Xcelerator · Digital Thread Demonstrator
It starts with a customer RFP — modeled on real U.S. Air Force solicitations — and ends with a flight-verified design. In between, an AI carried a single idea, the QX-250 quadcopter, unbroken through every discipline of a modern engineering enterprise: requirements, reliability and 6-DOF simulation, logical and electrical architecture, mechanical CAD, a managed program, the EBOM / MBOM and quality, and verification — inside the real Siemens Xcelerator tools — every tool operated autonomously by the AI, with no human hands on the tools. The AI Lifecycle Harness is what will steward that loop; this weekend a human still directed each step — but never operated a tool.
The moment
“Watson, come here — I want to see you.” In 1876 one sentence proved a whole new medium was real. This is that kind of threshold — the first time an AI carried a single design idea, unbroken, through every discipline of a modern engineering enterprise and committed it to the system of record.
The digital thread's ambition is to collapse the distance between a customer's requirement and the as-built, flight-verified configuration — across reliability, systems, electrical, software, manufacturing and quality engineering — so nothing is lost in translation between tools or teams. What follows is that thread, authored end-to-end by an agent and materialized in the live Teamcenter.
Where it begins
Every engineering program starts with a need. This one begins with a request for proposal modeled on real U.S. Air Force solicitations — a full acquisition document with a 41-page system spec — that the QX-250 is built to answer.
Grounding the story in a real RFP is the point: it makes the whole thread accountable. Every requirement derives from it, every design decision serves it, and verification proves the answer against it. The RFP & proposal is the origin; the AI Lifecycle Harness is what carries it forward — decomposing the RFP into a structured program of work and driving it across the tools to a flight-verified design.
Why the order matters
The single most important decision in this program was sequence: run the reliability and system-performance simulations first, and let them drive the design.
By the time late analysis finds a problem, the mechanical, electrical and software designs are already committed and every fix is a rework loop. On the QX-250 the analysis ran up front — and it materially changed the design in all three domains before any were frozen:
→ Electrical
MADe found 24 order-1 cut sets — every single fault drops the aircraft. Redundancy went onto the electronics: a 2nd battery, 2nd flight controller and a BMS/OR-ing block were added. Cheaper parts, yet more reliable (R = 0.9999784).
→ Software
The Amesim 6-DOF model showed the attitude loop recovers fast but is lightly damped — it rings (ζ ≈ 0.15) and holds attitude only. A control-law finding: it wants more rate-damping and an outer position/heading loop.
→ Mechanical
The sweeps proved geometry and control are coupled: shrinking the arm to 125 mm without re-tuning the controller destabilises hover (roll diverges to 67°). Geometry can't be changed in isolation.
The aircraft
A 5-inch freestyle-class quadcopter — the vehicle the whole thread describes.
single-string design point · ~650 g as-built (dual-battery redundancy, right-sized packs) · ≤ 700 g limit ✓
at the 500 g design point · 3.83:1 as-built (650 g, flight-verified) · ≥ 2:1 required at WOT
with redundancy, MTTF 4673 h (MADe)
lightly damped, ζ≈0.15 (Amesim 6-DOF)
Explore the thread
Every page — grouped the way the work actually flows, from the customer RFP through to a flight-verified design. The top menu carries the same groups.
FMECA, fault tree and an RBD with dual-redundant electronics. Two MIL-STD reports. The failure physics that seeds the design.
Open →A purpose-built 6-DOF model with a cascade flight controller — holds position, rejects a gust, flies a mission, re-verified at the as-built mass.
Open →A 51-requirement SRS with derive links, 13 functions and a 13-block logical architecture — live in Teamcenter, shown as relation graphs.
Open →The NX geometry flying the validated Amesim 6-DOF trajectory — form driven by behavior, live in your browser.
Open →The physical layer: the EBOM split make-vs-buy, standard parts as neutral CAD, and the whole redundant QX-250 assembled in NX — 53 components on one parametric datum.
Open →The functional architecture and a synthesized logic design, authored natively in Capital.
Open →An AI-authored bundle materialized into Capital with zero manual placement: 13 components → 13 devices, wired with routed pathways, schematic synthesized.
Open →The hardware timing/safety layer: eight SystemVerilog blocks with self-checking testbenches, a formal safety proof, and the real RTL to read.
Open →PX4-based flight software — every RFLP function given a software home, the reliability loop as running code, and a SITL-first path.
Open →A managed Program with approval gates, risks and deliverables, and a 12-task Schedule Manager plan.
Open →A 22-item engineering BOM, a stage-based MBOM, a 10-op process plan and a PFMEA / control plan.
Open →A 51-requirement → verification matrix, live in Teamcenter — native V&V objects with target → measured → PASS (51 / 51).
Open →24 RFP standards materialized in Teamcenter with section-level links, tied to the requirements that cite them.
Open →The steward: reads the RFP, frames the agentic WBS, orchestrates the tools and measures itself — RFP to flight-verified. Human-prompted this weekend; built to steward the loop.
Open →A critical re-scoring of the whole suite against one bar: can an agent deliver a complete, simulation-validated design — with full program management — in a day? Scored on headless, MCP-native, agent-legible and orchestratable — plus the systems-level changes required to get there.
Open scorecard →The quantitative counterpart: instrument the pipeline — time by domain, human-touch count, iterations, rework escapes, seam cost, verification coverage — computed by a stdlib aggregator from an event log. This build is the baseline to beat.
Open dashboard →