A continuous annealing line only gets one pass at every meter of strip. If the heating zone runs a few degrees hot, if the soaking time drifts, or if the protective atmosphere's dew point creeps upward for even ten minutes, the mechanical properties of that coil are set before anyone in the lab sees a tensile result. Traditional PID loops hold a setpoint well, but they cannot see the strip's actual thermal state or predict how a coming gauge change will interact with the furnace's thermal lag. AI-based profile control changes that by modeling the entire heating-soaking-cooling curve continuously and adjusting zone power and gas dosing before a deviation ever reaches the strip. Process teams evaluating this approach can Book a Demo to see live profile tracking against a real CAL dataset.
Hold the Thermal Profile That Sets Your Steel's Properties.
iFactory's AI continuously models strip temperature and atmosphere composition across every zone, correcting drift before it reaches the coil.
Why Zone Setpoints Alone Don't Guarantee Strip Temperature
A furnace zone can hold its setpoint perfectly while the strip itself runs a different temperature entirely. Line speed changes, gauge transitions, coil-to-coil width variation, and radiant heat lag all mean the relationship between zone air temperature and actual strip temperature keeps shifting. A PID loop reacts to what it measures at the thermocouple; it has no model of how the strip is actually absorbing heat several meters downstream.
The protective atmosphere adds a second variable that is just as unforgiving. A rising dew point in the hydrogen-nitrogen mix changes the surface chemistry available at the strip during annealing, and that shift shows up later as inconsistent coating adhesion or surface staining rather than an immediate alarm. By the time a lab result flags the problem, several coils have already gone through the same drifted conditions.
The Thermal Journey That Defines Final Mechanical Properties
Every coil moves through the same three thermal stages, and each stage locks in a different part of the steel's final microstructure. Getting any one of them wrong shows up later as an out-of-spec yield strength, elongation, or surface condition.
Recrystallization begins as the strip is ramped through the critical temperature range at a controlled rate.
Grain structure equalizes across the strip's width and thickness during the hold at peak temperature.
Controlled descent sets the overaging response for dual-phase and bake-hardenable grades.
Fast quench locks in the martensite or ferrite-martensite ratio that defines final strength.
Turning Sensor Data Into a Predictive Thermal Model
Rather than reacting to a single thermocouple reading, the AI model builds a continuous estimate of the strip's actual temperature at every point along the line and adjusts before the deviation becomes visible in the product.
Multi-Point Sensor Fusion
Zone thermocouples, pyrometer readings, line speed, and gauge data feed a single model of strip temperature rather than being read in isolation.
Forward Profile Prediction
The model projects strip temperature several meters ahead, accounting for thermal lag, so corrections happen before the strip reaches the affected zone.
Closed-Loop Zone Adjustment
Zone power and burner ratios are adjusted continuously against the predicted profile instead of a fixed setpoint alone.
Atmosphere Dosing Correction
Hydrogen and nitrogen dosing is adjusted in tandem with the thermal profile to keep dew point inside the range each grade requires.
Which Grades Are Most Sensitive to Profile Drift
Not every steel grade responds to a thermal deviation the same way. The table below reflects how tightly each common CAL product depends on precise zone control.
| Grade Category | Profile Sensitivity | Primary Risk of Drift |
|---|---|---|
| Automotive Exposed | High | Surface staining, coating adhesion loss |
| Dual-Phase (DP) | Very High | Incorrect martensite fraction, strength variance |
| Structural / HSLA | Moderate | Yield strength inconsistency across coil length |
| Bake-Hardenable | High | Reduced paint-bake strength response |
What Mills Track After Adding Profile AI
Typical strip-width temperature spread maintained across the soaking zone during steady-state running.
Reduction in coils flagged for out-of-spec tensile or elongation after profile control is deployed.
Typical dew point control band held across grade transitions and speed changes.
From Historical Coil Data to Live Profile Control
Deployment starts with historical coil, thermocouple, and lab result data rather than a blank model. The AI is trained against your line's actual thermal lag and grade mix, validated against past coils with known lab outcomes, then run in shadow mode alongside the existing PID control before taking closed-loop authority over zone power and atmosphere dosing. Most lines move from historical validation to live shadow mode within four to six weeks, with full closed-loop control following a further supervised run once the model's predictions consistently match lab results.
Continuous Annealing AI Control — Common Questions
Does the AI replace the existing PID controllers on the furnace?
No, the model works alongside existing PID loops rather than replacing them. It supplies an optimized setpoint trajectory and atmosphere target that the PID controllers then execute, so the underlying control hardware and safety interlocks remain exactly as they are today. This makes the rollout far less disruptive than a full control system replacement.
How does the model account for gauge and width changes mid-coil?
Gauge and width transitions are among the highest-risk moments for profile drift, since thermal mass changes abruptly while the line is still moving. The model factors line speed, incoming gauge schedule, and thermal lag into its forward prediction, adjusting zone power ahead of the transition rather than after the strip has already passed through it.
What data does the mill need to have available before starting?
A useful starting point is roughly six to twelve months of zone thermocouple logs, line speed and gauge schedule history, and the corresponding lab tensile and elongation results for those coils. Teams can review exactly what a data extract should include with iFactory Support before kickoff.
Can the same model handle multiple grade families on one line?
Yes, the model is trained across the full grade mix a line typically runs, since grade-to-grade transitions are exactly where manual control tends to lag. Automotive exposed, dual-phase, and structural grades each get their own target profile within the same model rather than requiring separate systems.
How is the improvement validated before full rollout?
The model runs in shadow mode first, generating recommended setpoints that are logged and compared against actual outcomes without taking control authority. Once its recommendations consistently track lab results across the grade mix, teams can Book a Demo to walk through a shadow-mode report from a comparable line.
See Your Line's Thermal Profile Modeled Live.
Bring a sample of coil and thermocouple data and watch iFactory's model predict the profile against your actual lab results.
