Waveform-Feature-Driven Diagnosis and Physics-Constrained Self-Correction of Representative EMT Component Implementation Faults: A Multi-Agent Feasibility Study.

Authors

Jieran Zhang, Jie Zhang, Pan Wu, Jin Xu, Keyou Wang

Published in

Sensors (Basel, Switzerland). Volume 26. Issue 18. Sep 12, 2026. Epub Sep 12, 2026.

Abstract

Latent faults in electromagnetic transient (EMT) components can alter initialization, switching-event, and history-state semantics while producing sparse or small waveform discrepancies. This study formulates a reference-model-based verification as a closed-set, waveform-guided diagnosis and constrained source-repair problem. Global, local-window, and event-level evidence rank fault mechanisms or trigger abstention. Repairs must pass compile-feasibility, behavioral-conformance, and applicability-aware physics-and-discretization audit gates. A multi-agent workflow implements this process. InvSqrt and single-phase-breaker faults were corrected to exported-waveform precision; the breaker record contained only two nonzero differences among 20,000 samples. The VARRL correction yielded 0.1% residual node-voltage RMS differences. A within-case VARRL extension across four matched conditions and three source variants exposed condition-dependent activation, a weakly separated effect of 0.018% RMS relative difference, a smaller inseparable effect routed to expert review, and a mixed-snapshot variant that reduced reference error but violated a discretization contract. An author-constructed ten-candidate challenge set spanning three component structures compiled in RSCAD FX 2.3 CBuilder; the deterministic audit returned Pass for three designed semantics-preserving refactorings, Reject for six isolated rule violations, and Expert review for one underspecified temporal convention. These results provide mechanism-level and preliminary audit-coverage evidence rather than population-level fault frequency, independent classification accuracy, general rule sufficiency, or formal verification.

PMID:
42817359
Bibliographic data and abstract were imported from PubMed on 01 Oct 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.