D. R. Jensen, N. A. HATCHER and M. D. BAILEY, Optimized Gas Treating Inc., Buda, Texas; and M. COADY, Delta Controls Corp., Shreveport, Louisiana
Of all the fundamental measured process variables, temperature is perhaps the most reliable. In most process applications, temperature is measured by a thermocouple, which consists of a relatively simple junction between two dissimilar metals. The junction produces a small voltage proportional to the difference in temperature between the junction where the temperature is being measured and the point where the reference temperature is being imposed. Thermocouples are so inexpensive, accurate and abundant that their use is normal in almost any application. However, the thermal reactor of a Claus sulfur recovery unit (SRU) is not a typical application.
The thermal reactor is characterized by extremely high temperatures along with the presence of various sulfur compounds that can react with many metals and cause corrosion. The metals used in most thermocouples are particularly vulnerable to sulfidic corrosion, quickly rendering them unusable. In fact, the useful life of a typical thermocouple at thermal reactor conditions is generally measured in minutes or hours, making their long-term use in this application impossible. However, temperature is still a critical process parameter to monitor in Claus SRU thermal reactors.
The refractory lining in a thermal reactor will fail if overheated (FIG. 1). Additionally, certain contaminant species [ammonia, and benzene, toluene, ethylbenzene and xylenes (BTEX)] can cause plugging/sooting problems downstream in the SRU if sufficient temperature is not maintained. These compounds require temperatures at the higher end of the normal operating range of 1,150°C–1,260°C (2,100°F–2,300°F) to be destroyed in the furnace. Therefore, special technologies and strategies are needed to measure and confirm the temperature in the thermal reactor.
This article discusses in detail the importance of temperature to the thermal reactor refractory lining; details available options for field temperature measurement along with their drawbacks and limitations; and introduces a new semi-empirical tool developed by the authors’ company to quickly, efficiently and accurately determine thermal reactor temperature based on process parameters. Finally, two case studies are presented in which the new tool is used to estimate thermal reactor temperature in an operating facility before, during and after a process upset, and the estimates are compared with operating data.
Thermal reactor refractory lining. Thermal reactor temperatures can readily reach the 1,500°C–1,650°C (2,732°F–3,000°F) range, which is near or above the melting point for most carbon steels. Also, well before the melting point is reached, steel will enter a region of excessive corrosion due to the high hydrogen sulfide (H2S) content of the process gas. Ceramic refractory linings are therefore required to prevent equipment damage due to both these effects. These linings are engineered to maintain the steel shell in a temperature range of 200°C–345°C (392°F–653°F) to maximize equipment life.
Although resilient, the thermal reactor refractory cannot withstand the most severe conditions that can exist in the reactor. Excessive heating can cause phase changes in the refractory that make it less strong and, in the case of monolithic (castable) refractory, melt the stainless-steel anchors providing support to the lining. This ultimately leads to failure of both the lining and eventually the steel shell. Given extreme overheating of ~1,875ׄ°C (~3,407°F), the refractory itself can melt and flow like lava. While atypical, it is possible to achieve these temperatures if near-stoichiometric combustion (of high H2S content acid gas or fuel gas) is sustained for significant periods of time without appropriate mitigation (e.g., quench gas/steam injection). Extreme overheating can also be a side effect of the contamination of process gas with heavy hydrocarbons, especially in the case of an oxygen-enriched Claus process.
Unfortunately, the dangers do not lie only at the very top of the temperature operating range. If the thermal reactor skin temperature drops too low [190°C (375°F)], then it is possible for sulfuric and sulfurous acid to condense, quickly corroding the carbon-steel shell of most thermal reactors and leading to rapid equipment failure. Operability and reliability concerns also exist with poor contaminant destruction at low process gas temperature. So, it is important for the responsible maintenance of equipment to know the thermal reactor temperature and be able to measure—or at least infer—the steel skin temperature.
