At burgeon, we have developed a range of optimization software. Our flagship product is called IMIT, version 9.2 is now available. Other products include a modified genetic algorithm and a genetic programming software, as well as a range of conventional optimization techniques like SQP, random search, etc., etc.

The optimization system IMIT Version 9.2 is able to tackle real-life design optimization problems, many of which have the same common features:

• The objective and constraint functions are evaluated as a result of expensive numerical computations, e.g. using a FE or a CFD package;
• Function values and/or their derivatives may contain some level of numerical noise. It means that function values, corresponding to two slightly different sets of design variables, may have some random variation.
• There can be situations when for some combinations of the design variables it is not possible to calculate the value of some of the functions in the optimization problem, e.g. when the numerical simulation model used for the response analysis goes beyond the limits of its applicability producing an abnormal stop without calculating any response value.

All these features are covered by the optimization system IMIT Version 9.2. The method behind the system is called MARS - Multipoint Approximations based on Response Surface fitting. The interface of the system IMIT-9.2 with external programs, which are used to return the values of the objective and constraint functions as a response to a current set of design variables (most typically, a commercial finite element package like ANSYS, ABAQUS, or CFD packages, etc.), is organized via ASCII files.

Here is a description of an optimization loop (a batch file on a PC or a run file in UNIX):

• the optimizer is called, it produces a current set of design variables as an ASCII file and stops;
• the simulation software (e.g. FEA or CFD package) reads this set of variables, does the analysis thus producing the values of the objective and constraint functions, and writes them to an ASCII file;
• the optimizer starts again, reads the values of the objective and constraint functions and generates a new set of design variables;
• this is repeated until the convergence is achieved.

Accordingly, a user does not need to recompile the system IMIT-9.2 for any new application. If you are familiar with the data structure of your simulation software, it would typically take an hour to link it to IMIT. It the past it has been linked to commercial FEA codes ANSYS, ABAQUS, a commercial CFD code BVGK and many other codes.

Here are the main features of IMIT-9.2:

• it dramatically reduces the number of calls for the simulation software returning the objective and constraint function values in comparison with conventional optimization techniques;
• it does not require derivatives of the objective and constraint functions but makes efficient use of them when the derivatives are available from a design sensitivity analysis;
• when a number of design variables NV grows, the number of calls for the objective and constraint function values NC grows no faster than linearly. Here is an estimation for NC:
• when only the function values are available (i.e. no derivatives used), NC = NI * (N+1) where NI is the number of iterations to converge (typically about 8-12)
• when, in addition to the function values, the derivatives are available, NC = NI+1 where NI is the same as above but each call returns the value of each of the function and N first order derivatives with respect to the design variables.
• it tolerates a considerable level of numerical noise;
• it works even when for some combinations of the design variables an external simulation code (say, FEM or CFD) does not return any value of the functions.

Here are some of the recent applications of IMIT-9.2:

• Optimization of an aerofoil (collaboration with BAe)
• Optimization of a robotic assembly process (collaboration with TU Delft)
• Optimization of ride characteristics of DAF truck (collaboration with TU Delft)
• Optimization of a flight simulator (collaboration with TU Delft)
• Optimization of a shell roof structure (collaboration with TU Delft)
• Optimization of thermodynamic characteristics of a Stirling engine (collaboration with The Technical University of Denmark).

Interested? Contact us and we will send you a trial program for a PC which is fully operational but the number of design variables is limited to five. Before getting the program, you will be requested to confirm its non-commercial use by signing an agreement.

For those interested in the use of a modified genetic algorithm, here is an example of its application to design optimization of structural steelwork to BS5950.