goatai.io/steel
Steel Operations Intelligence
Steel is where the reasoning engine is deepest — three decades of plant automation behind it. Industrial perception, process physics, and engineering constraints, reasoning together inside the plant. In active engagement with Jindal Steel, Angul.
Deployed at Scale
1,000+
DFMS · heavy vehicles
Plant-Automation Heritage
30+ yrs
Steel floor experience
Lead Steel Client
Jindal
Angul, Odisha
Vision & Safety Systems
5
EAF · BOF · ladle · crane · yard
Proof of Scale
Deployed, not in pilot.
DFMS — GOATAI’s operator fatigue-monitoring platform — runs across 1,000+ heavy vehicles in active mining operations today, producing DGMS-aligned safety evidence. Hardware, edge AI, and cloud, designed and built in-house. The same cognitive-risk architecture now extends to the ladle-crane cabin in the melt shop.
Heavy vehicles
1,000+
Active fleets
Live operations
Mining
Producing safety evidence
Aligned evidence
DGMS
Regulatory direction
HW · edge · cloud
In-house
Designed & built
It answers the silent question behind every pre-deployment system on this page — has any of this survived a real site? DFMS has.
See DFMS in the field ↗The Grounding
Not a vertical. The grounding the rest is built on.
The old failure mode is to present several unrelated domains; the opposite failure is a reasoning engine in the abstract that a plant head cannot picture. Steel is the concrete answer — the place where physics-grounded reasoning has the deepest roots, and the proof that the architecture generalizes.
Perception
Crane, yard, furnace, and ladle vision under real plant conditions — dust, heat, glare, steam.
Process physics
Metallurgical and thermal models for the converter and caster, not generic analytics.
Engineering constraints
The clearances, endpoints, and safe envelopes a plant engineer already designs against.
Crane-Safety Convergence
The ladle crane is the most dangerous operation in the plant. We instrument all of it.
A suspended ladle of molten steel leaves a sub-two-second reaction window. The risk is not in one component — it is in the machine, the operator, and the load at once. GOATAI instruments all three surfaces of that single lift with one reasoning architecture.
The machine
In developmentHookVision
Recognition-based clearance management — a camera + radar custodian of the clearances the bay was designed around. It directs attention; the deterministic barrier owns the stop.
The operator
Deployed at fleet scaleDFMS — crane mode
Multimodal cognitive-risk monitoring built for the crane cabin, not automotive logic — behavioral entropy against crane phase, mmWave sensing through PPE occlusion, edge processing for connectivity-dead zones.
The ladle
In designLVS
Ladle identification, lining-condition monitoring, and lift-and-landing safety across the bay — the load itself, instrumented stand by stand.
Machine geometry, operator cognition, and the ladle itself — three risk surfaces of one lift, one reasoning architecture.
The Systems
Five systems, one engine
Each is in active engagement with Jindal Steel, Angul — at honest, distinct stages. Specifications are drawn from live engineering work, not aspiration.
Crane anti-collision
HookVision
Anti-collision reframed as dynamic clearance management — a continuous custodian of the clearances the designer already assumed. The system learns what a bay looks like when all is well and flags deviation from normal, rather than chasing a catalogue of collision cases.
- Recognition-based perception map — novelty detection over the bay’s normal state
- Camera + radar anti-correlated fusion — each sees where the other goes blind
- Fixed parallel eyes, no PTZ — preserves the recognition premise
- An attention-direction layer over a deterministic geometric barrier — it never owns the stop
Slab & coil yard intelligence
YardVision
Real-time crane position and load identification across the slab and coil yard, from a fixed bracket-mounted optical network — built to feed yard management and safe-zone enforcement.
- Fixed-optical design, bracket-mounted — no moving pan-tilt-zoom
- Crane position inference from visual tracking
- Load identification — slab vs. coil, dimensions, stacking position
- Monitoring today — physical interlock is the path to remote yard operation
EAF operating-pulpit visibility
HeatVue
Control-room digital visibility for the electric-arc-furnace pulpit — a multi-camera video engine covering the vessel mouth, lance-interaction zones, slag splash, and furnace shell, mapped onto an operating-pulpit display.
- 4 × 5 MP GMSL2 cameras, dual-live hero configuration
- ≤60 ms latency, lens to vision-engine output
- HDMI 2.0 output, 4992 × 1728 @ 60 Hz, 1:1 mapped
- Hero zone ≥700 px on the furnace shell · Jindal Angul EAF
Ladle Visualization System
LVS
Ladle identification, lining-condition monitoring, crane lift-and-landing safety, and slag/skull detection across the ladle bay — combining runway-side fixed cameras with crane-cabin coverage.
- DE bay: 450 m long travel × 30 m, 15 m hook height
- 36 ladle stands · 3 ladle cranes (400 / 400 / 450 T)
- Per-stand visibility: ID, lining, lift/landing, slag & skull
- Stand-by-stand coverage plan · Jindal Angul BOF #1
Process-physics reasoning
BOF & caster surrogates
Physics reduced-order models coupled with ML surrogates for endpoint and quality reasoning over the BOF converter and continuous caster — the process-physics layer the perception systems reason against.
- Physics ROMs + ML surrogates for the converter and caster
- Endpoint and quality reasoning
- Grounded in real plant geometry (Angul: two BOF converters, shared pulpit)
- The metallurgical and thermal physics layer of the engine
The Engine, in Steel
The same cycle, grounded in metallurgy
Every system above runs one reasoning cycle over a steel-specific physics layer. The architecture is constant; the physics is grounded to the plant.
Monitoring
Furnace and ladle video, crane geometry, and plant telemetry as live physical state.
Prediction
Metallurgical and thermal models forecast endpoint, wear, and approach to limits.
Reasoning
Inference over the clearances, endpoints, and safe envelopes the plant already assumes.
Decision
Prioritized, explainable alerts and recommendations routed to the operators who act.
Knowledge
We explain the hard problems in the open.
The durable advantage is demonstrating the engineering, not asserting it. Our knowledge hub works through the physics and human factors behind these systems — the crane-cabin analysis is written from the melt shop.
The Engagement
Built with the plant, not sold to it
These systems have no functional equivalent on the market. The engagement is structured as co-creation — a nominated technology partnership and staged joint proof-of-concept — between a founder with 30+ years on the steel floor and the plant’s own operations and automation teams.
Explore a co-creation engagement
Technical deep-dive · Site assessment · Staged joint PoC