"""
==============================================================================
DVGeo: CST Parameterisation
==============================================================================
@Author : Eytan Adler, Alasdair Christison Gray
@Description : A DVGeo implementation based on the Class-Shape Transformation method
"""
# External modules
from mpi4py import MPI
import numpy as np
from scipy.special import factorial
try:
# External modules
from prefoil.airfoil import Airfoil
from prefoil.utils import readCoordFile
prefoilInstalled = True
except ImportError:
prefoilInstalled = False
try:
# External modules
import matplotlib.pyplot as plt
pltImport = True
except ImportError:
pltImport = False
# Local modules
from .BaseDVGeo import BaseDVGeometry
from .designVars import cstDV
[docs]
class DVGeometryCST(BaseDVGeometry):
r"""
This class implements a 2D geometry parameterisation based on Brenda Kulfan's CST (Class-Shape Transformation) method.
This class can work with 3D coordinates but will only change the point coordinates in one direction.
The CST equation is as follows:
:math:`y(x) = C(x) * S(x) + y_\text{te}x`
Where C is the class function:
:math:`C(x) = (x^{N1} + (1 - x)^{N2})`
And S is the shape function, in this case a summation of Bernstein polynomials:
:math:`S(x) = \sum_i^n w_i \binom{n}{i}x^i(1-x)^{n-i}`
Here x is the normalized chordwise coordinate, ranging from 0 to 1 from front to the rear of the shape.
Assumptions about the point sets being added:
- Dat file is ordered continuously around the airfoil and the beginning and end of
the list is the trailing edge (no jumping around, but CW vs. CCW does not matter)
- Geometry is exclusively an extruded shape (no spanwise changes allowed)
- Airfoil's leading edge is on the left (min x) and trailing edge is on the right (max x)
- Airfoil is not rotated (trailing edge and leading edge are close to y equals zero)
Parameters
----------
datFile : str
Filename of dat file that represents the initial airfoil. The coordinates in this file will be used to
determine the camber line, which is the dividing line to distinguish upper and lower surface points.
numCST : int or list of two ints
Number of CST parameters to use for the initial fit and the DVs (if DVs with type ``"upper"`` or ``"lower"``
are added). If ``numCST`` is an int, the value will be used for both upper and lower. If it is a two-item list,
the first value defines the number of upper CST coefficients and the second is the number of lower coefficients,
by default 8.
idxChord : int, optional
Index of the column in the point set to use as the chordwise (x in CST) coordinates, by default 0
idxVertical : int, optional
Index of the column in the point set to use as the vertical (y in CST) airfoil coordinates, by default 1
comm : MPI communicator, optional
Communicator for DVGeometryCST instance, by default MPI.COMM_WORLD
isComplex : bool, optional
Initialize variables to complex types where necessary, by default False
debug : bool, optional
Show plots when addPointSet is called to visually verify that it is correctly splitting
the upper and lower surfaces of the airfoil points, by default False
tolTE : float, optional
Tolerance used to detect trailing edge corners on the airfoil. The value represents the angle difference
in degrees between adjacent edges of the airfoil, by default 60 deg.
name : string, optional
Name of this DVGeo object, not necessary unless multiple DVGeos are used in one optimization.
"""
def __init__(
self,
datFile,
numCST=8,
idxChord=0,
idxVertical=1,
comm=MPI.COMM_WORLD,
isComplex=False,
debug=False,
tolTE=60.0,
name=None,
):
# Check if preFoil is installed before initializing.
if not prefoilInstalled:
raise ImportError("preFoil is not installed and is required to use DVGeometryCST.")
super().__init__(datFile, name=name)
self.xIdx = idxChord
self.yIdx = idxVertical
self.comm = comm
self.isComplex = isComplex
if isComplex:
self.dtype = complex
self.dtypeMPI = MPI.DOUBLE_COMPLEX
else:
self.dtype = float
self.dtypeMPI = MPI.DOUBLE
self.debug = debug
if debug and not pltImport:
raise ImportError("matplotlib.pyplot could not be imported and is required for DVGeoCST debug mode")
# Error check the numCST input
if isinstance(numCST, int):
self.nCSTUpper = numCST
self.nCSTLower = numCST
else:
if isinstance(numCST, list):
if len(numCST) != 2 or not isinstance(numCST[0], int) or not isinstance(numCST[1], int):
raise ValueError(f"numCST input of {numCST} is incorrect; must be int or list of two ints")
else:
self.nCSTUpper = numCST[0]
self.nCSTLower = numCST[1]
else:
raise ValueError(f"numCST input of type {type(numCST)} is incorrect; must be int or list of two ints")
# Store the DVs and flags to determine if the limited options have already been specified
self.DVs = {}
self.DVExists = {
"upper": False,
"lower": False,
"n1_upper": False,
"n2_upper": False,
"n1_lower": False,
"n2_lower": False,
"chord": False,
}
# Default DVs to be copied for each point set
self.defaultDV = {
"upper": 0.1 * np.ones(self.nCSTUpper, dtype=self.dtype),
"lower": -0.1 * np.ones(self.nCSTLower, dtype=self.dtype),
"n1_upper": np.array([0.5], dtype=self.dtype),
"n2_upper": np.array([1.0], dtype=self.dtype),
"n1_lower": np.array([0.5], dtype=self.dtype),
"n2_lower": np.array([1.0], dtype=self.dtype),
"n1": np.array([0.5], dtype=self.dtype),
"n2": np.array([1.0], dtype=self.dtype),