Temperature measurement devices. Since, as discussed above, standard thermocouples are not an option for temperature measurement in the thermal reactor, what options are available for this service? Two conventional choices—optical pyrometers, and ceramic-shielded, gas-purged thermocouples—are available. Each choice has benefits and drawbacks to its use.
Optical pyrometers, like that shown in FIG. 2, are devices that are positioned to observe the radiation emitted from an object, and then analyze the intensity of that radiation to infer the temperature of the object using the Stefan-Boltzman law. In the case of thermal reactors, the optical pyrometer is typically set up to observe infrared band electromagnetic (EM) radiation over a target area, which is usually a section of refractory opposite the pyrometer in the thermal reactor. The choice of specific frequencies is important because gases in the furnace can absorb some EM radiation and therefore affect the resulting indicated temperature. Careful choice of frequencies, along with certain compositional requirements, can allow the pyrometer to indicate the approximate gas temperature in the thermal reactor, as opposed to the refractory temperature. This can seem useful since the large mass of the refractory acts as a thermal capacitor and keeps operators from seeing the full effect of process changes over a reasonable time scale. However, because the refractory brick, in general, has low emissivity and high reflectivity, the temperature indicated when in refractory mode is more representative of average internal temperature. Therefore, it is often recommended to use the refractory temperature mode unless other concerns dictate the use of the gas temperature mode.
There are a few drawbacks to using an optical pyrometer to measure thermal reactor temperature. The first is that because all emitted radiation that is “viewed” by the pyrometer is used to infer temperature; the device naturally indicates an average temperature along the view path of the pyrometer. While useful, the average temperature can disguise the actual maximum temperature, which might be a point of concern when operating a unit on the edge of its temperature envelope.
Another drawback is that the pyrometer can only give information about what the radiation detector can “see.” This has a couple of practical implications. First, if there is an obstruction in the field of view, the obstruction’s temperature will be averaged into the calculation, even if the object is at ambient temperature. This can lead to a far cooler observed vs. actual temperature reading. Common causes of this type of error involve the misalignment of the pyrometer where part of the nitrogen-purged nozzle is in the pyrometers’ view path. In the most extreme case, the temperature error can be due to the lens of the pyrometer being completely occluded with sulfur or soot, leading to a false indication of unrealistically cool temperatures. The use of a two-color pyrometer system can help alleviate the problem of lens occlusion since it uses the ratio of two different wavelengths to infer temperature. However, operating a two-color pyrometer in this way usually means sacrificing the ability to measure both gas and refractory temperature simultaneously.
Second, the pyrometer can only give information about the temperature in the region it can see. If the burner is not providing good mixing, or if any other effect creates hot and/or cold spots in the thermal reactor, then the pyrometer will only provide information to the extent that the spot occurs inside the device’s view.
Finally, the configuration and calibration of the pyrometer can be tricky. Any pyrometer that attempts to measure gas temperature must assume a water concentration. This is often problematic as the actual water concentration changes with the process and ambient conditions. Also, to indicate the correct temperature, the effective emissivity of the system must be known. Determining this quantity is not a trivial exercise because the effective emissivity varies with chosen wavelength, furnace gas composition and refractory surface conditions. To approximate, a thermocouple is often inserted into the pyrometer nozzle, a reading is taken, then the emissivity is adjusted to make the pyrometer-indicated temperature match the thermocouple.
This approach has many sources of error. For example, pyrometer nozzles are purged with nitrogen (or less commonly air) at ambient temperature with the purge gas exiting at the nozzle right where the thermocouple is taking its reference measurement. Even if neglecting the purge, the thermocouple is taking a point measurement of the gas temperature very near the point the pyrometer nozzle joins the thermal reactor shell. This is not the average temperature across the entire line of sight of the pyrometer. Therefore, it is common for there to be disagreement between theoretical calculations of temperature and indicated values of 1.5%–2% of the indicated reading.
