Guidelines for accurate cleanroom CFD modeling
All relevant physics and heat sources must be considered
By Byron Blackmore, thermal and applications engineer, and Weiran Xu, thermal engineer and support supervisor, at Flomerics
Computational fluid dynamics (CFD) has proven its value and its validity in simulating cleanrooms. However, the user’s engineering judgment is still crucial to a successful CFD modeling job. When applying CFD software, caution must be taken to include all the relevant physics in the model. For example, in a cleanroom application, diffusers and major flow obstacles must be represented. In addition, heat sources such as heat generated by human occupants and equipment must also be included to produce an accurate solution. To ensure accurate results, general guidelines for simulating cleanroom design, some of which are listed below, should be followed.
The “building envelope” can be thought of as everything that separates the interior of a building from the outdoor environment, including the windows, walls, foundation, basement slab, ceiling, roof, and insulation. In the context of the CFD model, consideration of the envelope-in particular, what aspects of the cleanroom envelope are to be included in the model-is an important first step in the creation of the model. The volume considered can be termed in CFD as the solution “domain”-namely, the volume of the cleanroom in which the velocity, temperature, and so forth are to be calculated. The extent of the domain is determined by the objectives the person is trying to achieve with the CFD results, as well as the practical consideration of the available data.
It is recommended in the setting of the domain that the solid-surface aspects of the building envelope are used to construct the extent of the domain. The reason for this is that the setting of CFD boundary conditions is typically more easily applied to solid surfaces than to fluid planes or surfaces where there can be wide variation of conditions. In the case of the solid surfaces, the variation of conditions is usually much smaller. This results from the fact that cleanrooms are typically located internal to the overall building structure and are not subject to variations that are present from external conditions-for example, solar loading.
Figure 1: This graphics illustrates the different cleanroom scenarios that can be considered in a CFD model. Image courtesy of Flomerics.
The choice of domain can be explained more clearly in the following example. Let us consider a simplified representation of a cleanroom with a fully active ceiling, a raised-floor plenum, and a return exhaust duct, as shown in Figure 1. There are three obvious cleanroom scenarios to consider as CFD models. These different scenarios are shown by the dashed line boxes in the graphic.
In Scenario 1, the extent of the domain is from the ceiling of the cleanroom to the bottom of the floor slab. The main assumption here is that the flow conditions over the ceiling can be considered uniform, namely, the variation of the flows through the HEPA filters in the ceiling is ignored, and a known, constant flow rate is defined at each of the HEPA-filter locations. In the floor plenum, the returns are either defined with a constant exhaust flow rate or present a given resistance to the airflow. In the latter case, a pressure boundary condition also has to be set to represent the conditions beyond the exhaust and would be the preferred option if the flow rate through the returns was unknown. This is probably the most common domain that would be set up for a CFD model of a cleanroom. The typical objective of this type of model is to determine the variation of the airflow, temperature, pressure, and so forth in the cleanroom area itself.
In Scenario 2, the extent of the domain is from the ceiling of the plenum to the bottom of the floor slab. In this case, the definition of the flow conditions is made at the supply or supplies into the ceiling plenum. Here, the model would also yield the variation of the flow rates through the HEPA filters, which would be represented as appropriate resistances to airflow. Such a model could be used to determine what kind of adjustments would have to be made to the HEPA-filter resistances in order to equalize the flow through them for creating the desired ceiling supply conditions.
In Scenario 3, the extent of the domain also includes the return duct connecting the floor and ceiling plenums. In this case, the supply or supplies feeding into the ceiling plenum, or exhaust or exhausts returning out of the floor plenum, would be defined as variable flow fans. The other set of grilles would be defined as appropriate resistances to airflow. Here, the model would also yield the variation of the flow conditions entering and exiting the floor and ceiling plenums. Such a model could be used to determine what kind of adjustments would have to be made to the system to ensure the correct distribution of air in the system.
