Resilient service-oriented architectures

Software-defined systems are changing the market

© Fraunhofer IKS / Andreas Jacob

Because they react independently and intelligently, cognitive systems offer new opportunities in many different areas of life. These types of autonomous or cooperative systems therefore must be able to interact closely with their surroundings and adapt quickly to changing conditions. The software in today’s technical systems, such as automobiles and machines, was historically developed first and foremost from the fields of machine and electronics engineering and created as a proprietary solution that was enhanced was over time. For this reason, these solutions no longer represent state-of-the-art software technologies. When developing autonomous vehicles for instance, technology companies build them from the ground up as software systems to create crucial competitive advantages. One hears the term »software-defined vehicles« or generally speaking »software-defined systems«.

Opportunities through adaptable and adaptive architectures

Such cognitive systems are based on the use of artificial intelligence (AI), machine learning and other modern software technologies. The goal is to design previously rigid and inflexible systems in an adaptive and cognitive manner in order to enable applications such as over-the-air updates, micro services or virtual software execution. However, this also requires balancing the opportunities of modern software technology with the necessary quality characteristics, particularly safety, dependability and availability.

Fulfilling these quality characteristics yields a wide range of possibilities:

  • Cloud controls: flexible software architectures can be used to separate physical processes from the controls. The controls are then located outside of the physical machine, such as in an edge or cloud network.
  • End-to-end architectures: end-to-end architectures with flexibly reconfigurable subsystems permit heterogeneous subsystems to exhibit dynamic, cooperative behavior. This requires ensuring safe and dependable distribution in order to guarantee availability.
  • Service-oriented architectures: vehicles no longer consist of scores of distributed electronic control units and instead are made of service-oriented architectures. This structure enables features such as add-on functions and thus new business models.
  • Graceful degradation: adaptive architectures enable fail-operational approaches such as graceful degradation. When malfunctions arise, the functional scope or functional quality is gradually reduced to maintain safe operation despite the occurrence of malfunctions.

In order to satisfy the requirements of future autonomous systems and cognitive networks, the Fraunhofer Institute for Cognitive Systems IKS conducts research into new architecture concepts, in particular approaches that are designed for the development, analysis and run-time management of cognitive systems, thus making them flexible and resilient at the same time.

Designing, analyzing and validating architectures

© Frauhofer IKS / Andreas Jacob

Fraunhofer IKS is working on the possibility of exploring and validating the adaptation space in cognitive systems as early as the design phase. Particularly when it comes to risks, adaptations must be available in order to enable an adequate level of safety. If necessary, the space for potential adaptations must be restricted as well

 This calls for new tools since adaptable and adaptive software architectures exhibit a high degree of complexity. In order to validate safety-relevant, complex functions, design enhancements such as the automatic generation of configurations and runtime models can be taken under consideration.

Fraunhofer IKS is furthermore examining new planning and evaluation approaches that enable the optimal and dependable distribution of safety-critical functions and additional failure mechanisms in embedded systems, as well as open systems. Within end-to-end architectures, these mechanisms also react decentrally and at different levels. With this in mind, for these architectures Fraunhofer IKS is conducting research into monitoring and control mechanisms that allow cognitive functions to exhibit resilient behavior even beyond the limits of the subsystems.

Artificial intelligence (AI) fault recognition and management

A key factor in the utilization of intelligent systems in safety-critical applications is fault handling and uncertainty management. Conventional software systems provide warnings when faulty input occurs. This doesn’t happen with AI systems. On the contrary, modern methods such as DNNs continue to supply seemingly safe predictions. This is especially critical given that AI systems are heavily based on the underlying data and the resulting assumptions. For this reason unknown input data (out-of-distribution samples) or changes in the data distribution (dataset shift) lead to undependable, or in some cases, even safety-critical results.

The Fraunhofer Institute for Cognitive Systems IKS is therefore working on two approaches. One entails examining the input for deviations to the expected behavior. This encompasses data distribution that was not covered by the training or tests for example. Secondly, researchers are developing uncertainty quantification methods that make dependable statements regarding the uncertainty of the AI results. Based on the uncertainty, the decision-making process can be adapted or alternatives such as further data sources or other subsystems used.

References

 

Resilient Platforms for Autonomous Cyber-Physical Systems

In this project, Hitachi and Fraunhofer IKS developed a resilient architecture for cloud-based control systems based on the example of an automated valet parking service in a parking garage.

 

Continental and Fraunhofer IKS make autonomous vehicles safer

Together with Continental, Fraunhofer IKS was able to create a concept for the dynamic distribution of vehicle functions and develop a technical safety concept that describes an implementation of the identified safety requirements.