Matthias Jüttner

Iterative Software Agent Based Solution of Multiphysics Problems

A novel approach is presented using software agents for an iterative and distributed solution of multiphysics problems. Overall convergence is achieved by using the individual capabilities of interworking agents. Every agent solves a partial single physics problem based on specialized, commercial or in-house code. The autonomy of each agent allows a physics adapted solution process without the need of a predefined solver sequence. The applied software agents are described in detail. Here, we focus on weak uni- and bidirectional field coupled multiphysics problems. This framework can also be used for node or boundary coupling as well as for optimising partial physics simulation. A coupled 3D electromagnetic wave propagation and heat transfer problem inside a waveguide is examined as numerical example.

A Neural Network Based Recommendation System for Solvers and Preconditioners for Systems of Linear Equations

Numerical simulation methods like the finite element method lead to large systems of linear equations solved with well-known methods. Their performance varies depending on the considered simulation (discretization and physics) and the available hardware. To predict a suitable method including the solver and a well performing preconditioner, a feed-forward neural network is used. It computes performance ratings for each reasonable combination of solver and preconditioner depending on selected properties of the system of linear equations and on the provided hardware. Details about the designed and the applied training methods are given. A statistic as well as a specific evaluation show the performance of different neural networks as recommendation systems.

Harmonic balanced Jiles-Atherton hysteresis implementation for finite element simulation

Purpose The high calculation effort for accurate material loss simulation prevents its observation in most magnetic devices. This paper aims at reducing this effort for time periodic applications and so for the steady state of such devices. Design/methodology/approach The vectorized Jiles-Atherton hysteresis model is chosen for the accurate material losses calculation. It is transformed in the frequency domain and coupled with a harmonic balanced finite element solver. The beneficial Jacobian matrix of the material model in the frequency domain is assembled based on Fourier transforms of the Jacobian matrix in the time domain. A three-phase transformer is simulated to verify this method and to examine the multi-harmonic coupling. Findings A fast method to calculate the linearization of non-trivial material models in the frequency domain is shown. The inter-harmonic coupling is moderate, and so, a separated harmonic balanced solver is favored. The additional calculation effort compared to a saturation material model without losses is low. The overall calculation time is much lower than a time-dependent simulation. Research limitations/implications A moderate working point is chosen, so highly saturated materials may lead to a worse coupling. A single material model is evaluated. Researchers are encouraged to evaluate the suggested method on different material models. Frequency domain approaches should be in favor for all kinds of periodic steady-state applications. Practical implications Because of the reduced calculation effort, the simulation of accurate material losses becomes reasonable. This leads to a more accurate development of magnetic devices. Originality/value This paper proposes a new efficient method to calculate complex material models like the Jiles-Atherton hysteresis and their Jacobian matrices in the frequency domain.

A neural network based recommendation system for solvers and preconditioners for systems of linear equations

Numerical simulation methods, like the finite element method, lead to large systems of equations. Well-known and highly optimized methods are applied to solve equation systems. Their performance varies depending on the considered simulation (discretization and physics) and the available hardware. Choosing a suitable method includes the selection of a well performing solver and preconditioner which is rarely obvious. Here, a case study is presented, where recommendations are given based on provided training data by feed-forward neural networks. The neural networks compute performance ratings for each reasonable combination of solver and preconditioner, depending on selected properties of the system of linear equations and on the provided hardware. Details about the designed and the applied training methods are given. A statistic as well as a specific evaluation shows the performance of different evaluated neural networks. Results show the effort of using a neural network as recommendation system for solvers and preconditioners for linear equation systems.

Intelligent and Distributed Solving of Multiphysics Problems Coordinated by Software Agents - An Intelligent Approach for Decentralized Simulations

This paper presents an intelligent approach to support engineers with performing computational simulation of new developments and prototypes. With multiple interacting physical effects and large three dimensional models the choice of the right solution strategy is crucial for a correct solution and an acceptable calculation time. The presented multi-agent system can solve these simulation tasks using distributed heterogeneous computation resources with the objective to reduce the calculation time. An important factor for the criterion time is the choice of the linear solver. Here a case-based reasoning concept is introduced to improve the decisions in the multi-agent system. Allowing each agent to solve its problem part by using appropriate solution methods, a decentralized architecture with autonomous software agents is provided.

A DISTRIBUTED METHOD FOR TRANSIENT SIMULATIONS THAT DYNAMICALLY CONSIDERS SUPPLEMENTARY RESULTS FROM AUTONOMOUS SOFTWARE AGENTS

Based on autonomous software agents capable of calculating individual numerical field problems, a distributed method for solving transient field problems is presented. The software agents are running on distributed resources connected via a network and represent a dynamic calculation environment. Communication and data exchange between multiple agents enables their collaboration and allows decisions based on distributed overall knowledge. As unique characteristics, no central unit influences the solution process at any time. The presented simulation example and its evaluated calculation process proves the method to benefit from redundant resources.

Experimental evaluation of numerical errors for multi-physics coupling methods using disparate meshes

Purpose Multiphysics problems are solved either with monolithic or segregated approaches. Accomplishing contrary discretisation requirements of the physics, disparate meshes are essential. Here, experimental results of different interpolation methods for a segregated coupling are compared with monolithic approaches, implemented using a global and a local nearest neighbour method. The results show the significant influence of discretisation for multiphysics simulation. Design/methodology/approach Applying disparate meshes to the monolithic as well as the segregated calculation of finite element problems and evaluating the related numerical error is content of the contribution. This is done by an experimental evaluation of a source and a material coupling applied to a multiphysics problem. After an introduction to the topic, the evaluated model is described based on two bidirectional coupled multiphysics problems and its finite element representation. Afterwards the considered methods for approximating the ...

Softwareagenten zur zuverlässigen Durchführung dezentraler multiphysikalischer Simulationen

Zusammenfassung Zur Ermittlung des Systemverhaltens werden beim Engineering in Kombination mit der modellgestützten Entwicklung häufig Simulationen eingesetzt. Dieser Beitrag stellt einen agentenkoordinierten Ansatz für die parallele Simulation von multiphysikalischen Problemstellungen auf heterogenen Rechenressourcen vor. Durch die unterschiedlichen eingesetzten Simulationsmethoden steigt hierbei die Zuverlässigkeit beim Erreichen der Ergebnisse.

Coupled multiphysics problems as market place for competing autonomous software agents

A novel approach is presented applying capabilities of software agent-based programming to numerical field computations. Distributed and autonomous calculation resources are used in analogy to a machine-to-machine system application for setting up and solving coupled simulations respecting the individual machine capabilities. Within the presented system, the involved machines interact cooperatively and comparatively. No central unit takes any influence to the solution process. Redundant resources do automatically and proactively add variance to the solution process by modifying their partial tasks based on negotiations between the involved resources. In that way, different numerical methods, solution strategies, and solver configurations are automatically evaluated as examples of the variety within coupled simulations. Partial results are dynamically integrated into an overall solver sequence, demonstrated on the coupled electric field computation of a patch antenna array. A quick provision of results is shown contemplating changes at calculation resource and vague user decisions for a nonlinear and bidirectionally coupled simulation of a 3-D multiphysics problem.