Configuration file format¶
The working example below illustrates the general layout of a turbigen configuration file. More specimens are available in the examples directory.
# Annotated minimal example configuration to run a turbine cascade
#
# All files relating to the case are held in a working directory
workdir: runs/cascade
# Perfect gas inlet state
inlet:
Po: 1e5 # Stagnation pressure [Pa]
To: 300. # Stagnation temperature [K]
cp: 1005. # Specific heat capacity at constant pressure [J/kg/K]
mu: 1.8e-5 # Working fluid dynamic viscosity [kg/m/s]
gamma: 1.4 # Working fluid ratio of specific heats [-]
# Mean-line design
mean_line:
type: cascade # Choose turbomachine from turbigen.meanline
# Aerodynamic parameters required for mean-line design
span1: 0.1
span2: 0.12
Alpha1: 40.
Alpha2: -65.
Ma2: 0.6
Yh: 0.1
htr: 0.9
# Annulus shape using meridional aspect ratios
annulus:
AR_gap: [0.8, 0.9] # Span to inlet/exit boundary distance
AR_chord: 1.6 # Span to chord
# Blade shapes
blades:
# List of rows
- Co: 0.7 # Set number of blades using circulation coeff
sections:
# List of sections for this row
- spf: 0.5 # Span fraction
# Vector of thickness parameters
q_thick: [0.05, 0.12, 0.3, 0.02, 0.02, 0.18]
# Vector of camber parameters
qstar_camber: [0., 0., 1.0, 1.0, 0.0]
# - ...
# - ...
# Computational mesh generation
mesh:
type: h # Mesh topology
yplus: 30.0 # Non-dimensional wall distance
resolution_factor: 0.5 # Double grid spacing in each direction
# CFD solver
solver:
type: ts3 # Which solver
# Solver-specific configuration settings
nstep: 5000
nstep_avg: 1000
rtol_mdot: 0.02
# ...
post_process:
# contour_traverse:
# mnorm: [2.1,]
# coord_sys: yz
# var: ['Ys', 'Vm']
plot_nose:
row_spf: [[0.6,],]
# plot_section:
# row_spf: [[0.2,0.5,0.8],]
# coord_sys: xrt
# plot_pressure_distributions:
# row_spf: [[0.6,],]
# write_raw: true
# plot_incidence:
# plot_annulus:
# show_control_points: false
# write_raw: true
Input data is specified in a text-based YAML format. Comments are prefixed with #, and indentation is significant. A configuration file may contain the following top-level keys:
The following sections give more detail on each of the keys.
workdir¶
Specifies relatice or absolute path to a working directory for the run. If present, * is replaced with a new numbered directory on each invocation of the program. Examples:
workdir: runs/axial_turbine
workdir: /home/james/turbigen/runs/radial
workdir: design_*
# produces design_0000, design_0001, ... each time turbigen is run
inlet¶
The thermodynamic state at inlet to the turbomachine is set by fixing stagnation pressure, Po, and stagnation temperature To. For perfect gases, the fluid properties cp, mu, and gamma can be directly prescribed. For real gases, only the name of the working fluid in CoolProp nomenclature is required, and the fluid properties are obtained from tabulations. Examples:
# Perfect gas
inlet:
Po: 1e5 # Stagnation pressure [Pa]
To: 300. # Stagnation temperature [K]
cp: 1005. # Specific heat capacity at constant pressure [J/kg/K]
mu: 1.8e-5 # Working fluid dynamic viscosity [kg/m/s]
gamma: 1.4 # Working fluid ratio of specific heats [-]
# Real gas
inlet:
Po: 1e5 # Stagnation pressure [Pa]
To: 400. # Stagnation temperature [K]
fluid_name: water # Name of working fluid
For more information on how turbigen abstracts the working fluid, see turbigen.fluid
mean_line¶
Under the mean_line key, a type key is required to specify the architecture
or topology of turbomachine to design. This can either be: one the built-in
mean-lines listed in in turbigen.meanline, or a file path to a
user-written mean-line module that provides the correct functions. Also
specified are the aerodynamic or geometric paramters needed to solve for the
flow along the mean-line for the current turbomachine type. The design
variables are fed directly into the forward mean-line design functions in
turbigen.meanline (see the Mean line page for more
detail) The configuration file might look like:
mean_line:
type: axial_turbine # Single-stage axial turbine
Alpha1: 0. # Inlet yaw angle [deg]
phi: 0.8 # Flow coefficient
psi: 1.6 # Stage loading coefficient
