China Sliding Window Quality Control: Nine Operations, Four of Them Invisible in the Finished Window

We have argued elsewhere that a defect rate is the output of a measurement protocol rather than a property of a factory. That was an uncomfortable conclusion, and it is still the right one. But it answers only half of the question a buyer actually has.

Counting tells you how many. It cannot tell you what was done. A sliding window is assembled in roughly nine operations, and four of them leave nothing on the finished unit that anyone can inspect afterwards. You can measure the frame, operate the sash, look at the glass and still be holding a window whose four most consequential operations were performed badly — because the evidence for all four was buried by the operation that came next.

This matters more for sliders than for almost any other fenestration product. The three complaints that dominate the category — water standing in the track, a sash that has started to drag, fog between the panes — are not dimension problems. Nothing you could have measured on the finished window would have predicted them. Each one traces back to an operation whose result stopped being visible before the window was crated.

So here is the working definition this article will use: quality control on a sliding window is not the inspection at the end. It is the discipline of being present — or having a named representative present — at the operations whose work disappears.

Nine operations, and what survives each one

The table below is the whole article in compressed form. The third column is the one that matters: after this station, can anyone still check the work?

Four rows say No. Those four are the subject of this article, and they are also the four that appear in a quotation least often, because they are the four a photograph cannot show.

Station 2: the coating is decided in a tank you will never see

Powder coating is the one part of a window whose quality is not determined by the powder, the gun or the oven display. Published process guidance is blunt about this: the great majority of coating adhesion performance is decided in pretreatment. When a factory reports that "this batch of powder is no good" and the problem survives a powder change, the cause is almost always upstream, in a bath nobody logged.

The variables are ordinary and unglamorous, which is exactly why they go unrecorded:

• Degrease temperature. Public process guidance places the working band at roughly 40 to 55 degrees Celsius, and notes that below 40 the degreasing efficiency falls off sharply.
• Free alkalinity of the degrease bath. Commonly given as 8 to 15 points by standard titration; too high attacks the aluminium, too low leaves oil behind.
• Rinse water conductivity. Published guidance recommends the final pure-water rinse be held below about 50 microsiemens per centimetre, and that water above 100 must be changed. Another published source sets a tighter figure of 30. Two sources, two numbers — which is itself the point: this is a controlled process variable, not a fixed constant, and you should be asking which one the line works to.
• The water-break test. A field method worth naming in your plan: after the final rinse, a clean surface holds a continuous water film that does not break up for about 30 seconds. If the film beads or breaks in patches, oil or oxide is still there. It costs nothing and it catches the failure mode that produces blisters weeks later.
• Conversion-coat weight. Chromate and chrome-free conversion layers are commonly specified in the region of 40 to 200 milligrams per square metre. Too thin and corrosion resistance is inadequate; too thick and a powdery layer forms that reduces adhesion.
• Degrease duration. Published practice is of the order of 3 to 5 minutes. Longer is not cleaner — alkaline degrease etches aluminium, and over-etched stock comes out grey or blotchy and bonds worse, not better.

Cure is the other half. Polyester powder is commonly cured in the region of 180 to 200 degrees Celsius for roughly 15 to 20 minutes, and both directions fail differently: under-cured film is soft and poorly bonded, over-cured film is brittle and splits later under environmental stress rather than at the moment anyone is watching.

One more counter-intuitive point that is worth having in the plan. Public auditing commentary on powder-coated aluminium argues that past roughly 80 micrometres the risk profile flips: thin film risks corrosive breakthrough, thick film accumulates residual shrinkage stress during cure and becomes prone to cracking and delamination. Thicker is not automatically better, and a supplier who quotes film build as a virtue without quoting a range is quoting the wrong variable.

Now the QC consequence. The cross-cut adhesion test exists, and published aluminium finishing guidance treats it as the standard check. But it is destructive, so it is done on a sample; and it is done at the end, so it passes on a unit whose pretreatment was already wrong. The cross-cut test is the last station's opinion about the second station's work. If you want evidence about station 2, you ask for the bath log, not for the adhesion grade.

Station 4: the composite test is a process test, not a product test

A thermal-break profile is two aluminium halves and a polyamide strip that have been roll-crimped into a single structural body. Once crimped, the teeth are underneath and invisible. Nothing about the finished profile tells you how well it was crimped, and nothing about the finished window tells you either.

