BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.cam.ac.uk//v3//EN
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CATEGORIES:Fluids Group Seminar (CUED)
SUMMARY:Physics-informed Neural Networks for Simultaneous
Surrogate Modelling and Aerodynamic Optimization
- Dr. Yubiao Sun
DTSTART;TZID=Europe/London:20220513T123000
DTEND;TZID=Europe/London:20220513T133000
UID:TALK172946AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/172946
DESCRIPTION:Optimizing multiple variable problems is an exceed
ingly difficult task due to the curse of dimension
ality. This is particularly true for airfoil shape
optimization as remeshing or deformation of exist
ing meshes is required\, which is computationally
expensive. Our study aims to overcome this challen
ge by introducing a deep learning-based framework
that can handle multiple optimization problems eff
iciently. The essence of this framework is using s
urrogate modelling to produce high-fidelity soluti
ons and then perform gradient-based optimizations
for high-dimensional problems. The starting point
is to use PINN to construct surrogate models that
output flow fields for airfoils of varied configur
ations. The key feature of PINN is the incorporati
on of physical problem description\, including the
governing laws of physics\, domain geometry\, and
boundary conditions\, which enables neural networ
ks to solve underlying differential equations (e.g
.\, Navier--Stokes equations) as a learning proble
m. Thus\, the surrogate models can efficiently gen
erate flow fields as no labelled training data fro
m a separate high-fidelity simulation is required.
More importantly\, we extend the employed surroga
te models by including design parameters as inputs
to PINN. In the optimization process\, a quasi-Ne
wton algorithm is used and further accelerated by
automatic differentiation\, a popular algorithm de
signed to efficiently compute the gradients of obj
ective functions with respect to design variables.
Two examples have been presented to demonstrate t
he feasibility of using PINN-based surrogate model
ling for aerodynamic optimization\, both single pa
rameter and multiple parameter problems. The propo
sed method is straightforward to implement and com
putationally efficient\, providing a promising alt
ernative for computationally intensive optimizatio
n problems.
LOCATION:CUED\, LR3B
CONTACT:
END:VEVENT
END:VCALENDAR