VisionAery Analytic · Flagship
Liquid Leak Detection
Continuous edge-AI computer vision for visible puddles, pressurized spray, and visible vapor plumes — trained on real oilfield imagery, AVA LLD-verified before alarming, and wired into the VMS and SCADA paths your operators already trust.
Edge Hardware
Platforms LLD runs on
The model is portable across the edge hardware that's already common on industrial sites. Pick what fits the budget, power envelope, and existing camera fleet — we'll deploy it.
_1779202116804-C31fGfYV.png){kind=logo}
ARM SoC · GPU + DLA
NVIDIA Jetson Orin
Fanless on-pad inference box. The reference platform — best performance per watt for remote leak-detection deployments.
Intel iGPU (OpenVINO) or NVIDIA GPU
x86 server
Rack-mount or industrial PC in the control building. Runs on the integrated Intel GPU via OpenVINO, or a discrete NVIDIA GPU for higher channel counts.
On-camera inference
Axis ARTPEC-8 / 9
LLD runs directly on Axis ARTPEC-8 and 9 cameras — note this on-camera build supports puddle detection only and runs at reduced accuracy compared with the edge-box deployment.
On-camera inference
i-PRO smart cameras
Full LLD analytic runs directly on supported i-PRO smart cameras — no separate edge box required.
Partner Integrations
...and the platforms your sites already run
Native integrations into the control, edge-orchestration, and imaging platforms already on your sites — no rip-and-replace, no extra middleware.
Emerson PAC Edge
Native integration with Emerson IPCs (6010, 7010, 8010 series) and PAC Edge — alarms and analytic events feed directly into the same control platform your operations team already runs.
Download Emerson flyer (PDF)
Armada Atlas
Edge-orchestration integration with Armada's Atlas platform for remote fleet visibility, deployment, and lifecycle management of the inference boxes in the field.
Viper Imaging
Integration with the Viper Imaging platform — thermal and visible imaging streams from Viper-managed cameras flow into VisionAery analytics with no extra middleware.
VisionAery LLD in the field
Real detections from real production sites
Every overlay below was generated in real time by VisionAery LLD running on the site's edge inference box — visible puddles, pressurized spray events, and visible vapor plumes, on the cameras and in the lighting our customers actually run.
Trained on real oilfield imagery only — never synthetic data, and we never will. Every model in the VisionAery suite is built and refined on footage captured at active production sites — the conditions, equipment, and lighting our customers actually run.
Puddle
Puddle
Puddle
Spray
Spray
Puddle
Vapor
Puddle
PuddleLive model output
Short loops captured on production sites — overlays drawn by the LLD model frame-by-frame, no post-processing.
See it for yourself
Want to see LLD run on your site?
Try the live demo with your own image or video, or talk to our team about a POC on a tank battery, wellhead, or gathering header you already have a camera on.
AVA LLD · Alarm Verification Agent for Liquid Leak Detection
A reasoned verification on every alarm
The moment LLD detects a release, AVA LLD — our Alarm Verification Agent for Liquid Leak Detection — reasons about the frame and returns a verdict: confirmed, low-confidence, or dismissed. Only confirmed releases reach on-call. Everything else still bookmarks in the VMS for review, but it stays quiet. Deployed correctly, AVA raises LLD accuracy to 95%+ in alarm detections — higher signal, less alarm fatigue, and an analytic operators will actually leave running on every camera.
Step 1 · LLD
Detect
VisionAery LLD fires on visible puddle, spray, or vapor.
Step 2 · AVA LLD
Verify
AVA takes a second look at the alarm frame and returns a verdict.
Step 3
Alarm
Confirmed releases alarm VMS, SCADA, and on-call notifications.
AVA LLD
Alarm Verification Agent · online
0/1500 · ⌘/Ctrl + Enter to send
AVA LLD is AI and can make mistakes. Answers may be inaccurate — for production guidance, contact our team.
VisionAery field crew validating an LLD deployment on a live tank battery.
Field-testedSchedule a POC
See LLD prove itself on your sites
A VisionAery field crew comes to a representative pad of yours, stands up LLD on existing or temporary cameras, and runs the analytic against your real scenes — your tank batteries, your lighting, your weather, your operating cadence. You see the detections, the AVA verdicts, and the VMS/SCADA alarming end-to-end before any commitment.
