Yarn Evenness Reports: Read CVm, Imperfections and Spectrograms

Start with the question behind the report

Two yarn lots can have the same mass CV and still deserve different decisions. One may contain a repeating fault. The other may have scattered small changes with no clear cycle. Their average variation is alike, but the yarn is not arranged in the same way along its length.

A useful yarn evenness test therefore answers several questions. How much does mass vary? How many events cross the chosen limits? Does a fault repeat at a certain distance? How does the result compare with a relevant reference group?

Keep those questions separate while reading the report. A low overall CV does not make an unusual spectrogram peak disappear. A high imperfection count does not, by itself, name the machine part that needs repair. The aim is to connect each result to a specific production or fabric concern.

This guide concerns mass evenness. Hairiness, breaking force and yarn count describe other properties. They can still add useful evidence. Check what each sensor measured, then check which yarn lengths it tested.

Confirm what the instrument measured

A capacitive tester senses an electrical change as yarn passes through it. The response indicates how mass per unit length varies. The tester does not weigh each tiny yarn section on a balance.

ASTM D1425/D1425M-14(2020) covers this indirect method. Its public scope applies to suitable nonconductive textile strands. The guidance also explains why fibre blends need care. Fibres can differ in their electrical properties. A changing blend can therefore affect the apparent mass variation. A signal change need not come from mass alone.

Moisture and specimen preparation also belong in the comparison. Use the conditioning and test atmosphere required by the agreed method. ISO 139 addresses textile conditioning and testing atmospheres, including agreed alternatives. Its public scope does not justify a single conditioning time for every yarn.

Nor is mass variation the same as visible diameter variation. Carvalho and colleagues described a research system with separate sensors. It used both capacitive and optical measurements. Its image route inferred mass from diameter. That model needs assumptions about the yarn; an optical outline is not a direct weighing result.

This matters when one laboratory reports a mass CV and another supplies a diameter CV. Do not treat the two percentages as interchangeable just because their labels both contain “CV.” Ask for the measured property and method first.

Check the actual equipment specification as well. Current Uster Tester 6 technical data describes its capacitive sensor. This measures mass variation and imperfections. Optical functions have separate entries. Do not assume that an evenness tester cannot detect thick and thin places. Nor does every sensor measure hairiness.

Read four parts of the result together

The following questions provide a quick route through a report. They are prompts for interpretation, not acceptance limits.

Read the sample identity before the numbers. Confirm the fibre blend, yarn structure, spinning route and process stage. A cop sample and a wound-package sample are not automatically the same population. If the complaint concerns one position or production period, a report from unrelated packages cannot resolve it.

Keep each specimen result visible during an investigation. An overall average is useful for a summary, but it can hide the one package that produced the disputed fabric. You need both the lot view and a way back to the tested yarn.

Understand CVm and U without using a fixed conversion

CVm expresses the standard deviation of measured mass relative to its mean. U expresses the mean absolute deviation relative to the mean. Both are commonly reported as percentages. The Uster parameter definitions explain both measures. They also distinguish results for different cut lengths.

In compact form:

CVm (%) = 100 × standard deviation of mass / mean mass

U (%) = 100 × mean absolute mass deviation / mean mass

The NIST definition of coefficient of variation uses the same standard-deviation-to-mean principle. Note the calculation convention. A sample estimate often uses a divisor of n − 1 for variance. The complete toy profiles below use n because every value in each invented profile is included. They illustrate the ideas; they do not reproduce a tester's signal processing.

A complete eight-point example

Imagine eight equal-length sections with normalized mass values:

80, 100, 100, 100, 100, 100, 100, 120

The mean is 100. The two nonzero absolute deviations are 20 and 20, so their sum is 40. Dividing by eight gives a mean absolute deviation of 5.

U = 100 × 5 / 100 = 5%

The squared deviations total 20² + 20² = 800. The complete-profile variance is 800 / 8 = 100, and its square root is 10.

CVm = 100 × 10 / 100 = 10%

Here, CVm is twice U. Now consider four values of 90 followed by four values of 110. The mean is still 100, but every absolute deviation is 10. Both U and CVm are 10%.

The ratio has changed, although neither calculation is wrong. A convenient conversion factor may fit a particular assumed distribution. It is not a universal identity that can replace the reported result.

The examples also show why equal CV values deserve a closer look. The first profile has two larger departures and six values at the mean. The second has a smaller departure at every point. CVm alone does not describe that difference.

Preserve the order of the measurements

Suppose each of the following invented values represents the mean mass of one metre. Both profiles contain four low and four high values:

Grouped: 90, 90, 90, 90, 110, 110, 110, 110

Alternating: 90, 110, 90, 110, 90, 110, 90, 110

Both have a mean of 100, U of 10% and CVm of 10%. A summary based only on those numbers cannot distinguish them. Their traces would look quite different.

Now average each adjacent pair of metres. The grouped profile becomes 90, 90, 110, 110. Its CV remains 10%. The alternating profile becomes 100, 100, 100, 100, with a CV of zero at that averaging length.

No fibre moved and no defect was repaired. Averaging simply hid the alternating change. This is why a report's measuring length and cut-length results matter. A value based on longer sections answers a different question from one based on shorter sections.

The example is deliberately small. Real traces contain many more observations. The instrument defines how it processes them. This exercise does not imply a universal one-metre measuring zone. Nor can it recreate a commercial CVm(L) result from a screenshot.

Also separate evenness from mean count. Two ideal strands could each be perfectly uniform within themselves, while one is 20 tex and the other is 21 tex. Both would have zero within-strand CV in that ideal case. Yet the second is 5% coarser by mass per length. A mass-variation percentage does not replace the yarn count calculation.

A package average can hide a count difference

Extend that ideal case with equal numbers of observations from each strand. Each observation covers the same length. The combined mean is 20.5 tex. Every value lies 0.5 tex from that mean. The complete combined profile therefore has a standard deviation of 0.5 tex.

Combined CV = 100 × 0.5 / 20.5 = 2.44%

Yet the average of the two within-strand CV values remains zero. These results answer different questions. One measures variation across all the combined observations, including the count difference. The other summarizes the variation within each strand.

This does not prescribe how a commercial report calculates its summary fields. It shows why their definitions matter. A supplier may send an average CV across packages. Ask where the report shows count variation between those packages. Do not assume it has been included just because the report covers several packages. The fabric receives their actual yarn counts. A laboratory summary does not remove those differences.

Decode imperfections before adding them

An imperfection count depends on the sensitivity level and event definition. The current Uster technical data lists standard thin, thick and nep levels of −50%, +50% and +200% for ring and air-jet yarns. Its standard open-end nep level is +280%. These are identified instrument settings, not universal buyer requirements.

The sign and reference mass matter. A +200% change adds twice the reference amount. The total is then three times the reference. For a reference mass value of 20 units, +200% corresponds to 60 units. It does not mean a total of 40.

A +50% level would correspond to 30 units, while −50% would correspond to 10. These arithmetic examples explain the percentages only. The instrument also applies its event and length rules. You cannot count every plotted point above a line as a separate nep.

Record the thresholds beside the results. A nep count at +200% and another at +280% do not describe the same test condition. Nor can a total IPI be interpreted if its three component settings are missing.

Convert observed counts to a common distance

Suppose an original trial records the following events over 1.5 km. All results use the same agreed settings.

Count per km = observed count / tested kilometres

The total is 18 per km, not 27 per km. The number 27 belongs to the full tested length. If another report uses a different length, compare normalized rates while retaining the raw counts and exposure.

Keep the three components as well as their sum. A total dominated by neps suggests a different line of inquiry from one dominated by thin places. The total alone discards that detail. It also cannot show whether events were scattered across many packages or clustered in one.

For a combined rate, add the events and the tested distances from all specimens. Then divide the total count by the total distance. Simply averaging their per-kilometre rates gives each specimen equal influence. This ignores differences in exposure. An equal-weight package summary may be useful, but label it clearly.

Raw counts can even reverse an apparent ranking. Suppose sample A has 36 events over 2 km, while sample B has 18 events over 0.5 km. A has twice as many observed events. Its rate, however, is 18/km; B's rate is 36/km. On this defined measure, B has twice the event rate.

Neither result alone establishes a lot's long-run rate. These are the events seen in the specified lengths. Before comparing them, confirm that the same categories and thresholds were used. Then examine how the specimens were selected. A faulty package chosen for study can answer one useful question. A random delivery sample can answer another. They do not provide equivalent estimates of their lots.

Test length is evidence, not a setting to overlook

At a constant test speed of 300 m/min, two minutes of valid running covers 600 m, or 0.6 km. Ten observed events would therefore give:

10 / 0.6 = 16.67 events/km

Use the length actually measured. Do not include threading, stopped time or a rejected run merely because it occupied part of the session. A test report should make its exposure clear enough for someone else to check the rate.

Longer exposure can reveal events that a short sample misses. A long length from one package still represents that package, though. Test length and coverage of the production lot solve different sampling problems.

A useful example appears in Uster's advanced-classification case. Routine imperfection results did not explain the fabric fault. They did not show the needed difference between yarn groups. More detailed classification of long faults supplied further evidence. The case does not show that laboratory testers cannot detect thick places.

For a rare fault complaint, ask whether the test was likely to encounter the disputed feature at all. Retain the affected yarn or fabric location where possible. A clean sampled length supports a statement about that test. It cannot prove that every metre of the delivery is fault-free.

Read a spectrogram as a pattern clue

A trace shows variation in its original sequence. Spectral analysis asks how strongly the signal varies at different cycles. NIST's spectral-plot explanation describes this general search for cyclic structure. Yarn instruments commonly plot the result against wavelength. Read the actual axes; do not assume a generic frequency plot.

Wavelength is the distance along the yarn over which a pattern repeats. A concentrated peak can suggest a repeated action. It does not identify a particular roller without further evidence. More than one source can produce a similar spacing, and later processing can change that spacing.

Check whether the horizontal scale is logarithmic. Equal distances across such a chart do not represent equal additions in wavelength. Check the vertical scale too before judging one peak as twice another from its printed height.

