PHYMOS - Proper Hybrid Models for Smarter Vehicles

The vehicle of the future is „smart“.

The ability of a vehicle to respond to a changing environment and changing constraints by adopting its behavior in an optimal way will be considered a commodity feature. Realizing such a flexible behavior in a vehicle requires a high degree of „self-awareness“, in other words the ability to predict the impact of its interaction with the environment. Creating models to describe the vehicle itself and its environment properly in terms of the best trade-off between fidelity and runtime performance in a short period of time and in a very cost effective way is a key success factor.

Classical model-based approaches are typically associated with high development efforts. Advances in the field of artificial intelligence open up new opportunities but depend on large amounts of data, besides other risks to reach a high confidence in the model.

In this project hybrid (data and physics-based) approaches shall be evaluated in concrete applications, aiming to incorporate existing physical knowledge in order to generate scalable “Proper Models” in a very data efficient way. These methods will enable the development and realization of competitive product properties and innovative new functionality for smart vehicles in siginificantly shorter time.

E-Graphs for symbolic simplification

Status of the New Backend

 •  Other languages: Deutsch
In the course of presenting the Status of the New OpenModelica Backend, basic concepts of E-Graphs were introduced and work in progress on symbolic simplification using E-Graphs was discussed. These simplifications can speed up the simulation process at the cost of higher compilation effort. However, as Modelica models are not scalarized, this compilation effort becomes more and more negligible for larger models. In the future, these techniques will be used to perform even more significant symbolic transformations to improve simulation speed. [Read More]

Calibrate Complex System Models with Deep Learning

Sensitivity-Guided Iterative Parameter Identification and Data Generation with BayesFlow and PELS-VAE for Model Calibration

 •  Other languages: Deutsch
Calibration of complex system models with a large number of parameters using standard optimization methods is often extremely time-consuming and not fully automated due to the reliance on all-inclusive expert knowledge. We propose a sensitivity-guided iterative parameter identification and data generation algorithm. The sensitivity analysis replaces manual intervention, the parameter identification is realized by BayesFlow allowing for uncertainty quantification, and the data generation with the Physics-Enhanced Latent Space Variational Autoencoder (PELS-VAE) between two iteration steps enables inference of weakly identifiable parameters. [Read More]

MEMS Swim in BayesFlow

Fast, Precise, and Reliable Parameter Identification in MEMS End-of-Line-Testing with BayesFlow

 •  Other languages: Deutsch
In micro-electro-mechanical systems (MEMS) testing both high precision and reliability are essential. Due to the additional requirement of runtime efficiency, machine learning methods have been investigated in recent years. However, these methods are often associated with inherent challenges concerning uncertainty quantification and guarantees of reliability. The goal of this paper is therefore to present a new machine learning approach in MEMS testing based on Bayesian inference to determine whether the estimation is trustworthy. [Read More]

How to train neural network based on detailed model

Hybrid physical-AI based system modeling and simulation approach demonstrated on an automotive fuel cell

 •  Other languages: Deutsch
This paper presents an approach on how to train a Neural Network model based on a detailed physical Modelica model. The necessary steps to generate training data from simulation will be explained as well as the generation process of a surrogate model. It will be shown, how the surrogate will be re-integrated into the Modelica system model. A benchmark based on accuracy and simulation performance will be performed. The tools used are Modelon Impact, an online modeling and simulation platform, the TensorFlow/Keras toolbox in a Jupyter Notebook which provides a Python-based interface for generating Neural Networks, and the Modelica Neural Network Library that provides functions for constructing Neural Networks within Modelica. [Read More]

Presented workflow to replace algebraic loops

Replacing Strong Components with Artificial Neural Network Surrogates in an Open-Source Modelica Compiler

 •  Other languages: Deutsch
The University of Applied Sciences Bielefeld presented first interim results from the PHyMoS project at the annual OpenModelica Workshop at Linköping University, Sweden. The presentation “Replacing Strong Components with Artificial Neural Network Surrogates in an Open-Source Modelica Compiler” showcases the replacement of computation intensive non-linear systems with artificial neural networks in the simulation environment OpenModelica. The general workflow to detect and replace expensive equation systems in the generated code of the OpenModelica compiler was presented. [Read More]

PHyMoS Kick-Off

  •  Other languages: Deutsch

In March 2021 the project started and now there are three years time to work on it.