"chord": np.array([1.0], dtype=self.dtype),
}
# ========== Process the input airfoil and set variables accordingly ==========
coords = readCoordFile(datFile)
self.foilCoords = np.zeros_like(coords, dtype=self.dtype)
self.foilCoords[:, self.xIdx] = coords[:, 0]
self.foilCoords[:, self.yIdx] = coords[:, 1]
# Determine if the dat file is closed (first and last points are the same); remove the duplicate point if so
distance = np.linalg.norm(self.foilCoords[0, :] - self.foilCoords[-1, :])
distTol = 1e-12
if distance < distTol:
self.foilCoords = self.foilCoords[:-1, :]
# Check if the coordinates are clockwise or counter-clockwise by taking the cross-product of
# vectors from adjacent points at the trailing edge
vec1 = np.append(self.foilCoords[0] - self.foilCoords[-1], 0)
vec2 = np.append(self.foilCoords[1] - self.foilCoords[0], 0)
crossProduct = np.cross(vec1, vec2)
# Flip the y-coordinates to counter-clockwise if they are clockwise (negative cross-product)
if crossProduct[2] < 0:
self.foilCoords = np.flip(self.foilCoords, self.xIdx)
# Set the leading and trailing edge x coordinates
self.xMin = np.min(self.foilCoords[:, self.xIdx])
self.xMax = np.max(self.foilCoords[:, self.xIdx])
# The airfoil file can have one or two points at the minimum x-coordinate
idxLE = np.where(self.foilCoords[:, self.xIdx] == self.xMin)[0]
if len(idxLE) > 2:
raise ValueError(f"There can only be one or two points at the minimum x-coordinate, not {len(idxLE)}")
# Check that the leading edge is at y = 0
for idx in idxLE:
yLE = self.foilCoords[idx, self.yIdx]
if abs(yLE) > 1e-2:
raise ValueError(f"Leading edge y (or idxVertical) value must equal zero, not {yLE}")
# If there is one leading edge point, add a duplicate index to make setting idxFoil easier
if len(idxLE) == 1:
idxLE = np.repeat(idxLE, 2)
# Traverse the airfoil surface to find the corner(s) defining the trailing edge (ignore anything in the front
# half, chordwise, of the airfoil)
cosTolTE = np.cos(np.deg2rad(tolTE))
cornerIdx = []
for idx in range(self.foilCoords.shape[0]):
pt = self.foilCoords[idx, :]
# Ignore if closer to the leading edge
if pt[self.xIdx] - self.xMin < self.xMax - pt[self.xIdx]:
continue
edgePrev = pt - self.foilCoords[idx - 1, :]
edgePrev /= np.linalg.norm(edgePrev)
edgeNext = self.foilCoords[(idx + 1) % self.foilCoords.shape[0], :] - pt
edgeNext /= np.linalg.norm(edgeNext)
if np.dot(edgePrev, edgeNext) < cosTolTE:
cornerIdx.append(idx)
if len(cornerIdx) > 2:
raise RuntimeError(
"More than two corners in the airfoil identified when looking for the "
+ "trailing edge. If the actual airfoil geometry in the dat file has more "
+ "than two corners it is not supported. If not, try reducing the tolTE input."
)
elif len(cornerIdx) == 0:
raise RuntimeError(
"Zero corners in the airfoil identified when looking for the "
+ "trailing edge. If the actual airfoil geometry in the dat file has zero corners "
+ "(is a circle??) it is not supported. If not, try increasing the tolTE input."
)
# Airfoil is sharp if only one corner is detected
self.sharp = len(cornerIdx) == 1
# Save the trailing edge coordinates and surface indices
self.idxFoil = {}
if self.sharp:
self.yUpperTE = np.array([0.0])
self.yLowerTE = np.array([0.0])
self.idxFoil["upper"] = np.arange(cornerIdx[0], idxLE[0] + 1)
self.idxFoil["lower"] = np.concatenate(
(np.arange(idxLE[1], self.foilCoords.shape[0]), np.array([cornerIdx[0]]))
)
else:
self.coordUpperTE = self.foilCoords[cornerIdx[0]]
self.coordLowerTE = self.foilCoords[cornerIdx[1]]
self.yUpperTE = self.coordUpperTE[self.yIdx]
self.yLowerTE = self.coordLowerTE[self.yIdx]
self.idxFoil["upper"] = np.arange(cornerIdx[0], idxLE[0] + 1)
self.idxFoil["lower"] = np.arange(idxLE[1], cornerIdx[1] + 1)
# Compute splines for the upper and lower surfaces (used to split the foil in addPointSet).
# preFoil defines the leading edge as the point furthest from the trailing edge
self.foil = Airfoil(coords)
self.upperSpline, self.lowerSpline = self.foil.splitAirfoil()
# Fit CST parameters to the airfoil's upper and lower surface
chord = self.xMax - self.xMin
self.defaultDV["chord"][0] = chord
if self.comm.rank == 0:
print(f"######## Fitting CST coefficients to coordinates in {datFile} ########")
for dvType in ["upper", "lower"]:
if self.comm.rank == 0:
xPts = self.foilCoords[self.idxFoil[dvType], self.xIdx]
yPts = self.foilCoords[self.idxFoil[dvType], self.yIdx]
if dvType == "upper":
yTE = self.yUpperTE
else:
yTE = self.yLowerTE
self.defaultDV[dvType] = self.computeCSTfromCoords(
xPts,
yPts,
self.defaultDV[dvType].size,
N1=self.defaultDV[f"n1_{dvType}"],
N2=self.defaultDV[f"n2_{dvType}"],
dtype=self.dtype,
)
# Compute the quality of the fit by computing an L2 norm of the fit vs. the actual coordinates
ptsFit = chord * self.computeCSTCoordinates(
(xPts - self.xMin) / chord,
self.defaultDV["n1_lower"],
self.defaultDV["n2_lower"],
self.defaultDV[dvType],
yTE / chord,
dtype=self.dtype,
)
L2norm = np.sqrt(1 / ptsFit.size * np.sum((yPts - ptsFit) ** 2))
print(f"{dvType.capitalize()} surface")
print(f" L2 norm of coordinates in dat file versus fit coordinates: {L2norm}")
print(f" Fit CST coefficients: {self.defaultDV[dvType]}")
# Broadcast the fit DV to the rest of the procs
self.comm.Bcast([self.defaultDV[dvType], self.dtypeMPI])
[docs]
def addPointSet(self, points, ptName, boundTol=1e-10, **kwargs):
"""
Add a set of coordinates to DVGeometry.
