In smelting and hot rolling of high-end special steel such as premium bearing steel GCr15, a production line is no longer a simple stack of standalone machines. It is a tightly cascaded physical system that combines ultra-high-power electric arc heating, vacuum refining and degassing, multi-strand continuous casting, and temperature- and cooling-controlled hot rolling.
For technical experts and plant managers at special steel producers, production data is never in short supply. The real challenge is that even while the electric arc furnace, refining furnace, vacuum station, continuous caster, and hot rolling mill stream high-frequency time-series data in real time, the plant still finds it hard to answer a few critical business questions quickly: where is the faint precursor of a quality anomaly? How do temperature and moisture disturbances from upstream processes propagate downstream through the thermal chain and the solidifying cast flow? Is the situation a fine-tuning of a single heat, or is it a systemic scrap risk that already threatens the purity limits of the molten steel and the mechanical properties of the finished product?
This is the dividing line for intelligent monitoring of special steel production lines. What enterprises need is not just putting scattered equipment readings on a big screen, but restoring the full cascading causal chain of “scrap moisture → insufficient vacuum → lower degassing efficiency → white spot rejection” and “grid fluctuation → refining over-compensation → excessive casting superheat → center segregation → abnormal rolling force and cracking,” and using it to guide data-driven quality tracing and accurate prediction on the floor.
Special steel and bearing steel production lines: quality control challenges across a multi-stage long process
Bearing steel is called the “king of steels.” Its requirements for oxygen, hydrogen, and nitrogen content and for inclusion size are extreme. A special steel production line (electric arc furnace EAF → ladle refining furnace LF → vacuum degassing station VD → billet continuous caster CCM → hot rolling mill) involves extremely complex physical, chemical, and phase-transition processes.
Along the long process of “EAF melting → refining and fine-tuning → vacuum degassing → continuous casting → hot rolling,” process parameters between stages are tightly coupled. Initial hydrogen buildup from wet scrap in the EAF directly raises the degassing load at the VD station. And if a seal leak keeps the VD station from pulling the vacuum below 67 Pa, hydrogen atoms cannot escape from the molten steel and eventually form deadly “white spots” (hydrogen-induced cracks) inside the finished bearing steel. Similarly, when grid fluctuation makes the EAF tapping temperature low, the LF furnace over-compensates with arc heating to hold the downstream casting temperature, which overshoots the ladle temperature and drives the molten steel superheat in the continuous casting tundish far beyond spec. That not only worsens carbon segregation and center segregation inside the billet, but also leads to transverse cracking during rolling from incompatible deformation.
At an industrial site where multiphase flow, high-temperature stress, and a tight production cadence all interweave, traditional siloed equipment monitoring cannot support quality tracing at all. When an analyzer at the quality control center detects excess hydrogen or cracked finished product, technicians usually have to cross-check reports from the EAF, the refining furnace, and the rolling mill over and over, and it is extremely difficult to sort out this physical reaction chain, kilometers long, within a few hours.
Many alerts but slow judgment: O&M pain points in special steel smelting and rolling
The same fluctuations can already be monitored, but the cause is still hard to explain
Under different heats and steel grades, the “normal baseline” of process parameters in special steel smelting shifts dynamically. Take a sharp rise in rolling force at the finishing mill in the rolling workshop. It could be normal physical resistance from an oversized billet cross-section, or it could be severe carbon segregation and hardness variation inside the billet (center segregation band hardness reaches 230 HBW, vs. a normal 198 HBW) caused by excessive superheat in the upstream casting stage and partially clogged secondary cooling nozzles. If rolling force only triggers a static alert, operators on the floor cannot tell that this is being driven by a superheat anomaly in continuous casting dozens of hours earlier.
A single-point alert has already fired, but the quality impact is hard to predict
Process defects in special steelmaking are extremely “hidden.” Excess hydrogen in molten steel from wet scrap or a leaking vacuum pump shows no deterioration in any process parameter during EAF melting and continuous casting; the billet looks completely “qualified.” This “silent fault” travels all the way through casting and hot rolling, and only erupts when hydrogen atoms gather and form white spots in the finished steel in the slow cooling warehouse, causing rejection of the entire batch. Unless the vacuum level and the [H] removal rate during degassing give an early warning, the plant pays a heavy price in rejected finished product.
The business consequences are not abstract: whole-batch rejection and delivery delays are costly
Slow judgment and misjudgment show up directly in the per-ton profitability and delivery lead time of a special steel plant.
- In the hydrogen-induced cracking anomaly (heats H-093 to H-100): wet scrap plus a leaking VD vacuum pump seal left the vacuum at just 180 Pa, and the hydrogen content of the molten steel rose from 0.9 ppm to 3.8 ppm. When ultrasonic testing of the finished steel picked up white spots, 4 consecutive heats of bearing steel, 400 tons in total, were all rejected, with direct losses exceeding about US$333,000.
