Prototype Manufacturing · PUF sandwich panels

Catch the bad panel at the line, not at the customer's site.

A hidden foam void looks perfect from outside. It shows up weeks later, when the panel bows on site or ices over in a cold room. Here a thermal camera finds it while the panel is still warm, a second camera checks the skin, and every panel leaves the plant with a record the buyer can scan, back to the chemical drums and steel coils it was made from.

01 · The problem

The worst defects are the ones you can't see.

A PUF panel is two steel skins with foam poured between them. If the mix runs a little cold or the ratio drifts, the foam leaves an air pocket or doesn't stick to the skin. From outside the panel looks fine. Today's checks, a tap test and a tape on a few panels per batch, mostly miss it.

Where a defect gets caught

  1. 01Mix & pour

    Mix drifts: flagged before bad panels pile up

  2. 02Press

  3. 03Saw & thermal

    Tape check, 1 panel in 20

    Void seen as a cold patch, every panel

  4. 04Skin check

    A glance, if someone's free

    Dents and scratches, every panel

  5. 05Dispatch

    Panel leaves with its record

  6. 06Site install

    Panel bows, joints don't close

  7. 07In use

    Sweating, ice patches, power bills

TodayWith line vision
At the line

One panel

Scrap or rework it. The cost is the panel.

At the site

Panel + truck + crew

A new panel, freight both ways, a crew to pull and re-fit it, and a contractor who now checks every load.

In a cold room

Panel + the client

Ice, sweating walls and higher power bills before anyone traces it back. Pharma and food clients ask for proof, and there isn't any.

02 · See it run

What the eye sees. What the camera sees.

Foam gives off heat as it sets, so a fresh panel is warm all over. An air pocket doesn't make heat, so it shows up as a cold patch. Flip the view to see the same panels the way the thermal camera does.

Checked
0
Passed
0
Held
0
A PUF panel line: foam poured between steel skins, a double-belt press, a saw, a thermal camera and a skin camera. Panels with faults are moved to a hold lane; good panels go to the stack. Foam pour Double-belt press Saw Thermal camera Skin camera Stack Hold lane: tap test 40–45 °C, foam sets

Line 1 · event log2:02 pm

    A simulation of the prototype's output, not footage from a plant. Panel sizes, temperatures and messages follow a real 80 mm wall-panel line.

    03 · Trace every panel

    From the chemical drum to the wall it's fixed on.

    A panel's foam comes from a drum of polyol, a drum of isocyanate and a blowing agent. Its skins come from two steel coils. Every lot is lab-tested before it's allowed near the line, every run is tested after, and every panel carries the whole family tree.

    Seven lots of polyol, isocyanate, blowing agent and steel coil pass a lab gate (one isocyanate lot is stopped there), feed three mix runs, which make three groups of panels sent to a potato cold store, a pharma plant and a dairy plant.

    Sample lots, results and sites. Specs are typical for 40 kg/m³ PUF wall panels; each plant sets its own.

    Gate 1 · every lot that comes in

    Polyol: water content
    Wet polyol foams unevenly and shrinks in the cold.
    Isocyanate: NCO content
    Too low and the foam stays soft and weak.
    Blowing agent: purity
    Decides how well the finished foam insulates.
    Steel coil: gauge, zinc, paint
    Dents and rust start here, years before anyone sees them.

    Gate 2 · samples from every run

    Core density
    Too light is weak; too heavy wastes chemical.
    Compressive and bond strength
    Skins that stay stuck don't bow, dent or peel.
    Thermal conductivity (λ)
    The number the cold-room designer sized the plant on.
    Shape at −20 °C
    Foam that holds its size keeps the joints closed.
    04 · Where it pays off

    A panel's real test is the room it ends up in.

    The same hidden fault costs different things in different rooms. Pick one.

    Batch running

    Floor plan of a sterile filling suite. Air pressure steps down from the Grade B filling room through Grade C and the airlock to the corridor, so air always flows out of the cleanest room. With one bad joint, air escapes through the wall, Grade B loses its pressure and particle counts rise past the limit.

    Sterile filling suite

    What goes wrong

    • An unsealed joint or an off-size panel leaks air, and the cleanest room can't hold its pressure.
    • A void behind the skin dents when a trolley hits it. The coating cracks, and microbes have a place to hide.
    • The auditor asks what's inside the walls. A test certificate per order doesn't say which panel.