The is the main way that geometry in the form of a coordinate list is given to DVGeometry
to be manipulated.
.. note::
Even if ``isComplex=True``, the imaginary portion of coordinates passed in here
is ignored when determining if a given point is on the upper or lower surface.
Parameters
----------
points : array, size (N,3)
The coordinates to embed.
ptName : str
A user supplied name to associate with the set of coordinates.
This name will need to be provided when updating the coordinates or when
getting the derivatives of the coordinates.
boundTol : float, optional
Small absolute deviation by which the airfoil coordinates can exceed the initial
minimum and maximum x coordinates, by default 1e-10.
\\*\\*kwargs
Any other parameters are ignored.
"""
# Convert points to the type specified at initialization (with isComplex) and store the points
points = points.astype(self.dtype)
# Check that all points are within the airfoil x bounds
if np.any(points[:, self.xIdx] < self.xMin - boundTol) or np.any(points[:, self.xIdx] > self.xMax + boundTol):
raise ValueError(
f'Points in the point set "{ptName}" have x coordinates outside'
+ f"the min and max x values in the initial dat file ({self.xMin} and {self.xMax})"
)
self.updated[ptName] = False
self.points[ptName] = {
"points": points,
"xMax": self.xMax.copy(),
"xMin": self.xMin.copy(),
"yUpperTE": self.yUpperTE.copy(),
"yLowerTE": self.yLowerTE.copy(),
}
# Determine which points are on the upper and lower surfaces
self.points[ptName]["upper"], self.points[ptName]["lower"] = self._splitUpperLower(points)
# If debug mode is on, plot the upper and lower surface points
if self.debug:
# Gather all the plotting data on the root proc
dataGlob = {}
for name in ["points", "upper", "lower"]:
# First, determine the sizes and displacements of the arrays from each proc for gather
vecFlatLoc = self.points[ptName][name].flatten()
if name in ["upper", "lower"]:
vecFlatLoc = vecFlatLoc.astype("intc")
numLoc = vecFlatLoc.size
sizes = np.array(self.comm.allgather(numLoc), dtype="intc")
disp = np.array([np.sum(sizes[:i]) for i in range(self.comm.size)], dtype="intc")
if name == "points":
dispPoints = disp.copy()
numGlob = np.sum(sizes)
# Send coordinates to root proc
dtype = "intc"
dtypeMPI = MPI.INT
if name == "points":
dtype = self.dtype
dtypeMPI = self.dtypeMPI
dataGlob[name] = np.zeros(numGlob, dtype=dtype) # recv buffer
# Shift the data by the displacement if it is local index data
if name in ["upper", "lower"]:
vecFlatLoc += dispPoints[self.comm.rank] // 3
# Finally, collect the data on the root proc
self.comm.Gatherv([vecFlatLoc, numLoc], [dataGlob[name], sizes, disp, dtypeMPI])
if self.comm.rank == 0:
# Reshape the flatted coordinates
coords = dataGlob["points"].reshape((dataGlob["points"].size // 3, 3))
fig = plt.figure()
plt.scatter(
coords[:, self.xIdx][dataGlob["upper"]],
coords[:, self.yIdx][dataGlob["upper"]],
c="b",
)
plt.scatter(
coords[:, self.xIdx][dataGlob["lower"]],
coords[:, self.yIdx][dataGlob["lower"]],
c="r",
)
plt.scatter(
coords[:, self.xIdx],
coords[:, self.yIdx],
s=3,
c="k",
zorder=3,
)
plt.legend(["Upper", "Lower"])
plt.show()
plt.close(fig)
self.comm.Barrier()
[docs]
def addDV(self, dvName, dvType, lowerBound=None, upperBound=None, scale=1.0, default=None):
"""
Add design variables to the DVGeometryCST object. For upper and lower CST coefficient DVs,
the number of design variables is defined using the ``numCST`` parameter in DVGeoCST's
init function.
Parameters
----------
dvName : str
A unique name to be given to this design variable group
dvType : str
Define the type of CST design variable being added. Either the upper/lower surface class shape
parameter DV can be defined (e.g., ``"N1_upper"``), or the DV for both the upper and lower surfaces' class shape
parameter can be defined (e.g., ``"N1"``), but not both. The options (not case sensitive) are
- ``"upper"``: upper surface CST coefficients (specify ``dvNum`` to define how many)
- ``"lower"``: lower surface CST coefficients (specify ``dvNum`` to define how many)
- ``"N1"``: first class shape parameter for both upper and lower surfaces (adds a single DV)
- ``"N2"``: second class shape parameter for both upper and lower surfaces (adds a single DV)
- ``"N1_upper"``: first class shape parameters for upper surface (adds a single DV)
- ``"N1_lower"``: first class shape parameters for lower surface (adds a single DV)
- ``"N2_upper"``: second class shape parameters for upper surface (adds a single DV)
- ``"N2_lower"``: second class shape parameters for lower surface (adds a single DV)
- ``"chord"``: chord length in whatever units the point set length is defined and scaled
to keep the leading edge at the same position (adds a single DV)
lowerBound : float or ndarray, optional
The lower bound for the variable(s). This will be applied to
all shape variables
upperBound : float or ndarray, optional
The upper bound for the variable(s). This will be applied to
all shape variables
scale : float, optional
The scaling of the variables. A good approximate scale to
start with is approximately 1.0/(upper-lower). This gives
variables that are of order ~1.0.
default : ndarray, optional
Default value for design variable (must be same length as number of DVs added).
Returns
-------
N : int
The number of design variables added.