Process equipment and operators
With the overall solution domain of the CFD model set, the focus can then be turned to the objects that populate the cleanroom. One form of model to be considered is one in which nothing populates the cleanroom or clean corridor, and the system is balanced as an empty unit. However, the reality is that there are usually significant amounts of equipment and operators in these clean spaces, thereby ensuring that the conditions in the room are significantly different from the empty-space scenario.
The issue of equipment and operators must, therefore, be addressed in most cases. To include every object in detail is not feasible from the viewpoint of computational tractability, nor is it necessary. In reality, the only objects that need to be included are those that present a significant blockage to airflow or those that dissipate heat, as these are the items which will affect the desired laminar flow conditions within the cleanroom. To illustrate this point, the following section will consider a typical CFD cleanroom model from the viewpoint of objects.
Figures 2A and 2B show a vial filling machine in a cleanroom, with the protective curtain tent both shown and hidden. All the objects that would be expected to provide either significant blockage to airflow or are sources of heat generation have been added. Items such as chains and wires are not generally included, as the computational overhead in terms of a grid would be prohibitive. Generally, the items of blockage are considered statically in their worst mechanical position-for example, directly over a critical process.
If there is significant movement of objects, there are two options in terms of their modeling:
Figure 2A: This drawing shows a typical vial filling machine with the protective curtain shown. Image courtesy of Wyeth.
■ The object is considered statically in the worst positions in separate CFD calculations.
■ A moving boundary condition is applied to the object; in particular, a velocity component is attached to the object.
In terms of the addition of occupants (or, in this case, operators), there are two potential approaches:
■ The operators are modeled relatively simplistically; in particular, only blocks representing the head and torso are modeled, or even just the heat source associated with the operators.
■ A more complicated representation is used, as demonstrated in Figure 2B.
Figure 2B: This drawing shows a typical vial filling machine with protective the curtain hidden. Image courtesy of Wyeth.
The use of the latter representation is appropriate if the operator is going to be close to a critical process.
The representation of HEPA filters is necessary when the details of the ceiling plenum in the cleanroom are included in the model. In particular, a representation is required if the flow distribution over the ceiling into the cleanroom needs to be evaluated-for example, for those models defined in Scenario 2 and 3.
Figure 3: This chart represents typical HEPA-filter performance data. Image courtesy of Flomerics.
The representation relies on the conversion of the pressure drop versus airflow rate curve for the HEPA filter, shown in Figure 3, into appropriate CFD boundary conditions. A standard means of representing a resistance to airflow is to equate the pressure drop to the velocity through the HEPA filter, as such:
Δp = pressure drop across HEPA filter
ƒ = loss coefficient
ρ = air density
ν = air velocity (Note: In the majority of cases, this is the air velocity approaching the HEPA filter.)
The main issue is then the representation of the loss coefficient, ƒ. This is handled in different ways by the various CFD codes. The curve for most HEPA-filter types can be essentially thought of as providing a linear pressure drop with increasing flow rate-for instance, Δp α ν. There should be enough flexibility in the CFD software to include such a relationship.
Care should be taken in evaluating the manufacturer’s data-specifically, to check whether the performance data refers to the media making up the filter, or the whole unit. In the former case, additional detail may be required to ensure an accurate representation of the filter.
Note that a representation of a perforated plate will also have to be added to the model immediately underneath the HEPA filter. This will add an extra pressure drop to the system represented by the ceiling-in this case, the pressure drop will follow a more traditional Δp α ν2 relationship.
The representation of diffusers is similar to that of HEPA filters: it should be developed with available manufacturer’s data for accuracy. Typically, the manufacturer’s data is displayed to give the distance from a reference point to where an isovel (line of constant speed) is reached. This “throw” data, for example, provides the distances from the face of the diffuser to the 50, 100, and 150 fpm isovels.
To obtain an accurate CFD model representation of a diffuser so it behaves in the same way as the physical diffuser, a sub-model should be built that replicates the test conditions in which the diffuser was tested. However, the person would need to take care to ensure that: the volume of the room is set accurately, the supply diffuser and return grilles are located correctly, and that the method by which the room is heated (for thermal applications of the diffuser) is accurately defined.