Lam: 0.5 # Degree of reaction
Ma2: 0.6 # Vane exit Mach number
eta: 0.92 # Guess of polytropic efficiency
loss_split: 0.4 # Fraction of entropy rise in stator
htr: 0.9 # Hub-to-tip radius ratio
Omega: 314.159 # Shaft angular velocity [rad/s]
mean_line:
type: radial_impeller # Radial compressor with vaneless diffuser
Alpha1: 0.0 # Inlet yaw angle [deg]
Alpharel2: -50.0 # Rotor exit yaw angle [deg]
DHdiff: 0.75 # Diffuser de Haller
DHimp: 0.9 # Rotor de Haller
Marel1: 0.6 # Inlet relative Mach number
PR_tt: 2.0 # Total-to-total pressure ratio
eta_tt: 0.91 # Total-to-total isentropic efficiency
htr1: 0.5 # Inlet hub-to-tip radius ratio
loss_split: 0.8 # Fraction of entropy rise in rotor
mdot: 5.0 # Mass flow rate [kg/s]
phi1: 0.6 # Inlet flow coefficient
annulus¶
With mean radii and areas fixed by the mean-line deisgn, meridional aspect ratios set the length of the machine. For a turbomachine with N rows: there are N+1 aspect ratios defining the spacing between rows and the inlet and exit boundaries set in AR_gap; and N aspect ratios defining the blade chords in AR_chord. turbigen generates a smooth curvature continuous annulus between these defining points. For example, in a single-stage axial cases:
annulus:
AR_gap: [0.6, 2., 0.7]
AR_chord: [1.6, 1.6]
There are two special values that the meridional aspect ratios can take. For fully radial flow with a pitch angle of \(\beta=\pm 90 ^\circ\), blade chords are already fixed by the inlet and exit radii from mean-line design, so must be set to .nan in the configuration file. To optimise axial length to minimise integrated curvature over the hub and casing, set a negative aspect ratio.
For a radial compressor, we want to optimise the rotor length to minimise curvature, but the radial extent of the stator and vaneless space is already determined by the mean-line design. This would be specified as:
annulus:
AR_chord: [-1.0, .nan]
AR_gap: [0.5, .nan, 0.5]
blades¶
Under the blades key is a list of data for each blade row, then a section key specifies geometry for each section. The section data requires a span fraction spf and vectors of thickness and camber parameters, q_thick and qstar_camber:
# Span fraction for this section
spf: 0.5
# Vector of thickness parameters
q_thick: [0.05, 0.12, 0.3, 0.02, 0.02, 0.18]
# Vector of camber parameters
qstar_camber: [0., 0., 1.0, 1.0, 0.0]
- q_thick is a vector with 6 elements:
Leading-edge radius
Maximum thickness
Location of maximum thickness
Curvature at maximum thickness
Trailing edge thickness
Trailing edge wedge angle tangent
- qstar_camber is a vector with 5 elements:
Leading-edge recamber with respect to flow angle
Trailing-edge recamber with respect to flow angle
Camber line slope at leading edge
Camber line slope at trailing edge
Camber line curvature at mid-chord
All lengths are normalised by the meridional chord.
The number of blades can be set via three choices: directly using Nb, a non-dimensional circulation using Co, or the Lieblein diffusion factor DFL. Thickness and camber vectors are interpolated between each of the specified span fractions.
Tip gaps can be modelled by adding a tip key to the row configuration. The gap is normalised by the mean of the inlet and exit span.
A complete specification of the blade geometry is:
blades:
# First row
- Nb: 65 # Set number of blades directly
# Set blade geometry on two sections
sections:
- spf: 0.2 # Hub section
q_thick: [0.05, 0.12, 0.3, 0.02, 0.02, 0.18]
qstar_camber: [0., 0., 1.0, 1.0, 0.0]
- spf: 0.8 # Tip section
q_thick: [0.05, 0.12, 0.3, 0.02, 0.02, 0.18]
qstar_camber: [0., 0., 1.0, 1.0, 0.0]
# Second row
- Co: 0.7 # Set number of blades using circulation coeff
tip: 0.01 # 1% of span tip gap
# Prismatic blade section
sections:
- spf: 0.5 # Midspan
q_thick: [0.05, 0.12, 0.3, 0.02, 0.02, 0.18]
qstar_camber: [0., 0., 1.0, 1.0, 0.0]
mesh¶
To generate a computational mesh, the topology must be specified in a type key; currently h and oh are the only supported values. Following the type, settings to control, for example, mesh resolution and clustering can optionally be specified. The possible settings and their default values are listed in Meshing.