"""
# Do some error checking
if dvType.lower() not in [
"upper",
"lower",
"n1",
"n2",
"n1_upper",
"n1_lower",
"n2_upper",
"n2_lower",
"chord",
]:
raise ValueError(
'dvType must be one of "upper", "lower", "N1", "N2", "N1_upper", "N1_lower", '
+ f'"N2_upper", "N2_lower", or "chord", not {dvType}'
)
dvType = dvType.lower()
if dvType == "upper":
dvNum = self.nCSTUpper
elif dvType == "lower":
dvNum = self.nCSTLower
else:
dvNum = 1
# Check that a duplicate DV doesn't already exist
if dvType in ["n1", "n2", "n1_upper", "n1_lower", "n2_upper", "n2_lower"]:
if dvType in ["n1", "n2"]: # if either of these is added, the individual lower and upper params can't be
if self.DVExists[dvType + "_lower"]:
raise ValueError(f'"{dvType}" cannot be added when "{dvType}_lower" already exists')
elif self.DVExists[dvType + "_upper"]:
raise ValueError(f'"{dvType}" cannot be added when "{dvType}_upper" already exists')
else:
self.DVExists[dvType + "_lower"] = True
self.DVExists[dvType + "_upper"] = True
else: # the parameter that controls both the upper and lower surfaces simultaneously can't be added
param = dvType.split("_")[0] # either N1 or N2
if self.DVExists[dvType]:
raise ValueError(f'"{dvType}" cannot be added when "{param}" or "{dvType}" already exist')
else:
self.DVExists[dvType] = True
else:
if self.DVExists[dvType]:
raise ValueError(f'"{dvType}" design variable already exists')
else:
self.DVExists[dvType] = True
if dvName in self.DVs.keys():
raise ValueError(f'A design variable with the name "{dvName}" already exists')
# Set the default value
if default is None:
default = self.defaultDV[dvType]
else:
if not isinstance(default, np.ndarray):
raise ValueError(f"The default value for the {dvName} DV must be a NumPy array, not a {type(default)}")
default = default.flatten()
if default.size != dvNum:
raise ValueError(
f"The default value for the {dvName} DV must have a length of {dvNum}, not {default.size}"
)
# Set new default
self.defaultDV[dvType] = default.astype(self.dtype)
if dvType in ["n1", "n2"]:
self.defaultDV[f"{dvType}_lower"] = default.astype(self.dtype)
self.defaultDV[f"{dvType}_upper"] = default.astype(self.dtype)
# Add the DV to the internally-stored list
self.DVs[dvName] = cstDV(
name=dvName,
value=default.astype(self.dtype),
nVal=dvNum,
lower=lowerBound,
upper=upperBound,
scale=scale,
dvType=dvType,
)
return dvNum
[docs]
def setDesignVars(self, dvDict):
"""
Standard routine for setting design variables from a design variable dictionary.
Parameters
----------
dvDict : dict
Dictionary of design variables. The keys of the dictionary must correspond to the design variable names.
Any additional keys in the dictionary are simply ignored.
"""
for dvName, dvVal in dvDict.items():
if dvName in self.DVs:
if dvVal.shape != self.DVs[dvName].value.shape:
raise ValueError(
f'Input shape of {dvVal.shape} for the DV named "{dvName}" does '
+ f"not match the DV's shape of {self.DVs[dvName].value.shape}"
)
self.DVs[dvName].value = dvVal.astype(self.dtype)
# Flag all the pointSets as not being up to date
for pointSet in self.updated:
self.updated[pointSet] = False
[docs]
def getValues(self):
"""
Generic routine to return the current set of design variables.
Values are returned in a dictionary format that would be suitable for a subsequent call to setValues()
Returns
-------
dvDict : dict
Dictionary of design variables
"""
# Format the dictonary into the desired shape
DVs = {}
for dvName in self.DVs.keys():
DVs[dvName] = self.DVs[dvName].value
return DVs
[docs]
def getVarNames(self, **kwargs):
"""
Return a list of the design variable names. This is typically used when specifying a wrt= argument for pyOptSparse.
Examples
--------
>>> optProb.addCon(.....wrt=DVGeo.getVarNames())
"""
return list(self.DVs.keys())
[docs]
def totalSensitivity(self, dIdpt, ptSetName, comm=None, **kwargs):
r"""
This function computes sensitivity information.
Specifically, it computes the following:
:math:`\frac{dX_{pt}}{dX_{DV}}^T \frac{dI}{d_{pt}}`
Parameters
----------
dIdpt : array of size (Npt, 3) or (N, Npt, 3)
This is the total derivative of the objective or function of interest with respect to the coordinates in 'ptSetName'.
This can be a single array of size (Npt, 3) **or** a group of N vectors of size (N, Npt, 3).
If you have many to do, it is faster to do many at once.
ptSetName : str
The name of set of points we are dealing with
\\*\\*kwargs
Any other parameters ignored, but this is maintained to allow the same
interface as other DVGeo implementations.