- On-pad camera survey and AOI selection by a VisionAery engineer
- Live LLD inference with AVA verification on your actual scenes
- VMS bookmarks + SCADA alarm preview wired into your existing systems
- POC report with detection samples, false-positive notes, and a deployment plan
What it does
Seven capabilities that define VisionAery LLD
One model, seven things it does well — mapped end to end.
VisionAery
LLD
Edge Compute
On-site inference with no cloud dependency in the alarm path.
Puddle Detection
Detects pooled liquid releases the moment they appear on screen.
Pressurized Spray
Catches directional spray and early plume morphology from pressurized releases.
Visible Vapor Plumes
Flags visible vapor plumes that SCADA gauges never report.
Day, Night & All-Weather
Runs continuously through day, night, weather, and lighting changes.
AVA LLD Verification
A reasoned second look verifies every alarm before it reaches on-call.
VMS, SCADA & On-Call
Publishes confirmed alarms into VMS, SCADA, and on-call notification systems in real time.
VisionAery
LLD
Edge Compute
On-site inference with no cloud dependency in the alarm path.
Puddle Detection
Detects pooled liquid releases the moment they appear on screen.
Pressurized Spray
Catches directional spray and early plume morphology from pressurized releases.
Visible Vapor Plumes
Flags visible vapor plumes that SCADA gauges never report.
Day, Night & All-Weather
Runs continuously through day, night, weather, and lighting changes.
AVA LLD Verification
A reasoned second look verifies every alarm before it reaches on-call.
VMS, SCADA & On-Call
Publishes confirmed alarms into VMS, SCADA, and on-call notification systems in real time.
Edge Compute
VisionAery LLD runs on a fanless edge inference box deployed at the site — on the tower, on the skid, or in the comms shelter. Every model decision is made locally in sub-second latency, with no dependency on cloud reachability. If the WAN drops, the analytic keeps detecting, AVA keeps verifying, and the alarm continues to publish into the local VMS the moment the link returns. Operators get the same precision and the same response time regardless of backhaul quality.
Puddle Detection
The most common release mode on a tank battery is a slow drip that pools on dirt, gravel, or the dike floor. VisionAery Liquid Leak Detection is trained on thousands of real-world puddle scenes — fresh crude on caliche, condensate sheen on packed dirt, water-cut emulsion on liner — and fires the moment a pool is on screen. Detection is independent of vapor reading, gas chromatography, or operator presence; the model sees the pixels and acts.
Pressurized Spray
Pressurized releases — stuffing-box failures, header pinholes, control-valve packing leaks, manifold gasket failures — produce a directional spray or plume rather than a puddle. VisionAery is trained to detect the spray pattern and the early plume morphology, surfacing pressurized leaks within seconds of release rather than after hours of accumulation. Critical for sites where a release escalates quickly into a regulatory or safety event.
Visible Vapor Plumes
Where the camera and conditions support it, VisionAery flags visible vapor plumes — the kind a trained pumper sees on walkdown but a SCADA gauge never reports. Useful as an early-warning layer for scenes where a vapor release precedes the visible liquid release, and as a complement to OGI surveys and Method 21 monitoring programs.
Day, Night & All-Weather
Production sites do not pause for weather. VisionAery is trained against day, night-with-IR-illumination, dust, fog, light snow, rain, glare, and flare-light conditions. The model is benchmarked against a curated test set of failure-mode scenes — a fence-shadow at sunset, a rain-puddle on the access road, a glare-blowout on a stainless tank — so the precision-recall curve holds in the real world, not just on a cherry-picked highlight reel.
AVA LLD Verification
When the analytic fires, AVA LLD — the VisionAery Alarm Verification Agent for Liquid Leak Detection — takes a reasoning-driven second look at the alarm frame and returns a structured verdict (confirmed / dismissed / low-confidence) before the alert dispatches. The result is a sharply lower nuisance-alarm rate without sacrificing recall on real releases. Operators stop chasing false positives, and real leaks stop getting buried in alarm fatigue.
VMS, SCADA & On-Call
Confirmed alarms publish into the major VMS platforms (Genetec, Milestone, Avigilon, ExacqVision, Axis) as live alarms with a frame-bookmarked clip, into SCADA via MQTT, OPC UA, Modbus, or REST, and into PagerDuty / OpsGenie / SMS / email on-call notifications with the analytic frame embedded. The same alarm reaches every system that has to know about the release — at the same time, with the same evidence.