13.11.2018

Defence Sebastian Reiter

on Friday, 23 November 2018, at 13:00 in room A104, Sand 1

Safety assessment of networked embedded systems by means of error effect simulation

Reviewer 1: Prof. Dr. rer. nat. Oliver Bringmann
Reviewer 2: Prof. Dr. rer. nat. Wolfgang Rosenstiel

The acceptance and therefore the dissemination of embedded systems are continuously increasing. For the single systems, an increased complexity and interconnection can be observed. This does not only affect the effort for system development but also the effort for system analysis. Taking into account that an increasing amount of these systems realizes safety critical tasks, the analysis is an important factor. An illustrative example are the advanced driver assistance systems. These systems made considerable progress over the past years and already enable highly automated driving. With these systems, a failure or a wrongly delivered service can have significant negative effects on people or the environment. A safety assessment is therefore a mandatory task. However, because of the highly interconnection of the different systems, an isolated analysis of a single system is not sufficient. Therefore, not only the increased complexity of the single system, but also the interconnection with other systems have to be taken into account during analysis. Current standards often recommend techniques such as brainstorming, failure mode and effects analysis or fault tree analysis for safety assessment. The success of these techniques depends strongly on the involved people and their knowledge about the system. They have to assess the increased complexity and interconnections during the safety assessment.
The talk presents an approach to support current safety assessments. The required system knowledge is transferred from the involved people to a simulation model. Basis for the safety assessment is the determination of the effects of a fault in the system. By the provision of a simulation model, the error effects are automatically determined. Meaning the system complexity and the system dependencies are described automatically by the system model. The talk presents an approach for fault injection into simulation models to enable such an assessment. One important aspect is the support of different abstraction levels, especially the support of abstract system-level models. Additionally, a general approach for fault specification is presented. This approach enables the specification of faults, with regard to different abstraction levels, the automatic interpretation and control of the fault injectors. Besides the injection of faults, the effect monitoring and classification is an important part of the work. The overall approach is finished by a model-driven specification that easies the integration of the approach into a model-driven design flow. Provided approaches for code generation reduces the effort to execute the error effect simulation.
Der Dekan