The usual diffuser type defined in cleanrooms, namely the laminar diffuser, is readily defined as a single-fixed flow device. However, laminar diffusers typically have low free-area ratio faces, on the order of 10 to 20 percent. Therefore, it is important that the entrainment characteristics-namely, the way that the flow at the diffuser face ingests the air from the rest of the room-are handled correctly. This means that the CFD code should have some means of prescribing local acceleration of the airflow through the face of the diffuser. A typical laminar flow diffuser validation plot is shown in Figure 4.
Figure 4: This image is illustrative of a typical validation graph for a laminar flow diffuser. Image courtesy of Flomerics.
Other diffuser types are usually more problematic. For example, radial diffusers require that the agreement with the manufacturer’s data be made for both the vertical and horizontal directions. Otherwise, the flow from the diffuser tends to “dump” more than it actually does in reality. Care should be taken so that the representation matches the manufacturer’s data for a range of different flow rates and heat-load conditions.
Typically, returns are defined as being either boundary conditions, which extract a certain amount of airflow from the cleanroom, or, if the amount is not known, as resistances to airflow. In the latter case, the resistance presented to the airflow is defined using a similar technique to that prescribed for HEPA filters. The pressure beyond the return-for example, the pressure condition in the return duct-is also required so that the CFD code can calculate the correct flow rate through the return.
Leakage and pressurization
The issue of leakage and pressurization also needs to be carefully considered in the definition of the cleanroom model, since even relatively small flows through the fabric of the cleanroom can impact the flow conditions within the room.
One obvious path for leakage or pressurization in the cleanroom is the door crack. The relationship between the pressure drop and flow rate through the door crack is determined by a variation of Equation 1, namely:
Δp = pressure difference across the door crack
ρ = air density
ν = airflow rate through door crack
A = area of door crack
C = flow coefficient, usually in the range 0.6 to 0.7
This equation can also be applied to process holes in the fabric of the cleanroom, with the flow coefficient dependent on the shape of the hole. The user can, therefore, set the flow coefficient for the crack and the pressure beyond the room, which, in turn, calculates the flow rate, or set the flow rate at that location.
If no pressure boundary condition is set in the cleanroom, the values of pressure that are calculated by the CFD model are variations in pressure about an arbitrary base value. In order to determine actual pressure values in the cleanroom, there must be at least one pressure boundary condition in the cleanroom, which acts as a datum point.
The following should be considered in terms of gridding in the cleanroom:
■ The grid should be refined in regions where high gradients are expected. In the cleanroom, for example, there are several areas where the gradients of particular physical values are high:
1. Regions close to flow devices or cracks in the cleanroom fabric; for example, diffusers, door cracks, etc.
2. Regions close to physical objects or objects that dissipate heat; for example, pieces of equipment, technicians, etc.
3. Sources of contamination; for example, spillages, technicians.
■ The grid should be continually refined until the point at which the result is grid-independent. At grid independence, the results of the calculation do not change when more grid cells are added.
Figure 5A: This image shows a fully structured grid associated with piece of cleanroom equipment.
The concept of “embedded” or “localized” grids is demonstrated in Figures 5A and 5B. Figure 5A shows the grid associated with a piece of equipment, using a fully structured grid system. Here, three items of interest are the turntables highlighted in red. The CFD software places grid cell faces at the edges of the different equipment items in the cleanroom. The presence of the three turntables results in high-density grid regions if the turntables are to be modeled accurately, as well as small grid cell sizes in regions away from the turntables that are, from a computational view, wasteful.
Figure 5B: This image shows an equivalent system with localized grid regions. Figures courtesy of Wyeth.
Figure 5B shows the grid for a case in which the grid around the turntables has been localized; in particular, the grid lines in the latter case no longer extend to the end of the solution domain, resulting in a reduction in the number of grid cells. In this particular example, the number of grid cells decreased from 260,000 to 153,000 just through this modification.
The following section describes an example of the optimization of the airflow conditions in a clean corridor. The images and cases were provided courtesy of Wyeth. The primary objective of this particular example was to improve the uniformity of the airflow through the filters from the ceiling plenum.
