Hybrid Veranstaltung / 21. Juni 2022 - 23. Juni 2022
embedded world Conference 2022
embedded world – die globale Plattform der Embedded-Community
Ob Sicherheit elektronischer Systeme, verteilte Intelligenz, das Internet der Dinge oder E-Mobility und Energieeffizienz – die Fachmesse embedded world lässt Sie die ganze Welt der eingebetteten Systeme erleben.
Das Fraunhofer IESE nimmt mit folgenden Vorträgen daran teil:
Dienstag, 21.06.2022, 16:00 - 17:45 Uhr; Session 8.3
Machine Learning Assurance With SafeML
Dr. Ioannis Sorokos
Machine Learning (ML) promises major technological advancements that may radically change society. Towards this end, means of establishing trust regarding the performance of ML must first be acquired and applied. For ML to be trusted, its behavioral uncertainty must be managed, and appropriate assurance provided to convince its benefits to all stakeholders. In this talk, we will expand on the three main challenges. First, the dataset integrity: the training dataset must provide sufficient coverage across the application domain. The training stage of ML models is critical, and must be supported by appropriate dataset for the application. Second, model integrity: an ML model and its development process should be rigorously evaluated before deployment. The recent UL 4600 standard advocates the use of assurance cases for ML assurance. Such cases structure the arguments linking high-level requirements with evidence e.g. of ML model training and testing quality. Third, operational integrity: when placed into operation, ML models must be monitored with respect to the uncertainty of their response. When such uncertainty is detected to exceed acceptable thresholds, they cannot be trusted and alternative systems must be employed. We further address the possible solutions to these challenges, including our ‘SafeML’ approach and how it can be applied towards supporting ML data synthesis, ML assurance case construction, and ML uncertainty monitoring.
Donnerstag, 23.06.2022, 14:45 - 15:15 Uhr; Session 6.13
Assuring the Fault Tolerance of System Architectures by Virtual Validation
Development and test processes for complex technical systems requires efficient techniques for assuring the manifold quality properties and customer requirements. By using model-based solutions, virtualization and simulation, the early validation of selected properties is enabled. Virtual validation as a key asset of these novel QA processes uses a fully virtual set-up for the validation of system properties, i.e., all components of the validation scenarios, the test object and its environment are realized by software or executable models. The challenge lies in the creation and availability of suitable models, virtual prototypes and execution environments.
This article presents the approach of virtual validation of fault tolerance with the simulation and virtual engineering framework FERAL (Focused Evaluation on Requirements and Architecture Level), which has been extended for the systematic fault injection on the levels of communication protocols, component interfaces, and component implementation. The approach has been evaluated in cooperation with a partner from industrial automation domain. Dedicated adapters for supporting domain-specific communication protocols such as MQTT and OPC-UA and customer-specific design and validation tools have been developed. In the project, specific flaws in the architecture design of the system could be detected and corrected. Evaluation results will be presented.