Returns
-------
dIdxDict : dict
The dictionary containing the derivatives, suitable for pyOptSparse
"""
# Unpack some useful variables
desVars = self._unpackDVs()
ptsX = self.points[ptSetName]["points"][:, self.xIdx]
xMax = self.points[ptSetName]["xMax"]
xMin = self.points[ptSetName]["xMin"]
scaledX = (ptsX - xMin) / (xMax - xMin)
idxUpper = self.points[ptSetName]["upper"]
idxLower = self.points[ptSetName]["lower"]
funcSens_local = {}
# If dIdpt is a group of vectors, reorder the axes so it
# is handled properly by the matrix multiplies
dim = dIdpt.shape
if len(dim) == 3:
dIdpt = np.moveaxis(dIdpt, 0, -1)
for dvName, DV in self.DVs.items():
dvType = DV.type
if dvType == "upper":
dydUpperCST = self.computeCSTdydw(
scaledX[idxUpper], desVars["n1_upper"], desVars["n2_upper"], desVars["upper"], dtype=self.dtype
)
dydUpperCST *= desVars["chord"]
funcSens_local[dvName] = dydUpperCST @ dIdpt[idxUpper, self.yIdx]
elif dvType == "lower":
dydLowerCST = self.computeCSTdydw(
scaledX[idxLower], desVars["n1_lower"], desVars["n2_lower"], desVars["lower"], dtype=self.dtype
)
dydLowerCST *= desVars["chord"]
funcSens_local[dvName] = dydLowerCST @ dIdpt[idxLower, self.yIdx]
elif dvType == "n1_upper":
funcSens_local[dvName] = (
desVars["chord"]
* self.computeCSTdydN1(
scaledX[idxUpper], desVars["n1_upper"], desVars["n2_upper"], desVars["upper"], dtype=self.dtype
)
[docs]
@ dIdpt[idxUpper, self.yIdx]
)
elif dvType == "n2_upper":
funcSens_local[dvName] = (
desVars["chord"]
* self.computeCSTdydN2(
scaledX[idxUpper], desVars["n1_upper"], desVars["n2_upper"], desVars["upper"], dtype=self.dtype
)
@ dIdpt[idxUpper, self.yIdx]
)
elif dvType == "n1_lower":
funcSens_local[dvName] = (
desVars["chord"]
* self.computeCSTdydN1(
scaledX[idxLower], desVars["n1_lower"], desVars["n2_lower"], desVars["lower"], dtype=self.dtype
)
@ dIdpt[idxLower, self.yIdx]
)
elif dvType == "n2_lower":
funcSens_local[dvName] = (
desVars["chord"]
* self.computeCSTdydN2(
scaledX[idxLower], desVars["n1_lower"], desVars["n2_lower"], desVars["lower"], dtype=self.dtype
)
@ dIdpt[idxLower, self.yIdx]
)
elif dvType == "n1":
funcSens_local[dvName] = (
desVars["chord"]
* self.computeCSTdydN1(
scaledX[idxUpper], desVars["n1_upper"], desVars["n2_upper"], desVars["upper"], dtype=self.dtype
)
@ dIdpt[idxUpper, self.yIdx]
)
funcSens_local[dvName] += (
desVars["chord"]
* self.computeCSTdydN1(
scaledX[idxLower], desVars["n1_lower"], desVars["n2_lower"], desVars["lower"], dtype=self.dtype
)
@ dIdpt[idxLower, self.yIdx]
)
elif dvType == "n2":
funcSens_local[dvName] = (
desVars["chord"]
* self.computeCSTdydN2(
scaledX[idxUpper], desVars["n1_upper"], desVars["n2_upper"], desVars["upper"], dtype=self.dtype
)
@ dIdpt[idxUpper, self.yIdx]
)
funcSens_local[dvName] += (
desVars["chord"]
* self.computeCSTdydN2(
scaledX[idxLower], desVars["n1_lower"], desVars["n2_lower"], desVars["lower"], dtype=self.dtype
)
@ dIdpt[idxLower, self.yIdx]
)
else: # chord
dydchord = self.points[ptSetName]["points"][:, self.yIdx] / desVars["chord"]
dxdchord = (ptsX - xMin) / desVars["chord"]
funcSens_local[dvName] = dxdchord @ dIdpt[:, self.xIdx] + dydchord @ dIdpt[:, self.yIdx]
# If the axes were reordered to handle a group of dIdpt vectors,
# switch them back to the expected order for output
if len(dim) == 3:
for dvName in funcSens_local.keys():
funcSens_local[dvName] = np.moveaxis(np.atleast_2d(funcSens_local[dvName]), 0, -1)
if comm:
funcSens = {}
for dvName in funcSens_local.keys():
funcSens[dvName] = comm.allreduce(funcSens_local[dvName], op=MPI.SUM)
else:
funcSens = funcSens_local
return funcSens
def totalSensitivityProd(self, vec, ptSetName, **kwargs):
r"""
This function computes sensitivity information.
Specifically, it computes the following:
:math:`\frac{dX_{pt}}{dX_{DV}} \times\mathrm{vec}`.
This is useful for forward AD mode.
Parameters
----------
vec : dictionary whose keys are the design variable names, and whose
values are the derivative seeds of the corresponding design variable.
ptSetName : str
The name of set of points we are dealing with
\\*\\*kwargs
Any other parameters ignored, but this is maintained to allow the same
interface as other DVGeo implementations.
Returns
-------
xsdot : array (Nx3)
Array with derivative seeds of the surface nodes.
