Fraunhofer Institute for Algorithms and Scientific Computing SCAI
Fraunhofer Institute for Algorithms and Scientific Computing SCAI

DesParO

Exploration, analysis, and optimization of parameter-dependent problems

Computer-aided simulations of technical processes and products commonly depend on many parameters, e.g., geometrical, material, and process-controlling parameters. Engineers are interested in configurations of these parameters, which optimize the production process, product features, and quality.

DesParO is a software package for intuitive exploration, automatic analysis, and optimization of such parametrized problems. DesParO can be coupled with simulation packages or used for measurement data. In particular, it focuses on keeping the number of simulations/experiments needed for analysis or optimization small. Moreover, simulations/experiments follow an experimental design (design-of-experiment, DoE) and, therefore, can be performed in parallel. Hence, DesParO is particularly useful for time- or resource-intensive simulation runs or costly physical experiments.

In addition to DesParO licenses, training, and studies, SCAI offers consulting for interpolation, statistics, and optimization and helps set up strategies and find suitable methods and tools.

© Fraunhofer SCAI
Visualization of optimal design in DesParO.

Methods

Robust tolerance prediction
DesParO predicts the value of the design objective and the tolerance limits on the objective. For noisy objectives, this allows the satisfaction of constraints safely, including tolerance and achieving the best possible robust design.

Global correlation analysis
DesParO automatically recognizes a pattern of interdependencies between the optimization criteria and design variables and represents it as an easily readable color-coded diagram. The diagram indicates the most influencing design variables and most sensitive optimization criteria and shows a sign of dependency: increase (red), decrease (blue), or non-monotony (black).

Interpolation of FEM data
DesParO also provides interpolation of bulky data, such as FEM data files containing the results of numerical simulation, to the values of design variables specified by the user. This allows DesParO to visualize a complete solution immediately and inspect the obtained optimal design in detail.

© Fraunhofer SCAI
Two clusters were found by the SimExplore tool for CFD simulations of the velocity field, with two input parameter variations (each point represents one simulation). Two surrogate models were built using DesParO based on this information.
© Fraunhofer SCAI
a) Velocity distribution used for testing prediction. b) The difference between real and predicted values is based on one surrogate model generated for all the simulations. c) Difference of real and predicted values using one of the surrogate models from the many generated per cluster. The result is improved when the surrogate model of a respective cluster is used.
Correlation matrix in DesParO.
© Fraunhofer SCAI
Correlation matrix in DesParO.

Optimization with DesParO

DesParO offers a classical, (semi-)automatic global-local optimization and an interactive visual variant. The solution of an optimization problem using this interactive variant proceeds through the following steps:

Step 1: Set a desired region for optimization criteria. By setting small values of crash intrusion and total mass, the several design alternatives in the parameter space become visible.
© Fraunhofer SCAI
Step 1: Set a desired region for optimization criteria. By setting small values of crash intrusion and total mass, the several design alternatives in the parameter space become visible.
Step 2: Test the alternatives. One optimum corresponds to a larger, the other one to a smaller value of maximal velocity.
© Fraunhofer SCAI
Step 2: Test the alternatives. One optimum corresponds to a larger, the other one to a smaller value of maximal velocity.
Step 3: Identify the best possible design.
© Fraunhofer SCAI
Step 3: Identify the best possible design.

Interactive environment for the optimization of design parameters

DesParO is a unique system for multiobjective optimi­zation, providing users with complete control of opti­mization processes. Contrary to other automatic optimization tools, DesParO allows users to explore the whole space of design variables interactively and find the optimal region concerning multiple design objectives. DesParO provides the users with a global view of the design space, reveals a complete set of alternative solutions, and allows users to select the optimal design.

Consequently, DesParO is free of common drawbacks inherent to automatic optimization tools, such as a solution stuck in a local optimum, typical for differential methods, or the exhaustive numerical experimentation of Monte Carlo-like strategies.

The optimization algorithm of DesParO has been tested on several real-life multidimensional problems in automotive design and showed excellent results for many design variables and design objectives.

DesParO is available as a stand-alone application for Windows and Linux platforms or as a documented SDK for integration into other optimization software.

Left: DesParo GUI; right: Visualization of optimal design in DesParO
© Fraunhofer SCAI
Left: DesParo GUI; right: Visualization of optimal design in DesParO

Optional integration into the SimExplore workflow

Once users have obtained a structural organization of the simulation results into clusters using SimExplore, DesParO can build surrogate models for each derived cluster based on the corresponding subset of simulation results to represent the dependency between input parameters and deformations or mesh functions. This highly simplifies further investigations of the underlying correlations between measures and crash behavior. Combining the clustering approach in SimExplore and surrogate models per cluster overcomes the limits of the state-of-the-art proper orthogonal decomposition method used for predicting simulation results, especially in nonlinearities, and improves the local prediction quality. Additionally, this approach can efficiently handle nonlinearities with DesParO’s correlation measures and its local tolerance and parameter sensitivity estimation.

Related Links: