Predicting how long a refractory lining will last has traditionally come down to experience: a senior engineer who has seen enough campaigns to make an educated guess based on heat count and a gut sense of how the current slag chemistry compares to past ones. That approach works reasonably well until conditions shift, a new steel grade enters the mix, or that engineer moves on, and then the estimate becomes a lot less reliable exactly when the plant needs it most. AI-based wear rate modeling replaces that gut sense with a model that actually learns from the combination of variables driving wear, and it is the approach behind iFactory's campaign prediction tools.
Refractory Lifecycle · Campaign Prediction
Refractory Campaign Prediction: What AI Wear Rate Modeling Actually Does
A single-variable estimate based on heat count alone misses most of what actually drives refractory wear. AI wear rate modeling correlates slag chemistry, operating conditions, and historical wear data together, producing a campaign life prediction that adjusts as conditions change instead of staying fixed from day one.
What the Model Actually Looks At
The Variables Behind an Accurate Wear Prediction
Slag Chemistry Trends
Basicity, FeO content, and MnO levels over time, since these directly drive chemical dissolution rate against the specific refractory grade in use.
Operating Temperature and Holding Time
Higher temperatures and longer holding periods accelerate both chemical attack and thermal stress on the lining, and both vary heat to heat.
Historical Wear Rate by Zone
Past thickness measurement trends for the same vessel and zone provide the baseline the model adjusts against as new data comes in.
Steel Grade and Production Mix
Different steel grades produce different slag chemistries and process conditions, so a shift in production mix changes the wear picture even with the same refractory installed.
How the Prediction Gets Built
From Raw Data to a Campaign Life Estimate
1
Collect Slag and Process Data
2
Correlate Against Wear History
3
Generate Live Wear Rate Estimate
4
Update as New Heats Complete
5
Flag Deviation From Prediction
The Practical Difference
Traditional Estimate vs. AI Wear Rate Modeling
| Factor | Traditional Estimate | AI Wear Rate Model |
|---|---|---|
| Basis | Heat count and engineer experience | Multi-variable correlation across live data |
| Updates when conditions change | Only when someone manually re-estimates | Continuously, as new data arrives |
| Depends on one person's experience | Yes, heavily | No, model persists regardless of staff changes |
| Catches early deviation from expected wear | Rarely, until thickness check confirms it | Flags the deviation as it develops |
See What Your Data Actually Predicts
Run Your Slag and Thickness History Through a Wear Rate Model
Bring your available slag chemistry and thickness data and we will show you what a correlated wear rate model reveals about your current campaign life estimates.
What Undermines a Prediction Model
Common Mistakes in Campaign Prediction
Relying on Heat Count Alone
Using a single variable to predict campaign life while ignoring slag chemistry and operating condition shifts that actually drive wear rate changes.
Feeding the Model Incomplete Data
Leaving gaps in slag chemistry or thickness logs that weaken the correlation the model can actually draw from your own operating history.
Not Updating Predictions as Conditions Shift
Treating a campaign life estimate as fixed from day one instead of letting it adjust as new slag and wear data accumulate through the campaign.
Ignoring Deviation Alerts
Letting a flagged deviation from the predicted wear curve sit unreviewed instead of investigating the process change likely driving it.
A Composite Scenario
A BOF Campaign, Two Prediction Approaches
Before
A BOF vessel's expected campaign life was set at the start based on heat count targets from the previous campaign. Midway through, a shift in scrap mix pushed slag FeO content higher, accelerating wear, but the fixed estimate wasn't revisited until a scheduled thickness check confirmed the vessel was well behind its expected life.
After
With a live wear rate model correlating slag chemistry against wear trend, the FeO shift and its impact on predicted campaign life showed up within days of the scrap mix change. The team adjusted slag conditioning practice promptly, and the campaign recovered close to its original planned life.
Before You Build a Model
Readiness Checklist for AI Wear Rate Modeling
Confirm slag chemistry data and refractory thickness data are both being captured consistently
Check that historical data spans enough campaigns to give the model a meaningful baseline
Identify which vessels have the most variable wear behavior and would benefit most from modeling first
Name who reviews deviation alerts and decides what process action to take
Common Questions
AI Wear Rate Modeling — FAQ
How much historical data does a wear rate model need to be useful?
More history generally improves accuracy, but a model can start producing directionally useful predictions from a handful of past campaigns, refining further as more data accumulates. Waiting for years of perfect historical records before starting usually means missing value the model could already be providing. Our team can assess what your existing data can already support.
Does AI modeling replace the need for experienced refractory engineers?
No, it supports their judgment rather than replacing it. The model surfaces correlations across more variables than a person can track manually, but interpreting what a flagged deviation means for a specific vessel and deciding on the right process response still benefits from experienced engineering judgment.
How accurate are AI-based campaign predictions compared to traditional estimates?
Accuracy improves as the model correlates more variables against your own operating history rather than relying on a single heat-count assumption, and it also updates continuously rather than staying fixed once set. Book a demo to see prediction accuracy against your own historical campaigns.
What happens when the model flags a deviation from the predicted wear curve?
A flagged deviation signals that actual wear is tracking faster or slower than expected given current conditions, prompting a review of recent slag chemistry or process changes that might explain it, so a response can be planned before the deviation becomes a scheduling emergency.
Can this modeling approach work across multiple vessel types at once?
Yes, though each vessel type, ladle, BOF, tundish, or EAF, tends to have different dominant wear variables, so models are typically built and tuned separately per vessel type rather than as one generic model applied across all of them.
Stop Predicting Campaign Life on Heat Count Alone
Correlate Slag Chemistry and Wear Data Into a Living Prediction
iFactory builds a wear rate model from your own slag chemistry and thickness history, updating campaign life predictions as conditions actually change.
