A narrow band across a woven roll may point to a problem that lasted only a few picks. Changing the running pick density can hide that band in one trial and create a new fault in the next. The first task is to connect the cloth to what happened as the loom stopped, stood still and restarted.

Stop marks and starting marks are unwanted changes in fabric appearance around these events. They are different from a coloured thread or other marker used to locate a trial. Use those terms clearly in fault reports. The marker locates a place; the defect is the unwanted result at that place.

This guide develops a practical investigation for woven fabric. All numerical records are invented examples. They show how to measure and compare the cloth. They give no required loom settings, stopping times or acceptance limits.

Describe the band before deciding what caused it

Begin with the cloth under a defined viewing condition. Note the face, direction of travel, width position and state of the fabric. Cloth held under tension on the loom may change its appearance when it relaxes or is finished. Preserve both observations if both matter to the product.

An open-looking band may contain unusually wide spaces between picks. A dense-looking band may contain compressed or crowded picks. An uneven or wavy band needs examination across the width as well as along the length. These descriptions help locate the change, but they do not identify a unique setting error.

Look closely enough to distinguish spacing from yarn identity. Check for the wrong weft, missing or damaged picks, and changes in colour or yarn thickness. Any of these may alter the appearance. Record which yarns are present before calling every visible line a restart mark.

In their 1997 study of start-up marks, Wimalaweera and Tang examined several weaving factors. They assessed the marks through changes in pick spacing. That supports measuring the local structure. It does not make one spacing limit suitable for every fabric.

Keep an acceptable nearby section as a reference. It should match the construction, yarn lot and measured cloth state. A reference from another roll may carry a different baseline and make the local change look larger or smaller than it is.

Reconstruct the event in the cloth

The stop code is a useful lead. It records what the machine detected, such as a warp or filling fault. It does not prove that the detector identified the first cause or that the repair restored the correct fabric sequence.

Build a short event record around one known stop. Connect the loom identity and time to a marked section of cloth. Try to identify the last retained pick before the stop and the first one after restart. Use the approved tracking method to establish those picks. Record uncertainty instead of assigning a boundary by eye.

Repairs complicate this record. A pick may be removed or the pattern moved backwards. The machine may also make a controlled movement that leaves no new pick in the cloth. A raw shaft-turn count therefore cannot always identify the number or order of retained picks.

Keep the removed-pick and repair history beside the sample map. A later photograph may show a wide interval. Was a pick inserted badly or removed, or did its spacing change? The repair record helps answer that question. Those possibilities require different follow-up work.

Use a marking method approved for the equipment and product. Do not place a hand, loose tool or improvised marker into a moving shed. An identifying yarn may alter the structure. Keep that area out of the baseline measurement and record how the yarn was introduced.

The UK HSE guidance for weaving machinery addresses guarding, safe access and isolation for repairs. It is UK guidance, not a statement of Bangladesh law. Local requirements and the maker's safe procedures govern the actual work, including any planned stop used in a trial.

Measure the local spacing that an average can hide

A general thread-count result is useful, but it can average across a narrow band. Keep a local spacing record as well when that is the fault under review. The measuring direction must follow the fabric length, across the weft rows, rather than along a weft yarn.

Define the points being measured. In the following example, spacing means the distance from one pick centre to the next. It does not mean the visible gap between the yarn edges. A change in yarn thickness can alter that visible gap even if centre spacing stays constant.

Suppose an acceptable section has a reference centre spacing of 0.40 mm. For evenly spaced picks, its reciprocal corresponds to 10 ÷ 0.40 = 25 picks/cm. The factor 10 converts the millimetre interval to a centimetre basis.

Now label the last retained pick before one stop as 0. Positive labels identify successive retained picks after restart. Negative labels identify picks made before the stop. A hypothetical image supplies the following eight intervals between nine visible pick centres.