The test method is a published one. Under the Chinese test method for the composite properties of thermal-break aluminium profiles, coupons are conditioned at roughly 23 degrees Celsius and 50 percent relative humidity for 48 hours, preloaded, then pulled at about 5 millimetres per minute, typically ten coupons per group with a characteristic value derived from the set. Published inspection guidance gives the acceptance values as a longitudinal shear characteristic value of at least 24 newtons per millimetre for the standard grade and at least 30 for the high-strength grade, a transverse tensile characteristic value of at least 24 newtons per millimetre, a shear retention loss after thermal cycling of no more than 15 percent, and deformation after a sustained high-temperature load of no more than 0.6 millimetres.

Two things follow, and both are useful to a buyer.

First, the test measures the composite, so it has to be run on the composite in its final finish state. A published interlaboratory study found that the same substrate gave a transverse tensile characteristic value of 87 newtons per millimetre untreated, and 95, 96, 104 and 92 respectively after anodising, electrophoretic coating, powder coating and paint. Every one of those is far above the 24 floor — which tells you something honest and slightly uncomfortable: passing that floor does not separate a good composite from a mediocre one. It separates a composite from no composite.

Second, and more important: the coupons are cut from a production length, not from your window. So a shear report is evidence about a process on a day. It is not evidence about the specific frames in your crates. This is not a criticism of the test; it is the nature of every destructive test in the industry. But it means the report you are shown and the product you receive are related by a production record, and if that record does not carry a date, a length and a heat number, the relationship is decorative.

There is also a sequence point worth knowing, because it turns into a legitimate question. The polyamide strip material is specified to survive working temperatures around 200 degrees Celsius precisely so that the composite profile can pass through the powder curing oven. That locks the order of operations: the profile is crimped first and coated second. So "was the coating applied before or after crimping" is not a curiosity — it is a question with a correct answer, and a supplier who hesitates has told you something.

Station 6: the corner, and the one accessory nobody orders

If you read only one section, read this one. Corner assembly on an aluminium window is where the structural load path and the water management system meet, and it is the station with the widest gap between what two products look like and what they are.

Three processes are in common use. An active corner key uses a sprung cleat and screws, is demountable and site-friendly, but depends on a spring that can corrode and lose tension if water reaches it. Crimping closes a 45-degree joint onto a cleat under high pressure in a factory machine, giving a tight mechanical bite with no adhesive. Glue-injection crimping uses a cleat with internal flow channels; after crimping, a two-part corner adhesive is injected through a process hole until it emerges from the relief hole, adding a bonded seal to the mechanical one.

Published construction acceptance criteria describe the sequence in useful detail: a flatness plate is used to keep the 45-degree joint plane; a deflector is fitted to the corner cleat before it goes into the chamber; the two-part adhesive is injected until it appears at the relief hole, which is the signal that the cavity is full; excess is cleaned before it cures; the relief hole is then sealed; the cut ends of the profile receive an end-face sealant; and mullion connections are made with dedicated aluminium or zinc-aluminium connectors bedded in silicone.

Three of those clauses are worth pulling out, because they are the entire ballgame.

The deflector

The small plate fitted to the corner cleat exists for one reason: to stop the injected adhesive from running into the drainage chamber. Published industry commentary says plainly that many factories omit it, and that when they do, adhesive flows into the drainage path and the window loses its drainage — the failure surfaces months later as water that has nowhere to go.

This is the sentence to put in your order: glue injection and drainage are two faces of the same decision. You inject adhesive so the corner does not leak, and the wrong injection is exactly how the corner stops draining. A sliding window drains through its frame. Anything that makes the corner tighter also makes it a better dam, and the deflector is what separates the two outcomes.

Fake injection

Published trade commentary describes a practice worth knowing by name: the corner is described as glue-injected, adhesive is applied to the visible joint line, and nothing is injected into the cavity. The finished corner looks identical to a properly injected one from every angle available on a finished window. The stated way to tell them apart is to look for the injection process hole on the side of the profile — a real injection leaves one.

Note what this means for inspection timing. Fake injection is undetectable at final inspection, undetectable at arrival, and detectable in about four seconds at the corner station. It is the cleanest illustration in the whole category of why QC is a schedule rather than a headcount.

