Engineers at the Idaho National Engineering and Environmental Laboratory (INEEL) used computational fluid dynamics (CFD) analysis to determine the cause of a catastrophic failure in a recuperative flameless thermal oxidizer (RFTO), used for the destruction of hazardous compounds. When the unit's feed tubes melted, resulting in system failure, it was unclear whether the cause was a design flaw or operator error. Physical examination of the unit could not conclusively resolve the issue, so INEEL engineers were asked to recreate the failure scenario as a computer simulation. Working from data in the operation logs, they analyzed the last run of the oxidizer using CFD. The results clearly showed why the tubes melted. "The design of this particular model was such that there were recirculation eddies in the inlet plenum," explains Mike Rynearson, a Principal Engineer at INEEL. "These eddies caused below normal air flow into one of the feed tubes. That, combined with higher than normal propane levels at startup and the hot soak bed condition, caused the oxidation front to move into the low flow tube, causing it to overheat and fail. Once it melted, the others quickly followed.

Costly Tube Failure

The recuperative flameless thermal oxidizer under investigation was used to remediate volatile organic compounds (VOCs) contamination in the vadose zone. The system extracted organic contaminants from soil under vacuum using a positive displacement blower. Contaminant vapor was then fed to the inlet of the oxidizer vessel under positive pressure as shown in Figure 1. The gaseous material entered a plenum it through a side-mounted inlet pipe near the base of the vessel. From the plenum, feed flow was diverted upward through twelve vertically mounted tubes packed with low void fraction ceramic balls in the bottom section and high void fraction saddles in the upper section. At the top of the tubes, the flow exited, reversed direction, and passed through the ceramic saddle packed oxidation chamber surrounding the tubes before being exhausted through a side-mounted pipe opposite the inlet. When the unit was working to specification, the combustion front was maintained in the upper section of the oxidation chamber, outside of the feed tubes. Heat flux from the oxidation reaction through the tube walls was imparted to the inlet gas effecting feed preheat and heat recovery. The process oxidized more than 99.5 percent of hazardous VOCs before releasing exhaust to the atmosphere. However, the operator occasionally experienced out of specification operation, and three times the unit failed catastrophically. "On those occasions, the heat exchanger tubes melted and had to be rebuilt," explains Thomas Foust, Advisory Engineer at INEEL.

The second time the heat exchanger tubes melted, INEEL performed a detailed failure analysis to determine the root cause of the failure, because it was unclear whether the tube failures were caused by operating problems, design problems, or a combination of both. One characteristic of this oxidizer was that it had a very narrow window of acceptable operating conditions. Propane feed was required between a lower limit necessary to sustain the oxidization reaction and an upper limit for safety reasons. This translated to operating propane feeds between 30 and 50 percent of the lower flammability limit (LFL) of propane in air. It was theorized that if propane levels exceeded this limit, the oxidation front, which was supposed to be localized downstream of the tubes in the outer packed bed region, could enter the tubes, rapidly causing an over-temperature condition and tube damage. If the oxidation front entered the top of the tube, melting would be observed in that area.

Examination of the operator logs following the third tube failure did not indicate excessively high propane feed levels. In addition, tube damage occurred in the lower rather than the upper portion of the tubes, inconsistent with the expected result of excessive propane feed.

It was not feasible to study the failure on a physical model due to cost and instrumentation limitations. INEEL used CFD analysis instead to simulate the operation of the unit. A CFD simulation provides fluid velocity, temperature, and species concentration values throughout the solution domain for problems with complex geometries and boundary conditions. As part of the analysis, a researcher may change the geometry of the system or the boundary conditions such as inlet velocity, and view the effect on fluid flow patterns, temperatures, or concentration distributions. INEEL engineers used the FLUENT CFD program from Fluent Incorporated, Lebanon, New Hampshire. "We use FLUENT because the program is user-friendly. Also, FLUENT has a well respected combustion model that also handles oxidation well," says Foust.

Failure Scenario Reconstruction

The engineers started the investigation by simulating the inlet plenum in the lower part of the oxidizer vessel, focusing on the airflow from there into the feed tubes. They wanted to learn whether or not there was an even distribution of feed into the 12 tubes. "Uneven flow would be a problem in this system because there was a definitive specification that the flow had to exceed 16 CFM for each tube," explains Foust. "If the flow rate dropped below that, propane would remain in the tube longer. That would raise the heat flux, ultimately allowing the combustion front to back into the feed tubes. The manufacturer assumed even flow into each tube, but development of turbulence and secondary eddies was likely as the incoming gas flowed from the inlet pipe into the large open plenum. We didn't feel it was safe to assume uniform flow."

