INTERNAL METHOD · UNSIGNED TRAINING PARTNERS

Train engineers on real evidence

The course stack joins UAV architecture, Python, YOLO, flight control, Linux/ROS, embedded deploy, fused localisation and system delivery. Project evidence gates are the course acceptance gates. This page is not an open-enrolment advert.

The academy is the operating system for copying capability inside the programme. Confined-network inspection mixes mechanics, flight control, perception, gas, data and work permits. Training that “only flies” fails at the well-head. Hours, assignments and graduation therefore map to G1 / G3 / G5: structure can be repaired, data can be traced, a user flow can be compared.

Physical pipe flight needs a compliant site, supervision, kit and risk control. Simulation and rig first, then flight — so failure happens where it can be repeated, measured and stopped. A student who keeps only success clips does not pass, and those clips do not enter public promotion.

LEARNING STACK

Four capability layers

From system judgement to mechanical / flight-control / AI integration, a student must explain limits, replay a test, keep failures and abort safely.

L1 / 57H

Foundation, Python and vision

About 30 hours on AI UAV architecture and option comparison; 18 hours Python; 9 hours YOLO detection. Output: a system architecture and a replayable experiment. Comparison objects include crawlers, tethered UAVs and collision-tolerant platforms, so YOLO is not treated as the product on day one.

L2 / 60H

Flight platform and deploy

Assembly and flight control, Linux/ROS, model deploy from PC to embedded, scripted control and safety takeover. Takeover sits above trajectory tracking: loss of link, low battery, gas threshold and localisation confidence can all interrupt.

L3 / ADV

Fusion, mapping and navigation

IMU, vision, LiDAR and gas time-sync; GNSS-denied localisation, avoidance, confidence and failure recovery. Long straight pipe with repeating texture, wet walls and dust are required degradation tests, not electives.

L4 / 96H

Mechanical systems engineering

Structure, aero, thermal, sealing, DFM, FMEA, test and design review, forming a P1 prototype. Aligns with the mechanical centre’s 16-week course so two drawing sets do not appear.

SIMULATION FIRST

Simulation and rig, then flight

Before a physical pipe: digital environment, software-in-the-loop, hardware-in-the-loop, power rig, cage collision and recovery drill. Physical flight still needs a compliant site, supervision, kit and risk control.

EVIDENCE FIRST

Failure is a formal deliverable

Keep full logs, failure video, anomalies, versions, environment and measurement uncertainty. Success-only clips do not pass the course and do not enter public promotion.

CAPSTONE

Graduation is not “it flew”

A mixed group must finish an auditable pipe-task pack. The scoring sheet is public and maps to evidence gates, not a judge’s impression.

G1

Structure evidence

Collision, repair, rotor isolation, mass and assembly.

G3

Data evidence

Sensor sync, calibration, segment index and raw files.

G5

User evidence

A controlled scene, a comparison flow, a value indicator and a review.

GOVERNANCE

How the course serves filings and the product

Assignments feed the configuration library: CDR, FMEA, test reports, data dictionary. That gives TSSSU/ESS-class R&D filings an auditable process file, instead of a slide deck written afterwards.

The academy does not promise employment numbers and does not sell a certificate for a course that is not running. If a formal training agreement with BIAA or another body is later signed, the scope will be published on its own. Until then this page describes PipeGuardian’s internal capability-building method only. It is not a public cross-link wall and not a signed joint programme.

WHAT A WEEK LOOKS LIKE

One week closes one replayable question

The academy refuses twelve “innovation topics” at once. Confined-space training moves by closing, not by opening counts.

Example: this week only answers “un-caged, fixed cage and roll-cage blade damage at three speeds”. Output is a table, video hashes and a failure list. Next week may ask field of view. A YOLO class without a segment-level label rule stays on synthetic data and must not claim field defect recognition. A ROS class must include a loss-of-link topic and safety takeover, not only a circular trajectory in simulation.

That rhythm matches the evidence gates and the filing pack: process files appear as the work happens. The academy does not promise a certificate employment rate, and does not rewrite internal training as a certified training product.