A break log reads the tail only above its fault floor
Worth reading first: A warp breaks in a tail no tensile test reaches · What the shed costs, in newtons · The back shaft works hardest.
A warp breaks in a tail no tensile test reaches. A 500-millimetre test samples about eighteen staple lengths of yarn; a warp of four thousand ends by a thousand metres holds a hundred and forty-three million of them, and an end breaks wherever one is weaker than the shed’s tension. Three laws fitted to the same two tensile tests agreed within five per cent at every gauge a laboratory can clamp and disagreed by a factor of 375,000 on the breaks at the back shaft.
That essay ended by pointing at the one instrument that does sample the tail. A loom already keeps it. A break log that records each warp break by shaft is a sample of the yarn’s weakest places at twenty-four tensions at once, because the shed’s geometry sets each shaft’s tension, from 14 per cent of the breaking load at the front to 52 at the back. Plot the breaks against the shaft and the tail’s slope should be there to read.
It also named the obstacle. Not every break comes from the spread. A slub, a bad piecing or a fly woven in is a discrete fault, much weaker than the yarn around it, and it breaks on whatever shaft it reaches. This essay puts the two populations into one log and asks what the log can still read.
Two populations, two shapes
The spread’s breaks rise along the harness. Each shaft carries a twenty-fourth of the warp, a sixth of a million staple lengths, and its tension rises steadily from the front. The back shaft works hardest because its lift is the largest, and on the cliff of a strength distribution a steady rise in tension is a steep rise in breaks. The tail essay found the last four shafts taking 86 to 99.95 per cent of the spread’s breaks, depending on the law.
That climb is what a yarn breaks at its thinnest place looks like when the gauge is a whole shaft’s share of a warp rather than half a metre. Each shaft is a tensile test with a hundred and sixty-seven kilometres between its clamps and its own tension, and the harness runs twenty-four of them side by side. The spread behind it comes from counting, since a yarn’s section holds a countable number of fibres and its strength scatters as that count does. The three laws disagree about the height of the climb and not about its direction, which is why a log’s direction says little and its height and steepness say a great deal.
The faults’ breaks are flat. A fault that survives the winder’s clearer and still breaks on a loom is by construction far weaker than any tension the shed applies, so every one of them breaks wherever it arrives. A shaft’s share of the faults is its share of the yarn, which is the same for every shaft. Over one beam each shaft carries 167 kilometres of yarn, so a fault rate of one in a hundred kilometres puts 1.7 breaks on every shaft, 40 on the beam.
So a log is a floor with a cliff standing on it. The floor is the fault rate. The cliff is the tail, and where it rises out of the floor depends on how many breaks the tail makes against how many faults the yarn carries.
How large the floor is
Under the three laws, one beam’s spread breaks are 1.7 for the normal yarn, 0.0015 for the lognormal and 63 for the Weibull. Those are the totals over all twenty-four shafts, and the tail essay’s numbers for the back shaft are their largest parts.
A fault rate of one in a hundred kilometres makes forty. That one rate is twenty-four times the normal yarn’s whole spread and twenty-seven thousand times the lognormal’s, and it is not an extravagant rate: an evenness tester counts thin and thick places in the hundreds per kilometre, and a fault classifier counts the rare large ones per hundred kilometres. How many of those survive clearing and break on a loom is a number this arithmetic cannot supply, so the essay sweeps it from a thousandth to a hundred per hundred kilometres rather than assuming one.
The normal yarn shows what the floor does. With no faults, its log is a cliff: nothing on the front shafts and 93 per cent of all breaks on the last four. At a tenth of a fault per hundred kilometres the back shafts still stand out, holding 39 per cent. At one, the last four shafts hold 20 per cent of the breaks, against the 17 per cent they would hold if the breaks were spread evenly. At ten, the log is flat to within five per cent, and nothing in it says the yarn has a tail at all.
The slope a log reports
The quantity the tail essay wanted from a log was its steepness: how fast the breaks rise with tension between two shafts, which is the slope of the tail in the one region the loom works in. Take the back shaft against the eighteenth, the natural log of their ratio, per unit of tension between them.
With no faults the log reports the spread’s own steepness exactly: 39.3 for the normal yarn, 29.5 for the Weibull and 122 for the lognormal. Those are the numbers that decide how a warp’s breaks respond to a tension change, and the tail essay found them out of reach of any tensile test.
