Towards Formal Design of FDIR Components with AI

The development of accurate, reliable and effective FDIR (Fault Detection, Identification and Recovery) components is essential in several application domains, to meet the dependability constraints and to accomplish the higher degree of autonomy required in future missions. In this work, we report on an ongoing activity that addresses the formal design, development and validation of FDIR integrating rule-based components with components based on Machine Learning (ML) and Deep Learning (DL). We show that the integration of symbolic and AI techniques can substantially improve the effectiveness and efficiency of FDIR management functions, while formal tool-supported verification and validation can provide a formal guarantee of the quality of the FDIR systems before they are implemented and deployed. This activity is being carried out within the AIFDIR study, funded by the Italian Space Agency (ASI) under the “Innovative Space Technologies” initiative. The AIFDIR methodology will be implemented and demonstrated using TASTE, a tool developed by the European Space Agency (ESA), which follows the MBSE (Model-Based System Engineering) approach. TASTE has been recently extended to enable the modeling of HW components and their possible failures, and the verification and validation using automated techniques based on model checking. TASTE will be further extended to allow for modeling and verification of systems including both symbolic and ML/DL-based components, and to support the deployment on the target hardware. A further contribution of the project is the development of a reference architecture for AIFDIR and its demonstration on case studies of interest.