KOMET

In the KOMET project, fortiss is developing innovative, automated methods for analyzing and improving the quality of system models. Products such as cars, airplanes, or medical devices are characterized by complex, interacting requirements, functions, and subsystems from different disciplines (mechanics, electronics, software, etc.). In order to achieve market-ready products quickly, companies often rely on model-based systems engineering (MBSE). As changes in large models have a significant impact and manual reviews are time-consuming, the costs of creating and maintaining high-quality models can be very high.

The focus of the project is on predicting the effects of model changes and automating model reviews using machine learning. This allows us to significantly reduce the costs and effort involved in maintaining high-quality models. Thanks to the reduction in effort, model reviews can also be carried out more frequently for large models, which means that quality degradation can be countered earlier and more effectively. Quality management of models is not only demanded by safety standards, but is also a prerequisite for their efficient collaborative development and reuse.

By collaborating with industrial partners, we ensure that our solutions are practical and directly applicable. The results of the KOMET project enable industrial users of MBSE to work more efficiently and cost-effectively by benefiting from improved model quality and reduced maintenance costs.

In the KOMET project, fortiss is developing innovative methods that significantly reduce the effort and costs involved in the quality management of system models. The research is based on fortiss' MBSE tool AutoFOCUS3 and is evaluated using systems engineering models from the fortiss Mobility Lab as well as industrial models. The project results will be incorporated into the KOMET platform developed in collaboration with qwitto GmbH. As part of this, fortiss will research and prototype the following solution approaches:

1. Methods for the automated prediction of the effects of model changes are investigated by analyzing typical progressions and statistical correlations from the change history.
2. A machine learning-based approach for automatic model reviews is developed. The method generalizes expert assessments of historical model revisions to current models by correlating them with automatically determinable model metrics.

The research aims to reduce the maintenance effort and improve the quality of the models. The results are incorporated into the open source tool AutoFOCUS3, which ensures the transfer of technology into practice.

Federal Ministry of Education and Research (BMBF)

FKZ: 01IS24012B

Funding program: KMU Innovativ

01.05.2024 - 30.04.2026