With faults, the log reports something smaller. The normal yarn’s 39.3 becomes 16.6 at a tenth of a fault per hundred kilometres, 3.8 at one and half a unit at ten. The Weibull’s 29.5 holds up better, 21.7 at one and 8.2 at ten, because its spread makes enough breaks to stand above the floor. The lognormal’s tail never appears in the log at all. Its spread makes about a fifteen-hundredth of a break a beam, so a thousandth of a fault per hundred kilometres, which is a twenty-fifth of a fault in a whole beam, is already more than twenty times as many.
So a flat log does not mean a gentle tail. It means a tail too far out to be seen over the faults. The two readings are the same log, and only the fault rate decides which one it is.
What the front shafts are for
The floor is a nuisance for reading the tail, and it is also the cleanest measurement of the fault rate that a mill could have. On the front shafts the spread makes essentially no breaks under any of the three laws, because the tension there is 14 per cent of the breaking load and the weakest place in a whole warp is not that weak.
Every break on the front shafts is a fault, under every law and at every fault rate drawn. So the front of the harness is a fault counter running at the loom’s own tension on the loom’s own yarn, which is a better instrument for the purpose than a fault classifier’s thresholds, since it counts exactly the faults that break.
That gives the log a natural division of labour. The front shafts read the fault rate; the back shafts read the tail above it. Where the one hands over to the other is itself a measurement. For a Weibull yarn the spread outnumbers the faults from the fifteenth shaft at a tenth of a fault per hundred kilometres, from the nineteenth at one, and only on the back shaft at ten. For a normal yarn it happens on the twenty-second shaft at a tenth, and on no shaft at all from one upward.
How many beams decide a law
The question a mill would actually ask of a log is which law its yarn follows. A log is twenty-four counts, each scattered as chance scatters a count of rare events, so the question has a definite answer: how many beams of log does it take before the true law is favoured over another by a hundred to one?
The answer splits sharply in two. Any law against the Weibull is settled within a fraction of a beam, a tenth of one at low fault rates and under a third at ten per hundred kilometres. A Weibull yarn makes sixty-three spread breaks a beam, most of them on the back shafts, and no other law comes close to that.
The normal against the lognormal is the hard case. Both make so few spread breaks that the log is mostly floor, and the difference between them is a handful of breaks on the last few shafts over many beams. At a tenth of a fault per hundred kilometres the log decides in 5.5 beams; at one, 22; at ten, 182. A weaving shed running a style for a season might log a few dozen beams, so the question is answerable when faults are rare and out of reach when they are common.
That is the tail essay’s conclusion turned round. The tensile certificate could not choose between the laws because it sampled the middle. A break log samples the tail, and it can choose, provided the yarn’s discrete faults are rare enough not to bury the tail’s breaks.
A lopsided log is itself a reading
Weavers know the back shafts break more. The harness has a depth, and a deep harness is felt as breakage at the back. The arithmetic says what that observation is evidence of.
A log in which the last four shafts take most of the breaks has to come from a spread that outnumbers its faults on those shafts. At one fault in a hundred kilometres, neither the normal yarn nor the lognormal can produce it: their last four shafts hold 20 and 17 per cent of the log, close to an even spread. The Weibull yarn can, with 59 per cent at that rate. So the familiar lopsided log, if it is found at a fault rate of that order, says that the yarn’s tail is heavier than a normal’s. If the faults are rarer, a tenth per hundred kilometres, the normal yarn’s last four shafts hold 39 per cent, and lopsidedness alone no longer decides.
So the two quantities have to be read together. The front shafts give the fault rate. Given that, the share at the back is a reading of how heavy the tail is, and the steepness between the back shafts is a reading of its slope, as far as the floor allows it to be read.
What sizing, evenness and tension now mean
The tail essay found that a size coat raising every section’s strength by a tenth cut the spread’s breaks by a factor of four to two hundred. It does nothing to the faults. A fault far weaker than every shed tension is still weaker after a ten per cent gain, so it still breaks. On a log, sizing lowers the cliff and leaves the floor. A mill that sizes a warp and sees its breaks fall by a quarter, not by four times, is seeing a log that was mostly floor.
Evenness does the same. A finer yarn is a worse yarn because fewer fibres in a section mean a wider spread, and the tail essay found an index change from 1.2 to 1.6 multiplying the spread’s breaks by thirteen to eleven thousand. An improvement in the spinning index lowers the cliff; an improvement in clearing lowers the floor. The log says which one a mill has bought.
And the tension. A small rise in the let-off moves every shaft up the cliff and multiplies the spread’s breaks, the familiar cliff that the shed is an extension explains. It moves the floor not at all. So a warp whose breaks hardly respond to the let-off is a warp whose breaks are faults, and a warp that becomes unweavable a few per cent higher is one whose breaks were the tail’s. A weaver who turns the let-off a notch and watches the back shafts has performed the experiment.
