PHARMACEUTICAL PROCESSING | JANUARY/FEBRUARY 2015 27 n
Results: Overall Pressure Distribution
Ideal flow conditions just above the conveyor level consist of
a perfectly vertical flow. Pressure distribution in the horizontal
plane is thus very important and must be as uniform as possible inside the RABS. The first CFD simulation of the cleanroom
showed a small pressure gradient that was sufficient to induce a
longitudinal component to the velocity vectors inside the RABS.
Laporte engineers designed deflectors to redistribute the
pressure in the capping section and removed the partition that
provoked a pressure increase in the vials accumulation section.
Combined with the modular adjustment of the HEPA filtration,
these modifications significantly improved the pressure distribution in the RABS. The CFD simulation correlates well with the
smoke tests performed with the new design and confirmed the
With the longitudinal flow corrected, Laporte and Creaform
focused on transverse velocity components in the vicinity of non-sterile machine components. The CFD simulations highlighted
two similar, undesired situations: one around the needles holder
and one around the capping arm.
Both components are non-sterile and the air draft from
underneath the physical barrier induces a significant transverse velocity component. This phenomenon drives particles
in contact with the arm directly toward the vials that are
conveyed at the level of the toothed plate. The aerodynamic
deflector was tested in simulation and provoked the shift of
the air draft towards the machine floor.
It caused the streamlines in the vicinity of the arm to reach
the underside of the conveyor, keeping the potentially contami-
nated particles far from the vials. A similar defector was used
drawings are incomplete or the geometry has changed over time.
This situation brings the engineers to constantly question the
CFD results. Creaform, by supplying efficient 3D scanning solutions, ensures high-quality numerical reproduction.
Creaform also manufactures portable and easy to use scanners that provide metrology-grade accuracy and resolution. For
CFD applications in the HVAC industry, it is quite common for
Creaform’s engineers to combine the scan of an entire room
acquired by a mid-range scanner with the precise scan of specific
parts using a handheld scanners. The result is a clean STL file
such as the one built for this pharmaceutical cleanroom.
The numerical geometry of the cleanroom includes the walls
and furniture, the HEPA filtration, the HVAC system, the physical
barrier with gloved access (windows surrounding the production
line) and the control panels. It also includes the RABS itself with
the accumulation table for vials, the conveyor, the filling needles,
the capping machine and many measurement instruments, all of
which were accounted for in the CFD simulations thanks to the
wrapping capabilities of STAR-CCM+®.
Precise and representative boundary conditions are critical for
the cleanroom simulation. They were carefully determined using
very recent data acquisition:
• Laminar flow equipment performance evaluation providing
air velocity profiles for each HEPA filtration diffuser.
• Ventilation balancing measurements for the HVAC system,
including return ducts.
• Precise pressure gaging in adjacent rooms for secondary
air flow rates through wall openings for conveyors and the
Turbulence modeling was achieved with the RANS approach
and more specifically with the SST (Menter) k- model, thus limiting the results to steady state. The All y+ Wall Treatment was
used because many near wall cells fell within the buffer region of
the boundary layer. The control over the entire surfaces to force
viscous sublayer resolution was computationally expensive and
Indeed, calculation of viscous forces is not required and flow
separation occurs at cutting edges, so its prediction is trivial.
Consequently, the mesh is polyhedral and does not make use of
prism layers. The prioritized cell refinement was the one allowing
to capture the surface details of the machine components, resulting in a cell count of 5. 6 million for initial runs and of 18. 4 million
for final runs. Simulations made use of the coupled flow model
with a second order discretization.