FAQ

Here is a list of frequently asked questions I've put together based on
feedback from the Xoptfoil thread on RCGroups:

http://www.rcgroups.com/forums/showthread.php?t=2095696

The questions are just listed directly below for reference.  The full questions/
answers are below that.

1) Where do I get detailed information on setting up and running Xoptfoil?
2) On Windows, When I double-click the Xoptfoil executable to run it, a window 
   pops up but then closes right away.  What should I do?
3) What's the difference between min-drag, min-sink, max-lift, max-xtr, and
   max-lift-slope?
4) My output airfoils look wavy or bumpy.  What can I do?
5) What if my seed airfoil doesn't pass all the constraints?
6) What are all the different output files for?
7) Can I stop an optimization that's taking a long time and restart it later?
8) How do I plot the results?
9) How do shape functions work?
10) What's the difference between Hicks-Henne and NACA shape functions?
11) How does flap optimization work?
12) What if I just want to find out the optimal flap settings for an existing
    airfoil, without modifying its shape?
13) What's the difference between particle swarm and genetic algorithm?
14) What's the difference between global and local optimization?
15) How do I optimize for high lift?
16) What if most of the initial designs can't be initialized in the feasible
    space?

================================================================================

1) Where do I get detailed information on setting up and running Xoptfoil?

   You can find out pretty much everything you need to know by looking at the
   user guide, which can be found in the doc directory.

2) On Windows, When I double-click the Xoptfoil executable to run it, a window 
   pops up but then closes right away.  What should I do?

   It probably means that there is an error in your input file, or maybe one of 
   the required input files is not present.  You should run Xoptfoil from the 
   command prompt, which will prevent the window from closing, so you'll be able
   to see the what the error was.  You can find instructions for how to run it 
   from the command prompt in the user guide (see Section 2: Running Xoptfoil).
   This is the recommended way to run it.

3) What's the difference between min-drag, min-sink, max-lift, max-xtr, and
   max-lift-slope?

   Min-drag minimizes Cd at a given operating point.  Max-lift maximizes Cl.
   Min-sink maximizes Cl^1.5/Cd.  Max-xtr moves the turbulent transition point
   as far aft as possible. Max-lift-slope attempts to maximize the lift curve
   slope at the operating point where it is applied. Note that max-lift and
   min-sink are equivalent to min-drag for any operating points where Cl is
   specified (using op_mode = 'spec-cl'), because Cl is no longer a variable.
   Since in most cases the designer knows what Cl should be, but not necessarily
   the angle of attack, most of the time min-drag is the best choice.  One
   exception is described in question 13: "How do I optimize for high lift?" and
   there are probably others as well.

4) My output airfoils look wavy or bumpy.  What can I do?

   There is a constraint that can be used to limit the number of curvature
   reversals on the top and bottom surfaces.  The relevant settings are
   check_curvature, max_curv_reverse_top, max_curv_reverse_bot, and 
   curv_threshold.  You can find out more information about these in the user
   guide.  Curvature checks are disabled by default because they limit the
   design space, but sometimes they are necessary.  It's also possible to get
   small bumps if you use Hicks-Henne shape functions and you made 
   min_bump_width too small.

5) What if my seed airfoil doesn't pass all the constraints?

   It is best to have a seed airfoil that passes all the constraints.  In fact,
   it's best to start with a seed airfoil that is already well-suited for the
   operating points you are targeting.  However, you can still run an
   optimization even if the seed airfoil doesn't pass all the constraints, and
   in most cases it will still produce a good airfoil.  Be aware that the lift-
   drag polar may not be improved over the seed airfoil's, because you are
   imposing constraints on the optimization that the seed airfoil did not need
   to meet.  The optimization progress that is printed out to the screen may
   still indicate that the performance is improved by a large amount, however.  
   This is because large penalties are added to the objective function any time 
   a design violates a constraint, including the seed airfoil.

6) What are all the different output files for?

   All the different files are described in the user guide (see section 2:
   Output files).  One of them gives the optimized airfoil coordinates, one 
   gives a summary of the performance, and the rest are either for restarting
   or for plotting the optimization progress.

7) Can I stop an optimization that's taking a long time and restart it later?

   You can stop Xoptfoil by writing the command "stop" (without quotes) in the
   run_control file. After that, you can restart it by setting restart = .true.
   in the input file. 

8) How do I plot the results?

   You can use the included xoptfoil_visualizer tool.  You need to have Python
   installed to use it.  See section 7: Using the xoptfoil_visualizer in the
   user guide for more information. Of course, you can also use the data from
   the output files in whatever alternative plotting program you prefer.

9) How do shape functions work?

   You can think of shape functions as bumps that get added to the surface of
   the seed airfoil in different locations.  By placing these bumps with 
   particular sizes and shapes and in particular locations on the seed airfoil,
   a new airfoil is generated.  Xoptfoil varies the strengths of these bumps 
   (and, for Hicks-Henne shape functions, the widths and locations as well) to
   create new airfoils that optimize performance at the operating points you
   select.

10) What's the difference between Hicks-Henne and NACA shape functions?
 
    The main difference is that NACA shape functions are fixed in width and
    position once the number of shape functions is set.  Hicks-Henne, on the
    other hand, can move around and become wider or narrower.  Both can be
    made larger or smaller by changing the strength of the shape function.
    Because Hicks-Henne shape functions have more degrees of freedom, fewer of
    them are required to achieve the same or greater level of control in shaping
    new airfoils.  Typically, using 4 Hicks-Henne functions on both the top and 
    bottom surfaces works well, which results in the same number of degrees of
    freedom as (or "design variables") as 12 NACA functions.  Hicks-Henne is
    recommended because it tends to work better for the same number of design
    variables.

