Notes on Method Class Implementation

An implementation of the main interface method class (ISamplingMethod for DOE and IOptimizationMethod for optimization) is mandatory for a DesignXplorer extension to be valid. This method class is the core of the extension. It is the object that is delegated the responsibility to generate the sampling or solve the optimization study.

If you start a completely new implementation of the sampling or optimization algorithm, you might consider writing a C# class that derives directly from the method class, as provided by the PublicAPIs assembly.

If you start from an existing code, either in C/C++ or FORTRAN code, you should consider implementing your method class in IronPython as an adapter to your algorithm implementation, wrapping your existing classes and routines.

Both approaches are illustrated by the extension examples listed in DesignXplorer Extension Examples.

For full reference documentation of the API, see the "API Description" section in the ANSYS ACT API Reference Guide.

Notes on Monitoring

Notes on Sampling Monitoring

DesignXplorer provides user interface blocks that allow the user to monitor the progress of a sampling, in addition to the progress and log messages.

Notes on Optimization Monitoring

DesignXplorer provides user interface bocks that allow the user to monitor the progress of an optimization. In addition to the progress and log messages, the API allows the external optimizer to optionally provide DesignXplorer with progress data such as history and convergence values.

The history values are values of the input and output variables and show how the optimizer explores the parametric space. In the user interface, history values are rendered graphically, as a 2D chart per objective and input parameter, where the X axis represents points or iterations and the Y axis represents the parameter values. For more information, see "Using the History Chart" in the DesignXplorer User's Guide.

The X axis can be configured in the XML extension definition file to represent points or iterations using the attribute HistoryChartXAxisType for the optimizer block (byPoint or byIteration).

To provide history values during the execution of IOptimizationMethod.Run(), the optimization method can call IOptimizationServices.PushHistoryPoint(IOptimizationPoint point), where point contains a value for some or all enabled input and output variables.

The convergence values are independent from the variables. They are the values of one or more convergence metrics, specific to your optimizer and showing its convergence during the process. In the user interface, convergence values are rendered graphically, as a 2D chart with one or several curves. For more information, see "Using the Convergence Criteria Chart" in the ANSYS DesignXplorer User's Guide.

To provide convergence values during the execution of IOptimizationMethod.Run(), the optimization method can call IOptimizationServices.PushConvergenceData (IConvergenceData data), where data contains the values for one or several convergence criteria.

Notes on Results

Notes on Sampling Results

The DOE postprocessing is similar for all DOE methods. Once the DOE is generated, DesignXplorer extracts all results provided by the sampling to generate postprocessing tables and charts:

  • Sample points from the property ISamplingMethod.Samples

Notes on Optimization Results

The optimization postprocessing is similar for all optimization methods. Once the optimization is solved, DesignXplorer checks IOptimizationMethod.PostProcessingTypes and extracts all results provided by the optimizer to generate postprocessing tables and charts:

  • Candidate points from the property IOptimizationMethod.Candidates

  • Sample points from the property IOptimizationMethod.Samples

  • Pareto fronts from the property IOptimizationMethod.ParetoFrontIndex

In some cases, DesignXplorer builds missing results if one of these result types is not supported. For example, if the optimization method provides the sample points without the candidate points, DesignXplorer applies its own sorting logic to generate the candidate points based on objectives and constraints definitions. If the optimizer provides the sample points without the Pareto fronts indices, DesignXplorer builds missing data based on objectives and constraints definitions.


Release 18.2 - © ANSYS, Inc. All rights reserved.