ARES-Cert: Trustworthy Surrogate-Guided Verification for Reliable Electric Drive Design

Reliability verification for integrated electric drive systems is challenging because critical failure modes arise from coupled electrical, thermal, mechanical, and control effects across a large operating space, while exhaustive high-fidelity verification is computationally expensive. We introduce \textbf{A}ctive \textbf{R}eliability-aware \textbf{E}ngineering and \textbf{S}ynthesis for \textbf{Certification} (\sys), a verification-guided framework that uses learned multi-physics surrogates to incorporate reliability information directly into system design. \sys introduces Failure-Mode-Aware Active Verification (FMAV), which combines epistemic uncertainty with proximity to physical failure boundaries to prioritize high-fidelity evaluations across heterogeneous failure modes. It further represents operating-condition-dependent reliability through a Dynamic Credibility-Robustness Space (DCRS) and integrates this signal into design optimization. We evaluate surrogate trustworthiness through safety-critical prediction errors, failure-boundary behavior, uncertainty calibration, and distribution shift, while final certification is performed using independent high-fidelity evaluations. Across four electric drive configurations, \sys achieves safe-operation rates of $97.3\%$-$98.7\%$, improving by 3.2-6.1 percentage points over performance-only optimization and by 1.3-2.8 points over fixed-margin optimization. These results show how uncertainty-aware surrogate modeling and targeted physical verification can support reliable design without relying on learned predictions for final certification.