The adjoint library is enhanced with new functionality enabling
automated shape optimisation loops. A parameterisation scheme based on
volumetric B-Splines is introduced, the control points of which act as
the design variables in the optimisation loop [1, 2]. The control
points of the volumetric B-Splines boxes can be defined in either
Cartesian or cylindrical coordinates.
The entire loop (solution of the flow and adjoint equations, computation
of sensitivity derivatives, update of the design variables and mesh) is
run within adjointOptimisationFoam. A number of methods to update the
design variables are implemented, including popular Quasi-Newton methods
like BFGS and methods capable of handling constraints like loop using
the SQP or constraint projection.
The software was developed by PCOpt/NTUA and FOSS GP, with contributions from
Dr. Evangelos Papoutsis-Kiachagias,
Konstantinos Gkaragounis,
Professor Kyriakos Giannakoglou,
Andy Heather
[1] E.M. Papoutsis-Kiachagias, N. Magoulas, J. Mueller, C. Othmer,
K.C. Giannakoglou: 'Noise Reduction in Car Aerodynamics using a
Surrogate Objective Function and the Continuous Adjoint Method with
Wall Functions', Computers & Fluids, 122:223-232, 2015
[2] E. M. Papoutsis-Kiachagias, V. G. Asouti, K. C. Giannakoglou,
K. Gkagkas, S. Shimokawa, E. Itakura: ‘Multi-point aerodynamic shape
optimization of cars based on continuous adjoint’, Structural and
Multidisciplinary Optimization, 59(2):675–694, 2019
QRMatrix (i.e. QR decomposition, QR factorisation or orthogonal-triangular
decomposition) decomposes a scalar/complex matrix \c A into the following
matrix product:
\verbatim
A = Q*R,
\endverbatim
where
\c Q is a unitary similarity matrix,
\c R is an upper triangular matrix.
Usage
Input types:
- \c A can be a \c SquareMatrix<Type> or \c RectangularMatrix<Type>
Output types:
- \c Q is always of the type of the matrix \c A
- \c R is always of the type of the matrix \c A
Options for the output forms of \c QRMatrix (for an (m-by-n) input matrix
\c A with k = min(m, n)):
- outputTypes::FULL_R: computes only \c R (m-by-n)
- outputTypes::FULL_QR: computes both \c R and \c Q (m-by-m)
- outputTypes::REDUCED_R: computes only reduced \c R (k-by-n)
Options where to store \c R:
- storeMethods::IN_PLACE: replaces input matrix content with \c R
- storeMethods::OUT_OF_PLACE: creates new object of \c R
Options for the computation of column pivoting:
- colPivoting::FALSE: switches off column pivoting
- colPivoting::TRUE: switches on column pivoting
Direct solution of linear systems A x = b is possible by solve() alongside
the following limitations:
- \c A = a scalar square matrix
- output type = outputTypes::FULL_QR
- store method = storeMethods::IN_PLACE
Notes
- QR decomposition is not unique if \c R is not positive diagonal \c R.
- The option combination:
- outputTypes::REDUCED_R
- storeMethods::IN_PLACE
will not modify the rows of input matrix \c A after its nth row.
- Both FULL_R and REDUCED_R QR decompositions execute the same number of
operations. Yet REDUCED_R QR decomposition returns only the first n rows
of \c R if m > n for an input m-by-n matrix \c A.
- For m <= n, FULL_R and REDUCED_R will produce the same matrices
- Allows user-defined control of when the mesh motion occurs,
which can be especially useful in situations where the mesh motion
is much slower than any of the fluid physics.
For example, in constant/dynamicMeshDict:
updateControl runTime;
updateInterval 0.5;
to have mesh motion triggered every 1/2 second.
Note that the _exact_ time that the mesh motion actually occurs may
be slightly differently since the "runTime" triggering is fuzzy in
nature. It will trigger when the threshold has been crossed, which
will depend on the current time-step size.
COMP: delay evaluation of fieldToken enumeration types
- lazy evaluation at runTime instead of compile-time to make the code
independent of initialization order.
Otherwise triggers problems on gcc-4.8.5 on some systems where
glibc is the same age, or older.
