Partial discharge in power transformers: A critical review of acoustic emission and UHF-based detection and localization methods
Partial discharge (PD) is a principal precursor of insulation deterioration in power transformers, and its early detection and accurate localization are essential to prevent progressive degradation and catastrophic breakdown. This review compares acoustic emission (AE) sensing and ultra-high-frequency (UHF) electromagnetic sensing and establishes the conditions under which each technique, or their combination, is to be preferred. The paper is presented as a structured critical review conducted under a transparent, pre-defined search protocol. Six bibliographic databases were searched using documented Boolean strings for English-language records published between 1990 and 2025; records were screened against explicit inclusion and exclusion criteria, appraised for methodological quality, and reduced to a final evidence base whose selection is documented in a study-selection flow diagram (Figure 1). To remove the ambiguity of unqualified descriptors, every comparative judgement in this review is made against an explicit rating rubric (Table 3) in which detection sensitivity, localization accuracy, noise immunity, detection range and cost are given operational definitions with quantitative or procedural thresholds. Assessed against these definitions, AE and UHF do not resolve into a simple ordering; their relative performance is conditional on sensor placement and on system architecture. UHF sensing attains the lower detection threshold, and internally mounted UHF sensors benefit from the electromagnetic shielding afforded by the grounded tank, whereas externally mounted UHF sensors are exposed both to substation and broadcast interference and to aperture attenuation — which reconciles the apparently contradictory claims regarding UHF noise immunity. The corresponding contradiction concerning AE localization is likewise resolved by architecture rather than by physics: all-acoustic time-difference-of-arrival (TDOA) localization must solve for four unknowns and is vulnerable to structure-borne arrivals preceding the direct oil-borne path, whereas AE triggered by a simultaneous electrical or UHF reference reduces the problem to three unknowns and achieves reported errors of the order of 0.1 m. Localization uncertainty is shown to be dominated not by the choice of estimator but by acoustic-velocity uncertainty, sensor-position error and timing synchronization. The evidence supports a hybrid architecture — UHF for detection and time reference, AE for spatial localization — rather than a choice between the two. The review contributes an operationally defined comparison rubric, a reconciliation of contradictory performance claims in the existing literature, a rigorous statement of the second-order cone programming (SOCP) localization formulation together with its assumptions and sensor-count requirements, and an evidence-based comparison of offline, periodic-online and continuous monitoring architectures.