Qualifying a new material source: the test sequence that takes two years

Éprouvettes de traction et coupons polis alignés sur un banc de laboratoire de métallurgie

A second prepreg source, quoted at the same areal weight and with the same resin, a far shorter lead time, and a data sheet showing the same values as the material already carried in the certification file. The question put to the design office is not whether this product is good. It is what separates a data sheet from an allowable, and why that gap is closed by two years of testing rather than by a purchase order.

An allowable is a lower bound, not a performance figure

A data sheet gives typical values: means, sometimes with a standard deviation, obtained on whatever the manufacturer had to hand at the time of measurement. A substantiation file does not consume means. It consumes allowables, that is, lower tolerance bounds: a value that a given proportion of the population exceeds, at a given level of confidence.

Two bases coexist. A-basis covers ninety-nine per cent of the population with ninety-five per cent confidence; B-basis, ninety per cent with the same confidence. The choice between the two is not left to judgement: a structure in which the failure of one element transfers the load onto another path may be substantiated on B-basis, whereas a single-load-path element calls for A-basis. To these is added S-basis, which is not statistical at all: it is the minimum value guaranteed by the material specification, the one that receiving inspection verifies batch by batch.

The consequence governs everything else. Qualifying a source is not demonstrating that a sample is strong; it is estimating the lower tail of a population that has not yet been manufactured. A high mean helps only marginally; a wide scatter ruins it. And the allowable is still only a material property: the design value used in the stress report derives from it after applying the knock-downs specific to the structure — holes, allowable damage, environment, scale effects.

Position of the allowables within the measured distributionBell curve of the measured property, its peak corresponding to the mean. Two dashed vertical lines mark, far into the lower tail, B-basis and then, further left still, A-basis. The area under the curve to the left of B-basis is filled: it represents the share of the population that falls below the bound adopted.densityshare of the populationbelow the bound adoptedmeasured meanA-basisB-basismeasured property
The value that enters the stress report lives in the lower tail, not at the peak. Shifting the mean to the right gains little; tightening the scatter gains a great deal. This is why a higher-performing but more scattered material can yield a lower allowable than a more modest, better-controlled one.

An allowable is not about how strong a material usually is, but about the minimum strength you can be sure of. It’s like saying, “99% of the time, this material will be at least this strong.” The more spread out the test results, the lower the allowable, even if the average is high.

For a carbon fiber composite used in an aircraft wing, the tension allowable might be 600 MPa (A-basis). This means that 99% of the material will handle at least 600 MPa, even if some samples tested at 650 MPa. If another source has a wider scatter, its allowable might drop to 550 MPa, even if its average is 630 MPa.

What batches buy, and what specimens do not

The scatter of a material splits into two parts: what varies from one specimen to another within the same batch, and what varies from one batch to the next. The first is small and reassuring. The second carries everything that production holds in store: a heat whose chemistry drifts within the permitted range, a sizing bath renewed, a fibre precursor moved to another site, an autoclave cycle run in a different oven.

Hence the rule that surprises every newcomer: one hundred specimens taken from a single batch are not worth ten specimens taken from ten batches. The former estimate the population of one batch very precisely; the latter estimate, less precisely but correctly, the population that is actually needed. For this reason the direct computation methods for A- and B-basis impose a minimum number of distinct heats or batches — of the order of ten on the metals side — before the number of specimens is even discussed. On the composites side, a qualification typically rests on three distinct raw-material batches, with a few specimens per batch, per property and per environmental condition; pooling methods lighten the count when the data behave well, never when they behave badly.

This is also the first reason for the schedule. Three batches are not ordered on the same day, otherwise they are no more than a single batch cut into three: they must genuinely differ, hence come from separate campaigns, from different fibre spools or different heats. The waiting time between them is not lost time, it is the very thing being bought.

Basis value obtained as a function of the number of distinct batchesAt a constant total number of specimens, the measured mean, shown as a horizontal dashed line, does not move when the sampling is spread over more batches. The basis value obtained, shown as a solid line, starts very low with a single batch then rises quickly before flattening out: the gap between the two curves is the statistical price of the sampling.relative valueat constant specimen countmeasured mean (dashed)accepted gapbasis value obtained13610number of distinct batches
Adding specimens within the same batch does not raise the bound. The mean, for its part, is secured from the very first measurements. What is paid for by multiplying batches is the reduction of the uncertainty on the between-batch variance — the only one that describes future production.