"""
# Unpack some useful variables
desVars = self._unpackDVs()
ptsX = self.points[ptSetName]["points"][:, self.xIdx]
xMax = self.points[ptSetName]["xMax"]
xMin = self.points[ptSetName]["xMin"]
scaledX = (ptsX - xMin) / (xMax - xMin)
idxUpper = self.points[ptSetName]["upper"]
idxLower = self.points[ptSetName]["lower"]
idxTE = np.full((self.points[ptSetName]["points"].shape[0],), True, dtype=bool)
idxTE[idxUpper] = False
idxTE[idxLower] = False
xsdot = np.zeros_like(self.points[ptSetName]["points"], dtype=self.dtype)
for dvName, dvSeed in vec.items():
dvType = self.DVs[dvName].type
if dvType == "upper":
dydUpperCST = self.computeCSTdydw(
scaledX[idxUpper], desVars["n1_upper"], desVars["n2_upper"], desVars["upper"], dtype=self.dtype
)
dydUpperCST *= desVars["chord"]
xsdot[idxUpper, self.yIdx] += dydUpperCST.T @ dvSeed
if dvType == "lower":
dydLowerCST = self.computeCSTdydw(
scaledX[idxLower], desVars["n1_lower"], desVars["n2_lower"], desVars["lower"], dtype=self.dtype
)
dydLowerCST *= desVars["chord"]
xsdot[idxLower, self.yIdx] += dydLowerCST.T @ dvSeed
if dvType == "n1_upper" or dvType == "n1":
xsdot[idxUpper, self.yIdx] += (
dvSeed
* desVars["chord"]
* self.computeCSTdydN1(
scaledX[idxUpper], desVars["n1_upper"], desVars["n2_upper"], desVars["upper"], dtype=self.dtype
)
)
if dvType == "n2_upper" or dvType == "n2":
xsdot[idxUpper, self.yIdx] += (
dvSeed
* desVars["chord"]
* self.computeCSTdydN2(
scaledX[idxUpper], desVars["n1_upper"], desVars["n2_upper"], desVars["upper"], dtype=self.dtype
)
)
if dvType == "n1_lower" or dvType == "n1":
xsdot[idxLower, self.yIdx] += (
dvSeed
* desVars["chord"]
* self.computeCSTdydN1(
scaledX[idxLower], desVars["n1_lower"], desVars["n2_lower"], desVars["lower"], dtype=self.dtype
)
)
if dvType == "n2_lower" or dvType == "n2":
xsdot[idxLower, self.yIdx] += (
dvSeed
* desVars["chord"]
* self.computeCSTdydN2(
scaledX[idxLower], desVars["n1_lower"], desVars["n2_lower"], desVars["lower"], dtype=self.dtype
)
)
if dvType == "chord":
dydchord = self.points[ptSetName]["points"][:, self.yIdx] / desVars["chord"]
dxdchord = (ptsX - xMin) / desVars["chord"]
xsdot[:, self.yIdx] += dvSeed * dydchord
xsdot[:, self.xIdx] += dvSeed * dxdchord
return xsdot
[docs]
def addVariablesPyOpt(self, optProb):
"""
Add the current set of variables to the optProb object.
Parameters
----------
optProb : pyOpt_optimization class
Optimization problem definition to which variables are added
"""
# Add design variables to the problem
for DV in self.DVs.values():
optProb.addVarGroup(
DV.name,
DV.nVal,
"c",
value=DV.value,
lower=DV.lower,
upper=DV.upper,
scale=DV.scale,
)
[docs]
def update(self, ptSetName, **kwargs):
"""
This is the main routine for returning coordinates that have
been updated by design variables.
Parameters
----------
ptSetName : str
Name of point-set to return. This must match ones of the
given in an :func:`addPointSet()` call.
\\*\\*kwargs
Any other parameters ignored, but this is maintained to allow the same
interface as other DVGeo implementations.
Returns
-------
points : ndarray (N x 3)
Updated point set coordinates.
"""
desVars = self._unpackDVs()
# Unpack the points to make variable names more accessible
idxUpper = self.points[ptSetName]["upper"]
idxLower = self.points[ptSetName]["lower"]
idxTE = np.full((self.points[ptSetName]["points"].shape[0],), True, dtype=bool)
idxTE[idxUpper] = False
idxTE[idxLower] = False
points = self.points[ptSetName]["points"]
ptsX = points[:, self.xIdx]
ptsY = points[:, self.yIdx]
xMax = self.points[ptSetName]["xMax"]
xMin = self.points[ptSetName]["xMin"]
yUpperTE = self.points[ptSetName]["yUpperTE"]
yLowerTE = self.points[ptSetName]["yLowerTE"]
# Scale the airfoil to the range 0 to 1 in x direction
shift = xMin
chord = xMax - xMin
scaledX = (ptsX - shift) / chord
scaledYUpperTE = yUpperTE / chord
scaledYLowerTE = yLowerTE / chord
ptsY[idxUpper] = desVars["chord"] * self.computeCSTCoordinates(
scaledX[idxUpper],
desVars["n1_upper"],
desVars["n2_upper"],
desVars["upper"],
scaledYUpperTE,
dtype=self.dtype,
)
ptsY[idxLower] = desVars["chord"] * self.computeCSTCoordinates(
scaledX[idxLower],
desVars["n1_lower"],
desVars["n2_lower"],
desVars["lower"],
scaledYLowerTE,
dtype=self.dtype,
)
ptsY[idxTE] *= desVars["chord"] / chord
# Scale the chord according to the chord DV
points[:, self.xIdx] = (points[:, self.xIdx] - shift) * desVars["chord"] / chord + shift
# Scale the point set's properties based on the new chord length
self.points[ptSetName]["xMax"] = (xMax - shift) * desVars["chord"] / chord + shift
self.points[ptSetName]["yUpperTE"] *= desVars["chord"] / chord
self.points[ptSetName]["yLowerTE"] *= desVars["chord"] / chord
self.updated[ptSetName] = True
return points.copy()
[docs]
def getNDV(self):
"""
Return the total number of design variables this object has.
Returns
-------
nDV : int
Total number of design variables
"""
nDV = 0
for DV in self.DVs.values():
nDV += DV.nVal
return nDV
[docs]
def printDesignVariables(self):
"""
Print a formatted list of design variables to the screen
"""
if self.comm.rank == 0:
print("\nDVGeometryCST design variables")
print("==============================")
for DV in self.DVs.values():
print(f"{DV.name} ({DV.type} type): {DV.value}")
print("")
def _unpackDVs(self):
"""
Return the parameters needed for the airfoil shape calculation
based on the DVs and default values. This requires a few extra
checks to handle the multiple ways of parameterizing the class
shape variables.
Returns
-------
desVars : dict
Dictionary containing the following airfoil shape parameters:
`"upper"`: upper surface CST coefficients
`"lower"`: lower surface CST coefficients
`"n1_lower"`: first class shape parameter on lower surface
`"n2_lower"`: second class shape parameter on lower surface
`"n1_upper"`: first class shape parameter on upper surface
`"n2_upper"`: second class shape parameter on upper surface
`"chord"`: chord length
"""
desVars = {}
desVars["upper"] = self.defaultDV["upper"].copy()
desVars["lower"] = self.defaultDV["lower"].copy()
desVars["n1_upper"] = self.defaultDV["n1_upper"].copy()
desVars["n2_upper"] = self.defaultDV["n2_upper"].copy()
desVars["n1_lower"] = self.defaultDV["n1_lower"].copy()
desVars["n2_lower"] = self.defaultDV["n2_lower"].copy()
desVars["chord"] = self.defaultDV["chord"].copy()
for DV in self.DVs.values():
if DV.type in ["n1", "n2"]:
desVars[f"{DV.type}_upper"] = DV.value
desVars[f"{DV.type}_lower"] = DV.value
else:
desVars[DV.type] = DV.value
return desVars
def _splitUpperLower(self, points):
"""
Figure out the indices of points on the upper and lower
surfaces of the airfoil. This requires that the attributes
self.xMax, self.lowerSpline, self.upperSpline, self.xIdx,
and self.yIdx have already been set.
