AISTech 2019 — Proceedings of the Iron & Steel Technology Conference

6–9 May 2019, Pittsburgh, Pa., USA

DOI 10.1000.377.055

Implementation of a Tuyere Leak Detection Soft Sensor on USS Blast Furnace No. 14

Yasser Ghobara1, Ravi Pula1, Ian Cameron1, Michael Bodley1, John Busser1, Mitren Sukhram1,
John G. Grindey2, Ralph Albanese2
1
Hatch Ltd.
2800 Speakman Drive, Mississauga, ON, Canada, L5K 2R7
Phone : 905-855-7600
E-mail: Yasser.Ghobara@hatch.com
2
United States Steel Corporation
1 North Broadway, Gary, IN, United States, 46402
Phone : 219-888-1820
E-mail: RAlbanese@uss.com

Keywords: Blast Furnace, ironmaking, tuyere water leakage, process monitoring, soft sensor

INTRODUCTION
Blast Furnace No. 14 (BF14), the largest blast furnace at U. S. Steel Corporation, is equipped with open circuit water cooling
for all tuyere, plate and stave cooling elements. In 2016, BF14 experienced rapid hearth wall temperature increases under all
three tapholes.1 Water leakage can compromise the protective skull on the hearth wall leading to elevated hearth refractory
temperatures. In extreme cases, water leakage can oxidize the hearth carbon refractory. Improved leak detection from the
tuyere nose cooling circuits was one of several countermeasures identified to reduce hearth refractory temperature
excursions. In 2017, Hatch developed a leak detection soft sensor. In 2018, Hatch and USS implemented the soft sensor to
rapidly detect small water leaks using the existing tuyere nose circuit flow meters. BF14 has 34 tuyeres, further information
on the BF14 design and operation are available from the AIST Blast Furnace Round-up.2

BACKGROUND
In addition to damaging hearth refractory, water leaks can cause unstable burden descent, chill the blast furnace hearth, and
cause production losses. Early detection of water leaks is critical to avoid these issues. Historically, water leaks were detected
by plant personnel through routine blast furnace inspections, interpreting furnace operating conditions and via pressure tests
during planned stops. Instrumentation in various forms, was later employed to improve leak detection while the blast furnace
was in operation. Continuous leak detection, by comparing the water circuit inlet and outlet flow rates, is a popular system
for tuyere, plate and stave cooling circuits. A leak is detected when the return water flow is less than the supply flow
measurement. Figure 1 shows a schematic of such a system.

© 2019 by the Association for Iron & Steel Technology. 519

 Figure 1. Inlet and outlet flow meters used for water leak detection in blast furnace cooling systems3
A flow-based system has two advantages; it is continuous and can detect leaks within a given threshold of differential water
flow. The limitations of a flow-based system are measurement errors due to air in the water, a need for identical flow meter
installation, instrument calibration precision and data interpretation. To detect a small water leak, it is critical that the two
flow meters being compared are very precisely matched.3,4
In a closed loop cooling system, water leakage is easily detected based on water make-up rates, something that cannot be
done in an open loop system. With that said, it remains challenging to identify the exact tuyere or stave/plate cooler circuit
that is leaking within the closed system due to many individual cooling circuits. Agrawal et al. developed a model to detect
water leakage from tuyere nose cooling circuits in a closed loop system.5 A drop in the water tank level was used to signal a
leak event. The differential flow rate at each tuyere circuit was then assessed to identify the leaking tuyere circuit using a
dimensionless leak detection factor (LDF).
Blast Furnace No. 14 has magnetic inlet and outlet water flow meters on the tuyere nose circuits to detect water leaks. The
resulting data can be hard to interpret, and small leaks may be missed. If a tuyere leak is suspected, the supply pressure on
each tuyere nose circuit is systematically reduced to less than the BF14 internal pressure to allow CO rich bosh gas to enter
the water system should a leak be present. A personal CO gas monitor is positioned at the circuit discharge to measure any
CO gas that is exiting with the discharge water. If CO gas is detected, a planned shutdown is initiated and the leaking tuyere
replaced. At every blast furnace stop, the BF14 team pressure tests each tuyere nose circuit to assure that no leaking tuyeres
are present.
This paper describes the implementation of a soft-sensor at BF14 that can detect small tuyere leaks using the existing water
flow measurements. Early detection allows the BF14 team to replace leaking tuyeres before the water leak causes operational
problems.

