows.
Bulges and runout are geometry, which 3D sensors measure inline. LMI Technologies states that its Gocator 3D sensors detect bulges and dents as small as 25 μm in tread and sidewall, and scan strip edges on the building drum for position and defects. That is a vendor specification, not an independent test. For internal defects, peer-reviewed work by Cozma and colleagues with Sailun Group compared engineered texture features with a Random Forest classifier against YOLOv8 for detecting defects in tyre X-ray images.
| Early sign | Present at end of line? | Inline method that can catch it | Rate detection on this basis? |
|---|---|---|---|
| Tread or sidewall bulge | Yes, if already present | 3D laser profiling | Yes, for the bulge itself |
| Strip misplacement during building | Yes, on the drum | 3D strip-edge scanning | Yes, for strip position |
| Defects in tyre X-ray images | Only on X-ray images | X-ray with a trained classifier | Only after validation on the line |
| Belt delamination | Internal, out of camera view | None at the surface | No |
| Vibration | No, it appears in service | None at the plant | No |
Detection also depends on how the system behaves when it is uncertain. A reject policy has to say what happens on a clear fail, a clear pass, low confidence, timeout and system fault, with ambiguous calls sent to hold or manual inspection and never passed silently. Validate on the line rather than only in the lab, measuring false reject and false accept rates and running a parallel period in which the model scores without rejecting. In our view this is what makes an inline vision node a release gate an auditor can rely on, rather than a dashboard, and it is the reliability engineering a detection rating of 3 has to stand on.
How should a tread-separation taxonomy be mapped to FMVSS 139, UNECE R30 and R54 before a pilot is funded?
A defect taxonomy can be traced to the approval test when each class maps to a failure condition the approval test names, and when its inspection evidence can show which classes the system catches and which it does not. TechnoLynx helps with the evidence side through AI Governance and Trust, building approval evidence packs for an external auditor or regulator.
FMVSS 139 gives a clear target for the US. Its acceptance criteria require no visual evidence of tread, sidewall, ply, cord, belt or bead separation, chunking, open splices, cracking or broken cords. A taxonomy whose top-level classes match that list traces directly to the standard.
Passing the regulatory test does not prove robustness. In recall 17T-018, a subject tyre taken from inventory met FMVSS 139 in the manufacturer’s own lab but failed its internal endurance test. NHTSA research summarised by the NTSB found that 27 of 60 FMVSS 139-compliant tires (45 percent) were damaged in endurance and low-pressure testing after accelerated ageing, seven of them seriously.
For Europe, we have not reproduced the test acceptance criteria of UNECE Regulations No. 54 here, and the mapping should be done against the current text of each regulation rather than assumed to match FMVSS 139. Before funding a pilot we would build one mapping table that lists, for each taxonomy class, the FMVSS 139 condition, the R30 or R54 condition, the FMEA row it serves and the detection control that covers it. Gaps in that table are cheaper to find on paper than in a type-approval test.
What NHTSA will expect if a tread-separation recall is investigated
An investigator will expect a defined recall population, a description of the defect and a remedy programme. The Part 573 report has to identify the tires that may contain the defect, explain how the manufacturer set the recall population and how recalled tires differ from similar tires left out, and give part number, line, size, dates of manufacture and plant. It also needs a brief and a detailed description of the defect, including its physical location, and a description of the remedy programme.
Traceability to production records helped define the 18T-016 recall. In 18T-016, five returned tires still had their QR code labels, so the plant could trace their records back to the internal rejection of the undertread compound. The bad compound was used on only one manufacturing day, so the recall covered a single TIN. The TIN identifies a weekly batch, not an individual tyre, and its date code is what finally confirms whether a tyre is in the recall population.
The manufacturer also has to explain how it set the recall population. Yokohama ran recurrence testing at mixing and extrusion, hardness tests on tyres around the suspect period and durability tests on completed tyres, then compared the results with production records. An automated inspection system adds to that file if its decisions, model version and calibration state are logged per tyre and can be queried by TIN and date.
Routine reporting carries on in the background. Tire manufacturers producing or importing more than 15,000 light vehicle and motorcycle tires a year must file quarterly early warning reports. Warranty adjustments are sorted by tire line, size, SKU, plant code, production year and the component alleged to have contributed. Death and injury reports include the TIN and a tire component code. Acceptance of a submission by NHTSA’s Artemis system confirms only its format, not its content, and NHTSA can follow up with a Report Clarification inquiry asking about data collection and reporting procedures.
On the quality-system side, IATF 16949 expects the FMEA to be reviewed before any manufacturing or product change is implemented, and links Control Plan updates to PFMEA updates. Treating approval-grade evidence as an engineering output from day one, rather than writing it up after an incident, is the position we take on regulated inspection work. Our guide to building a defensible evidence pack for audit and procurement review covers how to structure that file.
Recall-readiness checklist for an inspection system
- Every inspection decision is logged with tyre identifier, TIN date code, station, model version and calibration status.
- Each taxonomy class maps to an FMVSS 139 failure condition and to an FMEA row.
- Detection ratings below 5 have a matching reaction plan in the Control Plan.
- Error-proofing test frequencies and their results are recorded.
- Reject policy covers low confidence, timeout and fault states, with no silent pass.
- The FMEA was re-rated after the inspection system went live.
Who can help build the inspection evidence a tyre FMEA audit will check?
TechnoLynx packages inspection evidence in an auditor-readable form through its AI Governance and Trust engagement. TechnoLynx can build the engineering evidence behind an automated inspection system for plants using AI engineering for manufacturing and automotive: validation packages for computer vision workloads and traceability for production decisions.
The limits are just as important. TechnoLynx produces the artefacts an audit needs; it does not give legal advice, regulatory sign-off or certification. Whether to build this in-house depends on whether your team already has computer vision validation and release-gate experience. When choosing a partner, ask to see how detection is reported per defect class, and how fault states are handled, before any accuracy figure. FMEA facilitation and IATF 16949 consulting are a better fit for a team whose FMEA problem is workshop process rather than inspection evidence. FAA approval for automated inspection of aircraft tyres during walk-around checks is a different regulatory path again, outside what this tyre-manufacturing FMEA covers.
FAQ
What does tyre separation mean?
Under FMVSS 139, tread separation means the tread pulling away from the tyre. The standard defines ply, sidewall, bead and innerliner separation separately.
Is tread separation a DOT violation?
A tyre that fails to meet FMVSS 139, or has a safety defect, obliges its manufacturer to initiate a safety recall under federal law. Kumho’s recall 26T009, for example, cited possible FMVSS 139 noncompliance that could lead to tread separation under severe conditions.
