Manufacturing control, yield and reproducibility

A process must make the intended structure repeatedly

Tube feedstock, dispersion, spinning, treatment and handling can change alignment, defects and impurities. A reported best specimen demonstrates a possibility under its conditions. Production evidence must also describe yield, variation and rejected material.

Imagine two runs that each collect 100 m of nanotube fibre. Both have the same average mass per length, but one contains several lengths outside the intended range and another loses an intact length during handling. The collected metres alone do not answer how much output meets the intended screen. Begin with the route, the state being checked and the object that must remain usable.

In one wet-spinning route, a characterized feedstock is dispersed or dissolved into a spinning fluid, shaped through an outlet, assembled into a strand during coagulation, then drawn, dried and handled. Coagulation changes the fluid environment so that an assembled strand forms; drawing changes the strand’s deformation and orientation. Direct spinning from a growing nanotube assembly is a different route. A processing name does not identify every state between input and finished fibre.

Tube length and contact arrangement affect available load-transfer paths. Residual catalyst and non-tube carbon change what the feedstock contains and can matter to spinning; purity is therefore a relevant recorded input, not a guarantee of strength. Treatment can change packing, alignment and intertube interactions together. The same nominal setting applied to a different incoming state need not leave the same final structure. A causal account connects input → process history → structure → measured output.

Control means observing and maintaining relevant states, rather than recording settings alone. For example, retain feedstock lot and composition evidence, the prepared-fluid state, collection order, treatment history, drying condition and handling breaks. Select measurements that can discriminate the proposed change. A narrower outer area after treatment, a better short tensile specimen and a more uniform long strand are different observations. Processing, voids and property variation prepares this route-to-state reasoning; the calculations below follow production records rather than a full contact or alignment law.

Retain position and continuity, not just a run average

The following records are entirely invented. Each run originally collects ten separate labelled pieces, each 10.0 m long, in the listed order. Linear density λ means dry mass divided by reference length: λ = m/L. The unit tex is grams per kilometre, so 1 tex = 10⁻⁶ kg/m. For each piece the table reports its mean λ over that full 10.0 m, not the value at every point.

Declare this illustrative first-pass screen before reading the table: a piece must still be one continuous 10.0 m length after handling, and its reported mean λ must lie from 0.360 through 0.440 tex, including both limits. B and C keep all pieces intact. A’s piece 8 breaks into 6.0 m and 4.0 m fragments, with no mass loss. Neither fragment meets the original length rule and joining is not allowed by this screen.

These chosen criteria act on the supplied central values. They are not a standard, a full uncertainty-based acceptance plan, a tensile-strength specification or a component qualification. A whole-piece mean inside the window also does not prove that every small region is uniform. No destructive tensile result is silently used to certify every metre.

Run A’s values all fall within the density window, but piece 8 fails continuity. A therefore retains nine full pieces, or 90.0 m. B loses three pieces to the upper density limit and retains 70.0 m. C retains only pieces 4–7, at 0.360, 0.380, 0.420 and 0.440 tex: 40.0 m. Keep the reason for each rejection; “rejected” does not mean the nanotube walls necessarily fractured.

Equal means can hide different length records

A mean is the sum of the ten reported values divided by ten. A and C both have mean λ = 0.400 tex. A spans only 0.380–0.420 tex, whereas C spans 0.300–0.500 tex. The observed ranges are 0.040 and 0.200 tex. Equal averages therefore coexist with different spreads and very different retained lengths. A’s narrow density range still does not prevent its handling break.

B has mean 0.440 tex and range 0.420–0.460 tex. Its range is as small as A’s, but the whole group is shifted upward by 0.040 tex relative to A’s corresponding values. Precision around a centre does not say whether that centre meets the intended condition. The inclusive upper limit passes 0.440 and rejects 0.450; there is no hidden rounding or probability rule in this arithmetic.

Collection order supplies further information. C steadily increases along its recorded pieces, while A fluctuates around its centre. Sorting the values or reporting only a mean would hide that ordered pattern. Investigate whether feed delivery, bath state, draw history, drying or measurement conditions changed with time. The pattern suggests a question; it does not by itself identify a unique cause or prove a stable long-term distribution.

Separate manufacturing variation from measurement variation. Reweighing one unchanged piece ten times probes that measurement arrangement. Measuring ten different pieces also includes differences in the manufactured material. Pieces from one run share feedstock and process events; they are not ten independent manufacturing runs. To examine run-to-run reproducibility, state which lots, days, operators and equipment were actually repeated or changed.

A common reference check and repeated observations can help investigate a measurement shift, while repeated independent runs at recorded states examine production variation. Neither short-term instrument repeatability nor repeated favourable pieces characterizes every source of variation. Calibration, uncertainty and repeatability explains why averaging cannot remove a shared offset. Here the finite counts and ranges describe these supplied records only; no normal distribution, tail probability or process-capability index has been fitted.