The other common option for thermal reactor temperature measurement is a purged thermocouple (FIG. 3). Since typical thermocouple/thermowell systems are not resilient enough for the thermal reactor operating atmosphere, the industry has developed thermal measurement systems consisting of a ceramic thermowell (which can survive the thermal reactor atmosphere) with an integral inert gas purge. This serves to protect the thermocouple junction, which is highly susceptible to corrosion upon exposure to even small amounts of thermal reactor process gases.
However, this method of temperature measurement also has its own set of drawbacks. First, the installation of the ceramic thermowell is a delicate and complicated process. Usually, a hole must be drilled for the thermowell through the completed furnace refractory lining without compromising its integrity. Then, the thermowell must be installed and affixed to the thermal reactor without damaging the ceramic thermowell itself, which is not as robust or forgiving of rough handling as a typical metallic material would be. Also, compared to typical thermowell materials, the ceramic thermowell has a relatively high heat transfer resistance. So, this means there is some delay between changes in the thermal reactor temperature and being able to observe the temperature change. Finally, thermocouples are a point measurement of temperature precisely at the hot junction, which is located (if installed correctly) at the hot face of the refractory brick. As with the optical pyrometer, the thermocouple will only be able to sense a temperature anomaly if it occurs at the precise location at which the thermocouple is installed.
Recently, the co-author’s company has offered a specially designed, unpurged thermocouple for use in SRU thermal reactors, which utilizes a proprietary set of featuresa (primary silicon carbide thermowell, secondary monocrystalline sapphire thermowell, along with other proprietary sealing technologies). The combination of features allows for thermocouple operation without access to a source of purge gas. In general, purged thermocouples are still preferred, but there is an alternative for installations unable to accommodate a purged thermocouple’s installation requirements.
Calculated temperatures. Given the multiple failure modes and sources of error present for each of the thermal reactor temperature measurement technologies, it is prudent to check the indicated values. The best check would be to completely model the chemistry of the thermal reactor and, as a result, determine the temperature. The model would only depend on various commonly monitored process parameters (compositions, flows, pressures and temperatures). Several process simulators are commercially available on the market that can claim to provide a solution that accomplishes the calculations described.
However, even the best simulators available only approximate reality. There are various assumptions implicit in any simulation model that impact its reliability and accuracy. For example, virtually all models assume that good mixing is taking place regardless of reality. Model calculations are also subject to plant metering and laboratory errors that affect the results of modeling. All these factors must be kept in mind when utilizing any computationally derived temperature values.
The various simulation models also differ in their approach and rigor. In the simplest case, thermal reactor chemical reactions are modeled as simple conversion reactions (either from correlation or specified). Slightly more sophisticated and flexible models might use a method to determine the complex chemical equilibrium of the various reactions of interest (e.g., Gibbs free energy minimization). Still, others may use rate-based chemical reaction kinetics to determine the conversion of reacting species as a function of reactor conditions and time spent in the reactor. Such kinetic models have seen significant steps forward in the last few years with a better understanding of and a model for complex side reactions in the thermal reactor (e.g., direct H2S cracking and recombination). Some hybrid combination of two or more of the above methods might also be used, depending on data availability and modeler’s preference.
Obviously, more rigorous methods are expected to deliver more accurate and reliable results. Unfortunately, more rigorous models often require more data and can be cumbersome to set up for real-time monitoring. Additionally, the computational cost for each of the methods also goes up with the level of rigor. For a single simulation run, the difference in time spent waiting for the model to calculate is negligible. However, when expanded over many data points, this multiplied differential computational time can become significant. For example, with a dataset of 2,000 points a difference in run time of only 5 sec per run leads to a difference of 2.75 hr. This is a long time to sit and wait for a computer.
Real-time RF monitoring toolb. Realizing the need for a straight-forward, fast and flexible thermal reactor temperature estimation tool, the author’s company has developed a focused Excel-based modelb (FIG. 4) to provide real-time feedback to plant operations. This tool uses field-measured temperatures, pressures and % of water saturation of the amine acid gas, sour water stripper gas, natural/fuel gas (for startup or co-firing), diluent/quench gas [nitrogen (N2), steam, etc.], ambient air and supplemental oxygen, along with assumed dry-basis compositions.
