Water-cooled copper tuyeres survive in cupola and blast furnace environments exceeding 1,093 °C (2,000 °F) only because the copper body (pure Cu melting point ~1,084 °C / 1,984 °F) is held below its softening point by a continuous water circuit; the moment that circuit develops a micro-leak, the local copper wall burns through and injects water into the hearth [S4][S3].
A modern blast furnace operates 20–40 tuyeres fed by parallel cooling circuits, and the leak detection problem is essentially a mass-balance problem: instrument the inlet and outlet of each circuit, compare flows at high accuracy, alarm on the differential before the leak grows [S1][S3].
How the differential flow-meter method works, and what accuracy is actually required
The reference installation, ArcelorMittal Florange in France, sized the alarm around a 150 L/hr flow differential on a 40 m3/hr design flow, which is approximately 0.375% of circuit flow, and required meters to deliver at least twice that resolution [S1]. Emerson Rosemount 8700 magnetic flow meters were specified in a high-accuracy configuration delivering 0.15% of reading, more than 2x the minimum accuracy called out in the specification, and were deployed on 22 tuyeres to read both inlet and outlet [S1]. A competing reference, the SmartMine processing-plant guide, states the same 1/4% (0.25%) accuracy band as a general rule for tuyere cooling leak detection flow meters [S7].
Detection latency, not absolute accuracy, is the dominant design driver, because a sustained leak ultimately generates hydrogen at the tuyere nose, chills the hearth, and erodes the furnace wall, an outcome operators treat as both a safety and a refractory-life event [S1][S3]. The Hatch pressure leak detection system, originally developed for copper coolers in electric arc smelting furnaces, demonstrated sensitivity to leaks of only a few drips per minute, an order of magnitude finer than what an inlet-outlet flow comparison can resolve on a 40 m3/hr circuit [S2].
Pressure-based and NDT-based detection for the drips-per-minute regime
Pressure-based leak detection works on a different physical principle: a closed, pressurised cooling element shows a slow pressure decay when a hairline crack is open, and the leak rate scales with crack area, so a few-drips-per-minute leak is observable as a controlled pressure drop over a defined test interval [S2]. The Hatch paper documents that pressure-based detection complements inlet-outlet flow comparison because flow comparison flags gross circuit imbalance, while pressure decay flags small through-wall cracks before they grow into the imbalance regime [S2].
Residual wall thickness on operating copper coolers is measured by Low Frequency Pulse Ultrasonic (LFPU) testing, a non-destructive method that maps remaining copper thickness from the outside of the cooler without removing it from the furnace [S2]. The two techniques are complementary rather than competing: pressure decay answers "is there a leak now", LFPU answers "how much copper is left before a leak becomes likely", and the flow-meter differential answers "is the circuit as a whole losing water to the furnace" [S2].
AI analytics: thermal load, wear score, and the slow-leak gap
High-frequency cooling water data (inlet and outlet temperature, flow, pressure) feeds an analytics layer that computes a per-tuyere "Wear Score" by correlating thermal load with blast pressure fluctuations and PCI (pulverized coal injection) rate, the dominant modern tuyere wear driver [S3]. The architecture described by iFactory treats static time-based replacement and manual peep-sight inspection as obsolete, replacing them with a digital lifecycle record that flags individual tuyeres for replacement during planned down-days rather than emergency blow-downs [S3].
The slow-leak failure mode is where the architecture adds the most value: a stress crack can leak slowly for hours or days, degrading coke combustion and lowering hearth temperature before the conventional flow-differential alarm fires, and AI detects the micro-fluctuations in thermal load minutes before a critical burnout [S3]. This is the same slow-leak regime that pressure-based detection targets, which is why the reference deployments are converging on a layered architecture: differential flow meters at the circuit level, pressure decay on critical coolers, and analytics on top of the high-frequency data [S1][S2][S3].
Patent landscape 2026: who is filing, and where the gaps sit
A patent-landscape analysis covering 2015–2026 (data cut-off 2026-08-31) found 124 published records on tuyere cooling water leak monitoring, almost entirely classified under C21B (blast furnace equipment), with secondary claims concentrated in F27D (furnace details, 10.5% of records) and F27B (furnaces and kilns, 7.3%); dedicated sensing (G01M) and digital processing (G06F) classes each carry under 3.5% of records [S5]. Annual filings rose from a single record in 2008 to a peak of 10 in 2023, and 2021 to 2024 grew roughly +11%; 2025 and 2026 numbers are still incomplete because publication lags filing by about 18 months [S5].
