As a result, quality is often confirmed after production rather than actively managed during it. This
separation between process and product remains one of the main barriers to achieving fully stable and
predictable rolling operations.
In recent years, advances in inline non-destructive testing (NDT), combined with progress in data
integration and machine learning, have begun to close this gap. Among the companies driving this
transition, ISEND direction is clear: moving from isolated defect detection toward continuous, processintegrated understanding of material behavior.
Structural limitations and the need for integrated quality control
Traditional mill inspection systems typically provide only partial coverage of the product. Offline measurements and manual checks capture limited snapshots of quality, while process data is often distributed across multiple systems with little synchronization.
This fragmented structure tends to delay the detection of deviations until they have already affected the final product. In operational terms, this leads to conservative production strategies such as wider tolerances, higher safety margins, and reactive process adjustments.
While these measures reduce risk, they also limit efficiency in terms of yield, energy consumption, and process stability.
A more integrated approach is based on continuous measurement of the material itself during production, enabling quality to be managed as part of the process rather than as a post-process verification step.
Inline NDT as a continuous source of process intelligence
EDDYeyes system (Source: ISEND)
EDDYeyes system (Source: ISEND)
Modern inline NDT systems extend beyond simple defect detection and increasingly act as real-time sources of process intelligence.
Our solution HOTdiscover, eddy current-based inspection system, allows continuous monitoring of surface integrity under demanding production conditions. Instead of isolated signals, they generate continuous patterns that reflect variations in material condition along the product.
When combined with high-speed optical inspection and machine learning models trained on plantspecific data, these signals can be further classified in terms of defect type, morphology, and severity. The combination of electromagnetic and visual data improves both detection robustness and interpretability.
In addition, computer vision techniques enable continuous measurement of geometric parameters and the identification of defect types that may not be visible through a single sensing modality. This multi-sensor approach improves consistency in defect interpretation and reduces dependence on manual assessment.
EDDYeyes system is the result of combining these technologies, synchronized for operation under extreme inspection conditions.
Understanding the process through mass and geometry measurement
Beyond surface inspection, ISEND’s MASSdiscover system introduces continuous measurement of section and mass throughout the rolling process. With high accuracy under harsh mill conditions, it provides real-time insight into material flow and deformation behavior between stands.
This capability is particularly relevant for identifying instabilities that are otherwise difficult to detect, such as progressive mass deviations, strand imbalances, or the combined effects of temperature and reduction inconsistencies. By observing these phenomena directly, operators can correlate process parameters with material response, moving from assumption-based to evidence-based control.
In practical applications, this type of measurement has been shown to reduce cobbles, improve rolling stability, and enable operation closer to specification limits without increasing risk.
From sensing to system-level understanding: the role of data integration
A key aspect of ISEND’s approach lies in the integration of these sensing technologies within a unified data environment. STARconnexion is the platform that synchronizes data from inline sensors with processes inputs from quality records.
This creates a coherent data structure in which each product can be traced back to its origin billet, deformation history, and associated quality indicators. More importantly, it enables the correlation of process conditions with quality outcomes across time and production campaigns.
Rather than isolated measurements, the system provides a continuous, data-consistent representation of the mill’s behaviour.
Enabling predictive process control
With structured and synchronized datasets, the application of AI extends naturally beyond classification into predictive analysis. Within ISEND’s framework, AI models are used to identify recurring patterns associated with defect formation, process instability, or performance degradation.
This allows early detection of conditions that typically precede quality issues, such as periodic defects linked to mechanical elements or mass deviations leading to instability events. In this way, the system supports a transition from reactive intervention to preventive action.
Importantly, these models are not developed in isolation but are built on plant-specific data and validated through operational experience, ensuring their relevance to real production conditions.
Industrial impact and operational validation
The integration of continuous measurement, data synchronization, and predictive analysis has a direct impact on plant operation.
Improved visibility of mass and section allows more precise control of tolerances, reducing systematic overproduction and improving material yield. At the same time, enhanced defect detection and classification support more consistent quality assessment and reduce downstream quality claims.
From a process perspective, increased transparency contributes to more stable operation, fewer unplanned interruptions, and improved utilization of energy and equipment.
In parallel, the availability of structured historical data strengthens root cause analysis and supports continuous improvement initiatives across production campaigns.
Knowledge integration and workforce implications
An additional benefit of integrated data systems lies in the preservation and structuring of operational knowledge.
Process expertise, often distributed across experienced operators and engineers, can be embedded into rules, thresholds, and interpretation frameworks within the system. This helps reduce dependency on individual experience and ensures more consistent decision-making across shifts and production lines.
At the same time, reducing the need for manual inspection in hazardous environments contributes to improved operational safety.
Concluding remarks
The transition from detection to prediction in steel quality control is not driven by a single technology, but by the integration of sensing, data, and interpretation within the production process.
ISEND’s portfolio, combining inline NDT, mass and geometry measurement, and operational data integration, provides a practical example of how this transition can be implemented in industrial environments. By enabling continuous, data-driven understanding of both material and process behaviour, it supports a shift towards more stable, efficient, and predictable rolling operations.
In this evolving landscape, the convergence of process control and quality control is becoming increasingly evident. The ability to observe, understand, and anticipate process behaviour in real time is no longer a theoretical objective, but an operational reality already being deployed in advanced rolling mills.
Author
Francisco Moreton (Source: ISEND)