    What the panel record proves

    Every wall panel's chemical lots, lab results, thermal scan and size check, filed with the room's qualification papers.

    Result for the plant

    Rooms that hold pressure and pass particle tests first time, and an audit question answered in minutes.

    Simulations, not footage. The ten-year curves are illustrative: a pilot measures panels at the plant, and site data would come from buyers who share it.

    More rooms that depend on the panel

    • Vaccine and medicine stores2–8 °C held for audits; a warm wall means a temperature excursion
    • Hospital operation theatresPressure, cleanable surfaces and a record for the hospital's audit
    • Fruit ripening chambersEven temperature and gas-tight walls ripen the whole room at once
    • Refrigerated truck bodiesEvery bump tests the bond between skin and foam
    • Server and battery roomsFire-rated foam: the lot record shows the fire retardant went in
    • Poultry and dairy shedsCooler animals, lower fan and cooling bills in summer
    05 · What people get

    Not a dashboard nobody opens. A message, a record, a report.

    Each output goes to the one person who acts on it, where they already are.

    Line 1 · QC

    Panel 1,284 held. Cold patch about 40 × 25 cm, 30 cm from the joint edge.

    2nd cold patch in 6 panels. Isocyanate at 19 °C, set point 22 °C, since 2:09 pm.

    → Check the iso heater.
    → Tap-test panels 1,280 to 1,285.

    Tap test: voidFalse alarm
    Shift supervisorOn their phone, within a minute of the panel leaving the press. The reply buttons teach the model what was real.
    Panel recordPanel 1,286Line 1 · 13 Nov 2026 · 2:21 pm
    Panel
    PUF wall panel, 80 mm, 0.5 / 0.5 mm PPGI
    Thermal
    Even, no cold patch
    Skin
    No dents or scratches
    Size
    6,000 mm long, 80.4 mm thick
    Foam batch
    40 kg/m³ (lab test, batch 2611-07)

    Thermal image and both faces attached

    The buyer, on siteScan the QR on the panel label. Pharma and food clients get the same records as one batch file for their audit.

    Shift B report13 Nov · 2 pm to 10 pm

    Made
    4122,470 m²
    Held
    97 real, 2 false
    Alert
    1:40median, min:sec

    What caused the holds

    • Iso temperature drop, 2:09 to 2:31 pm5
    • Skin dents at the stacker2
    • False alarm (wet patch on film)2
    Plant owner and QC headEvery morning on WhatsApp and email: what went wrong, when, and the cause behind most of it.

    Sample outputs from the prototype, with made-up panel numbers.

    Who it changes things for

    Six people, six different days.

    Plant owner

    Hears about a bad batch when a contractor calls, angry.

    Sees every hold the same day, with the cause behind it.

    Site complaints per month

    QC head

    Tap-tests a few panels per batch and hopes they're typical.

    Every panel checked; tap tests go only to the ones flagged.

    Panels checked per shift

    Line operator

    Finds out the mix ran cold after twenty bad panels.

    Told within two panels, with what to check first.

    Bad panels per drift

    Dispatch and sales

    "It was fine when it left" is all they can say.

    Pull up that exact panel's thermal image and photos.

    Disputes settled with proof

    Contractor on site

    Checks each load by eye, and still gets surprises.

    Scans the label. Knows what was checked and when.

    Panels sent back from site

    Pharma or food client

    Asks for quality records. Gets a test certificate per order.

    Gets a record per panel for their audit file.

    Orders where proof wins the bid

    06 · What it's worth

    Put in your line's numbers. See the money.

    Every number here is an assumption, filled with typical values. Change any of them. The sum is shown in full under the result, so you can check it.

    Saved a month

    ₹4.4 lakh

    About ₹53 lakh a year

    Faulty panels a month
    450 m²
    That reach a site today
    225 m²
    Caught at the line instead
    158 m²
    Saved on each one
    ₹2,800 per m²

    450 m² × 50% × 70% × ₹1,400 × (3 − 1)

    Not counted: the bigger gain. Catching a drifting mix early means fewer bad panels get made at all.

    How it works

    Four checks on the line. One small computer beside it.