EARLY INVESTIGATIONS
In August 2017, Hatch audited the leak detection performance of the magnetic flow meters (Rosemount 8732E) installed on
BF14’s 34 tuyere nose water cooling circuits. These flow meters have an accuracy of ± 0.25% at water flow rate of 1.2 ft/s.
This is equivalent to 0.3 US gallons per minute (gpm) at 120 gpm per circuit. A performance analysis indicated that a water
leak of ~2.5 gpm may be visible using the Rosemount 8732E flow meters and the existing presentation software. The best
leak detection rate with this system was 0.5 gpm with 1.0 gpm being acceptable; this requires precisely matched flow meters.
Generally, blast furnaces identify water leaks by subtracting outlet from inlet flow rates and alarming based on a preset
difference. The challenge with the magnetic flow meters installed at BF14 was that many differential flow rates did not equal
zero; the difference ranged from -3.5 to +7.3 gpm, see Figure 2. Identifying small leaks <1.0 gpm, from these differential
flow rates was challenging.

520 © 2019 by the Association for Iron & Steel Technology.

 Figure 2. August 2017 audit comparing outlet to inlet water flow rates for the 34 tuyere nose cooling circuits at
BF14 over 11 different 6-day intervals.
During the 2017 audit, the magnetic flow meter data was investigated for 11 events when a tuyere nose was burnt, water
leakage occurred and the tuyere was changed out. Only 2 out of the 11 events were detected in the PI system by comparing
the differential water flow rates. A detailed analysis of flow rates preceding each leak event was unsuccessful in identifying a
common leakage signature. Some were abrupt divergences of the inlet and outlet water flow rates but more commonly, the
differential water flows were variable. For 7 of the 11 tuyere leak events, a noisy pattern such as the one presented in Figure 3
was evident. Note that the outlet flow was greater than the inlet flow before and after the tuyere was replaced. Also, the
differential flow continued to exhibit variability once the leaking tuyere was replaced.

Figure 3. Differential flow measurements before and after a tuyere nose circuit leak on tuyere 24 experienced on
May 17, 2017. A similar pattern was observed on 7 of 11 tuyere leak events investigated
The audit highlighted the difficulty in detecting tuyere water leakage using single signatures generated from the magnetic
flow meter differential flow measurements. A deeper analysis was required.

SOFT SENSOR DEVELOPMENT

Various statistical signal processing techniques were tested to validate the most appropriate way to detect a change in the
differential water flow rate and potential leak. This section will discuss the software tool implementation, available plant
measurements and model calibration. For the balance of the paper, the differential flow refers to the inlet-outlet flow so that
water leakage will yield a positive value.

© 2019 by the Association for Iron & Steel Technology. 521

Inputs and available measurements
BF14 has an open circuit tuyere nose cooling system with four (4) headers feeding thirty-four (34) tuyeres. Each tuyere nose
circuit has two (2) magnetic flow meters installed that measure the inlet and outlet water flow rates across each circuit. The
accuracy of the magnetic flow meter is ± 0.3 US gallons per minute (gpm). The soft sensor uses inlet and outlet flow rates to
calculate the differential flow across each tuyere cooling circuit. Ideally, the differential flow should be zero. Any positive
deviation in the differential flow rate means that the outlet flow is less than the inlet flow. If the absolute differential flow
across the tuyere is used to detect water leaks, this will increase the frequency of false alarms. It is critical that the software
tool differentiate between a tuyere leak event and calibration issues. In Figure 4, one can see an example of two signatures
representing a calibration issue and a leak event on tuyere 15.

Process shift due to

calibration issues

Tuyere leak event

Figure 4. Differential flow across tuyere 15 showing both a calibration issue and a leak signature
To detect and filter tuyere leak signatures, two statistical approaches were applied; a mean centering approach using the
average differential water flow rate and a median centering approach where the median value (mid-point of the data) was
used.

Mean Centering Approach

With mean centering, the centers of each measurement are compared against the moving mean of previous measurements.
The sampling interval for each new measurement and the interval for the mean of previous measurements are considered as
tuning parameters in the model. With this approach, drifts in the differential flow rate that are unrelated to tuyere leakage (i.e.
flow meter calibration events/issues) become tuned out compared to using the absolute differential flow rate. Figure 5 shows
the performance of the mean centering approach on the two signatures that were previously discussed on tuyere 15.

522 © 2019 by the Association for Iron & Steel Technology.

 Process shift due to
calibration issues

Tuyere leak event

Process shift due to

calibration issues
Tuyere leak event

Figure 5. Mean centering the differential water flow rate for tuyere 15
If an operator just looked at the absolute differential flow, it appears that a leak occurred on November 12, 2017 where in fact
this was a process shift due to instrument calibration. With the mean centering approach, the signature declines immediately
after the November 12th spike providing evidence that the change was not a result of a water leak. In January 2018, when
tuyere 15 did indeed leak, the mean centered difference increased continuously as the leak volume increased.

Median Centering Approach

The median of a data population is a robust statistic as data outliers have less influence on median values. Using median
values will generate less false alarms compared to the mean centering approach. The median differential flow rates of recent
and historical moving windows are sampled at different intervals. Flow drifts are detected and reflect an instantaneous
deviation in the median centered differential flow. Figure 6 shows the performance of the median centering approach on the
same signatures that were previously discussed on tuyere 15.

