Industrial Safety

Overview

Particle and Thermal Visualization Across Plant Operations

A major Taiwanese petrochemical operator deployed Atomrock's Particle Visualization AI and Thermal Visualization AI side by side, across both control-room and machine-room environments and across production-floor plant areas. The two perception layers together turn what plant operators normally cannot see, fugitive dust and aerosols, and thermal anomalies on energized equipment, into discrete, time-stamped AI events with location, severity, and a video record.

Both layers run on the same Atomrock stack: the AEC300, Atomrock's latest high-performance Edge AI Computing device, sits at each camera location running the Perception Vigilance application profile, and a unified cloud streaming center receives the inference output, where operators monitor live feeds, configure alarm zones, and review event history. One platform, two complementary senses, deployed across the customer's most safety-critical interiors.

Challenge

Two Failure Modes the Human Eye Was Never Going to Catch

Petrochemical operations sit on top of two parallel risk profiles. Inside server rooms, switchgear cabinets, transformer enclosures, and process-control facilities, equipment hot spots can develop quietly hours before they trip a thermal sensor or a fire alarm, and by the time conventional monitoring catches them the failure mode is usually already a fire risk. Out on the plant floor and around storage and transfer infrastructure, fugitive particulate and aerosol release is a leading early indicator of leaks, abnormal venting, and dust accumulation in environments where ignition risk is real.

Standard CCTV recorded both, but acted on neither. Walkthrough inspections covered points, not continuous coverage. The operator needed two things at once: continuous thermal monitoring of energized equipment in critical interiors, and continuous particulate/aerosol monitoring across plant areas, both reduced to actionable, zone-based events that a small operations team could triage without watching a wall of video.

Solution

Zone-Based Events, Edge Inference, One Operator Console

Atomrock deployed thermal cameras and particle-visualization-capable cameras across the operator's high-value sites, each one paired with an AEC300 edge AI device running the Perception Vigilance profile. RTSP streams flow into the edge device, inference happens locally, and only events plus relevant video reach the cloud, where they surface in two unified consoles: the Thermal Visualization Stream Center and the Particle Visualization Stream Center.

Each camera view supports operator-defined alarm zones drawn directly on a reference image. Thermal zones use a temperature threshold (≥ or ≤ a target value, with adjustable color mapping for visual emphasis), a detection mode (immediate trigger, or duration-based — temperature must hold above threshold for a configurable interval), and a per-event suppression window to prevent alarm flooding from a single sustained anomaly. Particle zones use a particle-coverage percentage threshold within the drawn boundary; once coverage exceeds the configured rate, the system fires an event with a particle-overlay clip attached.

Operators work from a single console per modality. Live streams are relayed through the cloud, so client browsers never reach into the plant's local network. Multi-camera grid mode lets one shift cover dozens of cameras in parallel; event cards along the timeline let an operator jump straight to the moment an anomaly fired, replay the cloud recording in either standard or particle/thermal-overlay view, and securely download the clip via a 24-hour expiring link for handoff to plant safety, electrical maintenance, or environmental teams.

Result

Continuous Coverage, Faster Response, Repeatable Across Sites

Across machine rooms and plant areas, the two perception layers now run continuously where walkthrough inspections used to. Thermal anomalies on energized equipment surface as duration-qualified events long before they would have shown up on conventional sensors, giving electrical and facilities teams hours of lead time on developing faults instead of minutes. Particle and aerosol releases on the plant floor are caught at the zone level with a coverage threshold the operator can tune per area, separating routine activity from genuine excursion events.

Operationally, the platform compresses what used to be two specialist monitoring problems into one console pair that a regular control-room shift can run. Zone-based event volumes are tunable per area, suppression windows keep the queue manageable, and clip handoff is friction-free. The deployment pattern, edge AEC300 plus cloud stream center, has become the customer's reference architecture for rolling the same two perception layers out to additional facilities.

To learn how Atomrock's particle and thermal visualization AI can support petrochemical, energy, and heavy-industry deployments, contact us.