October 22, 2020

Optimization of a room ventilation system

Industry: Cleanroom Engineering | Software: pSeven, scSTREAM™ | Company: Faure QEI

Introduction

Heating, Ventilation and Air Conditioning (HVAC) systems are nowadays a routine yet important technology in many domains. The goal of a HVAC system is to supply fresh air into a room, while exhausting the contaminated or simply overheated air. The objective can be to ensure acceptable ambient temperature for people or equipment in the room, but also to maintain a required level of air cleanliness (for example, to provide a low level of airborne infectious particles per volume of air in cleanrooms).

Figure 1. Temperature inside a room filled with workers and machines that produce heat.

The cleanroom engineering industry addresses many key domains, such as the production of micro-systems (which are sensitive to airborne particles), the management of radioactive waste in the nuclear industry, and the control of infection threat in the healthcare sector. Very recently, the topic of cleanrooms has become even more popular with the current health crisis due to COVID, when lowering the viral load in hospital rooms has become the main issue to manage the contamination threat in saturated hospitals.

Challenges

High energy efficiency has become more and more important in designing HVAC systems for cleanrooms, especially with the recent ISO 14644-16 standard (2019). Hence the challenge is to reach the specified level of ventilation quality, while remaining within low ranges of energy consumption. To that end, Faure QEI has pioneered the use of Computational Fluid Dynamics (CFD) tools to build more efficient HVAC systems. As shown in Figure 1, numerical simulation is now a key tool to model and visualize the flow patterns in a ventilated room.

As explained in the ISO 14644-16 standard, the aim is to avoid the oversizing of cleanrooms. Inside a room, it can be effective to have areas with different levels of dust instead of requiring a low level of dust everywhere. Such fine optimization can afford huge savings on blowing and/or cooling. To achieve this, the numerical simulation framework enables the fine checking of the air flow fields, and especially the age-of-the-air field [1], which provides a cheap and simple yet relevant estimation of the level of contamination that can be expected inside the room.

Numerous scenarios can be studied, covering various practical goals:

• Installation of an air guidance system
• Cowling of an equipment
• Risks of contamination during maintenance, measurement at the source of contamination or heat
• Reduction of air transfers between workstations, tuning of supply/exhaust flow rates
• Study of the room occupants' exposure
• Verification of the ventilation recovery time in case of unexpected flow reduction
• Optimization of inlets and outlets positions (automated search for optimum).

Objective

The present study demonstrates how the implementation of numerical simulations, coupled with process automation and optimization algorithm, can speed-up the design of efficient HVAC systems. We consider a room filled with solid objects (obstacles) of diverse sizes and shapes. For the sake of simplicity, these objects have simple shapes (squares and cylinders, see Figure 2). Two ventilation inlets (inlet1 and inlet2) are placed on the ceiling to supply air into the room. One outlet (on the ceiling, too) exhausts the air from the room. The positions of the two inlets and of the outlet generate a pressure gradient which mechanically initiates the movement of the air (mechanical ventilation). The goal is to find the position of the inlets and outlet so that the ventilation efficiency is maximized. To evaluate the efficiency of the ventilation installation, we use the mean (volume averaged) age-of-the-air inside the room [1].

Figure 2 : CFD model of the present study. The two inlets are in white, the outlet is in red.
The obstacles (workers, devices, machines) are in grey.

Optimisation Problem Statement

The two inlets and the outlet positions are constrained to a plan (\(xy\) plane, the \(z\) axis is in the vertical direction. The ceiling is defined by \(z=cste\)). In total, we have \( (2+1)×2=6 \) variables. To avoid overlaps, we impose a minimum distance between the two inlets, in the form of an analytical (quadratic) constraint:

\(d(inlet1→inlet2)^2 > {d_{min}}^2 \)

We consider that the two inlets will naturally move away from the outlet (to maximize the ventilation efficiency), hence we do not impose any constraints on the distance between inlet and outlet. Also, due to symmetry conditions (the two inlets are identical), we impose a linear constraint:

\(x_{inlet1}<x_{inlet2} \)

to avoid the evaluation of symmetrical (identical) designs.

