Synergistic quantification of trace metal contamination on insulators: a gated fusion framework combining LIBS spectra and event-reconstructed plasma images

Abstract

Alongside conventional salt contamination, metallic particulate deposits on outdoor high-voltage insulators can aggravate pollution-induced flashover, demanding rapid in situ quantification of trace metals. Laser-induced breakdown spectroscopy (LIBS) is well suited for the rapid, in situ multi-element assessment of insulator contamination; however, its quantitative accuracy is often limited by plasma fluctuations and matrix effects. Prior works on insulator contamination reported only moderate calibration performance for several trace metals (e.g., Ni : R² = 0.773, Cu : R² = 0.747, and Mn : R² = 0.615). To address these limitations, a dynamic vision sensor (DVS) is integrated with LIBS to record plasma emission dynamics and generate event-reconstructed plasma images that complement the spectra. Simulated contamination samples were prepared with Ca contents of 10.59–22.47 wt% and trace Fe, Cu and Zn contents of 0.10–0.32, 0.05–0.37 and 0.32–1.24 wt%, respectively, yielding 2000 paired LIBS–DVS measurements. A sample-group-wise leave-one-out cross-validation strategy was adopted to evaluate the generalization to unseen sample groups. GFMT Net combines a transformer spectral encoder, a 2D-CNN image branch, gated feature fusion and multi-task regression for the simultaneous prediction of Ca, Fe, Cu and Zn. Under LOOCV, GFMT Net achieved R² values of 0.9736, 0.9649, 0.9712 and 0.9741 with RMSEs of 0.5440, 0.0111, 0.0159 and 0.0402 wt%, respectively. Compared with Transformer-LIBS, the best LIBS-only baseline, GFMT Net decreased the RMSEs by 63.28%, 58.10%, 63.41% and 61.24% for Ca, Fe, Cu and Zn, respectively. Ablation and interpretability analyses confirmed that DVS images provided complementary plasma-state information to LIBS spectra. These results demonstrate that event-assisted gated fusion improves robust trace metal quantification on power insulation equipment.