What the finished corner is allowed to look like

The published product standard gives the geometric limits that are checkable at the end: assembly gap no greater than 0.3 millimetres, and frame diagonal difference no greater than 2.0 millimetres up to 2000 millimetres and 3.0 millimetres above it. Both are worth measuring on arrival. Neither tells you whether the cavity is full.

Weep provision is checkable too, and published acceptance criteria give an unusually concrete rule that you can count on site: for a compartment length below 1500 millimetres, two weep openings; between 1500 and 2000, three; above 2000, four. The outer openings sit with their centreline at least 75 millimetres from the corner, intermediate openings are evenly distributed, and the opening is not smaller than 5 by 25 millimetres. Counting holes and measuring their distance from the corner is a five-minute check that a photograph of the finished window will never give you.

Station 8: the dew point tells you about today

Insulated glass is the component that fails latest and costs most to ｒｅｐｌａｃｅ, and it is the component whose critical work is entirely enclosed. What you can see on a finished unit is one bead of sealant.

The published product standard sets the dew point below minus 40 degrees Celsius, requires a moisture penetration index no greater than 0.25 with an average no greater than 0.20, sets an initial gas content of at least 85 percent by volume for gas-filled units and at least 80 percent after the durability sequence, and requires that after ultraviolet exposure the internal surfaces show no fogging, condensation or contamination and the sealant shows no significant deformation. It also specifies what the edge should look like: the inner seal continuous and uniform, the outer seal neat, fully bonded to the glass and not projecting beyond the glass edge; spacer material free of twist, with a clean surface free of staining, spotting or patchy oxidation; and no foreign matter in the cavity.

The failure model is worth understanding because it explains why the test result is weaker than people expect. A sealed unit is a dynamic balance between two capacities: the edge seal resisting inward moisture migration, and the desiccant absorbing whatever gets through. Moisture always gets through a little. The desiccant handles it. When the seal is poor enough that ingress outpaces adsorption, the desiccant saturates, the cavity dew point climbs, and the day the cavity surface temperature meets it, the unit fogs.

So the honest limitation, stated in published glass-industry commentary: a dew point result judges the sample as it is on the day of testing and does not say how long it will last. It is the moisture penetration index — which measures how much water the desiccant took up during the sequence — that speaks to remaining life. Ask for both. A dew point certificate alone is a statement about one afternoon.

Two more items belong in the plan. Desiccant fill quantity and spacer continuity are invisible after assembly, and a desiccant gap in the spacer is precisely the kind of localised defect that shows up as an edge-only dew point anomaly years later. And gas fill — argon is the usual claim — is not something you can verify non-destructively on a delivered unit without specialist equipment. "Argon filled" is therefore a claim you either accept on a documented initial content figure, or you do not accept at all. Decide which before you order, not after you uncrate.

The distinction that sits exactly on the thing you care about

The published aluminium window standard divides product inspection into two categories with separate item tables: exit-factory inspection and type inspection. Most buyers assume they are two intensities of the same thing. They are not, and the difference lands precisely on durability.

In the standard's inspection item table, the repeated open-and-close durability item is listed with one test unit, marked as required for type inspection and not as an exit-factory item. Durability is never part of what a batch is checked against before it ships. What exit-factory inspection does cover, per the published text: appearance and surface quality and assembly quality are subject to full inspection; frame and sash assembly dimensional deviation is sampled at 5 percent with a minimum of three units from every batch of 100, a batch of fewer than 100 still being a batch.

The acceptance logic is also published and is worth knowing verbatim in substance: if every sampled unit conforms, the batch passes; if more than one unit fails, the batch fails outright; if exactly one unit fails, a double-size sample may be retested on that item, and the batch passes only if the retest is clean.

Type inspection is a different animal. Published summaries describe it as being drawn from a qualified batch of at least 100 units, with three units for air, water and wind performance and one for durability. And it does not happen continuously — it happens when triggered: new product or transferred production qualification, a significant change in materials or process, resumption after a production stop, a periodic cycle, a material difference between exit-factory results and the last type test, or a request from a quality supervision body.

Here is the sentence to take away. When a quotation or a certificate says "inspected and passed", it usually means appearance and assembly were looked at in full and dimensions were sampled at 5 percent. It does not mean the durability of your batch was measured, because that measurement is not in the exit-factory programme at all. You are not buying a test result. You are buying a trigger condition, and the trigger may not have fired this year.