To begin the analysis, the geometry of the plenum and the openings to the heat exchanger tubes were modeled using Fluent's preprocessor, GAMBIT. After the geometry was created, the preprocessor generated an unstructured analysis mesh. The non-uniform mesh was constructed to account for differences in the size of the components being analyzed. For example, a very fine mesh was created around small, critical areas (such as the tubes) and was progressively coarsened to regions where the components were larger (such as the plenum).

The next step was specifying boundary conditions and solving the problem. A mass flow rate was specified at the inlet boundary where the propane/air mixture entered the oxidizer inlet pipe. The temperatures of the propane and air were determined from physical tests. Foust used the k-e turbulence model initially, and subsequently switched to the Reynolds stress model. The Reynolds stress model is better at picking up secondary flows such as eddies, but requires longer computation times than the k-e model . The results of the simulation showed that the airflow into the tubes was not evenly distributed. A plot of velocity vectors (Figure 2) traced the cause to a 90-degree bend in the inlet pipe. When the gas/air mixture hit the bend, its velocity was too high to make the turn uniformly. Some feed hit the wall at an angle, causing recirculation eddies that resulted in uneven flow into the heat exchanger tubes. Flow into one tube was clearly below the manufacturer's recommended 16 CFM minimum flow rate.

The next simulation focused on the upper part of the unit, and was designed to observe heat transfer in the tubes. Foust again created the geometry and mesh in GAMBIT, using a near-wall meshing technique for the heat exchanger tubes which allowed more accurate resolution of heat transfer through the tube walls than a wall function approach.

Because the oxidation vessel is a packed bed chamber, filled with non-catalytic, porous, inert ceramic material, Foust was able to use the porous media model in FLUENT for simulation. The tubes also contained this ceramic material, but in two different shapes: spheres in the bottom one-third of the tubes and saddles in the top two-thirds. Based on the void fractions for the regions filled with spheres and saddles, the Ergun equation was used to compute loss coefficients for use in the porous media model. Airflow results from the previous inlet plenum simulation provided the inlet boundary conditions. FLUENT's premixed combustion model and the k-e model for turbulence were used. "We didn't expect much secondary flow inside the tubes so the k- e model, which is less computationally intensive, was used," he says.

Foust ran multiple CFD analyses following various startup scenarios represented in the operator logs. "The logs showed a normal temperature profile in the oxidizer for a month after the first rebuild," Foust explains. "Then, for training purposes, there had been a shutdown and immediate system restart after which they were unable to attain a normal operating profile. I looked at the logs for the day of the training activities, constructed startup and operating scenarios, and ran the CFD model to see if any of those conditions could lead to the failure." The startup procedure required higher than normal propane feed, about 60 percent of LFL, for a few minutes and, due to the shutdown followed by immediate restart, the bed was in a hot soak condition. The operator noted in his logs that normal operating parameters were difficult to maintain after this startup with normal propane feed rates indicating that tube damage had occurred.

When Foust ran the CFD analysis with propane at 60 percent of LEL and the oxidizer bed in a hot soak condition to approximate startup conditions, the results showed that oxidation front established in the low flow tubes. "The oxidation front established itself in the low flow tube at the transition between the spheres and the saddles indicated by the distinct change in temperature at that location (Figure 3)," Foust explains. "That made sense because there was a drop off in velocity at that spot as air moved into the region containing the saddle-shaped material, which had a higher void fraction than the spherical material." Since the tubes were thin-walled, melting and tube failure occurred very quickly at the combustion front, then spread to the other tubes. With melting and failure of the wall in the first tube; feed flow short-circuited to the exhaust, a preferential flow path developed, and flow in the remaining tubes decreased with the end result of damage to the remaining tubes. "The CFD analysis was consistent with the physical evidence,” says Foust. "From the CFD simulation, we were able to determine the cause: uneven airflow through the plenum caused the flow rate into one tube to be too low. When propane levels were above the LFL range on startup, and the packing was in a hot soak condition, the oxidation front established in the tube, the tube overheated and the wall failed. The INEEL operated three RFTO units, two of which had encountered no operational problems. The two units, which remain in operation, were of a similar design to that of the failed unit, except 20 feed tubes were used instead of 12. The 20 tube units did not suffer from the recirculation eddy problem to the same extent as the 12 tube model. Thus, variations in feed flow to the tubes were less pronounced and below design specification flow in the tubes was not encountered.

These units were shut down after the second failure of the 12-tube unit. "The CFD analysis determined that the cause was not operator error, but rather the result of a critical design flaw in the 12 tube unit inlet plenum," says Rynearson. Following modification to the operating procedure to eliminate system restarts from a hot soak condition, the 20 tube units were restarted and are operating to date. In addition to answering the question of why the 12 tube oxidizer failed, CFD analysis allowed INEEL engineers to propose simple design changes that would prevent the problem from happening in the future. "CFD clearly showed what happened and why, and helped us settle an issue that physical testing could not resolve," Rynearson adds.