Almost all fields of application in information technology, from mobility to mechanical engineering to logistics, are dominated by cyber-physical systems. CPS are networked computer systems ("cyber") that interact directly with the environment ("physical") via sensors and actuators. To check complex CPS during design and operation (testing), to monitor them during use (monitoring) or to predict the system behavior (prediction) is extremely time-consuming. For all these applications, valid models of the systems are needed. However, subcomponents or even entire subsystems of modern CPS are often only available as a black box whose internal realization is not known or only known to a limited extent. In this case, models must first be created at great expense, which is often done manually or not at all. Likewise, there is a lack of quality metrics for the use and transferability of models in their applications as well as for the selection and comparison of different model types. To date, the necessary steps for modeling are poorly automated, so that testing, monitoring, and prediction increasingly cause bottlenecks in the design. Incomplete isolated solutions exist in various application fields, some of which are standardized there, but which are nevertheless used in isolation and are often in turn insufficiently automated.

The goal of the AGenC project is to develop a toolbox with uniformly usable methods and interoperable technology solutions for CPS from different application areas. Specifically, a framework of novel software methods and tools that create models for CPS will be realized. Therefore, generalized interfaces for use in diverse application domains are to be created in order to develop novel model learners based on Adversarial Resilience Learning on the one hand, and combining discrete and continuous models on the other hand, in order to address relevant aspects from interpretability to accuracy of the models simultaneously. This requires the development of hard criteria for assessing model quality and metrics for comparing and extending different models, as well as test case generators, system monitors and predictors that operate on the basis of models. In this context, the models in the applications are modular, i.e., interchangeable, comparative, or complementary.

PalaestrAI