Answers are grounded in this paper and cite their sources.
The integration of machine learning into safety-critical electrical systems is fundamentally constrained by certification requirements that demand deterministic behavior, traceable decision logic, and auditable safety compliance. Existing learning-based approaches predominantly rely on opaque black-box models or post-hoc explainability techniques, neither of which provide guarantees of reproducibility, causal traceability, or bounded behavior under disturbed and faulted operating conditions. This study introduces an explainable machine learning architecture in which transparency, auditability, and safety compliance are enforced as architectural invariants rather than auxiliary analytical features. The framework embeds physics-aligned feature representations, interpretable decision-path learning, and constraint-aware inference with safety-conditioned confidence estimation, ensuring deterministic traceability from electrical variables to final system decisions. Formal analysis establishes guarantees on explanation fidelity, decision determinism, and robustness under bounded perturbations, explicitly linking explainability to safety margins and operational constraints. Extensive experimental validation on representative safety-critical electrical system scenarios demonstrates that the proposed architecture achieves predictive performance comparable to state-of-the-art black-box and physics-informed models while exhibiting superior explainability stability, confidence consistency, and safety compliance near certified operating boundaries. The results demonstrate that explainability, when enforced architecturally, eliminates the traditional trade-off between transparency and accuracy, positioning the proposed framework as a certification-ready learning-based decision component for safety-critical electrical infrastructures.
No takes yet. Share an insight, caveat, or question.
SANKAR S (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: