Verifier-Gated Multi-Candidate Supervisory Selection for Constraint-Aware Flux-Weakening in Permanent Magnet Synchronous Motor Drives

Flux-weakening current-reference generation must balance torque tracking, current use, voltage reserve, and continuity. This paper develops a verifier-gated multi-candidate supervisor for permanent magnet synchronous motor drives. Six deterministic reference laws are screened before a history-dependent score selects an admissible candidate. For a nonempty verified set, the encoded nominal-model constraints hold independently of ranking weights. Offline cross-entropy method calibration leaves the verifier unchanged. Across five quasi-steady profiles, the supervisor reduces mean squared current by 3.46% relative to a margin-matched scalar comparator, while torque root mean square error increases by 4.4%. Equal-budget retuning preserves this trade-off. An additional continuous nonlinear-programming comparator achieves lower current use and torque error than the supervisor, with higher measured host latency. Candidate ablations expose redundant options on the chosen profiles. Fixed-reference vertex checks satisfy the voltage limit for a synthetic ±5% parameter box, whereas larger boxes expose margin failures. Sixty sampled electrical closed-loop simulations include switching inverters, delays, and modeled nonidealities; no current-limit violation occurs in that panel. These results establish ranking-independent model-side admissibility and quantify its operating limits and computational trade-offs without implying physical-drive or real-time certification.