Towards an Open-Source Modelica Compiler in Julia

This paper presents current work on our Modelica Compiler framework in Julia: OpenModelica.jl. 1 We provide a brief overview of this novel framework and its features, and we also present the latest addition to the possible backend options. We target ModelingToolkit.jl (MTK), a framework for symbolic-numerical computation and scientific machine learning. We evaluated the performance of our new backend using the ScalableTestsuite, a benchmark suite for Modelica Compilers. In our experiment, we demonstrate that MTK can be used as a backend with competitive simulation performance. In addition, using the scientific machine learning features of the Modeling toolkit, we were able to approximate models in the Scal-ableTestsuite using surrogate techniques and how such techniques can be used to accelerate the solving of nonlinear algebraic loops during tearing.Based on our experiments, we propose using this new framework to automatically generate surrogate components of a Modelica model during the simulation to increase performance. The experimental work presented here provides one of the first investigations concerning the integration of the symbolic-numerical abilities of Julia within a Modelica tool.

Section: Motivationmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

OpenModelica.jl: A modular and extensible Modelica compiler framework in Julia targeting ModelingToolkit.jl

Tinnerholm¹,

Pop²,

Heuermann³

et al. 2021

“…There exists several computational frameworks to handle VSS (Zimmer, 2010;Höger, 2019;Elmqvist et al, 2017). While they differ somewhat in implementation, they all allow the compiler to be self-recursive, either through interpretation or JIT.…”

Section: Architectural Principlesmentioning

confidence: 99%

“…However, according to (Höger, 2019), Mosilab did not get traction and had certain drawbacks such as a combinatorial explosion of modes. Zimmer proposed a language Sol inspired by Modelica that handles structural changes to the model through symbolic processing at runtime via interpretation (Zimmer, 2010). Höger has a similar approach a few years later but additionally provides a method for dynamic index reduction that improves performance during structural changes (Höger, 2014).…”

Section: Related Workmentioning

confidence: 99%

Proceedings of Asian Modelica Conference 2020, Tokyo, Japan, October 08-09, 2020

2020

The automobile industry is currently undergoing a dramatic transformation. Highperformance networking, deep learning technology, and cloud computing have radically transformed how individuals and businesses manage data, and is poised to disrupt the state-ofthe-art in the development of intelligent machines. This talk explores how automated driving and artificial intelligence based on connected, distributed computing enables future mobility systems that will impact the design and evolution of future cities.

“…For example, it enables using dual number representation for automatic differentiation and uncertainty information. Julia is also used for modeling in the Model-ingToolKit (Ma et al 2021) and experiments are made to use Julia instead of MetaModelica for the OpenModelica implementation (Tinnerholm et al 2020).…”

Section: Introductionmentioning

confidence: 99%

Modia - Equation Based Modeling and Domain Specific Algorithms

Elmqvist¹,

Otter²,

Neumayr³

et al. 2021

A new design of the Modia experimental modeling language based on Julia is presented. It has simple yet powerful syntax and semantics. A unified means of describing the fundamental semantics, which is similar to Modelica, is outlined. Furthermore, it is shown how domain specific algorithms can be combined with equation based modeling. It is demonstrated for multibody systems and enables more efficient translation since much repetitive analysis and transformations are avoided and faster simulation. The drive train of a robot model was automatically translated from Modelica to Modia. Modern simulation algorithms from the Julia community allow working with automatic differentiation and uncertainties.