The Excel modelb was constructed using hundreds of thermal reactor simulations using the rate-based kinetic model implemented in the authors’ company’s commercial sulfur plant simulatorc. Through the clever choice of independent variables, the Excel modelb attempts to correctly capture some of the finer points of determining the thermal reactor temperature. These include reactor feed stream superheating correction (for pre-heated feeds), water saturation effects for the acid gases, and providing a means to identify and correct questionable feed compositions based on air demand analyzer reading.
Because the industry is generally concerned with a furnace operating at the top of the safe operating window [980°C–1,650°C (1,800°F–3,000°F)], particular care was taken to improve the fit of the model in that range. TABLE 1 shows other limits of the correlation data range used to generate the model. Concentration percentages given are all on a molar basis.
Data outside these ranges will not necessarily generate incorrect results, but no information outside this range was used to correlate the data so the model fit might give unexpected or incorrect results. The focus of the model was not necessarily to develop extremely high precision with an exact match to simulated results, but to develop one that produced values with accuracy equivalent to field instrumentation (typically ±1% of value).
The following set of plots in FIGS. 5–7 display data for three different scenarios, each of which demonstrates the model’s performance for different operating conditions.
FIG. 5 shows that as hydrocarbon content of the combined Thermal reactor feeds goes up, the modeled temperature difference (dashed line) from the no hydrocarbon case temperature has a good fit to the simulator-generated data points. This shows that the real-time model explains 99.97% of the simulator Thermal reactor temperature data variation with feed hydrocarbon contents of up to 4 mole% (as C1 equivalent).
FIG. 6 shows how the ratio of thermal reactor ΔT varies with the molar flow ratio for an oxygen (O2)-enriched air inlet ratioed to a base ambient air case. As the molar flow ratio drops, the oxygen enrichment percentage goes up, and the temperature should rise accordingly. This shows that the Excel modelb handles a wide range of oxygen concentrations and is suitable for typical oxygen-enriched conditions. The real-time modelb accounts for 99.91% of the simulator data variation.
FIG. 7 shows the differences between simulation values and real-time model values for various percentages of stoichiometric air and different feed hydrocarbon content. One expected effect (given FIG. 5) is that hydrocarbon content has relatively little impact on the observed error. In fact, at this scale it is difficult for most of the points to distinguish between the different values of hydrocarbon content. The error vs. stoichiometric ratio (with 1 being airflow related to a perfect 2:1 H2S:SO2 ratio) shows that the further away from the stoichiometric point the base case model is pushed, the more error can be expected. However, all well-operated plants will operate near the stoichiometric point even when deliberately raising the ratio of H2S:SO2 to a value greater than 2:1 to protect the overall sulfur train from upsets. This is because the air demand analyzers typically used loosely indicate full scale (< 5%–> 5%) over approximately 10% of the inlet air flow. The plot above not only covers all normal operating cases with a good match to simulated data, but it also acceptably covers a very wide range that would likely correspond with most plant upsets.
Field case studies. The authors’ company had the opportunity to test the real-time modelb against two actual operating plants, complete with times of significant upset. Both units are refinery-style sulfur units. For the first data set, historian data were provided for essentially an entire year of operation and the unit was a typical SRU operating at low flow.
There are several notable items in the plot of FIG. 8 (Case study 1). First, the tool tracks well while indicating a ~150°F higher temperature than the field optical pyrometer. This is mostly likely due to the very low rates being processed by the unit during this time, leading to significant heat loss in the thermal reactor. This heat loss is not an effect presently integrated into the real-time model. Another interesting point is that the model tends to overshoot the measurements during excursions (both high and low). This is expected since the model calculates temperature changes immediately, as soon as the input compositions, conditions or flowrates change. In an actual thermal reactor, there is a great deal of mass that provides thermal capacitance and, therefore, resists a sudden change in temperature. Finally, in the last month or so of operation, a sudden threshold shift in the pyrometer temperature reading was observed while none of the other temperature measurements nor the real-time modelb show a similar shift. While in this case the thermocouples also point out the fact that the optical pyrometer is having issues, the real-time modelb could be used alone to call into question the sudden shift.