Parameters
----------
points : ndarray (Npts x 3)
Point array to separate upper and lower surfaces
Returns
-------
ndarray (1D)
Indices of upper surface points (correspond to rows in points)
ndarray (1D)
Indices of lower surface points (correspond to rows in points)
"""
# Determine which surface (either upper, lower, or trailing edge) each point is
# on based on which spline it is closest to
# Upper (if it's complex, ignore the imaginary part since the spline doesn't handle that)
_, upperDist = self.upperSpline.projectPoint(np.real(points[:, [self.xIdx, self.yIdx]]))
upperDist = np.linalg.norm(upperDist, axis=1)
# Lower
_, lowerDist = self.lowerSpline.projectPoint(np.real(points[:, [self.xIdx, self.yIdx]]))
lowerDist = np.linalg.norm(lowerDist, axis=1)
# Trailing edge
teDist = np.full_like(upperDist, np.inf)
if not self.sharp:
x0 = points[:, self.xIdx]
y0 = points[:, self.yIdx]
x1 = self.coordLowerTE[self.xIdx]
y1 = self.coordLowerTE[self.yIdx]
x2 = self.coordUpperTE[self.xIdx]
y2 = self.coordUpperTE[self.yIdx]
teDist = np.abs((x2 - x1) * (y1 - y0) - (x1 - x0) * (y2 - y1)) / np.sqrt((x2 - x1) ** 2 + (y2 - y1) ** 2)
# Determine if each point is on the upper surface if it's closer to the upper spline than
# either the lower spline or the trailing edge line (and do the same for the lower surface)
upperBool = np.logical_and(upperDist <= lowerDist, upperDist <= teDist)
lowerBool = np.logical_and(lowerDist < upperDist, lowerDist <= teDist)
return np.where(upperBool)[0], np.where(lowerBool)[0]
[docs]
@staticmethod
def computeCSTCoordinates(x, N1, N2, w, yte, dtype=float):
"""
Compute the vertical coordinates of a CST curve.
This function assumes x has been normalized to the range [0,1]
and that yte has been normalized by the chord.
Parameters
----------
x : ndarray (# pts,)
x coordinates at which to compute the CST curve height
N1 : float
First class shape parameter
N2 : float
Second class shape parameter
w : ndarray (# coeff,)
CST coefficient array
yte : float
y coordinate of the trailing edge (used to define trailing edge thickness).
Note that the trailing edge will be twice this thick, assuming the same ``yte``
value is used for both the upper and lower surfaces.
dtype : type, optional
Type for instantiated arrays, by default float
Returns
-------
ndarray (# pts,)
y coordinates of the CST curve
"""
C = DVGeometryCST.computeClassShape(x, N1, N2, dtype=dtype)
S = DVGeometryCST.computeShapeFunctions(x, w, dtype=dtype)
return C * S.sum(axis=0) + yte * x
[docs]
@staticmethod
def computeClassShape(x, N1, N2, dtype=float):
"""
Compute the class shape of a CST curve
Parameters
----------
x : ndarray (# pts,)
x coordinates at which to compute the CST curve height
N1 : float
First class shape parameter
N2 : float
Second class shape parameter
dtype : type, optional
Type for instantiated arrays, by default float
Returns
-------
ndarray (# pts,)
y coordinates of the class shape
"""
C = np.zeros_like(x, dtype=dtype)
# 0 to the power of a complex number is undefined, so anywhere
# x is 0 or 1, just keep C as zero (doesn't change the result for real)
mask = np.logical_and(x != 0.0, x != 1.0)
C[mask] = x[mask] ** N1 * (1.0 - x[mask]) ** N2
return C
[docs]
@staticmethod
def computeShapeFunctions(x, w, dtype=float):
"""Compute the Bernstein polynomial shape function of a CST curve
This function assumes x has been normalized to the range [0,1].
Parameters
----------
x : ndarray (# pts,)
x coordinates at which to compute the CST curve height
w : ndarray (# coeff,)
CST coefficient array
dtype : type, optional
Type for instantiated arrays, by default float
Returns
-------
ndarray (# coeff, # pts)
Bernstein polynomials for each CST coefficient
"""
numCoeffs = len(w)
order = numCoeffs - 1
S = np.zeros((numCoeffs, len(x)), dtype=dtype)
facts = factorial(np.arange(0, order + 1))
for i in range(numCoeffs):
binom = facts[-1] / (facts[i] * facts[order - i])
S[i] = w[i] * binom * x ** (i) * (1.0 - x) ** (order - i)
return S
[docs]
@staticmethod
def computeCSTdydw(x, N1, N2, w, dtype=float):
r"""Compute the derivatives of the height of a CST curve with respect to the shape function coefficients
Given :math:`y = C(x) * sum [w_i * p_i(x)]`
:math:`\frac{dy}{dw_i} = C(x) * p_i(x)`
This function assumes x has been normalized to the range [0,1].
Only the shape of w is used, not the values.
