Jens Pollmer

Fault tree synthesis from UML models for reliability analysis at early design stages

System reliability analysis is often neglected at early design stages when design decisions change the system architecture rigorously. This is because appropriate methods are time consuming and require an in-depth knowledge about the applied method. We propose a modeling approach that focuses on reusability and automatic fault tree synthesis of the models. We use UML to model application dependent and application independent views on the system and allocate steps of the application definition to architecture elements. In doing so various different system concepts can be investigated with minimal re-modeling effort. We identify capturing fault propagation and fault containment information as a major challenge in automatic fault tree synthesis and offer an application dependent and an application independent approach to modeling this kind of information. Then we introduce an algorithm that transforms the UML model into a fault tree representation of the respective system and validate our approach using an example from the automotive domain. The results from the validation highlight the validity of the generated trees, the efficiency with which different system concepts may be evaluated, and the degree to which the analysis results can be generalized

Discrete event simulation and analysis of timing problems in automotive embedded systems

Designing integrated electronic control units (ECU) in the automotive domain is challenging especially when legacy sub-systems are involved. Nevertheless, important design decisions have to be made at early design stages when some performance parameters of new sub-systems may not be validated yet. To account for uncertainty in the system development we use statistical modeling and discrete event simulation to perform sensitivity analysis from an average-case perspective. Average-case analysis is useful if performance aspects during normal system operation are of interest. The simulation model we implemented enables us to conduct holistic sensitivity analyses of systems consisting of multiple controllers, field-busses, sensors, actors, and applications. Applications are modeled as signal chains traversing the system model experiencing delay in each node along the predefined path. We present the strength of our approach by analyzing the latency of a distributed application in case of an asynchronous, synchronous, and event-triggered integrated ECU context. The results show that clock drift may result in high signal latencies in asynchronous ECUs and synchronous ECUs are superior in terms of signal latency and jitter. Event-triggered ECUs have to be designed carefully and come with trade-offs between performance of the legacy system and new applications.

Bounding the number of relevant objects in automotive environment perception systems

Multi-sensor data fusion systems for environment perception in the automotive domain are regarded as a promising instrument for obtaining dependable vehicular context information. Sensor data from remote sensing devices like radar or laserscanners is transmitted via intra-car networks to electronic control units (ECU) that enable an intelligent, context sensitive vehicle behavior depending on the current traffic situation. Although new bus systems, such as Flexray, offer increased data rates, the communication resources need to be utilized efficiently. In order to do so, two aspects have to be considered: 1) The size of a single object description and 2) the overall number of perceivable objects. In this paper we focus on the latter of the two aspects. We created a flexible discrete event simulation framework that allows for an in-depth analysis of various aspects of environment perception systems. Our simulation covers scenarios consisting of different sensor-sets, traffic scenarios, fusion benefits, and algorithms for context perception. Using this framework we were able to limit the number of objects a single sensor is allowed to perceive and analyze the impact of this limitation on the overall system performance without such restrictions. Our findings include: 1) Bounding the number of relevant objects to a number between 4–8 does not affect the false negative ratio of the system, 2) the overall error and false positive error ratio does not increase by bounding the number of relevant objects, 3) in safety-relevant environment perception systems the number of relevant objects can be reduced even further without compromising the system integrity, and 4) bounding the number of objects at an early stage of signal processing is superior to a reduction at a late stage of signal processing.

Schedulability Analysis in Time-Triggered Automotive Real-Time Systems

In embedded automotive real-time systems correct timing is required to effectively implement distributed applications, time-triggered communication, and integration of intelligent safety applications. We use formal schedulability analysis to validate the task timing for asynchronous real-time systems in an automotive context. Restricting the computational model to purely time-triggered task activation and applying application specific constraints enables us to perform exact response time analyses without pessimism. The analysis can be done efficiently since our computational model drastically limits the number of critical instant candidates to be considered for the worst-case scenario. Simulation results show that the original computational model incorporates a significant amount of response time overestimation and a high computational workload. The presented method can be applied in early design stages as well as in series development and helps to reduce development risks for electronic control units and embedded telematic systems.

Modeling and Analysis of Advanced Automotive ECU Architectures at Early Design Stages Using EMF and Model Transformation

In the automotive industry active safety functionsare deployed to better protect the passengers and vulnerable road users in case of an accident. To do so, advanced system architectures which enable an OEM to integrate this new functionality safely and efficiently have to be developed. In order to evaluate different possible concepts at an early stage in the development, model-based approaches are feasible. We propose a methodical approach to evaluate architecture concepts with regard to standard analysis methods as well as application specific evaluation techniques. We use UML to define the meta-model structure of the system architecture, model the instances of the meta-model using EMF and a generated graphical editor, and perform evaluation operations with the help of an MDE framework. The tool chain allows for an easy modeling of system models at early design stages and performing application specific analysis techniques. To validate the feasibility of the approach we apply real-time schedule simulation to two example models.

Scheduling complex automotive embedded real-time systems

The formal verification of task schedulability is getting more important in the automotive domain as more and more applications are integrated into domain specific controllers. Especially for safety critical applications an indepth knowledge about the worst-case response time of tasks is of utmost importance. Such systems often consist of subsystems that comprise of a collection of tasks, called transactions, which are triggered by external events. We adapt a well known response time analysis to analyze asynchronous task sets where subsets may be triggered from different clock references. Therefore we refine the original computational model from the literature to capture the information about transaction clock references. To determine the schedulability of the system we first analyze all task subsets which are triggered by the same clock reference. Then we combine the results to derive the response times for the complete task set. The approach incorporates no overestimation of the response times and therefore represets a necessary and sufficient schedulability test for such systems. The computational complexity of the presented approach is O(nm) with n being the number of tasks and m the number of clock references in the system. We apply the analysis to a real-world scenario from the automotive safety domain to confirm the computational