Figure 6: This image is illustrative of the geometry of a clean corridor. Image courtesy of Wyeth.
The geometry of this problem is shown in Figure 6. The overall size of the room is 20 (width) by 14 (high) by 94 feet (length). The ceiling plenum is 2.5 feet high. The CFD model of the clean corridor contains representations of ceiling plenum supply diffusers (with baffle plates immediately underneath), ULPA (Ultra Low Penetration Air) filters, perforated ceiling screens, and return grilles, located at floor level. The cutout shape of the room was created through the inclusion of a large cuboid block with the dimensions of 6 (width) by 14 (high) by 81 feet (length).
It was assumed in this case that there was no heat transfer into or out of the corridor through the wall, ceiling, or floor.
The flow rate into the ceiling plenum was 155,600 cfm provided by eight supply fans: two fans provided about 20,908 cfm each, while the other six provided 18,964 cfm each. The fans were all 40 by 54 inches in size. Underneath each supply is a baffle that is 56 by 70 inches in size, located 6 inches from the face of the supply fan.
The ceiling plenum also consisted of 172 ULPA filters. The overall box containing the filter is 20 by 12 by 44 inches in size, and contains a 6-inch section of filter media and a perforated plate at the bottom of the unit. The air in the clean corridor is exhausted via 14 returns of various sizes and flow rates. In particular, the returns vary from 15 by 15 inches of exhausting 1035 cfm to 373 18 inches of exhausting 25,726 cfm.
The only parameter that could be manipulated in this particular case was the free-area ratio of the baffle immediately underneath the supply fans. The flow rates and locations of all the supply and return fans were fixed. A series of cases were run in which the free-area ratio of the baffle was varied from 0 percent (completely closed), which was the baseline case, to 80 percent open. The parameter that was considered was the standard deviation of the flow through the ULPA filters.
The plot of the speed field immediately below the filters for the baseline case is shown in Figure 8. The non-uniformity through the filters is apparent. The maximum deviation from the mean flow rate for this case is just over 22 percent, while the standard deviation from the mean for this case is around 8.8 percent.
Figure 7: This graphic represents the standard deviation versus the baffle-free area ratio. Figure courtesy of Wyeth.
The effect of changing the free-area ratio (FAR) on the standard deviation is shown in Figure 7. The plot illustrates a minimum in the standard deviation for a FAR of 30 percent. The speed field plot for this case is shown in Figure 9. The maximum deviation from the mean flow rate for this case is just over 12 percent, while the standard deviation from the mean for this case is around 5.4 percent.
Figure 8: This graphic represents the speed fill plot immediately underneath filters for baseline case. Figure courtesy of Wyeth.
The revision to the baffle, therefore, leads to a clear improvement of the conditions provided by the supply plenum.
Figure 9: This image illustrates the speed fill plot immediately underneath filters for 30 percent FAR case. Figure courtesy of Wyeth.
As this example indicates, CFD simulation can indeed generate substantial improvements in a cleanroom design. This is accomplished by enabling engineers to identify problems and evaluation potential solutions long before equipment is purchased and construction begins. This article has discussed some of the critical issues involved in CFD simulation in order to ensure accurate and robust simulations.
Byron Blackmore is a thermal applications engineer at Flomerics, where he is responsible for supporting the company’s FLOTHERM, FLOPCB, and FLOVENT software and consulting services in North America. He has been with Flomerics for six years. He has a BS in mechanical engineering from the Technical University of Nova Scotia and an MS in Mechanical Engineering, with a concentration in fluid dynamics and heat transfer from the University of Alberta.
Weiran Xu obtained his BS and MS degrees from Tsinghua University in Beijing, China, and his Ph.D. from Massachusetts Institute of Technology in Cambridge, Massachusetts, in Building Technology. He is the thermal technical support supervisor at Flomerics, and has been working with customers on many thermal/CFD consulting projects and support cases. A member of ASME, he has also written for a number of industry publications.