One further detail that almost nobody writes down, and which is free to enforce. The standard specifies a test sequence rule: where the same specimen set is used for two or more items, the order must be chosen so that an earlier test does not compromise a later one — non-destructive before destructive. Where a unit is required to be both thermally insulating and fire-resisting, the thermal test and then the fire integrity test are run on the same specimen in that order. If a report you are handed does not state the sequence, you cannot tell whether the numbers in it are independent.

On wall thickness and the minimum performance floors, we will repeat what we have said in other articles rather than quietly pick a side. A widely quoted public summary of the current edition gives external window base material thickness raised to 1.8 millimetres and external door raised to 2.2 millimetres relative to the previous edition, alongside minimum water tightness of the order of 250 pascals for external windows and 150 for external doors. We have also found public sources giving two different figures for the external door minimum in the same edition. We are not going to choose for you. Read the edition your project adopts, and write the number into the order.

The sliding-window-specific check: the one thing you can still test at the end

Everything above applies to aluminium windows generally. This part is specific to sliders, and it is also the part of the QC plan that survives into final inspection — which is exactly why there is no excuse for skipping it.

Published factory inspection guidance states the requirement well: inspect the full path from the internal collection chambers through to the exterior weep outlets. Openings must be present, correctly located, clear of sealant, and protected in a way that does not block flow. A controlled water check is what reveals reversed components, blocked weeps, missing end dams, and sealant bridging — the case where someone sealed a joint and in doing so bridged and closed the drainage route.

Sealant bridging deserves a line of its own. It is not a manufacturing defect in the usual sense; it is a correct action taken in the wrong place, often by whoever sealed the perimeter, and it converts a frame that was designed to drain into a frame that stores water until it overflows inward. Nothing on a drawing shows it. Nothing in a dimension check catches it. Pouring water into the track does, in ninety seconds.

The geometry your method has to respect is product-specific, and ours is published: the two numbers on our top-hung side-press series are two tracks on one profile, not a tolerance band; our larger sliding series offer a 43 millimetre high track and an 8 millimetre low track as a choice rather than a variation; a stepped 45 millimetre sill with an optional 85 millimetre water baffle is a decision about how much water the opening has to hold before it matters; and one series uses a passive one-way drainage valve. A QC method that does not know which of those it is standing on is not a method. Ask for the section drawing with the drainage path drawn as a route — entry, travel, exit — rather than as a closed outline.

And one clarification that saves arguments: water standing in the outer track is the system working. The outer track is designed to receive and discharge. The defect is water crossing the inner barrier, and those two complaints are usually reported in the same sentence by whoever is standing in the room.

Three different strengths hide behind the word "inspection"

When a quotation says "100 percent inspected", it almost always refers to row one. That is a real and useful activity. It is not a statement about the four rows that matter most.

What you can actually write into the contract

The instrument is an inspection and test plan, and the structure is not something you have to invent. A published sample plan for aluminium windows from a national building authority uses eight columns: item number, activity, responsibility, inspection method, requirement reference, acceptance criterion, frequency, and record. You do not need a quality department to fill that in. You need to name four things for each row: what is checked, against what, how often, and who signs.

The gates worth naming are six:

• Pre-production. Approved drawings, bill of materials, opening direction, handing, glass make-up, hardware, finish code, screens, packing instruction. Published guidance adds the detail that makes this gate work: every production document carries a revision number so a superseded drawing cannot return to the line. Your opening dimensions go on the drawing in millimetres, and for a slider the slide travel goes on it too — the visible glass is not the hole you are filling.
• Incoming material. Profile system, temper, wall thickness, finish and batch; glass thickness, type, safety treatment, coating orientation, spacer and cavity; gasket groove match and condition; sealant identity and shelf life; hardware brand, model, finish and fastener material.
• First article. One completed unit, held before mass production starts.
• In-process. Cutting, machining, drainage, joints, glazing and operation, typically at around a third of the run.
• Final random inspection. Finished batch before packing.
• Shipment release. Corner protection, wrapping, crate marks, quantities and loading sequence.

The first article is not a sample

This is the distinction most procurement documents blur, and it is worth stating flatly. A sample is selected. A first article is the first one off the line. Published quality-management practice requires a first-article inspection at the start of every batch or shift and after any tool change or process parameter change, with production released only on confirmation by an authorised person.

