CORDIS - EU research results

Controlled Component and Assembly-Level Optimization of Industrial Devices

Periodic Report Summary 2 - CASOPT (Controlled Component and Assembly-Level Optimization of Industrial Devices)

CASOPT Project Summary
The numerical computation of electric fields applied to optimization of high voltage power devices has been initiated almost 50 years ago. Today, it has become a key part in designing high voltage switchgear, transformers and other electrical devices. It covers a multidisciplinary research area including optimization algorithms, computational geometry, boundary element method, computational electromagnetics, electric discharge modeling, and high performance computing. In order to achieve the next significant advance, the Casopt-Project team concentrates top experts on each of these areas. The objective of Casopt is focused on creating a new generation of numerical procedures for automatic shape optimization with respect to dielectric performance of industrial devices.

The basic step of the new optimization procedure consists of calculating the electrostatic field for a complex 3D geometry. A good efficiency of solving geometrically very complex models can be achieved by applying the Boundary Element Method (BEM) in connection with High Performance Computing (HPC) technologies, which have been included as two major work packages in Casopt. As a project result, three different BEM solvers have been created. In their implementation special attention has been devoted to the two following aspects: (1) the ability to solve very large models with a hundred thousand up to a few millions unknowns, and, (2) the high accuracy of the solution in case of large differences of material properties on interfaces between dielectric bodies. The first aspect became essential due to enhanced capabilities of mesh generators integrated with engineering design systems while the second is very important for direct current (DC) high voltage systems enabling efficient transfer of electric energy over large distances. The features of new solvers include compression techniques for large equation systems, multipole techniques for handling of far fields, domain decomposition approach, as well as distributed and shared memory parallelization. Within the scope of a separate benchmarking work package the new features of all three BEM solvers have been tested on several simplified and real life models. The results showed that significant progress could be achieved in solving electrostatic field problems using the BEM approach. Both essential requirements including solving large models and good accuracy for extreme differences of material properties could be met. The implementations of the new BEM solvers within the Casopt-Project provide a good foundation for the future generation of dielectric simulation tools applied in industry world-wide.

The BEM approach has a fundamental advantage for the parametric shape optimization: the meshing of the outer space (the so called "airbox") is not required. It enables fully automatic geometry regeneration as well as re-meshing after changing the values of geometrical parameters in the design system (CAD). The basic architecture of the optimization procedure has been studied on an example of gas insulated switchgear (GIS). This procedure has been significantly enhanced within the Casopt project (work package Assembly Level Optimization, ALO), and enabled optimization runs on real-life geometry using the local and global optimization methods supported by Kriging methods. It allows to fully automatically perform several hundreds of complex 3D computations for every simulated problem: typically, a GIS designer submits the prepared CAD/ProEngineer model to the new optimizer and gets a response within a few hours (or overnight) in form of an optimized geometrical shape of the device component. The calculation of the objective function for parametric optimization involves Design Criteria Evaluation (DCE), which has been defined as a separate work package of the Casopt-Project.

Another advantage of the BEM approach for dielectric simulations is the fact that the changes of shape are limited only to surfaces of the model. This feature provides the foundation for free form optimization investigated within the Casopt work packages Component Level Optimization (CLO) and Geometry/Analysis Interfaces (GAI). The geometry can be changed by the free form optimizer directly by moving the nodes on the surface mesh. Several approaches have been investigated by Casopt researchers: (1) direct and indirect approach based on non-gradient-driven node movement with subsequent mesh smoothing, (2) a gradient-based shape optimization using subdivision surfaces with a coarse geometry representation for geometry manipulation, and fine mesh representation for analysis, (3) another gradient-based approach using fine mesh for both geometry processing and analysis with a mesh smoothing/repairing procedure. In the investigated example a simple cylinder has been applied as the initial geometry. The free form optimizer has converted the cylinder to a new shape that is very similar to the result of the parametric procedure. The numerical procedure used in this example is based on a formulation of the “adjoint problem” (like uniform field strength on the electrode surface), which delivers the gradient information used by the optimizer for changing the mesh nodes coordinates. In contrast to the parametric approach the free form optimization does not require specification of geometrical parameters. The computer algorithm “invents” a new shape which significantly reduces the effort of designers for preparing the initial geometry.

The results of Casopt project can be classified in three stages of industrial maturity: (1) The results can be directly integrated into industrial environments, like the parametric optimization module that is currently being integrated into the ABB Simulation Toolbox by the ABB researchers; (2) The results can be used by ABB engineers for test purposes demonstrating the features of the new computational algorithms; to this category belong all 3 BEM solvers created within Casopt; (3) the results shows feasibility of the newly invented simulation approach for pilot examples, like free form optimization. In categories (2) and (3) a significant advance of knowledge has been achieved, which guarantees continuation of the close cooperation between ABB and the university partners.

Conclusion: The Casopt-Project can be considered a very successful example of cooperation between academia and industry. The international team including 30 researchers created results that can be directly used by industry and has opened opportunities for scientific work within the scope of new academic and industrial research projects.
From the academic side, university partners have gained very valuable experience in cooperating with industry so closely, which will definitely be of importance for future projects and research. A number of scientific problems could not have been solved without the support of Casopt and the knowledge transfer provided by the project.

Detailed information (including Figures):