Interval between labelled picks

Measured spacing

Change from 0.40 mm

−3 to −2

0.40 mm

0%

−2 to −1

0.40 mm

0%

−1 to 0

0.40 mm

0%

0 to +1

0.52 mm

+30%

+1 to +2

0.48 mm

+20%

+2 to +3

0.36 mm

−10%

+3 to +4

0.32 mm

−20%

+4 to +5

0.40 mm

0%

For each interval, calculate:

Spacing change = (measured spacing − reference spacing) ÷ reference spacing × 100.

The first post-stop interval is 30% wider than the reference. Later intervals become narrower. A report containing only the word “open” would lose part of this sequence.

The eight intervals total 3.28 mm. Their mean spacing is 3.28 ÷ 8 = 0.41 mm. The reciprocal of that mean gives an interval-based density of 10 ÷ 0.41 = 24.39 picks/cm over the measured span.

That overall value is close to the reference compared with the largest local change. For contrast, the reciprocals of the single 0.52 mm and 0.32 mm spacings are about 19.23 and 31.25 picks/cm. These are ways to express local pitch; they are not separate standard thread-count tests.

Do not divide nine by the first-to-last centre distance. Nine centres define eight intervals. There are two ways to handle reciprocals: average the individual reciprocal values, or take the reciprocal of the average spacing. These operations give different results. Do not present one as the other.

Do not let opposite changes cancel

The reference length for eight intervals is 8 × 0.40 = 3.20 mm. The observed span is only 0.08 mm longer. Relative to the reference span, this is 2.5%, even though one local interval is 30% wider.

The signed differences partly cancel. The wider intervals add 0.12 mm and 0.08 mm. The narrower ones subtract 0.04 mm and 0.08 mm. If the aim is to describe disturbance in either direction, their absolute differences total 0.32 mm. Dividing by eight gives 0.04 mm, or 10% of the reference spacing.

Thus the same profile has a 2.5% mean signed change, a 10% mean absolute change and a 30% maximum absolute change. Each statistic is correct, but each describes a different feature. Select the outcome before comparing trials, keep its name beside the value, and retain the full profile for interpretation. None of these invented values is an acceptance limit.

ISO 7211-2:2024 covers methods for counting threads per unit length in woven fabric. Use the full applicable method for a formal result. The image exercise above adds a separate record to help diagnose the fault. Its measurement rules are stated in the example.

Keep the image measurement honest

A photograph needs a scale that applies in the specimen plane. Digital zoom or a larger screen image does not create finer measurement. Keep the original file, the calibration record and the point selections so another person can check the result.

For a simple illustration, suppose a calibrated image contains 200 pixels per millimetre. One pixel then represents 0.005 mm. Assume each of the two selected pick centres can be wrong by one pixel in either direction. Their interval can then be wrong by as much as two pixels, or 0.010 mm.

That is a bound for this one assumed point-selection error. It is not the total measurement uncertainty. Other effects may come from the scale, lens distortion or movement of the cloth. The rule used to locate a yarn centre also matters.

Relative to the example's exact 0.40 mm reference, 0.010 mm is 2.5%. A change of only a few percent may need closer measurement. Do not attribute it to a setting until the measurement supports that conclusion. In real work, include uncertainty in the reference spacing too.

Use several width positions when the band is uneven. Keep their identities rather than pooling them immediately. An edge result and a centre result can show different responses to the same stop. Repeated measurements on one photograph improve the check of reading consistency; they do not create new production trials.

Check how far the band extends

Trace the same affected picks across the cloth where the structure allows it. A band can be clear at the centre and hard to see near an edge. That does not establish that the edge is unaffected; the local structure and viewing condition may conceal the change.

Keep a width map with each close view. Mark the measurement position and use the same pick labels throughout. If one image starts two picks later than another, their largest deviations may belong to different parts of the event.

For a curved or skewed line, measure spacing in a consistent local direction. A diagonal image distance can differ from the intended lengthwise spacing. Flatten and support the specimen under the agreed procedure without stretching the defect away. Record any condition that prevents a reliable measurement.

This map helps set the investigation boundary. It can show whether the same sequence extends across the width or whether a local region needs separate attention. It does not, by itself, prove that a temple, frame or tension setting caused the fault.

Define how long the loom was actually stopped

“A ninety-second stop” can mean several things. One log may measure from the fault signal to the start command. Another may measure from zero main-shaft speed to resumed movement. A third may end when accepted production resumes. Choose the definition that matches the question.

Consider an invented event. The stop request occurs at 14:00:00. The main shaft reaches rest at 14:00:02, resumes movement at 14:01:32, and the first retained pick is recorded at 14:01:34. The stationary-main-shaft interval is 90 seconds, while the request-to-first-retained-pick interval is 94 seconds.

If repair work lasts from 14:00:30 to 14:01:20, its duration is 50 seconds. That leaves 44 seconds elsewhere in the longer event interval. It would be wrong to call all 44 seconds avoidable waiting: the total also includes stopping and restarting transitions.

Record any controlled let-off, take-up or pattern movement during the main-shaft standstill. A stationary main shaft does not prove that every part of the yarn and cloth system stayed fixed. Likewise, a restart command does not prove that the first pick was accepted.

These distinctions help explain a change in results. A faster repair response and a different restart setting are separate interventions. If both change together, the cloth may improve without showing which intervention mattered.

Separate material relaxation from machine response

The cloth fell is the point where the next weft becomes part of the woven fabric. Its position and the yarn loads can change during a stop. On restart, the drives and yarn system must return to a suitable forming condition.

Ahmed, Alimuzzaman and Haque's 2020 study examined starting marks in nine fabric constructions on an air-jet loom. It links the marks to changes in pick spacing and cloth-fell position, material relaxation and the restart response. The public abstract describes trials with several settings. It also names the projection microscope used to measure the marks.

Use that evidence to frame the investigation. It does not establish one best shed angle or warp tension for every construction. If a trial changes several settings, the result belongs to that combination. Separate their effects before crediting one change with the improvement.

The stopped material and the starting machine deserve different observations. For the material, ask whether yarn and cloth position or load changes during the standstill. For the machine, ask whether the first forming cycles differ from stable running. Time-linked measurements help connect those observations.

An older 1970 study by Inui and Kurata studied set marks on a water-jet loom. It found effects from both the tension curve and changes in race bending. This is evidence from that equipment, not a modern repair prescription. It is a useful warning against explaining every band through warp tension alone.

Changing the shed also changes yarn paths and insertion clearance. Before making that change, identify the suspected mechanism and the possible side effects. A band may become less visible while end damage, insertion faults or other fabric defects increase.

Compare a candidate setting at matched stop durations

A useful trial asks a narrow question. For example, test one approved change to the restart control. Does it reduce local spacing disturbance after both short and longer standstills? Hold the construction and the other relevant settings constant as far as the equipment permits.

Suppose the team compares baseline A with candidate B. For illustration only, the planned main-shaft standstills are 30 seconds and 180 seconds. These durations represent two trial conditions, not recommended production delays.

The chosen outcome is the largest absolute spacing change in a defined local window. Use twelve intervals before the stop and twelve after it in this illustration. The latter group includes the interval from pick 0 to +1. Every specimen uses that same window and a matched stable-running reference. Keep the signed spacing profile too. The maximum absolute value alone does not show which intervals became wider or narrower.

There are three planned blocks. Each block contains all four combinations: A at each duration and B at each duration. That gives 2 settings × 2 durations × 3 blocks = 12 separate stop-and-restart events. Restore the defined running condition between events and retain the actual duration for each one.

NIST's guidance on blocked experiments explains how to group comparable trials. Such blocks can limit the effects of known nuisance factors. In this example, a block might be a short production period with the same material. Plan the order within blocks to limit time trends, using random order where it is practical and permitted.

The following invented observations are percentages for the defined local-spacing outcome.

Setting and standstill

Block 1

Block 2

Block 3

Mean

A, 30 seconds

14%

16%

15%

15%

B, 30 seconds

9%

11%

10%

10%

A, 180 seconds

29%

31%

30%

30%

B, 180 seconds

20%

22%

21%

21%

At 30 seconds, the mean changes from 15% to 10%, an improvement of 5 percentage points. Relative to A's value, that is 33.3%. At 180 seconds, the change from 30% to 21% is 9 percentage points, or 30.0% relative to A.

The candidate's numerical effect is therefore different across the two durations. Pooling all observations immediately would hide that detail. With this balanced example, A's overall mean is 22.5% and B's is 15.5%, a difference of 7 percentage points. Report the separate conditions beside any combined number.

The samples are small and invented. They establish neither statistical significance nor a suitable size for a production trial. They show how to avoid comparing a short-stop baseline with a long-stop candidate. An actual study needs suitable replication, a planned analysis and enough coverage of the conditions that matter.

Handle exceptions without quietly changing the question

Suppose one event lasts much longer than its assigned duration because the repair takes extra work. Keep the event and its explanation. Follow the agreed analysis plan. The event might stay in the comparison, enter a separate duration group, or need a follow-up trial.

A missing specimen is missing evidence. It should not become a zero-defect result. A specimen with the wrong weft or a failed repair may answer a different question from a clean restart. Record that state explicitly and retain the reason for any exclusion.

Check whether the measuring rule has changed. A wider image window includes more intervals. This gives its maximum deviation more chances to capture an extreme value. If one result uses the centre and another uses the edge, a difference may come from position. Match the window and location before comparing the numbers.

Keep stop frequency separate from damage per stop. In another invented monitoring example, six marked events among 100 assessed stops give 6%. Three marked events among 25 assessed stops give 12%. The second period has fewer marked events but a higher observed fraction per assessed stop. Neither figure alone reports loss per metre of production.

Use event rates, local severity and rejected cloth length for their own purposes. Reducing unnecessary stops can help production, but it does not prove that the remaining restarts are well controlled.

Use machine features as specific tools

Current makers provide different ways to manage the transition. The Toyota JAT910 brochure describes linked controls for let-off and take-up. It also describes starting acceleration in several steps. Its list also includes controls linked to stop cause and time. These are examples of model-specific functions, not a set of values to copy.

Picanol's PRA II Plus description describes how the system removes a faulty filling. After successful removal, it restarts the loom automatically. That is relevant to repair time and repeatability for the supported faults. It does not mean that every warp or filling stop can be handled in the same way.

Identify which function is actually installed, its software version and the maker's meaning for each field. A command expressed as a take-up correction can have a different sign or reference from a cloth-fell movement on another system. Confirm the convention before interpreting a saved setting.

Mechanical condition matters too. If actual movement differs from the command, a compensation value may mask wear or slip. Ask the responsible team to check the drive, cloth-gripping surfaces, guides and sensors. They must follow the approved maintenance procedure. Do not treat a larger correction as proof that the underlying mechanism is sound.

Confirm the result in the intended fabric state

A successful trial should improve the agreed fault measure while preserving the rest of the product requirements. Check the weave repeat, retained-pick sequence, yarn damage and widthwise appearance. Examine enough cloth around each event to see where the disturbance starts and ends.

Finishing can change how the mark looks. Where it does, compare paired observations before and after the intended route. Track the same event through processing. ApparelDiary's fabric shrinkage testing guide helps define the length states used for such comparisons.

Decide the release rule before judging the final samples. The buyer may care about a visible band even when the average thread count meets specification. Conversely, a development metric may improve without reaching the required appearance. Keep the measured result and the product decision distinct.

For each accepted change, retain the original setting, revised value, supported fabric range and evidence. A later yarn lot or construction may need a fresh check. ApparelDiary's quality assurance and quality control guide explains how process evidence supports release decisions.

Finish with an answer that others can trace. Show which events were compared and what changed in the local cloth. Record which other properties remained acceptable. Resolve any doubt about the event boundary or measurement before adopting the setting for production.

References