Distributed co-simulation framework for hardware- and software-in-the-loop testing of networked embedded real-time systems

Today's complex control systems such as trains, aircraft or cars are typically composed of multiple networked components which are developed by geographically distributed manufacturers. During their development process, integrating and testing the components are central steps. However, the manufacturers' locations complicate the process since the components must be shipped to a central place and intellectual property must be protected. Using a distributed co-simulation framework which supports Software- and Hardware-In-The-Loop (SIL/HIL) testing can solve these issues. It enables a virtual integration and testing of the components via the Internet and protects intellectual property if it operates on a network-centric abstraction level. In this case, it focuses on the data exchange between the components and knowledge about their internal implementation is not required.

In today's state-of-the-art, a framework that operates on a network-centric abstraction level and supports co-simulation, SIL and HIL testing together is not available yet. Besides that, HIL testing involves hardware devices with real-time requirements. Connecting those devices via public wide area networks such as the Internet, the accuracy of distributed HIL tests is limited by the determinism of the network’s communication delays. The available frameworks are mainly based on Quality of Service mechanisms such as differentiated services. However, the communication cycles of the System Under Test (SUT) might be smaller than the guaranteed latencies which leads to deadline misses. Hence, delay-management mechanisms are required which ensure a timely forwarding of input data.

This thesis proposes a distributed co-simulation framework which operates on a network-centric abstraction level and supports the above mentioned techniques. It synchronizes the components of the SUT, coordinates their data exchange and includes fault-injection to validate the dependability. By providing a generic component interface, heterogeneous simulation tools and physical devices are supported. The main contributions of the thesis are two delay-management mechanisms based on state-estimation and speculative execution. The first mechanism forwards estimated inputs to the component if (I) received inputs are delayed or (II) as intermediary inputs. This reduces the number of communication activities inside the framework. The second mechanism divides a simulation setup into several subsets. Those subsets execute independent tasks in advance to forward data to real-time devices in time. Using the mechanisms, the framework is able to connect simulations, software algorithms and real hardware via public communication networks while maintaining the SUT's real-time requirements. Hence, there is no need to make all simulation models or physical prototypes centrally available.

The evaluation using a distributed control application demonstrates the scalability of the framework. The time to execute a simulation setup increases linearly with the simulated time and is bounded by the growth of the component's number in larger setups. Furthermore, the evaluation shows the advantages of the delay-management mechanisms for distributed real-time tests. After determining a proper real-time configuration of the simulation host, the state-estimation mechanism can be used for a timely forwarding of inputs to the components. Using intermediary packets improves the accuracy of distributed real-time tests and makes it independent from the network delays. While the speculative execution enables real-time tests locally and in Local Area Networks, networks with larger delays (e.g., the Internet) require less stringent temporal requirements of the SUT. From a performance point of view, both mechanisms achieve significant speedups depending on the setup's size, the network topology and the communication period.