The second plot (Case study 2 in FIG. 9) shows operating data for an oxygen-enriched SRU. The data set covers 1.5 d of operations before, during and in the recovery from a low air flow excursion. The low flow excursion was caused by a loss of indication for one of the feed meters, causing feed forward air control to cut air flow unnecessarily.
This plot shows some of the same features as the previous plot—in particular, the overshooting of temperatures during excursions is still observed and for the same reason. Once again, a constant temperature is offset with the real-time model reading ~150°F hotter than the other temperature indications. In this case, the offset may be the result of a recycle flow (which is a feature of some oxygen-enriched process) for which no flowrate data are available. The recycle could be quenching the temperature from a value nearer the real-time modelb. Also, just as before, there could be unaccounted-for heat loss that possibly leads to a lower real-time model calculated temperature, and a better match with the field data. Whatever the case, the real-time model is within 4% of the thermocouple indicated temperatures without any corrections.
Takeaways. Thermal reactor temperature measurement is a complicated endeavor. Multiple options are available, and all of them are customized/specialized for use in an SRU. Unfortunately, all the technologies have drawbacks, which can lead to incorrect indication. The authors’ company has developed a tool to provide real-time temperature estimation to quickly, efficiently and accurately double check and diagnose other measurement technologies. The modelb has been verified against both a state-of-the-art simulation model and actual plant data from a field study consisting of two SRUs in different facilities and different configurations. This comparison shows that the real-time temperature modelb accurately and efficiently estimates thermal reactor temperature using only the process variables commonly monitored in operating facilities. This is a valued addition to the industry because it can serve as a double check to field instruments to protect equipment and can be used in the analysis of past plant performance data and planned future operations. HP
NOTES
a Delta Controls Corp.’s QSeal®
b Optimized Gas Treating’s ProBot™
c Optimized Gas Treating’s OGT|SulphurPro®
DARYL JENSEN joined OGT as a Development Engineer in 2021. He spent the previous 15 yr with Ortloff Engineers (now, Honeywell UOP), during which his responsibilities included process design/development, simulation, basic engineering design package project execution, process control design, and onsite startup/troubleshooting for gas processing and treating technologies, including sulfur recovery (refinery and gas plant), cryogenic NGL/LPG recovery, acid gas treating and sour water stripping. Jensen holds BS and MS degrees in electrical engineering from New Mexico State University and is a registered Professional Engineer in Texas.
NATE HATCHER joined OGT in 2009. For most of his career, he has been involved with various facets of gas treating or sulfur recovery. His career began with Black & Veatch Pritchard Inc. and then 7 yr with ConocoPhillips Inc. as Treating and Sulfur Processing Best Practices Network Lead. Hatcher worked briefly with Trimeric Corp. providing design and consulting services to the oil and gas sector. In 2022, he joined Phillips 66 as Director, Sulphur Processing. He holds a BS degree in chemical engineering from the University of Kansas, is a registered Professional Engineer in the state of Kansas, and is a member of the industry Amine Best Practices Group.
MATT BAILEY joined Optimized Gas Treating Inc. in business development in 2014, where he provides sales and marketing support. Bailey has 20 yr of industry experience in technical sales, creating safe engineered solutions for refineries, chemical plants and engineering companies. He has additional experience in relief valve studies with hydrocarbon units for Murphy Oil and Motiva. Safety has been a key component in driving his career. Bailey licensed with the Texas Board of Professional Engineers, and holds a BS degree in chemical engineering from Texas A&M University and a MBA from the University of Houston.
MATT COADY received a BS degree from Louisiana State University in 2012, after which he joined Delta Controls in technical sales, and was promoted to General Manager in 2019. Coady joined Raymond James in 2023.