© 2019 by the Association for Iron & Steel Technology. 523

 Process shift due to
calibration issues

Tuyere leak event

Process shift due to

calibration issues

Tuyere leak event

Figure 6. Median centering of differential water flow rates for tuyere 15

Median centering is not affected to the same degree by the instantaneous outliers in the raw data when compared to the mean
centering method. Using median centering, fast drifts in the differential flow rate can be detected while slow changes may be
missed. Slow drifts are better detected by the mean centering approach. Using both approaches provides the operator with a
robust way to detect both types of drifts in the differential water flow needed to identify a leaking tuyere.

Model Calibration
The soft sensor was calibrated against historical tuyere leak events at BF14 and the model tuning parameters were
determined. To calculate the soft sensor alarm limits, a stable period was selected for each tuyere. This period was defined as
1-2 months with no tuyere leaks or furnace shutdowns. The resulting variability in differential water flow was used to
calculate the upper control limit.
The same methodology was applied to calculate the control limits for the median centering approach. Based on the soft
sensor performance and the calculated statistical control limits, an alarm limit of 0.3 gpm was set for all tuyere nose cooling
circuits. This alarm limit was applied on both the mean centering and median centering trend charts.

RESULTS AND DISCUSSION

Implementation
The soft sensor was successfully implemented into the BF14 PI system in May 2018. Figure7 shows a summary chart that
displays the mean centered deviation for each tuyere and available trend graph for groups of 8 or 9 tuyeres.

524 © 2019 by the Association for Iron & Steel Technology.

 Figure 7. Leak detection summary table in the BF 14 ‘Key Parameters’ PI workbook displaying the mean centered leak
detection data and available trend graph for tuyeres 1-8.
This table was displayed with the BF14 ‘Key Parameters’ workbook available at all workstations. Trend graphs of tuyere
mean and median centered differential flows were also available to help the control room operators interpret the leak
detection model data.
An alarm limit was developed based on the mean centered differential flow exceeding 0.3 gpm. The median centered flow
trend chart was then examined to confirm that a water leak was present. An audible alarm based on the mean centered flow
data was later implemented after the first six (6) tuyere leak events were successfully detected. In these six (6) cases, the
mean and median centered deviations were all greater than 0.3 gpm throughout the duration of the leak events. Figure 8
shows the control action process when a leak detection alarm is sounded.

Figure 8. Response when the tuyere leak soft sensor detects a potential water leak

© 2019 by the Association for Iron & Steel Technology. 525

Once the audible alarm sounds, the control room operator identifies the leaking tuyere nose circuit and checks the median
centered differential trend for the same cooling circuit. If the median centered values also indicate a potential leak, the control
room operator will ask the shift area manager to perform a gas test on the suspected tuyere to confirm the presence of a water
leak. If CO gas is detected in the discharge water, preparations are made for a planned shutdown to replace the leaking
tuyere. If the pressure test is negative, BF14 will continue to operate and the shift area manager and control room operator
will note the event and have further discussions with the process engineers and day shift management. Experience indicated
that within a day or two, the leak rate would increase and ultimately a positive gas test would confirm that the tuyere needed
to be replaced.

Tuyere Leaks Detected Following Implementation

The first tuyere leak that was successfully detected with the soft sensor was tuyere 7 on May 13, 2018, two weeks after the
soft sensor system was implemented, see Figure 9.

Figure 9. First tuyere leak detected on tuyere 7 using the new leak detection soft sensor
On May 13, 2018 at 18:30, the mean centered differential water flow deviated and remained between 0.2 and 0.4 gpm. The
median centered differential flow deviated to 0.4 gpm on May 13, 2018 at 21:00 confirming the occurrence of a leak per the

526 © 2019 by the Association for Iron & Steel Technology.

soft sensor logic. The subsequent return of the median value to baseline while the mean differential flow continued at 0.2-0.4
gpm indicated that the leak rate was relatively constant.
An initial gas test did not confirm a water leak on tuyere 7. On May 16, 2018 at 11:30, a second gas test confirmed that
tuyere 7 was leaking. Being the first leak event detected using the soft sensor, there was some uncertainty that the observed
deviation of 0.2-0.4 gpm was indeed a leak and not an instrumentation error. The team was pleasantly surprised to see the
small leak when tuyere 7 was removed, see Figure 10.

Leak

Figure 10. A small water leak was found when tuyere 7 was removed on May 16, 2018
The response time to leak events detected by the soft sensor decreased as more experience and confidence was gained. An
alarm threshold of 0.3-0.4 gpm was established to signify a water leak event. Figure 11 details tuyere water leak events on
tuyeres 11 and 30 that were successfully detected using the soft sensor. In both cases, the water leaks were verified by the
USS BF14 team via gas testing, and the response time to replace each tuyere was less than 8 hours.

