Centerline segregation is one of those defects that doesn't show up on the caster floor. It shows up months later, in a fatigue crack on a rail, a failed hydrotest on a pipe, or a customer complaint about inconsistent hardness across a coil. By the time it's traced back to the strand, the heat is long gone and the only tool left is a stirrer setting that was probably wrong from the start. Electromagnetic stirring is supposed to fix this at the source, but most casters still run EMS current and frequency off a fixed setpoint per grade, set once during commissioning and rarely revisited even as steel chemistry, casting speed, and superheat shift heat to heat.
Fixed EMS settings can't chase a strand that changes every heat
iFactory adjusts EMS frequency, current, and timing heat by heat, matched to real casting conditions, to control equiaxed structure and cut centerline segregation without adding new hardware.
Where you stir the strand changes what you're actually fixing
Electromagnetic stirring isn't one setting, it's three separate interventions positioned at different points along the strand, and each one targets a different part of the solidification problem. Getting the zone right matters as much as getting the current right.
M-EMS
Installed at the mold, M-EMS stirs the meniscus and upper liquid pool. It reduces pinholes and subsurface porosity and helps refine the initial solidification structure before the shell has fully formed.
S-EMS
Positioned lower in the secondary cooling zone, S-EMS stirs the still-liquid core as the shell thickens. It widens the equiaxed grain zone but can worsen centerline segregation if current is set too high.
F-EMS
Applied near the very end of solidification, F-EMS specifically targets the concentrated solute pocket that forms at the centerline, and is generally the most effective single lever for reducing centerline segregation directly.
Research on combined stirring modes has found that M-EMS alone does relatively little to improve positive centerline segregation, while F-EMS is the zone that actually drives the improvement, and that pairing M-EMS with F-EMS produces a better result than either alone because the solute redistribution from the mold stage carries forward and amplifies what F-EMS accomplishes downstream.
Small current and frequency changes move the result a lot
| Parameter | Effect if too low | Effect if optimal | Effect if too high |
|---|---|---|---|
| Stirring current | Weak flow, segregation barely improves | Balanced equiaxed structure, tight segregation index | Over-stirring, negative segregation and white bands appear |
| Stirring frequency | Sluggish rotation, poor solute mixing | Consistent rotational flow across the section | Segregation degree can increase again at higher frequency |
| F-EMS position | Misses the final solidification pocket entirely | Targets the concentrated solute zone directly | Disturbs an already-solid shell, no added benefit |
| Casting speed match | EMS timing falls out of sync with shell growth | Stirring zone lines up with actual solidification front | Excess dwell time in the field increases segregation risk |
One industrial study on billet casting identified a narrow current-frequency combination as the clear best performer for that section size, while values just outside that window measurably worsened the outcome, which is exactly the kind of sensitivity that a static, once-set parameter table cannot track heat to heat.
Fixed EMS Setpoints
- One current and frequency value per grade, set at commissioning
- No adjustment for casting speed drift or superheat swings
- Segregation defects found downstream at rolling or in service
- Operators can't tell if a given heat needed more or less stirring
- Parameter table rarely revisited once EMS is installed
iFactory-Guided EMS
- Current and frequency recommended per heat from live strand data
- Speed and superheat changes trigger real-time parameter updates
- Segregation index estimated before the strand leaves the caster
- Every heat logged against its actual stirring parameters and outcome
- Model recalibrates continuously as grades and practices change
Segregation tolerance windows keep getting tighter
Automotive, energy, and rail customers have been steadily tightening acceptance criteria for internal soundness, particularly for higher-strength and heavy-section grades where a segregated band becomes a real fatigue-crack initiation site rather than a cosmetic issue. A mill that can show consistent, low centerline segregation across a full production run, not just on the samples pulled for a customer audit, has a genuine edge when bidding on segregation-sensitive contracts like heavy rail steel or pressure vessel plate.
At the same time, EMS equipment itself hasn't changed much in the last decade. Coils, power supplies, and stirrer positions are largely fixed capital investments, so the opportunity left on the table isn't in new hardware, it's in how intelligently the existing hardware is driven heat to heat. That's a software and data problem, not an equipment problem, and it's one of the reasons EMS optimization has become an attractive AI pilot for caster teams that already have the stirrers installed but are still running them off a static table.
There's also a scrap and downgrade cost that rarely gets attributed correctly. When centerline segregation causes a downstream reject at the rolling mill or a customer return after service failure, the cost usually gets booked against the rolling operation or the quality department, not against the caster parameter that actually caused it. Closing that attribution gap is often what finally gets EMS optimization funded.
Energy and maintenance considerations play a smaller but still relevant role. Running EMS current higher than necessary on grades that don't require it consumes power without any quality benefit, and over-stirred strands can accelerate wear on coil insulation and cooling systems over time. A parameter approach that only applies aggressive stirring where segregation risk actually justifies it tends to reduce unnecessary equipment stress alongside the quality improvement, which caster maintenance teams generally welcome even though it isn't the primary reason most mills start this kind of project.
