Computational models are an important tool for simulating phenomena in science and engineering.
One industry standard for exchanging these models is the Functional Mock-up Interface (FMI), which allows exporting models as a Functional Mockup Unit (FMU). This standard allows engineers to connect third-party models and perform co-simulation with specialized tools. However, it can be hard to find out whether a model produces the desired output, especially when it is a black-box model. Therefore, there is a need for good ways of testing the correct behavior of such a model. The curse of dimensionality especially complicates this, because it can lead to extremely high runtimes when sampling black-box models.
In this work, I want to find out whether sensitivity analysis methods are useful to analyze black-box models that are implemented via the FMI standard.
For the screening and the analysis, SAlib is used, which is an open-source Python library implementing the Sobol Method and the Morris Screening methods among others. These methods help us to determine the quality of a model for a broad range of inputs with a rough screening and to analyze the relationship between inputs and outputs in a quantitative fashion.
For testing the implementation, I used two different simulation models: First, a model of a one-dimensional spring equation and, second, a battery model provided by simercator GmbH for a case study. I use the language Modelica to create FMUs and the Python library fmpy to simulate them.