Enhancing EV Charging Station Reliability and RAS Safety

Abstract

The surge in electric vehicle (EV) adoption prompts companies to prioritize dependable charging station designs, despite hurdles in maintaining consistency. A newly proposed design, featuring 36 ports, employs both uniform and non-uniform arrangements, subjected to rigorous testing with systems ranging from 50 to 350 kW. Failure rates are projected through meticulous assessments based on MILHDBK217F and MILHBK-338B standards, employing binomial distribution and cost analysis to gauge port reliability and overall station success rates. This innovative design not only bolsters voltage stability but also curtails maintenance expenses by bolstering port reliability.In the realm of robotics and autonomous systems (RAS), Deep Reinforcement Learning (DRL) demonstrates exceptional prowess but grapples with the risk of unsafe policies, potentially resulting in perilous decisions. To address this concern, a novel study introduces a reliability evaluation framework tailored for DRL-driven systems, leveraging formal neural network analysis. This framework adopts a two-tiered verification strategy: firstly, by assessing safety locally using reachability tools, and secondly, by aggregating local safety metrics across various tasks to evaluate global safety. Empirical validation validates the efficacy of this framework in fortifying the safety of RAS.