How the log was computed
The yarn is the tail essay’s: a 20 tex ring-spun cotton at a spinning index of 1.35, a 28-millimetre staple, and three laws, normal, lognormal and Weibull, each fitted to reproduce the same expected breaking loads at 100 and 500 millimetres. The warp is 4,000 ends of 1,000 metres on twenty-four shafts, each shaft carrying a twenty-fourth of the ends, at a tension rising evenly from 14 to 52 per cent of the 500-millimetre breaking load, the shed essay’s figures for a 25 tex yarn.
Each shaft’s spread breaks are its staple lengths times the law’s probability below its tension. Each shaft’s fault breaks are its kilometres of yarn times the fault rate per hundred kilometres. The slope is the natural log of the back shaft’s total over the eighteenth’s, divided by the difference in their tensions. The beams needed are the natural log of a hundred over the expected log-likelihood ratio of one beam’s log under the true law against the other, which for twenty-four independent Poisson counts is the sum over shafts of .
Required of it, and not shown: that the fault floor is the same on every shaft and equal to the rate times a shaft’s yarn; that the spread rises shaft by shaft under every law; that at one fault in a hundred kilometres the front shaft’s breaks are more than ninety-five per cent faults under every law; that with no faults the reported slope is the spread’s own and that faults only lower it; that every pair of laws is told apart less well over a higher floor; and that a law cannot be told from itself.
What a floor of identical faults leaves out
Every fault is weaker than the front shaft. That is the definition of the floor and it is what makes the floor flat. A real fault population has its own spread of strengths, and faults near the shed’s working tensions would break preferentially at the back like the spread does. Such a population sits between the two drawn here: a second, gentler cliff under the first. The front shafts would then read only the weakest faults, and the reading of the tail would be confounded further.
The counts scatter only as chance does. A real log carries more variation than a Poisson count: beams differ in their sizing, in the yarn lot, in the weaver, and a break may be logged against the wrong shaft. All of that adds noise, so the beam counts above are lower bounds. A drafting roller’s fault is periodic rather than random, and a random error hides where a periodic one shows: a periodic weak place would break the same ends again and again, a pattern a Poisson floor cannot produce and a log could spot. And the shafts’ tensions are taken from geometry alone. What the shed costs in newtons gives each shaft’s strain from its lift, but the let-off adds a common tension to every shaft and the heddle eyes abrade the ends, and a bundle is weaker than its threads partly because damage gathers where the load does. A log’s back shafts are therefore further up the cliff than the arithmetic puts them.
Who worked out which part
Recording warp breaks by cause and position is long-standing weaving-mill practice, and so is the knowledge that the back shafts break more. Counting yarn faults per hundred kilometres is standard in yarn testing. The weakest-link reading of a yarn is Peirce’s, from 1926, and the extreme-value limits behind the three laws are Fisher and Tippett’s, from 1928, worked out at the same cotton research association. What is added here is the log as a sum of a flat fault floor and the spread’s cliff; the slope a log reports, and how fast a floor flattens it; that the front shafts are a fault counter at the loom’s own tension; and how many beams a log needs to tell two laws apart, from under one for the Weibull to hundreds for the normal against the lognormal when faults are common.
Still open: whether a staggered harness reads more
Every shaft here carries the same number of ends, because that is how a straight draft threads a harness. Nothing requires it. The quietest shafts belong at the back already found a reason to choose which ends go where, and a draft that puts more ends on the back shafts would put more of the warp’s tries in the region where the spread makes its breaks.
The prediction is that a harness threaded heavier at the back reads the tail faster, since the back shafts’ spread breaks rise with their ends while the floor on each shaft rises only with its yarn, so the cliff would stand higher above the floor exactly where the log reads it. Whether that buys enough to bring the normal-against-lognormal question inside a season’s beams is a calculation of the same kind as the one above, with the ends per shaft as a variable. It would turn the break log from a record a mill keeps into an instrument a mill could design, and it would cost nothing but a choice of draft.
What links here
Computed from the collection rather than written here: the essays that point at this one.
Reads more easily once this is understood
Essays that name this one as worth reading first.
Shares its objects with
Essays naming at least two of the same things, that neither author linked.
- A jacquard harness needs three half-spans of height — both name harness, shed
- A jacquard's harness has a depth after all — both name harness, shed
- A mispick is one row in the wrong place — both name fault, shed
- Three mistakes and the shape each one leaves — both name fault, harness
- Where the heddles go — both name harness, shed
Named objects
A flat tag is an object no other essay names yet.