    1. Thermal camera

      Right after the saw, while the foam is still warm. A cold patch means foam that didn't fill or didn't stick.

    2. Skin camera

      Looks at both faces for dents, scratches and coating damage, before the protective film hides them.

    3. Size check

      Laser sensors measure length, thickness and squareness on every panel, not one in twenty.

    4. Mixing machine data

      Reads temperatures, pressures and the mix ratio. When faults line up with a drift, it names the drift.

    For the engineer: hardware, models and integration
    Thermal
    One long-wave thermal camera over the line after the saw, 384 × 288 pixels or better. Images tied to panel IDs from the saw's cut signal.
    Skin
    Line-scan or area cameras with diffuse lighting, top and bottom, before the film applicator.
    Size
    Laser distance sensors for thickness at three points across the width; encoder and edge sensors for length and squareness.
    Models
    Anomaly detection trained on good panels first, so it works before many faults are collected. A defect classifier is added as the supervisor's tap-test replies label real faults.
    Root cause
    Hold events lined up against the dosing machine's log (component temperatures, pressures, ratio, belt speed) to find the drift that keeps coming before faults.
    Compute
    One industrial PC with a GPU at the line. Works with the internet down; reports and panel records sync when it's back.
    Connections
    PLC and dosing machine over Modbus or OPC UA, read-only. A label printer for the QR. WhatsApp and email for messages.
    Your data
    Images and logs stay on the plant computer. Only summaries and the panel records buyers open go to the cloud.
    07 · How it's measured

    Six weeks, one line. Targets agreed before we start.

    The pilot runs beside your current QC, not instead of it. Every hold gets a tap test or a cut sample, so the numbers are checked, not claimed.

    1. Week 1

      Sit on the line. Log faults the way your team finds them today. That's the baseline.

    2. Weeks 2 to 3

      Fit the cameras and sensors in a planned stop. Record every panel without flagging anything.

    3. Week 4

      Train on your panels. Tune it until false alarms are rare enough to trust.

    4. Weeks 5 to 6

      Run live, side by side with your QC. Count what each one caught.

    MeasureTargetHow we check it
    Hidden faults caught at the line70% or moreTap tests and cut samples on held panels and a random sample of passed ones
    False holds3 or fewer a shiftHolds the tap test clears
    Time from press to alertUnder 2 minutesTimestamps on every hold
    Panels leaving with a recordEvery oneRecords against the dispatch list
    Panels traced to lab-passed lotsEvery oneEach record links its chemical lots and coils to their test reports

    Results: published on this page when the first pilot ends, whatever they are.

    Same idea, other lines

    Anything made on a line can be checked this way.

    The cameras and models change; the idea doesn't. See every unit, find faults where they start, and send a record with the product.

    • Roofing and cladding sheetsCoating, roll-forming marks, profile height
    • Cold-room and cleanroom doorsFoam fill, hinge fit, gasket seating
    • Gypsum and cement boardsEdges, cracks, thickness
    • Tiles and stoneShade, chips, flatness
    • Coated steel coilsCoating gaps, colour, scratches
    • Packaging and printingSeal, print and label checks
    Questions

    What plant owners ask first.

    Can a camera really see inside the foam?

    Not directly. It sees heat. Setting foam is warm, and a void or a skin that hasn't bonded stays cooler, so it shows as a cold patch for a few minutes after the press. That's why the camera sits right after the saw. Very small or very deep voids can still be missed, which is why the pilot measures what it catches against tap tests.

    Our line is old. Does that matter?

    Usually not. The cameras and sensors mount over the line, and the machine data is read, never written. If the dosing machine has no data port, we start with the cameras alone.

    What happens when it's wrong?

    A false alarm costs a tap test, and the supervisor's "False alarm" reply teaches it. A missed fault is what the pilot is there to count. Nothing is scrapped automatically; a person decides.

    Do we need the internet at the plant?

    No. It runs on a computer beside the line. Messages and panel records go out when there's a connection.

    Who owns the images and data?

    You do. Images stay on your plant computer. We only use them to train your models.

    Run a panel or sheet line?

    Send us a month of rejects and complaints. We'll put your numbers in the model above and tell you honestly if a pilot is worth it.

    Talk to us about a pilot