To sum-up, the optimization problem has:

• 6 variables 
• 1 expensive objective to maximize (one CFD simulation takes about 30 minutes)
• 2 cheap constraints (defining the geometrical relations between the two inlets).

As we are dealing with a CFD model, the numerical simulation is expensive and may display noisy behavior (due to convergence issues or variable quality of the mesh). Our goal is also to obtain a global optimum solution over almost all possible positions of inlets and outlets. Hence this optimization case is well suited for the surrogate model-based optimization algorithm (SBO) technique [2].

Solution

Automation of the CFD simulation and post-processing

The air flow is modelled as steady, incompressible. Due to the relatively high Reynolds number \((\sim10^5)\), the flow simulation displays turbulence features, hence the application of the Reynolds Averaged Navier-Stokes (RANS) method to take turbulence into account.

The simulation is performed with scSTREAM™, a commercial CFD simulation tool by MSC Software. Simulation steps to be automated are typical for any CFD routine:

• Pre-processing: change the ventilation configuration, recreate the mesh with the new positions of the inlets and outlet Solver: run the simulation and monitor the convergence (a criterion on the residuals and a minimum number of time iterations are used)
• Post-processing: extract the results and compute the ventilation efficiency.

The integration of scSTREAM™ with pSeven is made in a workflow:

Figure 3. Integration of scSTREAM™ into pSeven. The full computational chain is automated.

which receives a set of 6 variables and returns the measure of the efficiency of ventilation (which is the objective we want to maximize). The interface to scSTREAM™ (both pre- and post-processing) is based upon the scSTREAM™ Virtual Basic API. The pSeven Text block allows to update the .vbs script which generates the geometry with a new set of 6 variables. Another instance of the pSeven Text block then reads the output file generated by the .vbs script which computes the ventilation efficiency based on the simulation results. The Program block of pSeven behaves as a shell to run solver execution command.

Figure 4. Top-level optimization workflow in pSeven

This integration workflow is encapsulated into a single CFD block (an instance of the Composite block in pSeven), which is connected to a Design Space Exploration block (DSE) on the top level of pSeven workflow. The optimization problem is entirely defined in the DSE block (bounds of the variables, constraints, and objective). Notice the C inlet position isolated block where the distance constraint between the two inlets is computed separately. The optimizer uses this feature to restore this quadratic constraint internally in advance and take advantage of this to satisfy the constraint efficiently all along the optimization process. This feature is key to saving expensive evaluations in areas of the design space where the two inlets overlap, and the model is not accurate.

Conclusion

This study shows that the combination of pSeven with a commercial CFD tool allows for fast and exhaustive trade-off analysis of a complex process. Batch execution, process automation, iterative learning of the objective to maximize, allow to reduce the design time tremendously compared to a manual process. Moreover, the systematic exploration of design choices brings additional valuable insight.

Figure 5. Illustration of the designs evaluated during the optimization process (parallel plot).

Data analysis and visualization is greatly improved using user-friendly visualization tools. Such tools are useful at the post-processing stage: for example, Figure 5 shows a parallel plot to visualize each design (set of parameters and corresponding responses). Such plot allows for cross-selection of designs. We see that if we impose a minimum value to the distance between inlets (C inlet position), the Ventilation efficiency is reduced. This visual trade-off for various combinations of parameters shows that these two responses are competing between each other.

The smart choice of design points by the SBO algorithm allows the efficient use of the available CPU cores and licenses of CFD software. Thanks to automation of the optimization process, a huge amount of engineering time is saved, and the search is exhaustive, which means that the probability of finding the best design is much higher compared to manual trade-off analysis. Of course, all the results are then available for engineering judgment and the ventilation engineer has the final word on the design choice.

By Martin Pauthenet and Mikhaïl Kozlov, Application Engineers at pSeven SAS and Pierre Bombardier, Project Engineer at FAURE QEI.

About Faure QEI

References

[1] Meiss, A. and Feijo-Munoz, J. and Garcia-Fuentes, M.: Age-of-the-air in rooms according to the environmental condition of temperature: A case study, Energy and Buildings, Vol. 67, 2013

[2] Surrogate Based Optimization guide, from pSeven Core documentation.