Parameters
----------
x : ndarray (# pts,)
x coordinates at which to compute the CST curve height
N1 : float
First class shape parameter
N2 : float
Second class shape parameter
w : ndarray (# coeff,)
CST coefficient array
dtype : type, optional
Type for instantiated arrays, by default float
Returns
-------
ndarray (# coeff, # pts)
Derivatives of the y coordinates with respect to the CST coefficients
"""
C = DVGeometryCST.computeClassShape(x, N1, N2, dtype=dtype)
S = DVGeometryCST.computeShapeFunctions(x, np.ones_like(w), dtype=dtype)
return C * S
[docs]
@staticmethod
def computeCSTdydN1(x, N1, N2, w, dtype=float):
r"""Compute the derivatives of the height of a CST curve with respect to N1
Given :math:`y = C(x, N1, N2) * S(x)`
:math:`\frac{dy}{dN1} = S(x) * \frac{dC}{dN1} = S(x) * C(x, N1, N2) * \ln{x}`
This function assumes x has been normalised to the range [0,1].
Parameters
----------
x : ndarray (# pts,)
x coordinates at which to compute the CST curve height
N1 : float
First class shape parameter
N2 : float
Second class shape parameter
w : ndarray (# coeff,)
CST coefficient array
dtype : type, optional
Type for instantiated arrays, by default float
Returns
-------
ndarray (# pts,)
Derivative of the y coordinates with respect to the first class shape parameter
"""
C = DVGeometryCST.computeClassShape(x[x != 0.0], N1, N2, dtype=dtype)
S = DVGeometryCST.computeShapeFunctions(x[x != 0.0], w, dtype=dtype)
dydN1 = np.zeros_like(x, dtype=dtype)
dydN1[x != 0.0] = np.sum(S, axis=0) * C * np.log(x[x != 0.0])
return dydN1
[docs]
@staticmethod
def computeCSTdydN2(x, N1, N2, w, dtype=float):
r"""Compute the derivatives of the height of a CST curve with respect to N2
Given :math:`y = C(x, N1, N2) * S(x)`
:math:`\frac{dy}{dN2} = S(x) * \frac{dC}{dN2} = S(x) * C(x, N1, N2) * \ln(1-x)`
This function assumes x has been normalised to the range [0,1].
Parameters
----------
x : ndarray (# pts,)
x coordinates at which to compute the CST curve height
N1 : float
First class shape parameter
N2 : float
Second class shape parameter
w : ndarray (# coeff,)
CST coefficient array
dtype : type, optional
Type for instantiated arrays, by default float
Returns
-------
ndarray (# pts,)
Derivative of the y coordinates with respect to the second class shape parameter
"""
C = DVGeometryCST.computeClassShape(x[x != 1.0], N1, N2, dtype=dtype)
S = DVGeometryCST.computeShapeFunctions(x[x != 1.0], w, dtype=dtype)
dydN2 = np.zeros_like(x, dtype=dtype)
dydN2[x != 1.0] = np.sum(S, axis=0) * C * np.log(1 - x[x != 1.0])
return dydN2
[docs]
@staticmethod
def computeCSTfromCoords(xCoord, yCoord, nCST, N1=0.5, N2=1.0, dtype=float):
"""
Compute the CST coefficients that fit a set of airfoil
coordinates (either for the upper or lower surface, not both).
This function internally normalizes the x and y-coordinates.
Parameters
----------
xCoord : ndarray
Upper or lower surface airfoil x-coordinates (same length
as yCoord vector).
yCoord : ndarray
Upper or lower surface airfoil y-coordinates (same length
as xCoord vector).
nCST : int
Number of CST coefficients to fit.
N1 : float, optional
First class shape parameter to assume in fitting, by default 0.5
N2 : float, optional
Second class shape parameter to assume in fitting, by default 1.0
dtype : type, optional
Type for instantiated arrays, by default float
Returns
-------
np.ndarray (nCST,)
CST coefficients fit to the airfoil surface.
"""
# Normalize x and y
chord = np.max(xCoord) - np.min(xCoord)
xCoord = (xCoord - np.min(xCoord)) / chord
yCoord = yCoord / chord
# Find the y-coordinate at the trailing edge
idxTE = np.argmax(xCoord)
yTE = yCoord[idxTE]
# Subtract the linear TE thickness function
yMinusTE = yCoord - yTE * xCoord
# Compute the coefficients via linear least squares
dydw = DVGeometryCST.computeCSTdydw(xCoord, N1, N2, np.ones(nCST), dtype=dtype)
w = np.linalg.lstsq(dydw.T, yMinusTE, rcond=None)[0]
return w
[docs]
@staticmethod
def plotCST(upperCoeff, lowerCoeff, yUpperTE=0.0, yLowerTE=0.0, N1=0.5, N2=1.0, nPts=100, ax=None, **kwargs):
"""Simple utility to generate a plot from CST coefficients.
Parameters
----------
upperCoeff : ndarray
One dimensional array of CST coefficients for the upper surface.
lowerCoeff : ndarray
One dimensional array of CST coefficients for the lower surface.
yUpperTE : float, optional
y-coordinate for the upper surface trailing edge point.
yLowerTE : float, optional
y-coordinate for the lower surface trailing edge point.
N1 : float, optional
First class shape parameter.
N2 : float, optional
Second class shape parameter.
nPts : int, optional
Number of coordinates to compute on each surface.
ax : matplotlib Axes, optional
Axes on which to plot airfoil.
\\*\\*kwargs
Keyword arguments passed to matplotlib.pyplot.plot
Returns
-------
matplotlib Axes
Axes with airfoil plotted
"""
if not pltImport:
raise ImportError("matplotlib could not be imported and is required for plotCST")
if ax is None:
_ = plt.figure()
ax = plt.gca()
x = np.linspace(0, 1, nPts)
yUpper = DVGeometryCST.computeCSTCoordinates(x, N1, N2, upperCoeff, yUpperTE)
yLower = DVGeometryCST.computeCSTCoordinates(x, N1, N2, lowerCoeff, yLowerTE)
ax.plot(x, yUpper, **kwargs)
ax.plot(x, yLower, **kwargs)
ax.set_aspect("equal")
return ax