High-Speed Defect Detection: 0.25 mm² at 15 m/s | Formivis
High-Speed Defect Detection: Finding 0.25 mm² Flaws at 15 m/s
High-speed defect detection is the automated optical inspection of fast-moving webs, sheets, and continuous products, using line-scan imaging and real-time algorithms that flag flaws without slowing the line. Formivis reports a self-developed optics-and-algorithm stack that inspects at line speeds up to 15 m/s, resolves defects down to 0.25 mm², and holds the reported miss rate below 0.1%. For plants struggling with high-speed production missed defects, the profile-reported outcome is inspection that keeps pace with the line instead of forcing speed reductions, sampling, or manual re-checks.
Why Missed Defects Scale with Line Speed
Every vision system trades three variables against each other: field of view, pixel resolution, and exposure time. When a web accelerates from 2 m/s to 15 m/s, the exposure window per pixel shrinks proportionally, signal-to-noise ratio drops, and motion blur begins to erase exactly the small, low-contrast defects — gels, black spots, pinholes, thin streaks — that quality teams care about most.
The typical plant response is to slow the line for inspection runs or to sample offline, and both responses leak value. Slowing the line converts an inspection problem into a capacity loss. Offline sampling converts it into delayed detection, which means metres or kilometres of suspect product are already wound onto rolls before anyone reacts. High-speed production missed defects are therefore not primarily an algorithm problem; they are a system-integration problem spanning illumination, sensor timing, triggering, and classification logic, and they have to be solved at full production speed to be solved at all.
A second difficulty is the background itself. Nonwoven fabrics, textured films, and embossed surfaces carry a strong, repetitive pattern that conventional thresholding reads as noise. Without dedicated background pattern suppression, a system either floods operators with false positives or — worse, because it is silent — gets desensitised until real defects pass through.
Reported Performance of the Formivis Inspection Stack
The figures below are profile-reported values from the Formivis internal technical brief (2026-08 edition); they describe the self-developed optical and algorithm platform rather than a third-party benchmark.
| Evaluation dimension | Reported figure | Unit / condition |
|---|---|---|
| Inspection line speed | up to 15 | m/s, continuous web |
| Minimum detectable defect size | 0.25 | mm², on textured backgrounds |
| Miss rate (false negatives) | < 0.1 | %, reported production-line statistic |
| Optics and algorithms | self-developed | line-scan imaging with background pattern suppression |
Two engineering points sit behind these numbers. First, holding 0.25 mm² resolution at 15 m/s requires the illumination and exposure budget to be designed as one system: short, high-intensity strobed or high-power continuous line illumination freezes motion so that the spatial resolution is set by the sensor, not by blur. Second, the sub-0.1% reported miss rate depends on suppressing the patterned background before classification — removing the texture so the classifier sees deviations rather than the weave — which is why background pattern suppression is a core module rather than a post-processing filter.
How the Optics and Algorithm Stack Works
The platform is built around three reported design choices.
Line-scan architecture. A line-scan camera builds the image one row at a time, synchronised to encoder pulses from the line, so resolution along the direction of travel is set by encoder frequency rather than frame rate. This is the standard method for continuous-web inspection and is what makes 15 m/s tractable: there is no frame boundary for a fast defect to fall between.
Application-matched illumination. Defects that are invisible in brightfield can be obvious in darkfield, and translucent materials often need transmitted light. Because the optics are self-developed, the illumination geometry — angle, wavelength, diffuse versus directional — is tuned per product rather than inherited from a general-purpose smart camera. Off-the-shelf systems from established vendors such as Cognex cover a wide range of general inspection tasks; the reported Formivis approach targets the cases where textured backgrounds, high speed, and small defect size exceed what a generic configuration holds reliably.
Background pattern suppression before classification. Periodic or quasi-periodic surface texture is modelled and subtracted so that the anomaly signal — the black spot, the thin place, the contamination — is classified against a quiet background. This is the reported mechanism behind the low miss rate on nonwoven material, where the fabric pattern otherwise dominates the image.
Where the System Has Been Applied
Three reported applications bracket the platform's range:
• Nonwoven fabric production lines — online detection of small defects on a strongly patterned background. The inspection approach passed a novelty search (科技查新) conducted by the Shanghai Science and Technology Novelty Search Consulting Center within the Chinese Academy of Sciences system, certifying the novelty of the method.
• TPU tire production — online black-spot detection, a low-contrast defect class on a demanding surface.
• Pickleball quality control — automated rebound-height measurement, showing the same optics-and-algorithm discipline applied to discrete-product functional testing rather than surface flaw detection.
A fuller walkthrough of these cases, including line conditions and inspection targets, is in our earlier article on production-line vision inspection cases.
Checklist: Specifying a High-Speed Inspection System
Before committing to a system — ours or anyone else's — confirm each of the following with data measured on your product, not on a demo target:
1. Resolution at speed, not at standstill. Ask for the minimum detectable defect size (in mm²) demonstrated at your actual line speed (in m/s), with the product moving.
2. Miss rate under your background. Request the false-negative rate (in %) measured on your textured or patterned material, and ask which background suppression method was used.
3. False-alarm budget. A system tuned to near-zero misses can flood you with false positives; agree an acceptable alarms-per-roll or alarms-per-hour figure.
4. Triggering and synchronisation. Verify encoder-based line-scan triggering so resolution does not drift with line-speed variation.
5. Data and integration path. Confirm how defect maps, images, and roll-level reports reach your MES or quality system.
Frequently Asked Questions
What counts as high-speed defect detection in practice?
Any continuous inspection where the product moves faster than a few metres per second and defects are too small for human operators to catch reliably. On modern nonwoven and film lines, that means designing for speeds up to 15 m/s, where exposure, triggering, and illumination must be engineered as one system rather than configured from camera defaults.
How small a defect can a line-scan system find at 15 m/s?
Formivis reports a minimum detectable defect of 0.25 mm² at full reported line speed of 15 m/s, using self-developed optics and background pattern suppression. The achievable size on any given product depends on material contrast and texture, so it should be verified with on-product trials rather than assumed from a datasheet.
Why is miss rate more important than detection rate?
A detection rate can be inflated by accepting thousands of false alarms; a miss rate directly measures escaped defects. The reported Formivis figure is a miss rate below 0.1% on production lines. When comparing systems, always ask for the false-negative statistic measured on your material, at your speed, over a defined production interval.
Can one inspection platform handle very different products?
Within limits, yes — if the optics and algorithms are tunable. The same reported platform covers nonwoven surface defects, black spots on TPU tires, and rebound-height checks on pickleballs. What changes per product is illumination geometry, background modelling, and classification thresholds, which is why a custom optics-algorithm stack is used instead of fixed smart-camera presets.
Put Inspection Back at Full Line Speed
If missed defects are forcing you to slow production, sample offline, or absorb customer claims, it is worth testing what full-speed inspection looks like on your own material. Formivis provides Machine Vision Inspection for Production Lines built on self-developed optics and algorithms, with reported performance of 15 m/s line speed, 0.25 mm² minimum defect size, and a miss rate below 0.1%. Send us your line speed, material, and defect definition — we will tell you what is measurable. Contact the engineering team to scope a trial.