Figure 11. Water leak events on tuyere 11 and 30 that were detected by deviations in the mean and
median centered differential flow signals

© 2019 by the Association for Iron & Steel Technology. 527

A total of seventeen (17) tuyere leaks were successfully detected in 2018 after the soft sensor was implemented, see Figure
12. No tuyere leak events were missed by the soft sensor during this period.

Figure 12. Tuyere nose leak events detected using the soft sensor from May 1 to December 31, 2018
The response time was defined as the time when a leak was confirmed by the soft sensor to the time BF14 was stopped to
change the tuyere out. The response time clearly decreased during 2018 as more leak events were captured. Of note is that 11
of 17 tuyeres (65%) experienced small leaks, < 1.0 gpm before being replaced. With confidence that a leak rate was small,
the BF14 management had a few hours to plan a tuyere change out without needing to rapidly shut BF14 down.
On July 17, 2018, the soft sensor detected a water leak on tuyere 1 and BF14 was shutdown to replace the tuyere even though
the gas test did not confirm the leak. When the tuyere was removed and a water hose inserted, a leak was not found even
though there was iron penetration on the tuyere nose. The tuyere casting was sectioned to see if a leak was present that may
have closed when the tuyere cooled down. Surprisingly, a crack was found between the body and the nose cooling water
channels that could allow high pressure nose circuit water to leak into the lower pressure body circuit, see Figure 13.

Figure 13. A crack found between the body and the nose circuit of tuyere 1, July 17, 2018
This was the only time where a leak was indicated by the soft sensor and not confirmed by the CO gas test. The presence of
the internal crack between the high and low-pressure water circuits indicated that the soft sensor did detect water leakage
between chambers. Following this, USS decided that all potential water leaks detected by the soft sensor must be confirmed
by a gas test prior to shutting BF14 down to replace a tuyere.
The soft sensor provides early warning of a water leak and demonstrated that small leaks, 0.3-1.0 gpm, could be identified
with confidence. The system eliminated the need to do many gas tests on the 34 tuyere discharges as the leaking tuyere
circuit was precisely identified. Tuyere replacement planning could be completed while the leak was small and before the
BF14 process parameters were impacted.

528 © 2019 by the Association for Iron & Steel Technology.

 CONCLUSIONS
A tuyere leak detection soft sensor was successfully configured and implemented at U.S. Steel Gary Works Blast Furnace
No. 14. The soft sensor identified small water leaks at the capability limit of the existing flow meters by using mean and
median centering of the differential water flow data that was readily available. Since May 2018, all leaking tuyeres were
identified by the soft sensor and 65% were found when the water leak rate was low, < 1.0 gpm. Gas testing time to confirm a
water leak was greatly reduced as the soft sensor definitively identified the leaking water circuit that needed to be tested. A
planned stop could be implemented while the leak rate was small and before the Blast Furnace No. 14 process was impacted.

ACKNOWLEDGEMENTS
The authors thank U. S. Steel Corporation for permission to publish this paper. We also acknowledge Hatch’s Jakob Janzen
and Rishi Midha for their help with the early investigations in 2017 when the soft sensor concept was in its initial stages of
development.

DISCLAIMER
The material in this paper is intended for general information only. Any use of this material in relation to any specific
application should be based on independent examination and verification of its unrestricted availability for such use and a
determination of suitability for the application by professionally qualified personnel. No license under any patents or other
proprietary interests is implied by the publication of this paper. Those making use of or relying upon the material assume all
risks and liability arising from such use or reliance.

REFERENCES
1. J. Entwistle et al., “Hearth Temperature Control at USS Blast Furnace No. 14,” AISTech 2019, Pittsburgh, PA, USA
2. ‘2018 AIST Blast Furnace Roundup’, Iron & Steel Technology, AIST, March 2018, pp. 270-271
3. T. Gerritsen, W. Braun, P. Gebski and A. Sadri, “Monitoring of Copper Cooling Components in Furnaces Using Leak
Detection and NDT Methods,” CINDE Journal, Vol. 37, No. 1, 2016, pp. 32-38
4. W. Braun, J. Janzen and B. Bussell, “Automatic Pressure Testing of Water Cooled Components in Furnaces,” EMC
2013
5. A. Agrawal, S. C. Kor, A. R. Choudhary, M. K. Agarwal, S. Kundu, and V. R. Tripathi. "A novel method to detect the
water leakage from tuyere nose cooling circuit in blast furnace." Ironmaking & Steelmaking, 43, No. 10, 2016, pp.
744-751

© 2019 by the Association for Iron & Steel Technology. 529