Resumo:
Assessing electric power quality (PQ) requires measuring spectral indicators, comparing
them against standard limits and issuing explainable diagnoses. Fragmented tools can
dissociate measurement conventions from the final verdict and compromise auditability.
This dissertation presents SQE-UNIFEI, a modular executable framework for simulation,
canonical measurement, multi-standard screening and PQ diagnostics, based on versioned
software contracts and regenerable evidence. Its six features span synthetic signal generation
through real-time monitoring and share the same measurement record across the
graphical interface, command line and embedded package. The measurement contract
marks harmonic orders not observed within the sampling window as unavailable rather
than introducing artificial zeros. The normative catalogue, version 3.5.2, brings together
PRODIST, IEEE, CIGRÉ, IEC and EN 50160 references, making limits, harmonic scopes,
percentage bases and boundary operators explicit, along with the verification status of
each clause. Adversarial verification confirmed 173 defects, 68 of which affected measured
values, verdicts or figures; corrective changes were checked by regression tests within a
suite of 4766 collected tests, 12 skipped and 0 failures. Native acceleration reduced harmonic
computation cost while preserving bitwise equivalence with the reference within
the evaluated environment. The computational evaluation used synthetic signals and oscillographic
records. Comparing a deterministic rule with a neural network incorporating
a differentiable physics layer on a corpus with disjoint source and event partitions,
component-wise ablation produced a confidence interval including zero, and the paired
McNemar test yielded 𝑏 = 0 disagreements favouring the network across the 4 standards
evaluated. On this corpus and under this rule, the learned branch in the served inference
path added no measurable accuracy to the normative layer, supporting explainable
and computationally efficient deterministic solutions within this scope. The framework
contributes by integrating simulation and multi-standard diagnostics in a verifiable workflow,
with cross-interface consistency and computational traceability of measurements
and decisions, providing a foundation for embedded applications and electrical system
monitoring.