Ask what the yield numerator and denominator contain

Yield is a fraction of a stated input or output that meets a stated criterion. A piece-count fraction is retained original pieces divided by original pieces considered. A length fraction is retained length divided by collected length. A mass fraction instead weights every piece by its mass. These fractions coincide only under additional equal-size or equal-mass conditions; choose the account needed by the production question.

For A, retained length divided by collected length is 90.0/100.0 = 90.0%. Piece-count yield is also 9/10 = 90.0%, because the original pieces all have equal length. Now calculate masses: a 10.0 m piece at 0.400 tex weighs 4.00 mg. Summing A’s ten λ values gives 4.000 tex, which multiplied by 10.0 m gives 40.0 mg of collected fibre. The broken piece’s original 0.390 tex gives 3.90 mg, still present in the rejected fragments. Retained fibre mass is 40.0 − 3.90 = 36.1 mg.

Retained mass divided by collected fibre mass is 36.1/40.0 = 90.25%, slightly different from the 90.0% length fraction. The broken piece is slightly lighter per metre than the run average. The mass-account figure follows a separate stipulated closed dry-carbon account: 50.0 mg enters, 40.0 mg is collected as fibre and 10.0 mg is recovered elsewhere as non-product material. There are no additives, uncounted inventory changes or mass changes during the later handling break.

Collected-fibre mass recovery from input is 40.0/50.0 = 80.0%. Retained product mass per input is 36.1/50.0 = 72.2%. Equivalently 0.800 × 0.9025 = 0.722. All these numbers are correct and answer different questions. A report that calls each simply “90% yield” conceals its basis. This account also does not measure energy use, cost, environmental impact or the strength of the retained pieces.

Retain original identities through rework. In a separate invented piece-count record, 100 original pieces give 82 first-pass conforming pieces and 18 held pieces. After recorded rework, 10 of those same 18 meet the same stated criterion and eight remain rejected. First-pass yield is 82%; final observed yield is 92%. The reworked pieces are not 10 new original pieces, and the 18 held pieces must not be added again to the starting denominator. These supplied outcomes are bookkeeping evidence, not a repair recipe.

Scale introduces new constraints

Throughput, length uniformity, joining and process control determine whether useful products can be made. Preserve batch identities and measurement protocols so changes in performance can be traced to actual input and process states.

Throughput is an amount per stated elapsed time. In A, collecting 100.0 m takes 2.00 h including setup, stops and assessment. Collected-length throughput is therefore 50.0 m/h; retained-length throughput under the declared screen is 90.0/2.00 = 45.0 m/h. A reported 100 m/h spinning speed during stable collection does not replace either result: one hour at that speed plus one hour of other run time is consistent with this ledger.

Continuity is a separate constraint. The 6 m and 4 m rejected fragments contain 10 m and 3.90 mg of fibre, but their sum is not an uninterrupted 10 m piece. A later rule permitting 6 m pieces could retain the 6 m fragment, raising retained length to 96 m. That would be a change of product requirement, not proof that first-pass production under the original 10 m criterion improved. A splice would introduce a new joint and needs its own evidence; it is not invisible continuity.

A destructive test coupon removed from a long product is also material removed from that product. Record where it came from and whether its removal changes the available continuous lengths. Its result samples that gauge and test history; it does not test every uncut location. A faster line, a longer collected coil and a short record-setting coupon therefore support distinct claims.

For scale-up, preserve the time and length histories that could explain lost output: startup, interruption, collection position, treatment, cutting, handling and inspection disposition. A stable-looking short interval does not establish uniform output across an entire shift, lot change or larger machine. The useful throughput must use the final stated product definition and the same accounting window.

Make the next production claim testable

A useful trace connects feedstock lot → prepared batch → run and settings/state record → ordered length or piece → measurement → disposition. Keep original readings and corrections, criterion version, rejects, holds and rework history. If a measurement or criterion changes, retain the preceding record and show exactly which conclusions were recomputed.

To test whether a new preparation route improves production, compare output under the same product definition and measurement model. Record changed feedstock, treatment, operator, equipment and elapsed time rather than attributing every difference to one selected setting. Runs A and C demonstrate that the same mean can accompany different length variation; they do not establish why C varies or how future runs will behave.

Choose sampling to address the intended population and its structure. A few adjacent pieces at the favourable centre of one run do not represent all startup, late-run and changed-lot material merely because the measurements are precise. Independent runs and relevant positions need actual coverage. More observations from the same favourable place cannot manufacture evidence for a missing condition.

