An internal defect in an aerospace composite part only shows up under ultrasound — and almost nobody cross-references it with the real autoclave curve. iLEAN does.

The problem

Three sources tell the story of the part — and they are almost never in the same report.

An aerospace composite part with no internal defect is the intersection of three realities that, in many plants, live in different systems:

1. The lay-up — which prepreg, how many plies, in what orientation, with which heat lot. It usually sits on a process sheet signed by the laminator.
2. The autoclave cycle — the real temperature, pressure and vacuum curve during cure. It lives in the autoclave SCADA, exported to PDF at the end of each cycle.
3. NDT — ultrasound (A-scan, C-scan) and surface vision. It lives in the inspection equipment, with a per-part report generated after the fact.

The process engineer knows it: when porosity above the limit shows up on a part, the root cause is in the autoclave curve — but the NDT report and the autoclave log are not in the same place, and linking them by hand for every part is a couple of hours of work that almost nobody does in series production. The defect does get detected, yes. The root cause goes unattributed. And as long as the root cause stays open, the pattern repeats — more rework, more marginal parts, more composite scrap cost.

How it fits the IRIS system

iLEAN does not add a fourth system — it stitches the lay-up, the autoclave and NDT into each part's dossier.

The problem with composite control is not a lack of information: it is information living on islands that, at the critical moment (the part comes out of the autoclave, NDT detects porosity), never reaches the person analyzing the cause already stitched together. iLEAN acts as the putty that fills those cracks, without asking you to change the autoclave, the NDT equipment or the laminator's procedure.

Edge captures the real autoclave cycle. Vision reads the surface and the C-scan. Brain correlates all three and proposes a root cause. The person signs — never the other way round.

The three iLEAN pieces applied to composite control:

• iLEAN Edge — a terminal on the autoclave that captures the thermocouples, the pressure and the vacuum per bag in real time. If the plant loses its network during a 4-hour cycle, Edge keeps capturing and links the complete log once the connection returns. With no black box on the autoclave, no later analysis is possible.
• iLEAN Vision — captures a surface image at the autoclave exit (a CNN trained to detect fiber bridging, excess resin, cosmetic defects) and also captures the C-scan from the ultrasound equipment you already have (from the screen, via camera, or by direct integration if the equipment allows it). It links it to the part through its traceability.
• iLEAN Brain (Agents) — agents that cross-reference the lay-up (prepreg heat lot, prepreg age in cold storage, ply orientation) + the real autoclave cycle + the NDT result, and identify the pattern: “the parts with porosity above 1% are the ones from prepreg lot X when the vacuum dropped below Y mbar at minute Z of the cycle”. The engineer receives the root cause; the agent never acts alone.

See the full IRIS architecture →

Before and after

Classic composite control vs. cross-referenced control with iLEAN

Aspect	Classic control (silos)	With iLEAN Edge + Vision + Brain
Autoclave cycle	PDF exported at the end, filed per cycle	Real-time capture, linked to each individual part
Prepreg traceability	Process sheet + cold storage card	Heat lot + prepreg age linked to the part
Surface defect	Visual inspector when the part is received	Vision in line, before it enters NDT
Ultrasonic C-scan	Per-part NDT report, read by the inspector	Pre-classified by Brain, the inspector validates and signs
Root cause of porosity	A meeting a week later, a qualitative conclusion	Pattern identified by Brain from real data
Per-part dossier for Part 25 / CofC	Rebuilt by hand	Alive from minute one, with evidence attached

Impact estimate

Impact estimate for your plant — to be validated with your numbers.

The block below is an estimate to be validated with the specific data of your plant. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• Tier-1 plant with manual lay-up or ATL/AFP, a large autoclave with several parts per cycle, in-house ultrasonic NDT. Volume in the order of tens to hundreds of parts/month.
• Edge pilot on the autoclave + Vision on the exit line + Brain for root cause correlation. First value expected within a few weeks: a cure dossier per individual part instead of a log per cycle.
• Indicative payback between 4 and 9 months. Levers: a reduction of parts in rework of ≥30%, a reduction of scrapped parts (the hard lever — a structural composite part is expensive), traceability ready for Part 25 with nothing to rebuild.
• First value within a few weeks: the first case of an unattributed root cause gets closed with the first month of cross-referenced data.

And the NDT lead's reasonable doubt

“What if the AI says a part is fine and it has a delamination it did not see?” — hallucination is a problem of free generation, not of anchored tasks. Brain does not “interpret” the C-scan: it pre-classifies regions by probability and proposes, and the qualified NDT operator reviews and signs. In anchored tasks, the best models brought error below 1.5% ^[1]. And the responsibility for the signature still belongs to the NDT operator — iLEAN's three safety rings guarantee that no agent can release a part without a person validating it. It is about assisting and simplifying, not replacing.

^[1] OpenAI paper “Why Language Models Hallucinate”, 2025 — on the reliability of AI in anchored tasks.