The pyramid: the coupon does not fail like the structure

Tests are organised in levels, from the most numerous to the most representative. The coupons measure the material in stress states as pure as possible: tension, compression, in-plane shear, interlaminar shear and, on the metallic side, fracture toughness and crack growth rate. The elements introduce the singularity — a hole, a fastener, a ply drop-off — and with it a failure mode the coupon could not reveal. The details and sub-components add the real geometry, the load introduction and the boundary conditions. At the apex, one or two tests validate the whole.

Two scientists test a material sample on a machine while recording data.
Material qualification requires rigorous, long-term testing.

The principle is the same at every tier: the analysis is calibrated at the bottom, where tests are numerous and cheap, then verified at the top, where they are rare and expensive. A discrepancy found at the upper level is not corrected at the upper level; it sends you back down, to the property or the model that produced it. A material qualification does not climb the whole pyramid — the upper tiers belong to the structural programme — but it must supply what is needed to climb it: the properties, their scatter, and the demonstration that the process that produced them is reproducible.

Test pyramid, from coupon to componentFive tiers stacked into a pyramid, from bottom to top: coupons, elements, details, sub-components, components. The width decreases towards the apex and the number of tests shown on the right goes from thousands of specimens down to one or two tests. A vertical arrow on the left marks the representativeness increasing upwards.representativegenericcouponsthousandselementshundredsdetailsdozenssub-componentsa few testscomponentsone or two
The number of tests and the representativeness vary in opposite directions. Each tier serves to validate the analysis method of the tier below, not to replace it. A surprise at the apex is never settled at the apex: it travels back down to the material property or to the faulty model.
What each test level brings, and what it cannot see
Level What is measured there What can still surprise above Order of magnitude
Coupons Material properties under simple stress states, scatter, environmental effects Any failure mode linked to a geometric singularity Thousands of specimens
Elements Holes, fasteners, ply drop-offs, joint strength Interaction between neighbouring singularities, scale effects Hundreds
Details and sub-components Real geometry, load introduction, boundary conditions Load distribution in the complete structure Dozens
Components and full scale Overall validation, static then fatigue What service reveals: repairs, real environment One or two

The environmental matrix decides the test volume

The number of specimens is not set by the number of properties, but by that number multiplied by the orientations, the thicknesses, the product forms and the environmental conditions. That is where the count takes off, and that is where the scope of a qualification is really negotiated.

On composites, the split is clear-cut. Fibre-dominated properties — tension along the ply axis — move little with temperature and moisture. Matrix-dominated properties — compression, compression after impact, shear, joint strength — collapse when the absorbed water lowers the glass transition temperature and that limit is approached. The maximum operating temperature adopted therefore sits below the glass transition measured in the wet condition, with a margin. Conditioning itself weighs on the schedule: since diffusion is governed by the square of the thickness, a thick laminate takes weeks, sometimes longer, to reach equilibrium — and the hot/wet test only begins afterwards.

On metals, the matrix changes name without changing nature: product form, thickness range, sampling orientation including the short transverse direction, thermal exposure, resistance to stress-corrosion cracking. The weak property is almost never the one the data sheet announces; it is the short transverse property of a heavy gauge, or the fracture toughness at the bottom of the scatter.

Property retention as temperature rises, conditioned materialTwo relative properties as a function of temperature. The solid line, a fibre-dominated property, stays almost horizontal over the whole range. The dashed line, a matrix-dominated property, decreases steadily then plunges as the wet glass transition is approached. A thin vertical line marks the maximum operating temperature adopted, placed short of that transition, the margin between the two being arrowed below the axis.relative propertyfibres: tension along the axismatrix: compression, shear (dashed)max. operatingwet Tgdry Tgmargintemperature, conditioned material
It is not tension that decides the operating envelope. The solid curve, fibre-dominated, would allow almost anything; the dashed curve, matrix-dominated, sets the limit. Moisture shifts the glass transition to the left, and therefore shifts the maximum temperature adopted as well.

