Case study · Utilities
Substations watchedaround the clock.
One of the world’s largest electric utilities, serving customers through a vast network of substations, transmission lines and distribution infrastructure. It operates tens of thousands of substations housing high-value, safety-critical equipment, where an undetected defect, and oil leakage in transformer equipment above all, can cascade into power supply interruption, production shutdowns and damage downstream. To get past the limits of manual patrol inspection it deployed an intelligent patrol system combining fixed cameras, drones, robot dogs and AI defect recognition.
Constraints and answers
A patrol that never sleeps, and never misses a meter.
Cameras, drones and robot dogs on one platform
Manual inspection left monitoring points uncovered
Manual patrol inspection was labor-intensive, inconsistent, and could not cover every monitoring point around the clock.
Too few defect samples to train on
Defect samples for oil leakage, cracks and water accumulation were often missing or too few to train models on.
Flight restrictions blocked drones at some sites
Drone deployment was blocked at some sites by airport proximity and flight approval restrictions.
Baseline standards required only 46 types
Traditional systems lacked scalable AI defect recognition, and baseline standards required only 46 recognition types.
One missed defect cascades down the chain
An undetected transformer defect risks a chain failure: oil leakage, explosion, substation paralysis, and supply interruption downstream.
Intelligent substation patrol system
Fixed cameras, autonomous drones and robot dogs integrated into one AIoT platform for continuous equipment monitoring across substations.
Multi-tier AI defect recognition engine
Pre-trained models for common defects, models fine-tuned to manufacturer-specific equipment variations and models trained from scratch for special cases such as animal intrusion and sandstorms.
Large-model enhanced analysis layer
Fine-tuned large language models deployed at 7B, 14B and 32B parameter scales, supporting decision-making, analysis and overall evaluation.
Real-time hot update architecture
Algorithm updates are pushed to production without interrupting system operation or the patrol schedule.
Different eyes on the equipment, and nothing waits for a schedule.
The results
66 defect types recognized, updated with zero downtime.
- 24/7
- CONTINUOUS MONITORING
- One AIoT platform of fixed cameras, drones and robot dogs, watching round the clock.
- 66
- DEFECT TYPES RECOGNIZED
- Plus 9 meter types and 4 switch types, across the full power chain.
- 40
- TRANSMISSION ALGORITHMS
- Tower corrosion, insulator defects and construction machinery detection.
- DOWNTIME ON UPDATES
- Algorithm updates pushed to production without interrupting patrols.
Hot updates are what keep a 66-model estate current without standing the patrol down.
The stack
What was deployed
Edge computing at the station, regional real-time analysis and centralized intelligence processing, each on GPUs sized for the tier.
Algorithms across generation, transmission, substation, distribution and consumption, from belt detection to load forecasting.
One platform joining fixed cameras, autonomous drones and robot dogs into a single monitoring and detection picture.
Fine-tuned large language models at 7B, 14B and 32B scales, carrying decision-making, analysis and evaluation across patrol operations.
역량에 관한 노트
본 사이트 전반에 소개된 역량, 자격, 배포 레퍼런스는 Amanah 핵심 엔지니어링 팀의 총체적인 작업을 반영합니다. 일부 프로젝트는 이전 또는 계열 법인 하에 수행되었습니다. 소버린 배포는 규정상 기밀이므로 개별 고객명은 공개되지 않습니다. 기반이 되는 레퍼런스는 NDA 하에 검증 가능합니다.
Request a Strategy Session
The same platform, inside your borders.
Book a working session with our sovereign-architecture team. We map your mandate, residency and audit requirements to a deployment plan your own people own and operate.
We reply within one business day. Your inquiry stays with our team, never a third party.