- In the superheat-out-of-control anomaly (heats H-071 to H-090): grid power shortfall kept the tapping temperature low, and LF arc over-compensation pushed CCM superheat as high as 46°C (target 22°C). Combined with abnormal mold vibration and clogged secondary cooling, the center segregation rating broke through grade 2.5. During hot rolling, finishing mill vibration surged to 5.0 mm/s, the wire cracked in a regular pattern, and the finished product scrap rate hit 8%, causing direct losses of tens of thousands of US dollars plus late-delivery penalties.
From passive monitoring to proactive judgment: reorganizing the data pipeline
Object-based modeling: scattered measurement points return to a unified operational view
To give multi-stage smelting data clear operational meaning, it needs to be mapped onto clear physical entities. TDengine uses Industrial Ontology modeling to organize the EAF-1 electric arc furnace, the LF-1/2 refining furnaces, the VD-1 vacuum degassing station, the CCM-1 continuous caster, the Reheat-1 reheating furnace, the rolling mills at each stage, and the OES-1/ONH-1 quality inspection spectrometers into a tree-shaped asset hierarchy.
Figure 1: A tree hierarchy organizes plant assets and measurement points into a unified operational view
Once the data carries ontology relations, every sensor’s time-series signal is tagged with heat number, steel grade, process stage, and equipment attribute. Following the lifecycle of a given heat, process experts can pull up its full set of parameters from EAF melting to final rolling with a single click, removing the information silos between process stages.
Real-time analysis and event linkage: alerts return to the full process context
On top of the asset model, TDengine’s real-time analytics engine runs online stream computation on indicators such as “power consumption per ton of steel, vacuum index decay rate, tundish superheat, per-stand speed ratio, and dynamic rolling force deviation.” When it detects abnormal deviation, the system packages an “event context” and pushes it to the dispatch screen.
The “smart query” and “proactive recommendation” services let process staff search for quality gaps in natural language. When the ONH analyzer records excess hydrogen in the molten steel, the system correlates the preceding VD station’s vacuum hold time and minimum negative pressure and feeds the result back to the operator, so billets at risk of white spots are intercepted before they leave the casting stage, avoiding wasted rolling and slow cooling downstream.
Figure 2: AI-recommended real-time analysis
Figure 3: General information settings for real-time analysis
Figure 4: Trigger conditions for real-time analysis
Process analysis and AI-assisted insight: judgment shifts from experience-driven to evidence-driven
Once multi-source smelting time-series data is modeled as objects and linked by anomaly, on-site process optimization breaks free of traditional experience and shifts to data investigation on a shared timeline.
The system provides intelligent process analysis and shift comparison. For quality anomalies such as billet center segregation or cracked finished product, process staff can pull up the relevant heat data and overlay it against a golden heat (such as H-110) for parameter-by-parameter comparison. AI measures the LF arc power overshoot, computes the integrated value of CCM superheat in the elevated range, quantifies the deviation slope of secondary cooling water pressure, and produces a one-click diagnostic report with causal logic and time-series charts, turning closed-loop quality improvement into practice.
Figure 5: AI interpretation and data mining on the analysis panel
Figure 6: AI event root cause analysis
The analysis loop in typical anomaly scenarios
Anomaly case 1: wet scrap and vacuum leak cause hydrogen-related cracking risk and whole-batch rejection
During heats H-093 to H-100, the EAF melted wet scrap stored in the open air. At H-093, the EAF cooling water temperature difference rose from 8°C to 9.5°C, and initial hydrogen pickup in the molten steel increased. By H-095, moisture in the scrap peaked, the EAF cooling water temperature difference reached 13°C (furnace wall heat load exceeded spec), and initial hydrogen content in the molten steel hit 4.0 ppm. After refining, the steel moved to the VD vacuum degassing station. Because the vacuum pump seals had aged and leaked, the VD station only reached 180 Pa (normal: below 67 Pa, which triggered a vacuum-exceeded alert), and the vacuum hold time had to be cut to 12 minutes. Insufficient degassing left final hydrogen content at 3.8 ppm. Although the casting process looked normal on the surface, white spots formed in the billets during slow cooling, and 4 heats of finished product were rejected, with losses of about US$333,000.
After ultrasonic testing found white spot defects, quality inspectors used the process analysis workbench for a three-step trace:
Step 1: Pull up final quality and composition results. Review the ONH-1 hydrogen content and the OES-1 ultrasonic test marks. The data shows that the finished hydrogen content of the four heats from H-095 to H-098 all fluctuated between 3.2 and 4.2 ppm (far above the 2.0 ppm control limit), and ultrasonic testing revealed clear white spot cracks, with defect_flag marked as 2 (rejected) in every case.
Step 2: Correlate the vacuum degassing stage. Overlay the VD station’s vacuum and degassing curves. The abnormal heats showed the vacuum chamber pressure deteriorating from a normal 50 Pa to 180 Pa during vacuum hold, and the degassing rate flattened markedly (the [H] exponential decay constant dropped from a normal -0.069 to -0.03). This confirmed that insufficient vacuum pump capacity was the main reason hydrogen atoms could not escape the molten steel.
