Admittance-spectrum-based dual-resonance feature prediction and target-oriented design of annular piezoelectric ceramics for conducted EMI suppression

Abstract

The capacitance of Y capacitors is limited by leakage-current requirements, whereas high-inductance common-mode chokes (CMCs) increase filter mass and volume. This study develops a measurement-informed closed-loop framework linking measured conducted electromagnetic interference (EMI) targets to dual-resonance prediction, inverse geometry design, commercial-size mapping, and experimental validation of annular lead zirconate titanate (PZT) ceramics. A full-factorial finite-element method (FEM) dataset of 1232 geometries was used to model the first two resonance frequencies and peak admittances using six surrogates under strict nested fivefold cross-validation. An output-specific ensemble of least-squares boosting (LSBoost) and a multilayer perceptron (MLP) achieved pooled mean absolute percentage errors of 0.161%, 0.464%, 0.326%, and 1.292% for f₁, Y₁, f₂, and Y₂, respectively. For measured targets at 162 and 486 kHz, particle swarm optimization identified an inverse-design solution that was mapped to a commercial PZT ring of 4.50/1.50/2.00 mm (outer radius/inner radius/thickness). Independent FEM and admittance measurements yielded resonance-frequency errors of 0.28% and 0.29%, with peak-admittance discrepancies of 12.76% and 9.65%. Without an external CMC, the PZT ring provided 1.45 and 1.35 dB greater attenuation than a 680-pF Y capacitor at the target frequencies; two additional nominally identical rings showed the same advantage. In complete-filter tests, the 2.2-mH CMC–PZT configuration provided 2.81 and 5.14 dB greater attenuation than the 20-mH CMC–Y-capacitor reference. Because the CMC values differed, these are configuration-level results; for the implemented passive-component sets, the PZT-assisted set was 73.6% lighter and had a 61.8% smaller estimated volume.