And here is why the first article is the most valuable event in the whole plan: it is the only unit anyone will let you destroy. Cutting a corner open, running a cross-cut on the coating, pulling thermal-break coupons, taking a dew point — every one of the four invisible stations has a real test, and every one of those tests is destructive. Destructive testing is only possible on the unit that has not been promised to a building yet. After that, you are holding a report about something that is gone.

One caution about outsourcing this. A third-party inspector can be your eyes at a gate, but only if the gate is named in the contract and dated in the production schedule. An inspection that is not tied to a specific operation is a photograph, and a photograph of a finished window proves nothing about any of the four rows above.

Finally, a scope boundary that saves time. A QC plan is written per product family. The plan you write for a sliding window will not serve a folding door — including an application as different as folding doors interior — and it certainly will not serve an industrial swing door, where the governing regime is a cycle count on something opened all day by people who are not thinking about it. Different load paths, different failure signatures, different tests. If one QC document is offered to cover all of it, read it as marketing. The same logic applies upstream at the drawing stage: a nominal call-out such as a 36 x 80 entry door names a family, not a manufacturing dimension, and the approved drawing has to carry the actual frame size in millimetres.

What we publish, and what you still have to ask for

A QC plan needs acceptance criteria, and acceptance criteria need declared values. The honest version of our own catalogue is below, including the gaps, because a gap in a published table is itself a fact you should plan around.

Ten lines to put in the enquiry, and the answers that mean nothing

Fill these in before you ask for a price. The right-hand column is the valuable one, because a plausible-sounding answer that carries no information is more expensive than no answer.

Six questions we get asked

Can a third-party inspector ｒｅｐｌａｃｅ all of this?

No. A third party can sample and count at the end, and can witness if the witness point is named and dated. What it cannot do is create a witness point that was never scheduled. Book the operation, not the visit.

Is a thicker coating better?

Up to a point, and then it reverses. Published auditing commentary warns that beyond roughly 80 micrometres the risk shifts from corrosive breakthrough to cure-stress cracking and delamination. Ask for a range, not a maximum.

Why will no one give me a failure rate?

Because a rate depends on who counts, over what denominator, at which door. We wrote about that at length elsewhere. The shorter version for a QC conversation: what you can get instead of a rate is a set of dated records, and dated records are more useful than a percentage anyway.

Is fogging between the panes always a defect?

Almost always, yes, and it is the one complaint in the category that nearly always is. It is also worth separating from condensation on the room-side surface, which is a ventilation question and a completely different claim. The remedy for the first is a new sealed unit, not a new window.

Does a type-test report cover my order?

It covers a configuration that was tested on a date. Compare the specimen size, the glass make-up, the opening form and the edition of the standard against your order. If any of those differ, you have a report about something else. And check that it was triggered at all.

What is the single cheapest thing to add to a first order?

A first article that you are allowed to destroy, plus a pour test on every unit. One costs a delay of a day. The other costs ninety seconds per window. Between them they cover the two invisible stations most likely to generate a complaint.

The short version

• Counting defects is a different activity from controlling quality. Counting happens at the end; control happens at nine stations, four of which leave nothing to inspect.
• The coating is decided in a bath. Ask for the bath log, not the adhesion grade.
• The thermal-break test is destructive and run on coupons. It is evidence about a process on a day, not about your frames.
• Glue injection and drainage are one decision. Ask whether the deflector is fitted.
• Fake injection is invisible at final inspection and takes four seconds to spot at the corner station. Look for the process hole.
• A dew point result describes one afternoon. Ask for the moisture penetration index alongside it.
• Durability is a type-inspection item, not an exit-factory item. "Inspected and passed" does not include it.
• Test sequence is specified. Non-destructive before destructive. A report that omits the sequence cannot be read.
• For sliders specifically: pour water into the track on every unit. It is the one hidden failure the final inspection can still catch.
• Weep count and offset are published rules you can check on site in five minutes.
• The first article is the only unit anyone will let you destroy. Use it.
• Without a declared performance class, the word "failed" has no definition. Ask for the class before you need it.

If you want the section drawings, the drainage route drawn end to end, and a first article you are allowed to cut open, say so in the first email rather than the fourth. It changes what gets scheduled, and scheduling is what quality control actually is.