NeuralFMU: Presenting a Workflow for Integrating Hybrid NeuralODEs into Real-World Applications

The term NeuralODE describes the structural combination of an Artificial Neural Network (ANN) and a numerical solver for Ordinary Differential Equations (ODE), former acts as right-hand side ODE to be solved. This concept was further extended by black-box model in form Functional Mock-up Unit (FMU) obtain subclass NeuralODEs, named NeuralFMUs. resulting structure features advantages first-principle data-driven modeling approaches one single simulation model: higher prediction accuracy compared conventional First-Principle Models (FPMs) also lower training effort purely models. We present intuitive workflow set up use NeuralFMUs, enabling encapsulation reuse existing models exported from common tools. Moreover, we exemplify this deploying NeuralFMU consumption based on Vehicle Longitudinal Dynamics Model (VLDM), which is typical case automotive industry. Related challenges that are often neglected scientific cases, such real measurements (e.g., noise), unknown system state or high-frequency discontinuities, handled contribution. To build hybrid with quality than original FPM, briefly highlight two open-source libraries: FMI.jl, allows import FMUs into Julia programming language, well library FMIFlux.jl, enables integration neural network topologies NeuralFMU.