11) How does flap optimization work?

    If you want to optimize flaps, you must set use_flap = .true. in the
    operating_conditions namelist.  You then select a single x- and y-location
    for the flap, and at each operating point say whether you want the flap
    angle to be specified or optimized.  (Note that, even if you are optimizing
    the flap angle for a particular operating point, the value set in the
    flap_degrees variable is still used for the seed airfoil design.)  The sign
    convention is positive down, negative up.  Optimized flap angles are subject
    to the min_flap_degrees and max_flap_degrees constraints.  Each flap angle
    is associated to an individual operating point, whereas the shape of a given
    airfoil design applies to all operating points.

    To apply a flap deflection, the Xfoil subroutine FLAP is called.  This
    subroutine determines where the break on the upper and lower surfaces occurs
    based on the flap angle and the flap hinge location, rotates the portions of
    the airfoil aft of the break, and then "patches" the breaks in a smooth way.

12) What if I just want to find out the optimal flap settings for an existing
    airfoil, without modifying its shape?

    You can do this by setting both nfunctions_top and nfunctions_bot equal to
    0, setting use_flap = .true., and setting flap_selection = 'optimize' for at
    least one of the operating points.  Then, the optimized flap setting(s) will
    be the only design variable(s), and the optimized airfoil will look the same
    as the seed airfoil.

13) What's the difference between particle swarm and genetic algorithm?

    In some ways they are similar, but they are also very different.  Both use a
    population of designs that are initially generated randomly.  These designs
    then interact with each other to form better designs.  In particle swarm
    optimization, each "particle" has a velocity, an inertial parameter, and
    is attracted towards both the best design it has found and the best design
    that the entire swarm has found to different degrees.  Eventually, all of
    the particles in the population will converge on a single design.

    In the genetic algorithm, a number of designs are selected at each iteration
    to reproduce.  These are typically chosen in a way such that better designs
    have a better probability of reproducing.  These "parent" designs are paired
    off, and traits from both are combined to produce "offspring."  If the
    offspring perform better than some of the existing members of the
    population, those poorly-performing designs will be discarded.  This process
    also tends to converge on a single design, given enough iterations.

    As for which is better, there is no clear answer.  In testing, particle
    swarm optimization with the "exhuastive" profile finds the best solutions,
    on average.  The genetic algorithm sometimes performs as well or better, but
    in testing it has resulted in slightly poorer-performing optimized airfoils
    on average.  Both these options are much better (again, on average), than
    particle swarm with the "quick" convergence profile, but the latter still
    gives good results and converges in many fewer iterations.

14) What's the difference between global and local optimization?

    Traditional optimization methods use the local gradient of the objective 
    function to determine the direction to search.  Essentially, this means that
    they look around the current design and determine which small changes will
    improve the performance.  This approach has serious limitations for design
    spaces with many local minima (locations in the design space that may not be
    the global minimum, but nevertheless making any small changes to the design
    will hurt the performance).  Airfoil optimization is like that.  Therefore,
    instead of using a gradient-based local search method, global search methods
    such as particle swarm optimization and genetic algorithms are utilized
    instead.  These methods are able to ignore local minima by exchanging
    information amongst many candidate designs scattered all over the design
    space.
    
    Because airfoil optimization problems have many local minima, the global 
    search methods are recommended.  Xoptfoil does have a local search method as
    well, which sometimes can hone in on the minimum, once it's in the right
    vicinity, more efficiently than the global search methods.  The local search
    method implemented in Xoptfoil is the Nelder-Mead simplex algorithm, which
    is not actually gradient-based but still is a local search method.  It is
    not recommended to only use the local search when optimizing an airfoil
    design (therefore, use search_type = 'global_and_local' or 'global', but not
    'local').  If you do use only the local search, it is likely that the final
    airfoil created by the optimizer will not be much different (or much better
    than) the seed airfoil.
    
15) How do I optimize for high lift?

    Optimizing for high lift is a tough problem, but it can be done.  The reason
    that it is difficult is that it explores the limits of the airfoil's
    functional range; the lift peak occurs just before stall.  Since Xoptfoil
    penalizes operating points that don't converge, and stalled flows usually
    don't converge in Xfoil, it is hard for it to explore this boundary.
    Another practical difficulty is that in trying to maximize Clmax, the
    designer doesn't know beforehand either the alpha or Cl where this occurs.

    This background aside, it is still possible to maximize for high lift with
    Xoptfoil.  The following method is thanks to weps on RCGroups:

    -- Determine the angle of attack where Clmax occurs for the seed airfoil
       using Xfoil (or one of the tools that packages Xfoil, including the
       xfoil_only program that comes with Xoptfoil).
    -- Set an operating point at the angle of attack determined in the previous
       step (be sure to set op_mode = 'spec-al' at this point). Use the max-lift
       optimization type for this operating point.
    -- Be careful with pitching moment constraints for the point where you are
       trying to maximize lift.  For example, if moment_constraint_type is
       'use_seed', you will constrain the pitching moment to be no more negative
       than the seed airfoil's at this point, which could limit the ability to
       really maximize lift, and it could also result in funny shapes, e.g.
       reflex being added to satisfy the pitching moment constraint.

16) What if most of the initial designs can't be initialized in the feasible
    space?
 
    This is usually due to having constraints that are too limiting.  If you are
    running into this problem, first try relaxing the geometric constraints such
    as the thickness and curvature constraints.  It can also be caused by
    setting unrealistic operating points; i.e., specifying a lift coefficient
    that is impossibly high.