Satellite component control with AI — the part flies only once. So does the file.

The problem

A one-off part, a file that has to be perfect, and a program that forgives nothing.

A satellite component is not reproduced: it is built once, it flies once, and the file that goes with it has to satisfy the ESA or NASA PA engineer down to the last detail. The space quality manager lives with three fronts at the same time:

1. Cleanroom inspection — ISO class 8/7, restricted personnel, restricted contact. Every manual inspection adds contamination risk and human error from fatigue. And yet a crack in a NASA-STD-8739.3 solder joint or visible particulate contamination is immediate scrap.
2. EEE parts traceability — every component comes in with its lot date code, its URO, its DPA, its ESCC QPL entry and its outgassing history. The master spreadsheet “knows” which manufacturer validated which lot, but the day the person in charge goes on vacation, the knowledge goes with them.
3. AIT dossier and ABCL — the As-Built Configuration List, the NCRs, the PFRs, the special process certificates, the bake-out records. When the program customer calls the review, the quality team spends weeks rebuilding what it already knew and had already done. Every review is a sprint, and every sprint costs money and nerves.

The classic system works — but it rests on the space veteran's eye and on a master spreadsheet that copes worse and worse with the pace of a real program. Losing the knowledge of a single senior engineer can stall an AIT campaign for weeks.

How it fits the IRIS system

iLEAN does not recertify your cleanroom — it seals the cracks between the PLM, MES, ERP and master spreadsheets you already have.

The problem in space is not a lack of data — it is data scattered across islands that gets rebuilt by hand on review day. iLEAN acts as the putty that fills the gaps between PLM (configuration), MES (operations), ERP (EEE parts purchasing) and the veteran's master spreadsheets, without asking you to touch your PA Plan or your approved QMS.

Edge looks at the part inside the cleanroom without putting extra people in. Connect captures the certificate whatever channel it arrives on. The agent keeps the ABCL live part by part. The person signs — always.

The three iLEAN pieces applied to satellite component control:

• Edge — a vision head with a convolutional neural network, mounted fixed or on an articulated arm inside the cleanroom. It reads space soldering (NASA-STD-8739.3), torque on critical joints, visible particulate contamination and geometric deviations in milliseconds. It works with no network: the critical cycle does not depend on WiFi in a room with a Faraday cage.
• Connect — captures the certificate of origin, lot date code and URO of every EEE part whether it comes by email from the qualified distributor, as a PDF from the foundry or in the space procurement lead's master spreadsheet. It also captures what comes from outside (GIDEP alerts, ESCC QPL changes, technical notes from the program customer) at second zero.
• Agents in Central — they cross-reference Edge images, lot traceability, bake-out / outgassing measurements and the MES work order to keep the ABCL live, draft consistent NCRs/PFRs and assemble the AIT procedure. The team reviews instead of transcribing. The three rings guarantee that ITAR/EAR data does not leave ring 1 without a human signature.

See the full IRIS architecture →

Before and after

Manual satellite component control vs. cross-referenced control with iLEAN

Aspect	Classic space AIT	With iLEAN Edge + Connect + Agents
Cleanroom inspection	Manual inspector + magnifier + paper checklist	Edge head with a trained CNN + the PA engineer's signature
EEE parts traceability	Master spreadsheet + PDFs in a shared folder	Captured at second zero + cross-referenced with GIDEP/ESCC QPL
As-Built Configuration List	Rebuilt the week before the review	Live per component, signed in line
Consistent NCRs and PFRs	Free text written by the engineer on shift	Agent template + photo/Edge evidence
AIT dossier for the program customer	A manual sprint lasting weeks	Generated; the team reviews
ITAR/EAR — data sovereignty	Agreements based on intent	Three physical rings + a signed file drop per plant

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 immersion diagnostic.

• A plant building electronic units for satellites with a class 8/7 cleanroom, a mix of ESA programs + commercial customers, EEE parts with an ESCC QPL entry and mandatory URO.
• Edge pilot on one critical AIT station + Connect for EEE parts certificates + an ABCL agent. First value expected within a few weeks (one NCR caught in line, one questionable lot blocked before assembly).
• Indicative payback between 5 and 9 months, depending on how often rework is triggered by contamination or incomplete traceability, and on the average time spent preparing the AIT dossier for each program review.
• The hard levers are one flight unit recovered before ATP and a ≥ 30% reduction in the time spent preparing the AIT dossier. One flight unit pays for the entire pilot.

And the PA manager's reasonable doubt

“What if the AI passes a space solder joint that is not good?” — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely compares an image against a reference trained on your own NASA-STD-8739.3 solder joints, the best models brought error below 1.5% ^[1]. And even so, what is critical is never decided alone: iLEAN holds the unit, the PA engineer signs, and ring 1 (where the program's sensitive ITAR/EAR data lives) never accepts anything that has not been validated, packaged in signed JSON and approved by a person.

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