Two years, item by item

Two years is an order of magnitude, not a rule: a well-framed process delta is settled in a few months, whereas an entirely new material system, on a line that has yet to deliver anything to the sector, takes longer. What does not vary is the nature of what fills the schedule. It is not filled with test machine hours, but with waiting periods in series, none of which can begin before the previous one has finished.

The material specification and the process specification are written first: without them, one does not know what is being qualified, and two batches made to unfrozen parameters are not two batches of the same product. The supplier audit and the accreditation of the test laboratory are run in parallel. Then come the manufacture of the batches, then that of the panels or semi-finished products, with their non-destructive inspection — a panel carrying porosity outside specification is not repaired, it is remade, and the next batch is not available for several months. Conditioning takes the following slot, then the tests themselves, whose rate depends on the capacity of the accredited laboratory far more than on the number of specimens.

The last slice of the schedule contains no test at all: it is the data reduction, the statistical analysis, the drafting of the specification and of its receiving inspection plan, then the first monitored production run. Yet this is where the fate of the file is decided, because an allowable without an inspection plan capable of maintaining it is worth nothing.

Sequence of the items over roughly two yearsFive horizontal bars spread over twenty-four months: specifications and audit at the start, manufacture of the batches next, preceded by a campaign wait marked by a dashed edging, then conditioning and coupon tests in the middle, element tests towards the end, data reduction and receiving specification up to the edge of the schedule.material specification and supplier auditmanufacture of batches and panelscampaign waitconditioning then coupon testsconditioningelements and detailsstatistics, specification, receiving06121824months elapsed
The items that dominate the schedule are not tests. The wait between two material campaigns and the conditioning of the specimens, marked by a dashed edging, are not accelerated by any budget. Only the central part, the tests themselves, can be compressed by buying laboratory capacity.

Full qualification, equivalence, process delta

The three regimes are distinguished not by their rigour but by what they are entitled to assume already known. A full qualification assumes nothing: it builds the database. An equivalence assumes that the material system is the same and that only the source, the line or the site changes; it compares a reduced sample with the existing database through statistical tests bearing both on the mean and on the individual minimum value, property by property. A process delta assumes the material unchanged and examines only the properties sensitive to the modified parameter.

The point to remember fits in one sentence: equivalence is not a negotiated discount. It fails property by property, and a single property falling outside the domain sends you back to a full qualification for that property, or to accepting a knock-down in the stress report. Between two steps, the difference in cost and lead time is counted in orders of magnitude, which explains the energy spent upstream to avoid triggering the higher step.

Three qualification regimes, and what each is entitled to assume known
Regime Situation What is demonstrated Sampling What tips it into the higher regime
Full qualification New material system, no existing database The database itself: means, scatter, allowables Several distinct batches, all the properties, the whole environmental matrix Not applicable: this is already the full regime
Equivalence Second source, another line or another site for the same system That the sample belongs to the already qualified population, mean and individual minimum included Reduced, focused on the discriminating properties and the most severe conditions A single statistical test failing on one property
Process delta Manufacturing parameter modified, material unchanged That the change does not affect the properties sensitive to the parameter concerned Targeted, often a few conditions and one batch A measured discrepancy, or a key parameter outside the frozen domain

What qualification does not guarantee

A qualification is dated. It describes what a process produced over a few months, on selected batches. What extends it is not the certificate but the pair formed by the material specification and the process specification, with the receiving inspection plan that follows from them: the control limits applied to each delivered batch come out of the qualification database, and have no other legitimate origin.

This is why the key process parameters are frozen and then declared: composition, fibre precursor and sizing, resin batch, cure cycle and its tolerances, tooling, manufacturing site. Any change to one of them must be notified, and it is the customer — not the supplier — who decides whether it calls for delta testing. A supplier who treats sizing as an undeclarable trade secret has, in effect, made its qualification unverifiable.

The classic trap. The qualification batches come off the pilot line, with the development tooling and the best operators, because that is the line that is available. The file passes. The first production batch, made on the final tooling, falls outside the receiving inspection limits — limits established, precisely, on a scatter that production will never have. There is then no short way out: either the limits are widened, which means demonstrating that the allowables still hold, or batches are remade. Both cost months, and the second restarts the count from zero.

