Data-Driven Maintenance Decision Analytics Using Multimodal Structural Health Indicators for Composite Asset Fleets
Main article
Abstract
The safety, airworthiness, and lifecycle cost efficiency of composite asset fleets depend critically on the ability to derive reliable, interpretable health indicators (HIs) that integrate heterogeneous sensing modalities. Traditional single-modal approaches based on acoustic emission (AE) or guided-wave (GW) interrogation alone suffer from limited representativeness, poor generalization across fleet units, and insufficient monotonicity under stochastic damage accumulation. This study presents a comprehensive data-driven maintenance decision analytics framework that constructs multimodal structural health indicators (SHIs) by fusing passive acoustic emission signals and active guided-wave damage indices. The framework encompasses four integrated layers: multimodal data acquisition and synchronization, modality-specific feature engineering, semi-supervised inter-modality fusion, and fleet-level maintenance decision analytics. Prognostic quality criteria—including monotonicity (Mo), prognosability (Pr), and trendability (Tr)—are embedded directly into the learning objectives to yield health trajectories suitable for remaining useful life (RUL) estimation and maintenance scheduling. Experimental results on a composite panel fatigue dataset demonstrate that the proposed multimodal SHI achieves composite prognostic scores exceeding 0.88, substantially outperforming single-modal baselines. Fleet-level decision analytics translates SHI outputs into condition-based maintenance triggers, demonstrating a 23.4% reduction in unplanned downtime relative to fixed-interval scheduling. The findings underscore the practical value of multimodal fusion for industrial prognostics and health management (PHM) of composite structures.