Step 3: Trace back to wet scrap at the melting source. Pull up the EAF cooling water temperature difference and furnace load. As early as H-093, the EAF cooling water temperature difference had already climbed from 8°C to 13°C, which proves that the wet scrap released large amounts of water vapor during EAF melting and drove initial hydrogen pickup in the molten steel higher.
Figure 7: Judgment path from wet scrap and excess VD vacuum to rejection on excess hydrogen in finished product
Through these three steps, the plant clarified the fault logic: wet scrap → high initial hydrogen pickup in the EAF molten steel → VD vacuum seal leak keeping the vacuum below 67 Pa → insufficient vacuum hold time and collapsed degassing efficiency → hydrogen accumulation in the finished steel cracking during slow cooling.
Anomaly case 2: grid fluctuation and refining over-compensation cause continuous casting superheat loss of control and rolling cracking
During heats H-071 to H-090, grid voltage fluctuation cut the EAF input power to 72,000 kW (normal: 90,000 kW), and the tapping temperature dropped to 1560°C. To compensate for the low temperature, the LF furnace ran heating at full power of 20,000 kW (normal: 14,000 kW), which overshot the ladle discharge temperature to 1620°C. In casting, CCM tundish superheat blew far past spec to 46°C (normal: 22°C). At this high superheat, secondary cooling water pressure fell from 3.5 bar to 1.9 bar at H-080 because of clogged nozzles, and mold vibration frequency went abnormal at H-083, pushing the center segregation rating inside the billets past grade 2.5. In the final hot rolling stage, insufficient soaking raised the roughing mill rolling force to 10,500 kN, finishing mill vibration jumped to 5.0 mm/s, and the wire surface cracked in a regular pattern.
After spotting the rolling cracks, the hot rolling plant immediately joined with the steelmaking plant for a full-process review:
Step 1: Pin down the rolling anomaly. Review the Roughing-1 rolling force and Finishing-1 vibration trends. When heat H-086 was rolled, the roughing force jumped from 8,000 kN to 10,500 kN and finishing vibration spiked from 1.5 mm/s to 5.0 mm/s. This confirmed a sudden change in the steel’s resistance to deformation, with the segregation band and the matrix deforming badly out of sync.
Step 2: Trace back to the casting solidification and heat transfer state. Overlay the CCM superheat, mold vibration, and secondary cooling water pressure curves. The abnormal heats kept tundish superheat in the very high range of 42 to 48°C for a long time, and secondary cooling water pressure dropped to 1.9 bar during H-080 to H-082. High superheat plus insufficient cooling lengthened the liquid pool in the solidifying billet and drove solute elements heavily toward the center.
Step 3: Reach the root cause in refining and primary melting. Walk back through the LF heating power and EAF temperature trends. Grid fluctuation lowered EAF power, and the tapped steel came out at just 1560°C. The LF furnace over-applied arc heating during alloying (20,000 kW) and overshot the discharge temperature.
Figure 8: Judgment path from grid fluctuation and low tapping temperature through LF over-compensation to CCM superheat loss of control and rolling cracking
This case shows the following multi-stage causal chain: grid fluctuation → insufficient EAF power and low tapping temperature → LF over-compensation with heating → continuous casting tundish superheat exceeding the limit at 46°C → lengthened liquid pool in mold solidification → billet center segregation rating past grade 2.5 → uneven hardness in hot rolling with surging rolling force and vibration → downgraded finished wire with cracking.
Intelligent monitoring of special steel production lines: the real need is causal visibility through full-process time-series data
In steelmaking with high temperature, near-continuous operation, and many heats running in parallel, the core value of an intelligent system is to physically chain the scattered time-series data from the EAF, refining, casting, and rolling through ontology objects, so operators can see downstream quality risk at the onset stage of white spot or segregation defects.
With its ultra-high-performance time-series storage engine, multi-dimensional ontology modeling, and agile process workbench, TDengine helps special steel enterprises remove the barriers between process stages and the quality department, enabling precise tracing of steel quality and continuous optimization of production processes.
TDengine comes with a high-performance, distributed time-series database, Industrial Ontology modeling, and an Industrial Agent Runtime, providing a full-stack solution for industrial data streams from collection and storage to real-time analytics, visualization, event management, and root-cause analysis. To learn more about TDengine, visit www.tdengine.com and try it for free.
Try it yourself
Install and deploy TDengine Visit the TDengine Download Center, select TDengine All-in-One, choose the deployment platform and architecture that matches your environment, and follow the guided steps to complete the installation.
Load the sample data
On first activation, choose Steel Continuous Casting Process Optimization on the sample data loading screen and wait for it to finish loading.
If you have already activated the product, click your avatar in the top-right corner, select Management Console, choose Sample Data on the left, then select Steel Continuous Casting Process Optimization. Wait a few minutes for the data to load.
Steel Continuous Casting Process Optimization