Return to the original trace when a peak seems surprising. Check whether the unusual section covers the whole run or a short portion of it. Look for a sudden start, a step in the mean, or a repeated change. Preserve the untrimmed run alongside any enlarged section. Cropping and rescaling can make a feature easier to see. They can also hide whether it is local or persists through the run.

Work from a process cycle to a possible wavelength

Consider an invented stage delivering strand at 12 m/min. Suppose a disturbance acts 80 times per minute and creates one mark per cycle.

Spacing at that stage = 12 / 80 = 0.15 m

Each cycle lasts 60 / 80 = 0.75 seconds. During that interval, the strand advances 0.15 m. Both calculations describe the same physical spacing.

If the marked strand then receives a uniform draft of 20, the ideal final spacing becomes:

Final wavelength = 0.15 × 20 = 3 m

This estimate assumes the mark survives, the draft is uniform and no slip changes the spacing. It is a hypothesis for investigation. Check the speed used in the first calculation. If it already describes the final yarn, do not apply the downstream draft again.

The tester's speed is a separate matter. Scan the finished three-metre pattern at 300 m/min. It would pass the sensor 100 times per minute. The pattern remains three metres long. A change in scanning speed changes how often it passes. The spacing already present in the yarn stays the same.

Keep laboratory running speed separate from the production speed that created the fault. Confusing them can point the investigation towards the wrong cause. Trace the suspected feature from its point of origin through each later draft.

Confirm the cause beyond the chart

In Uster's roving-front-roller case, a spectrogram and related evidence directed attention towards roving. Physical inspection found a defective front top roller. The investigation then checked the result after replacement. The confirmed cause came from that chain of evidence, not from a peak label alone.

Apply the same logic without assuming the same faulty part. Compare the observed spacing with plausible cycles. Inspect the relevant position. After a justified repair, repeat the test under comparable conditions and check the affected product. A disappearing peak is useful evidence, but it should agree with the original complaint.

Use percentiles to frame a decision

Uster Statistics provides industry benchmarking across defined textile groups. A percentile locates a result within a reference distribution. It is not a percentage of defective yarn and does not state the probability that a fabric will fail.

Uster's 2024 practical case on statistical comparisons explains the 50% reference. It marks a midpoint, with half of the benchmark population on each side. The examples also show why material and process categories matter.

For a property where lower values are preferred, a lower numerical result may occupy a better benchmark position. Even then, the comparison is useful only when the reference group fits the yarn. Check material, count, production route, process stage, end use and metric, along with the data edition.

A percentile does not become a purchase limit by itself. A buyer may choose a benchmark-based requirement, but the contract must identify it. A laboratory's coloured warning can also reflect a configured limit. Ask what boundary produced the colour before treating it as an industry rule.

No live benchmark values are assumed in the examples here. Obtain the applicable chart through the service and retain its category details with the comparison. Do not use an old chart's publication year as evidence that it is the latest available dataset.

Turn the report into a focused investigation

Start with the observed problem. If a fabric shows a repeated band, retain its direction and spacing and compare them with the yarn evidence. If the concern is isolated thick faults, examine event classes and exposure. If average count is wrong, a low CV cannot answer that complaint.

Next, rule out a mismatch in measurement. ASTM's public guidance identifies several conditions that affect comparison. These include the statistic, measuring zone, total length and speed. Climate and specimen preparation also matter. Align these conditions before blaming production for a difference. The full agreed method and instrument instructions still govern the laboratory procedure.

Then choose a test that can separate the competing explanations. For a suspected position fault, compare yarn from that position with matched unaffected positions. For a stage effect, retain linked material before and after that stage. Select enough coverage for the decision rather than making the easiest available package stand for the whole lot.

Record predictions before changing equipment. If a suspected repeating action is responsible, what wavelength should appear after the later drafts? Which trace or event class should improve when the cause is removed? A clear prediction makes the follow-up more informative than a general search for a lower number.

Over time, comparable results can support statistical process control. Keep process-change evidence separate from the buyer's acceptance decision. A stable process can still miss a requirement, while a single acceptable result may conceal a developing fault.

The most useful report preserves the connection between sample, settings, signal and product. Read the CV, counts, trace and benchmark together. Make the narrowest decision that the evidence supports. Where the cause remains uncertain, choose a specific follow-up to resolve it.

References

• ASTM D1425/D1425M-14(2020): Evenness of textile strands using capacitance testing equipment, public scope and significance.

• ISO 139:2005: Standard atmospheres for conditioning and testing, public scope and current status.

• Carvalho, Gonçalves, Soares, Belsley and Vasconcelos: An Overview Over Yarn Mass Parameterization Methods, SENSORDEVICES 2011.

• Uster Tester 6 technical data, April 2026.

• Uster SE-562: Description of quality parameters measured by fibre and yarn testing equipment, 2014 definitions.

• NIST: Coefficient of variation.

• Uster practical case 01: Advanced classification of long yarn faults.

• NIST Engineering Statistics Handbook: Spectral plot.

• Uster practical case 13: Roving-frame front-roller fault investigation.

• Uster Statistics: Official service overview.

• Uster practical case 20: Statistical comparisons across process and material groups, 2024.