Top assignees by record count are Japanese and Korean integrated mills (Nippon Kokan, Kawasaki Steel, Pohang Iron & Steel, Kobe Steel, Sumitomo Metal Ind, Nippon Steel, JFE Steel) together with Chinese producers (Xinjiang Bayi, Shandong Iron & Steel, Ansteel Engineering), and the practical reading is that detection is mostly claimed as a feature of a furnace or cooling-system patent, not as a freestanding sensor invention [S5]. A Chinese patent CN101775454B (Google Patents) exemplifies the genre, claiming a tuyere leakage monitoring and alarming system that combines flow and pressure signals into a single trip output, filed as part of the blast furnace monitor category [S8].
Selection criteria and trade-offs: flow, pressure, analytics, calorizing
Four practical options exist, and the choice is set by what leak size the plant needs to catch and how much capex the cooling skid can absorb. A typical cupola or small blast furnace plant can compare the options against the decision criteria below: [S1]
Differential magnetic flow meters (e.g., Rosemount 8700 high-accuracy configuration, 0.15% of reading) are the baseline for any new build, deliver 0.375% differential alarm on a 40 m3/hr circuit, and integrate cleanly on Profibus PA or HART to existing DCS [S1]. Pressure-decay leak detection (Hatch-type) catches a few drips per minute, which is roughly an order of magnitude finer than flow comparison on a 40 m3/hr circuit, and is best deployed on copper coolers serving high-risk zones [S2]. AI analytics on top of the flow and temperature stream adds slow-leak and wear-life prediction, at the cost of a per-tuyere data historian and a wear-score model that the plant has to maintain [S3]. Calorizing (aluminising) the inner copper surface addresses the underlying wear mechanism by forming an intermetallic alloy layer that resists molten iron and slag attack, and is the benchmark for high-production blast furnaces in North America, but it is a manufacturing-side decision rather than an instrumentation decision [S4].
For cupola operators specifically, where tuyere counts are typically lower (often under 20) and operating temperatures and PCI are lower than a large blast furnace, the lower-cost option is usually a differential magnetic flow meter pair with the 0.25% accuracy band on inlet and outlet, supplemented by a manual LFPU thickness check on a planned maintenance interval, rather than a full pressure-decay system [S2][S4][S7]. A related but distinct issue, copper rod and wire mill capacity additions, is covered in this 2026 capacity update, and the broader question of how copper material selection interacts with calorizing and water-side corrosion is treated in the encyclopedia entry.
Failure modes, alarm tuning, and the start-up check
The two failure modes that drive the specification are nuisance trips and missed slow leaks, and they pull in opposite directions. Nuisance trips come from meter drift on a circuit that has not actually leaked, and the 0.15% accuracy spec (more than 2x the alarm threshold resolution) is exactly there to keep the false-trip rate down on a 40 m3/hr circuit [S1]. Missed slow leaks come from a stress crack that leaks below the flow-differential resolution for hours before growing into the alarm band, and the layered response is pressure decay on critical copper coolers plus analytics on the high-frequency stream [S1][S2][S3].
Operational discipline during start-up is the cheapest reliability gain, and the standard sequence is to reconnect the cooling water circuit, verify flow on both inlet and outlet meters, and physically check for leaks at all connections before blast-on, because a connection leak at start-up shows up as a false-positive differential and a real connection leak at the same time [S6]. A useful side benefit of the magnetic flow meter architecture is that the meters are self-checking via advanced diagnostics, which catches a failed meter (a third failure mode) before it either masks a real leak or causes a nuisance trip [S1].
Applicable standards and instrumentation context
No single ISO or IEC standard governs tuyere leak detection end-to-end, because the design is a process-safety choice driven by the metallurgy of the copper cooler and the physics of two-phase water-into-molten-iron interaction. What is standardised is the instrumentation layer: magnetic flow meters typically carry a Metrology Institute custody-transfer accuracy statement, and the flow meter reference in the encyclopedia covers the magmeter and Coriolis accuracy classes a specifier should expect on cooling-water service. Process-side pressure and leak detection instrumentation is covered in the gas detection and online water analyzer entries, which describe the pressure-transducer and conductivity-monitoring techniques that complement flow comparison on the same circuit. Where a furnace controls upgrade also touches burners, refractories, or downstream molten-metal handling, the pressure transmitter and ballast water treatment encyclopedia entries give the specifier the boundary conditions for adjacent instrumentation classes. [S1]
Trackable next signals: filings published in 2026 under C21B with secondary G01M or G06F claims, which would indicate a shift from furnace-incorporated detection claims toward freestanding sensor inventions [S5]; and the next round of ArcelorMittal or comparable integrated-mill case studies reporting nuisance-trip rate after the 0.15% magmeter retrofit, which would quantify the false-alarm reduction that the Florange reference only describes qualitatively [S1].