Adaptive Fusion Networks for Cable Material Durability Assessment via Multimodal Data Integration

Predicting cable durability is vital for safe and efficient electrical systems. This research proposes an Adaptive Fusion Network (AFN) that integrates normalized sensor data (e.g., partial discharge, corrosion) and encoded visual condition ratings (Good, Medium, Poor) via concatenation and processed through dense layers with ReLU activation. To address incomplete labeling, a pre-trained model annotated unlabeled data from 2,500 15-kV XLPE cable segments across multiple years, creating a diverse 10,000-sample dataset. The AFN achieved an MSE of 0.012547, MAE of 0.046415, and R2 of 0.991043, outperforming benchmarks like Random Forest (MSE 0.135725, R2 0.903107) by 89% in MSE reduction, highlighting its potential for real-time durability monitoring and predictive maintenance in power systems.