A successful wet-spun fibre specimen establishes a bounded material and process achievement. Reproducible production additionally needs records of what was made, rejected and retained across relevant runs. This lesson’s invented length/mass screen illustrates that accounting; it supplies no tensile, electrical or durability qualification. Evidence for macroscopic performance and limits continues the property-specific question, and Quality evidence and reproducible processing develops broader process-window and inspection reasoning.

Worked examples

Try the reasoning

1. Why report rejected output?

   Show the answer and reasoning

   It affects usable yield.

   A selected best sample hides process variability. Keep the amount, location and reject reason so that the retained fraction has an honest denominator and the process can be investigated.

2. Can feedstock purity matter at scale?

   Show the answer and reasoning

   Yes.

   Impurities can affect processing and structure. A feedstock lot contains more information than a carbon-nanotube label. Record the actual composition/state; no single purity number guarantees the final response.

3. Five accepted pieces are each 2 m long; one rejected piece is 10 m long. Calculate yield by piece and by collected length. Can the larger percentage be reported without its basis?

   Show the answer and reasoning

   5/6 ≈ 83.3% by piece; 10/20 = 50.0% by length.

   The five short pieces and one long piece are six objects but contain equal accepted and rejected total lengths. Piece counting weights each object once; length counting weights its actual metres. Neither percentage may silently replace the other.

4. Keep all A/B/C records and continuity states, but tighten the whole-piece mean-density window to 0.380–0.420 tex inclusive. How much length does each run retain?

   Show the answer and reasoning

   A 90.0 m; B 10.0 m; C 20.0 m.

   All A density values remain in range, but A piece 8 is still broken. Only B piece 5 at 0.420 passes. C pieces 5 and 6 at 0.380 and 0.420 pass. The changed yield is a changed criterion applied to the same records, not a physical improvement or deterioration.

5. A continuous 10 m piece has 0.300 tex in its first 5 m and 0.500 tex in its last 5 m. What whole-piece mean follows, and what does passing the original mean-density screen establish?

   Show the answer and reasoning

   0.400 tex; it passes that mean criterion but does not establish pointwise uniformity.

   The two halves weigh 1.50 and 2.50 mg, giving 4.00 mg over 10 m. Each half lies outside the original 0.360–0.440 tex window. A rule for the whole mean does not become a local rule after the calculation.

6. Someone randomly shuffles C’s ten values before plotting them. Which quantities stay unchanged, and which evidence is lost?

   Show the answer and reasoning

   Mean, min–max, and the density-only pass count stay unchanged; the observed collection-order trend is lost.

   Those summaries ignore order. The increasing original sequence links observations to time/position and can guide an investigation. Its loss cannot be repaired by an unchanged average.

7. In a separate revised-analysis case, a justified calibration correction subtracts 0.040 tex from every original B value, with all B lengths still continuous. Find the corrected mean and retained length under the original window. Does that establish a physical manufacturing improvement?

   Show the answer and reasoning

   0.400 tex and 100.0 m retained; no physical improvement follows from correcting the readout.

   The revised B values equal A’s density values, but B has no handling break. All ten B pieces pass the original density/continuity rule. Preserve the original readings and correction. A common measurement shift changes the inferred output classification; it is not itself a changed fibre-making route.

8. Among 100 original pieces, 20 fail the density criterion, 12 fail continuity and five fail both. Under a rule requiring both criteria, how many pass? Why is 68 wrong?

   Show the answer and reasoning

   73 pass; the union of failures contains 20 + 12 − 5 = 27 pieces.

   The five pieces failing both appear in both failure lists. Subtracting 20 and 12 separately from 100 rejects them twice. Keep traceable piece identities rather than adding overlapping category counts.

9. A new record retains the same 90.0 m but completes the full stated run window in 1.50 h. Compute retained-length throughput. Does the change establish a higher tensile strength?

   Show the answer and reasoning

   60.0 m/h; no tensile-strength conclusion follows.

   90.0/1.50 = 60.0, one third above 45.0 m/h. The output/time account changed. No new force, area, gauge or failure observation was supplied.

10. Keep A’s original density window but change the product definition to permit continuous pieces at least 6.0 m long. The broken piece has uniform λ = 0.390 tex in both retained fragments. How much total length now meets that revised definition?

    Show the answer and reasoning

    96.0 m, rather than the original 90.0 m.

    The 6.0 m fragment now passes at equality; the 4.0 m fragment does not. Add 6.0 to the original 90.0 m. No material was newly manufactured and the original 10.0 m first-pass criterion was not met by that broken piece.

What to carry forward

• Measure yield and variation alongside best results.
• Trace the complete production route.
• Specify whether yield counts original pieces, collected length, retained mass or input mass.
• Equal means do not preserve spread, collection order, continuity or reject causes.
• Do not count repeated readings as independent production runs or reworked identities as new original pieces.
• Elapsed-time throughput and the stated product definition belong with a scale-up claim.