Section: Partnerships and Cooperations

National Initiatives

Inria Project Lab (IPL): ModeliScale, Languages and Compilation for Cyber-Physical System Design

The project gathers researchers from three Inria teams, and from three other research labs in Grenoble and Paris area.

Table 1.

Name	Team	Inria Center or Laboratory
Vincent Acary	Tripop	Inria Grenoble Rhône Alpes
Bernard Brogliato
Alexandre Rocca
Albert Benveniste	Hycomes	Inria Rennes
Benoît Caillaud		Bretagne Atlantique
Khalil Ghorbal
Christelle Kozaily
Mathias Malandain
Benoît Vernay
Marc Pouzet	Parkas	ENS &
Tim Bourke		Inria Paris
Imsail Lakhim-Bennani
Goran Frehse	SSH	ENSTA Paris-Tech.
Antoine Girard		L2S-CNRS, Saclay
Eric Goubault	Cosynus	LIX, École Polytechnique,
Sylvie Putot		Saclay

The main objective of ModeliScale is to advance modeling technologies (languages, compile-time analyses, simulation techniques) for CPS combining physical interactions, communication layers and software components. We believe that mastering CPS comprising thousands to millions of components requires radical changes of paradigms. For instance, modeling techniques must be revised, especially when physics is involved. Modeling languages must be enhanced to cope with larger models. This can only be done by combining new compilation techniques (to master the structural complexity of models) with new mathematical tools (new numerical methods, in particular).

MiodeliScale gathers a broad scope of experts in programming language design and compilation (reactive synchronous programming), numerical solvers (nonsmooth dynamical systems) and hybrid systems modeling and analysis (guaranteed simulation, verification). The research program is carried out in close cooperation with the Modelica community as well as industrial partners, namely, Dassault Systèmes as a Modelica/FMI tool vendor, and EDF and Engie as end users.

In 2019, three general meetings have been organized, with presentations of the partners on new results related to hybrid systems modeling and verification.

Two PhDs are funded by the ModeliScale IPL. Both started in October 2018:

• Christelle Kozaily has started a PhD, under the supervision of Vincent Acary (TRIPOP team at Inria Grenoble), Benoît Caillaud, Khalil Ghorbal on the structural and numerical analysis of non-smooth DAE systems. She is located in the Hycomes team at Inria Rennes.

• Ismail Lahkim-Bennani has started a PhD under the supervision of Goran Frehse (ENSTA ParisTech.) and Marc Pouzet (PARKAS team, Inria/ENS Paris). His PhD topic is on random testing of hybrid systems, using techniques inspired by QuickCheck  [36].

FUI ModeliScale: Scalable Modeling and Simulation of Large Cyber-Physical Systems

Participants : Albert Benveniste, Benoît Caillaud, Khalil Ghorbal, Mathias Malandain.

FUI ModeliScale is a French national collaborative project coordinated by Dassault Systèmes. The partners of this project are: EDF and Engie as main industrial users; DPS, Eurobios and PhiMeca are SME providing mathematical modeling expertise; CEA INES (Chambéry) and Inria are the academic partners. The project started January 2018, for a maximal duration of 42 months. Three Inria teams are contributing to the project : Hycomes, Parkas (Inria Paris / ENS) and Tripop (Inria Grenoble / LJK).

The focus of the project is on the scalable analysis, compilation and simulation of large Modelica models. One of the main contributions expected from Inria are:

• A novel structural analysis algorithms for multimode DAE systems, capable of handling large systems of guarded equations, that do not depend on the enumeration of a possibly exponential number of modes.

• The partitioning and high-performance distributed co-simulation of large Modelica models, based on the results of the structural analysis.

In 2019, the effort has been put on the first objective, and two important milestones have been reached:

• The design of a novel algorithm for the structural analysis of multimode DAE systems. This algorithm is a generalization of the Pryce structural analysis method to the multimode case. The key feature of our method is that it works on implicit representations of the set of modes, and of the varying structure of the multimode DAE. In other words, it does not imply the enumeration of the system's modes. Performing the structural analysis at compile-time brings two decisive advantages: 1/ it allows to deliver to the user precise diagnostics about the model, and can be compared type-checking in programming languages; 2/ it is instrumental for the generation of efficient simulation code. Our algorithm is the first method enabling the compile-time analysis of systems with extremely large combinatorics of modes.

• Our multimode DAE structural analysis algorithm has been implemented in IsamDAE, a software comprizing an algorithmic library, to be used in modeling language compilers (Modelica tools) and a standalone tool, to be used independently of a complex Modelica toolset. IsamDAE has allowed to benchmark the method against several families of models, inspired by case-studies developed by industrial partners of the FUI ModeliScale project. Despite the tool is still under development, we have already been able to deal with models with up to ${10}^{23}$ modes.

On top of these two main results, the Hycomes team has started investigating the use of Quantized Space Systems (QSS), for the simulation of large DAE systems. QSSs simulation (QSS) was introduced in the early 2000’s by F. Cellier and E. Kofman as an alternative to time-based simulation, which is the dominant approach to ODE/DAE systems simulation. Rather than linking QSS to Discrete Event Simulation, we propose to relate it to Synchronous Programming and its continuous time extension Zelus. In the deliverable  [20], we expose our understanding of QSS and its variants, then we propose ideas toward a QSS-based cosimulation, by building on top of our knowledge on distributed executions of synchronous programs.

The plan for 2020 is to extend our structural analysis to cover impulsive mode changes and the consistent initialization problem, in the multimode case. A coupling of IsamDAE with Dymola (Dassault Système's commercial implementation of the Modelica language) is under development.

Another future development is to turn our structural analysis method to a compositional method, where large models could be considered by parts. This is a key problem in the Modelica language, as the compilation of a Modelica model is not modular.

Work on QSS methods will continue, and we envision to prototype a QSS-based distributed simulation method for hybrid ODE systems, based on the Zélus language.