Concrete deployment cases, with sensitive names removed.
The previous page only described broad scenarios. This version shows project-shaped cases: starting problem, deployment scope, AI role, evidence outputs, and measurable operating result. Customer names and part identifiers remain anonymised because the programs are regulated or commercially sensitive.
Case 01: aerospace supplier audit package.
Starting point.
A Tier-1 aerospace supplier had four additive manufacturing sites producing low-volume structural brackets. Audit evidence was assembled manually from slicer exports, printer logs, inspection PDFs, and operator records. The review team could prove conformance, but only after days of cross-checking.
Deployment.
3DCIPHER enrolled 28 printers with Vault3D, connected operator identity through the customer's IdP, and issued TwinCert records for one flight-qualified part family during the pilot.
AI involvement.
An evidence-classification model matched inspection records, material-lot files, and printer bundles to the correct TwinCert fields. An anomaly model flagged bundle fields that did not match normal site behavior, such as unusual slicer build drift or missing sensor envelopes.
Result.
The team exported a verifier-ready audit package for 8,400 parts. Internal evidence collation dropped from roughly two working weeks to under two days for the sampled audit cycle.
| modules | Vault3D, TwinCert, console, verifier-CLI |
|---|---|
| AI outputs | evidence matching, anomaly flags, audit-summary draft, missing-field detection |
| human control | Quality engineers approved every AI-suggested field mapping before certificate issuance. |
Case 02: patient-specific implant traceability.
Starting point.
A medical implant manufacturer needed a part-level record tying patient-specific geometry, titanium powder lot, printer chamber conditions, post-process inspection, and custody transfer to the surgical site.
Deployment.
The pilot connected one production cell, one quality system export, and the customer's archive. TwinCert became the record that quality and receiving teams could verify offline.
AI involvement.
AI normalized inspection result text from different lab templates, extracted material-lot references, and generated a review checklist showing which evidence fields were machine-derived, operator-entered, or missing.
Result.
The manufacturer moved from document-folder review to a single signed certificate per implant, with AI used as a preparation assistant rather than an autonomous compliance decision-maker.
| modules | Vault3D, TwinCert, customer console |
|---|---|
| AI outputs | document extraction, field normalization, missing-evidence checklist |
| human control | Release authority stayed with the customer's quality owner. |
Case 03: licensed spare-parts network.
Starting point.
An industrial OEM licensed out-of-warranty spare parts to regional print partners. The OEM needed to distinguish authorized production from counterfeit returns without forcing every receiving warehouse into a complex PKI workflow.
Deployment.
MeshGuard embedded per-licensee payloads into CAD variants. Receiving sites scanned inbound parts or supplier meshes and used verifier-CLI to resolve the payload against the customer manifest.
AI involvement.
The neural detector compared mesh and point-cloud features against the embedded payload. A triage model ranked suspicious returns by detector confidence, supplier history, part class, and mismatch severity.
Result.
Warehouse teams received a simple accept, review, or reject workflow. The OEM kept recall paths intact for licensed parts and moved disputed items into a documented evidence chain.
| modules | MeshGuard, verifier-CLI, optional TwinCert |
|---|---|
| AI outputs | watermark detection, confidence scoring, counterfeit triage ranking |
| human control | AI ranking prioritized review; it did not automatically reject suppliers. |
How AI is used in customer deployments.
| AI capability | where it runs | what it does not do |
|---|---|---|
| Neural mesh watermarking | CAD distribution and receiving-site scan flows | Does not prove mechanical conformance by itself. |
| Evidence extraction | Customer quality system or audit-builder daemon | Does not issue a certificate without customer approval policy. |
| Anomaly detection | Console, SIEM feed, or audit preflight | Does not override the signed manifest or HSM result. |
| Audit-summary generation | Customer-controlled audit export path | Does not replace the signed evidence bundle. |