Dr. Mihai A. Diaconeasa of the PRA group was awarded a U.S. Dept. of Energy grant for a collaborative project led by the Idaho National Laboratory on a simulation-based reliability methodology development for autonomous controls and adversarial human actions involved in fission battery designs.
Fission batteries are unique from a technological standpoint, given their proposed autonomous operation and tamper-proof features. Also, the actual operation of fission batteries will be novel in that local users may only have simple on/off control capabilities, while the manufacturer will need to be able to remotely monitor a fleet of units spread across different geographical regions. Furthermore, it must be self-detecting/protecting from human adversaries. The burden of local monitoring and control is therefore shifted to the autonomous technology and limited remote-monitoring capability of the manufacturer.
Reliability analysis for cyber-physical systems (CPS) such as fission batteries is challenging since modern CPS incorporate distributed and networked heterogeneous software, hardware, and physical components that operate and interact in tandem. Human actions, such as those of the adversary, can also play an important role and need to be considered in the design process. These ingredients yield highly structural and behavioral complexity for CPS models, making them computationally expensive to predict, model, and test. Consequently, highly sophisticated failure scenarios emerge, revealing new challenges for state-of-the-art quantitative reliability metrics and evaluation methods.
To execute our project, the risk modeling needed for autonomous operations will first require newly developed dynamic PRA methods, due to the self-diagnosis, self-adjustment, and duration-prediction capabilities needed for autonomous operations. Second, reliability modeling will require analyzing autonomous control, associated error-detection algorithms, and human actions for both cyber and tamper-proof designs. Finally, to perform the reliability/resilience evaluations, we will use Dual-Graph Error Propagation Models (DEPM) based on discrete-time Markov chain (DTMC) models.