An electric-arc furnace (EAF) now gets its most safety-critical inspection from a four-legged robot. ANYbotics' ANYmal captures thermal imagery of the furnace bottom while the furnace is active, checking for washout — the failure mode in which liquid steel penetrates the refractory lining and the steel shell and spills into the pit below.

The inspection task itself is not new. What changes is who, or rather what, performs it. Until now, a member of the operations team had to capture the thermal imagery under the operating furnace. ANYmal removes that person from the hazard zone entirely. The measurement — thermal imaging of the furnace bottom — stays the same; the exposure risk does not.

Why the measurement matters

The physics behind the inspection is straightforward but unforgiving. An EAF shell contains liquid steel at roughly melting-point temperatures, held back by a refractory lining that degrades over successive heats. Wear does not proceed uniformly. Localized thinning produces a characteristic thermal signature on the outside of the furnace bottom: a hotspot visible in infrared before the lining actually fails. Detect that hotspot early enough and maintenance can repair the lining on schedule. Miss it, and the consequence is washout — molten steel breaching the shell and pouring into the pit, an event that risks both severe injury and extended unplanned downtime.

That failure progression is what makes the thermal image actionable rather than merely informative. A hotspot is a precursor, not a failure. The value of the inspection lies in the lead time between precursor detection and lining breach.

From image to workflow

Thermal imagery alone does not prevent downtime; somebody has to interpret it and act. ANYbotics addresses that step with Shift, its operations platform. Shift identifies hotspots in ANYmal's imagery as potential washout precursors and flags them to the maintenance team. The intended outcome: the lining gets repaired before it fails, so downtime is prevented rather than managed.

Jacob Euler-Rolle, product management, workflow integration at ANYbotics, describes the deployment plainly: "This is our ANYmal, which performs autonomous thermal inspections of the furnace bottom in an electric-arc furnace monitoring for washout: liquid steel penetrating the refractory lining and the steel shell, spilling into the pit. ANYmal captures the thermal imagery under the active furnace, so nobody from the operations team has to. ANYbotics' operations platform, known as Shift, identifies hotspots in that imagery as potential washout precursors and flags them to the maintenance team, so the lining can be repaired before it fails. Downtime gets prevented, and a job becomes less hazardous for team members."

What is measured versus what is claimed

The deployment demonstrates a working robotic inspection under an active furnace — a claim few conventional monitoring approaches can match, since fixed instrumentation under a live EAF faces the same thermal and mechanical hazards that make human inspection dangerous. ANYbotics has not published detection rates, hotspot-classification accuracy, or the thermal resolution of the onboard imaging for this application, so hotspot identification performance in Shift should be treated as a vendor capability statement rather than a verified specification. The company also has not stated the inspection cadence or the interval at which ANYmal repeats its rounds, which would determine how much warning the maintenance team actually receives before a washout event.

What the source does establish: the robot operates autonomously, the imaging happens while the furnace is running, and the flagged hotspots feed a human maintenance workflow rather than an automated shutdown path.

The adoption question

For plant operators, the interesting question is not whether a quadruped can carry a thermal camera under a furnace — this deployment shows it can. The question is how ANYmal's precursor detection integrates with existing refractory lifecycle management: does a Shift flag carry enough diagnostic weight to justify pulling a furnace out of service for lining repair, and how does that decision compare on cost and risk against the operator's current wear-model thresholds? Until ANYbotics or its steelmaking customers publish detection performance against actual washout events, that integration will be negotiated plant by plant.