H-meshing requires less configuration and is more robust to consistently automate across a design space. With some extra settings, this could look like:
mesh:
type: h
yplus: 60.0
# Optional settings
dm_LE: 0.0001
nchord_relax: 0.5
resolution_factor: 0.5
# ...
OH-meshing, at present, requires SSH access to the proprietary AutoGrid software on a remote machine, and is more complicated to set up and debug. More information is given in OH-meshing.
mesh:
type: oh
yplus: 60.0
remote_host: gp-111 # Hostname of AutoGrid machine
# Optional settings
nk_omesh: 33
untwist_outlet: true
# ...
solver¶
The solver key can be omitted entirely to skip running CFD, and just post-process the initial guess solution, which can facilitate debugging. If present, the solver section requires a type key to indicate which CFD code to use, and then solver-specific configurations settings may be given, with default values used otherwise. Solvers gives more detail. Sample configurations for Turbostream 3 and 4 are:
# Turbostream 3
solver:
type: ts3
nstep: 5000
nstep_avg: 1000
# Optional settings
dampin: 10.
# ...
# Turbostream 4
solver:
type: ts4
nstep: 3000
nstep_avg: 500
# Optional settings
cfl_ramp_nstep: 2000
# ...
operating_point¶
To set the mass flow rate to the design value using a PID controller on exit static pressure, specify the PID constants under an operating_point key using:
operating_point:
mdot_pid: [0.5, 0.1, 0.0]
The constants are non-dimensionalised with the nominal duty, so values of order one are a good first guess. Smaller values are more robust at the expense of longer convergence times.
To run the CFD off-design, keys mass_adjust and rpm_adjust can be added to change the mass flow rate and shaft speed respectively, relative to the nominal values. For example, to throttle to 90% of the design mass flow at 110% of the design shaft speed:
operating_point:
mdot_pid: [0.5, 0.1, 0.0]
mass_adjust: -0.1 # 10% below design mass flow
rpm_adjust: 0.1 # 10% above design rpm
iterate¶
turbigen has the capability to iterate on unknown values of mean-line loss, deviation, and incidence until they match CFD results. Each of the three can be configured individually under the iterate key.
To match mean-line loss to CFD results, include a mean_line key. Under match_tolerance, for each design variable that should be adjusted, is an absolute tolerance on the difference between nominal mean-line and mixed-out CFD values. Every iteration, the change in the design variables is multiplied by a relaxation_factor, which may be reduced if the iterations are unstable. If all variables are within tolerance, the iteration terminates.
The example below is for a radial compressor, where mean-line design requires guesses of total-to-total efficiency, eta_tt, and rotor entropy rise fraction, loss_split:
iterate:
mean_line:
match_tolerance:
eta_tt: 0.005 # Efficiency to within 0.5%
loss_split: 0.05 # Loss fraction to within 5%
relaxation_factor: 0.5 # Change is half difference CFD-nominal
To match mean-line flow angle to the mixed-out CFD value, include a deviation key. The trailing-edge recamber of all rows is iterated simultaneously. An example configuration is:
iterate:
deviation:
clip: 5.0 # Maximum recamber in one step
relaxation_factor: 0.8 # Multiplier on changes to metal angle
tolerance: 0.5 # Absolute tolerance for termination in degrees
To locate the stagnation point on the nose of the blades, include an incidence key. This will recamber the leading edges of all blade sections defined for all rows simultaneously. The configuration is:
iterate:
incidence:
clip: 5.0 # Maximum recamber in one step
relaxation_factor: 0.2 # Multiplier on changes to metal angle
target: 0.0 # Desired stagnation point angle wrt camber line
tolerance: 5.0 # Absolute tolerance for termination in degrees
rtol_mdot: 0.05 # Recamber when mdot this close to nominal
In the event of instability, clip and relaxation_factor can be reduced for all three adjustments.
A complete iteration configuration looks something like this:
iterate:
deviation:
clip: 5.0
relaxation_factor: 2.0
tolerance: 1.0
incidence:
clip: 2.0
relaxation_factor: 0.2
target: 0.0
tolerance: 5.0
mean_line:
match_tolerance:
eta_tt: 0.005
loss_split: 0.05
relaxation_factor: 0.5
job¶
Automations are in place for running turbigen as a SLURM queue job using the sbatch command. Some parameters are hardcoded assuming the University of Cambridge Wilkes3 cluster and may require modifying for different systems. Sumbitting jobs requires the following configuration:
job:
account: NAME-SL2-GPU # SLURM account to charge
hours: 4 # Integer number of hours for job time limit
tasks: 2 # Number of GPUs