The right reflex. Write the receiving inspection plan at the same time as the allowables are computed, never afterwards, and enter in it at least one matrix-dominated property, measured in the conditioned state. An inspection limited to axial tension and to physical characteristics — areal weight, resin content, gel time — lets a change of sizing or precursor through without a murmur, one that will be paid for on interlaminar shear hot/wet, that is to say at element level, two years later. Keep retained samples of every qualification batch as well: the day a property drifts, that is the only physical reference available.

Before signing, what to re-read in-house

Three checks are made in one’s own documentation, without asking the supplier for anything. The first: does the allowable used in the stress report cover the product form, the thickness range and the orientation actually used? A database established on thin sheet does not extend to a heavy forging, and the short transverse direction is not a variant of the longitudinal one. The second: does the environmental matrix of the qualification encompass the intended operating envelope, including cases of prolonged exposure and hot zones? The third: who owns the data, and on what basis may it be used? A database shared between manufacturers may be used only after demonstrating that one’s own processing is equivalent to the one that produced it; without that demonstration, it is someone else’s data.

The grey area of the moment lies with the processes where the very notion of a batch is not settled. In metal additive manufacturing, is the relevant population the powder lot, the machine, the build campaign, the orientation of the part on the plate? Each answer changes the number of tests by substantial factors, and practices are not yet uniform from one customer to another. For a supplier preparing to enter the field, the consequence is first of all budgetary: it is not the material that costs, it is the hours of accredited laboratory, the conditioning, and the calendar time during which the line must be held stable without modification.

This text describes a general mechanism. The applicable certification specifications, the means of compliance accepted by the authority and the approved documentation of the manufacturer or the operator prevail over everything above, including over the orders of magnitude quoted here for guidance only.

Questions fréquentes

Can the values published by the material supplier be used directly?

No, because a data sheet publishes typical values, that is to say means obtained on a sample that is not under our control. A substantiation file requires lower tolerance bounds established on distinct batches and associated with a level of confidence. The data sheet serves to pre-select a candidate and to write the specification, never to feed a stress report.

Why multiply the batches rather than the specimens within a single batch?

Because the scatter that matters is the one separating one batch from the next, not the one separating two neighbouring specimens. One hundred specimens from a single heat describe that heat perfectly and teach almost nothing about future production. That is why the direct computation methods for A- and B-basis first set a minimum number of batches or heats, before any counting of specimens.

Does a second source require the whole test pyramid to be repeated?

In principle no: if the material system is identical and only the source or the line changes, an equivalence is conducted, which compares a reduced sample with the existing database. That comparison bears on the mean and on the individual minimum value, property by property, under the most severe environmental conditions. But equivalence fails property by property: a single discrepancy sends you back to a full qualification for that property or imposes a knock-down in the stress report.

What triggers a requalification at a supplier already approved?

Any change to a key parameter declared in the process specification: a change of site or line, of fibre precursor or sizing, of resin batch, of tooling, of cure cycle beyond the frozen tolerances. The supplier has an obligation to notify; it is the customer who decides the extent of the delta testing. A change left unnotified and then discovered at receiving inspection costs far more than the delta it would have called for.

Why does moisture conditioning weigh so heavily on the schedule?

Because the diffusion of water into a laminate is governed by the square of the thickness: doubling the thickness multiplies by four the time needed to reach equilibrium. On thick parts, the wait is counted in weeks, sometimes longer, and no hot/wet test can begin before then. This item is neither compressible by budget nor parallelisable; it can only be anticipated by starting conditioning as soon as the panels come out.

Can a qualification database shared between manufacturers be relied upon?

Yes, and it is indeed the way to avoid everyone repeating the same thousands of coupons, but its use is conditional. One must demonstrate that one’s own processing — tooling, cycle, inspections — is equivalent to the one that produced the database, through a documented equivalence programme. Without that demonstration, the values remain those of another process and cannot serve as allowables in a certification file.

Commentaires

Leave a Reply

Découvrez l’écosystème FL330

FL330, c’est toute l’aviation professionnelle — explorez nos autres espaces.