Quick Answer:
In perpetual vs. service-based slag detection, the key difference lies in how detection performance is maintained over time. Across industrial systems—including continuous casting—there is a clear shift from perpetual purchases to performance-based, service-based models.
In ladle slag detection, this reflects a practical reality: detection performance must continuously adapt to changing conditions. Traditional systems benefit from ongoing support, while machine learning–based systems require continuous model training and validation—making performance-based models the natural fit.
A Broader Shift in Industrial Technology
Across industries, technology is moving away from one-time purchases toward performance-based, service-based models.
This shift is not just about pricing—it reflects how modern systems behave. Systems are no longer static. They evolve with data, improve with use, and require ongoing refinement to maintain performance.
Continuous casting fits this pattern closely. It is a process defined by variability:
- Heat-to-heat changes
- Operator differences
- Changing ladle and signal conditions
In this environment, the idea that a system can be installed once and maintain performance indefinitely no longer holds.
A dynamic process requires systems that evolve with it—and are supported accordingly.
Slag Detection Is Not a Fixed Problem
Ladle slag detection is a real-time decision that directly impacts yield.
At the end of each heat, operators are balancing steel recovery against slag carryover risk. That decision is influenced by constantly changing conditions—ladle behavior, slide-gate control, plant noise, and operator timing.
Detection systems must interpret these conditions accurately and consistently. But those conditions do not repeat exactly from heat to heat.
When detection systems remain fixed, performance does not.
Two Types of Slag Detection Systems
In continuous casting, slag detection systems generally fall into two categories—each with different approaches to maintaining detection performance.
Classic Systems (Rule-Based)
Traditional systems rely on:
- Fixed thresholds
- Rule-based signal interpretation
- Initial setup with periodic adjustment
These systems can perform well, but only if they are actively maintained. Without ongoing tuning:
- Sensitivity drifts
- Detection timing becomes inconsistent
- Operators compensate manually
Machine Learning Systems (KT2000 GEN2)
Machine learning–based systems take a different approach.
KT2000 GEN2 uses:
- Acoustic pattern recognition
- Classification based on real casting data
- Models trained on plant-specific conditions
Instead of fixed thresholds, the system learns what slag onset looks like.
But this introduces a key requirement:
Machine learning systems must be continuously trained and validated—making a performance-based model essential.
Why This Matters: The Resource Reality
Maintaining detection performance—especially for machine learning systems—requires capabilities that are not part of normal plant operations.
It involves:
- Data preparation and labeling
- Model training and validation
- Signal analysis across varying conditions
- Ongoing performance monitoring
Steel plants are designed to:
- Run production
- Maintain equipment
- Control the process
They are not structured to:
- Continuously retrain machine learning models
- Maintain detection systems at that level
A service-based model provides these capabilities without requiring internal resources.
What Happens Without Continuous Support
As casting conditions evolve, the signals that detection systems rely on do not remain constant. Acoustic patterns shift, background noise changes, and ladle behavior varies from heat to heat. These changes directly affect how reliably a system can interpret slag onset at the end of the heat.
In practice, this shows up differently depending on the system type:
- Classic systems: fixed thresholds gradually become misaligned with actual conditions
- Machine learning systems: models become less representative of current operating behavior
In both cases, the outcome is the same. Detection performance begins to drift:
- Sensitivity becomes inconsistent
- Detection timing shifts later or becomes less stable
- Yield is impacted through early shutoff or missed slag events
These losses are rarely obvious or tied directly to the system. Instead, they appear gradually—through variability, operator compensation, and small amounts of lost steel that accumulate heat after heat.
Why Performance-Based Models Are Emerging
The shift toward performance-based models reflects how slag detection systems actually function in real plant conditions. Detection is not a one-time setup—it is a performance function that must be maintained as the process evolves.
A service-based approach allows that performance to be actively managed over time. In practice, this includes:
- For Classic systems: ongoing tuning to keep sensitivity aligned with current conditions
- For GEN2 (ML systems): continuous model training, validation, and updates
- Regular performance review as casting practices and conditions change
- Adjustments based on real plant data—not initial assumptions
Rather than treating detection as a fixed installation, the system becomes an ongoing performance tool—one that improves alongside the process instead of gradually falling behind it.
Alignment: A Different Way of Working
Another key driver of this shift is alignment between the plant and the system provider. The structure of the model changes who is responsible for long-term performance.
In a perpetual model:
- The system is delivered
- Responsibility for maintaining performance shifts entirely to the plant
In a performance-based, service-based model:
- The provider remains actively engaged
- Performance is continuously monitored, reviewed, and improved
This creates a more practical working relationship:
- The plant focuses on production, yield, and stability
- The provider focuses on maintaining and improving detection performance
The result is shared accountability for outcomes—where both parties are working toward consistent, high-performance casting rather than maintaining a system that gradually drifts over time.
Perpetual vs. Service-Based Slag Detection | Key Differences
The difference between perpetual vs. service-based slag detection becomes clear when you compare how each model maintains performance over time.
| Capability | Perpetual Model | Performance-Based Model |
| Detection Performance | Fixed at install | Continuously maintained |
| Classic System Tuning | Internal | Supported |
| ML Model Management (GEN2) | Not maintained | Continuously trained |
| Adaptation to Conditions | Limited | Ongoing |
| Long-Term Value | Declines | Improves |
The Result: Consistent Detection, Higher Yield
The difference between a maintained system and a static one shows up clearly in day-to-day casting performance. When detection is continuously supported and aligned with current conditions, operators can rely on the system to make consistent, confident end-of-heat decisions.
In those conditions:
- Slag is detected more reliably
- Operators can cast with greater confidence
- Steel recovery improves heat by heat
- Variability across heats is reduced
When detection performance is not maintained, the system gradually becomes less reliable. Operators begin to compensate—often without realizing it—by shutting down earlier or second-guessing system signals.
Over time, this leads to:
- Increasing inconsistency in detection behavior
- Greater dependence on operator judgment
- Quiet, cumulative yield loss that is rarely traced back to the system
The difference is not dramatic in a single heat—but over time, it becomes measurable.
The Next Step in Slag Detection Performance
The shift away from perpetual systems is not a pricing trend.
It reflects how slag detection actually performs in real casting conditions.
Detection is not a one-time setup.
It is a performance function—one that must evolve with the process.
- Classic systems require ongoing tuning to stay aligned
- Machine learning systems require continuous model training and validation
- Both require active performance management to deliver consistent results
Static systems fall behind.
Performance-based systems improve with every heat.
At Kiss Technologies, we approach slag detection as an ongoing performance responsibility—not a one-time installation.
We work alongside your team to:
- Continuously refine detection performance
- Align the system with real casting conditions
- Ensure consistent, reliable results heat after heat
The result: higher yield, reduced variability, and detection you can trust.
If you’re evaluating how your current system is performing—or where performance may be drifting—we can help you quantify the impact and identify where improvements can be made.