- replace stringOps::toScalar with a more generic stringOps::evaluate
method that handles scalars, vectors etc.
- improve #eval to handle various mathematical operations.
Previously only handled scalars. Now produce vectors, tensors etc
for the entries. These tokens are streamed directly into the entry.
- ITstream append() would previously have used the append from the
underlying tokenList, which leaves the tokenIndex untouched and
renders the freshly appended tokens effectively invisible if
interspersed with primitiveEntry::read() that itself uses tokenIndex
when building the list.
The new append() method makes this hidden ITstream bi-directionality
easier to manage. For efficiency, we only append lists
(not individual tokens) and support a 'lazy' resizing that allows
the final resizing to occur later when all tokens have been appended.
- The new ITstream seek() method provides a conveniently means to move
to the end of the list or reposition to the middle.
Using rewind() and using seek(0) are identical.
ENH: added OTstream to output directly to a list of tokens
---
BUG: List::newElem resized incorrectly
- had a simple doubling of the List size without checking that this
would indeed be sufficient for the requested index.
Bug was not triggered since primitiveEntry was the only class using
this call, and it added the tokens sequentially.
- allows use of Enum in more situations where a tiny Map/HashTable
replacement is desirable. The new methods can be combined with
null constructed for to have a simple low-weight caching system
for words/integers instead of fitting in a HashTable.
- silently deprecate 'startsWith', 'endsWith' methods
(added in 2016: 2b14360662), in favour of
'starts_with', 'ends_with' methods, corresponding to C++20 and
allowing us to cull then in a few years.
- handle single character versions of starts_with, ends_with.
- add single character version of removeEnd and silently deprecate
removeTrailing which did the same thing.
- drop the const versions of removeRepeated, removeTrailing.
Unused and with potential confusion.
STYLE: use shrink_to_fit(), erase()
- Now accept '/' when reading variables without requiring
a surrounding '{}'
- fix some degenerate parsing cases when the first character is
already bad.
Eg, $"abc" would have previously parsed as a <$"> variable, even
although a double quote is not a valid variable character.
Now emits a warning and parses as a '$' token and a string token.
- add toScalar evaluation, embedded as "${{EXPR}}".
For example,
"repeat ${{5 * 7}} times or ${{ pow(3, 10) }}"
- use direct string concatenation if primitive entry is only a string
type. This prevents spurious quotes from appearing in the expansion.
radius "(2+4)";
angle "3*15";
#eval "$radius*sin(degToRad($angle))";
We want to have
'(2+4)*sin(degToRad(3*15))'
and not
'"(2+4)"*sin(degToRad("3*15"))'
ENH: code refactoring
- refactored expansion code with low-level service routines now
belonging to file-scope. All expansion routines use a common
multi-parameter backend to handle with/without dictionary etc.
This removes a large amount of code duplication.
- add floor/ceil/round methods
- support evaluation of sub-strings
STYLE: add blockMeshDict1.calc, blockMeshDict1.eval test dictionaries
- useful for testing and simple demonstration of equivalence
- SubField and SubList assign from zero
- SubField +=, -=, *=, /= operators
- SubList construct from UList (as per SubField)
Note: constructing an anonymous SubField or SubList with a single
parameter should use '{} instead of '()' to avoid compiler
ambiguities.
- drop plugin support for Qt4 (old paraview)
- handle upcoming changes in VTK version naming in CMake files
* VTK_MAJOR_VERSION becomes VTK_VERSION_MAJOR etc.
- the #eval directive is similar to the #calc directive, but for evaluating
string expressions into scalar values. It uses an internal parser for
the evaluation instead of dynamic code compilation. This can make it
more suitable for 'quick' evaluations.
The evaluation supports the following:
- operations: - + * /
- functions: exp, log, log10, pow, sqrt, cbrt, sqr, mag, magSqr
- trigonometric: sin, cos, tan, asin, acos, atan, atan2, hypot
- hyperbolic: sinh, cosh, tanh
- conversions: degToRad, radToDeg
- constants: pi()
- misc: rand(), rand(seed)