Segregation risk isn't the same across your grade mix
Not every grade on a caster carries the same segregation exposure, and treating them all with the same fixed EMS table is part of why static parameters underperform. Rail steel and heavy-section bloom products tend to be the most segregation-sensitive because the larger cross-section takes longer to solidify, giving solute more time to concentrate at the centerline before the final shell closes. High-carbon and alloy tool steels carry similar risk, since carbon and alloying elements segregate more aggressively than the base iron matrix during solidification. Line pipe and pressure vessel plate grades sit in the middle of the risk range, but face some of the tightest customer specification windows, which means even a moderate segregation index can trigger a downgrade that a lower-spec grade would pass without issue.
Lower-carbon commodity grades are more forgiving, which is exactly why many caster teams have historically left EMS parameters alone for those products and focused optimization effort elsewhere. That's a reasonable starting instinct, but it also means the segregation-sensitive grades, the ones actually driving downgrade cost, are often running on the same generic setpoint as everything else. A grade-aware approach lets the model apply tighter, more conservative stirring parameters specifically where segregation risk and customer tolerance intersect, without over-engineering the setup for grades that don't need it.
Section size interacts with all of this too. A 150mm billet and a 300mm bloom don't just need different absolute current values, they need different relationships between casting speed and stirrer timing, because the solidification front moves through the section at a different rate. Casters running mixed section sizes on the same strand often see their segregation problems concentrate on whichever section size was least represented in the original commissioning tests, since that's usually the geometry the fixed table fits worst.
From strand conditions to a stirring recommendation
Ingest live strand data
Casting speed, superheat, mold level, and secondary cooling rates are pulled continuously from the existing caster control system.
Estimate solidification front position
A shell-growth model, calibrated to your section size and steel grade, tracks where the strand actually is in its solidification path.
Recommend EMS parameters
Current, frequency, and stirrer zone timing are recommended to align stirring with the true solidification front rather than a fixed schedule.
Score predicted segregation
Operators see an estimated segregation index before the strand exits, with enough lead time to adjust if the prediction is outside target.
Log and improve
Final sample results are matched back to the parameters that were actually used, continuously sharpening the model for your caster.
Most caster teams have never actually charted their segregation index against real stirring parameters heat by heat. Book a walkthrough and we'll build that chart with your own casting data.
What changes within one casting campaign
These figures come from campaigns where the model had at least six to eight weeks of shadow-mode validation before operators began acting on live recommendations, which is roughly the calibration window most caster teams need before the segregation index prediction is accurate enough to trust for setpoint decisions. Results improve further as more heats accumulate in the training set, particularly for grades that don't run frequently enough to have generated much historical data during commissioning.
See your own segregation data mapped to stirring parameters
We'll pull a sample of recent heats and show where fixed EMS settings are leaving segregation control on the table.
What a caster pilot actually involves
An EMS optimization pilot doesn't require new stirrer hardware or a caster shutdown. The model reads from sensors and control signals already present at most modern casters, and the deployment work is mostly calibration rather than installation. A typical pilot runs on one strand or one grade family first, uses historical heat records to build an initial shell-growth and segregation model, then validates in shadow mode where recommendations are logged but not yet acted on, before operators start using the recommended parameters directly. iFactory's team manages the model tuning throughout, so the caster engineering group isn't stuck maintaining machine learning infrastructure on top of running the strand. Once the first grade family shows a clear segregation improvement, most teams expand coverage to additional grades and, eventually, additional strands, using the first result as the internal case for the rollout.
Integration with existing Level 2 automation is usually the first technical question caster engineering teams raise, and it's a fair one, since nobody wants a parallel system fighting the control room for authority over stirrer setpoints. iFactory's model operates as a recommendation layer, reading from the same casting speed, mold level, and temperature signals the Level 2 system already uses, and passing recommended EMS parameters either as an operator display or, once trust is established, as a direct setpoint suggestion that requires operator confirmation. Nothing overrides the caster's existing safety interlocks or automation logic, and operators retain the ability to run fixed parameters at any time if they choose to.
Change management tends to matter more than the technical integration. Caster operators have often spent years learning to read subtle strand behavior and adjust manually within whatever the fixed table allows, and a system that starts recommending different parameters heat to heat can feel like it's second-guessing that experience. The teams that get the most out of an EMS pilot are usually the ones that involve senior operators early, show them the segregation data behind specific recommendations rather than just the recommendation itself, and let the shadow-mode comparison period build trust before anyone is asked to act on a model-suggested parameter live.
EMS optimization, explained plainly
Find out what fixed EMS settings are costing your segregation index
iFactory shows you, heat by heat, where stirring parameters are leaving quality on the table. Book a demo and we'll walk through it using your own casting data.
