Virtual Sensing Reliability Characterization for Next-Generation Green Aircraft Resilience

Accurate sensor data is essential for effective Structural Health Monitoring (SHM) in aerospace systems. However, sensor malfunctions and structural anomalies can lead to data corruption, complicating the distinction between physical damage and faulty measurements especially under high-dimensional and noisy conditions. This work introduces a hybrid model and data-driven framework designed to detect and correct sensor faults while continuously evaluating sensor reliability in evolving operational scenarios. The methodology features a two stage compression pipeline that first applies Proper Orthogonal Decomposition (POD) to extract dominant spatial patterns from high-dimensional strain data, followed by Self-Organising Maps (SOM) to cluster fault-relevant features in a topologically meaningful way. To emulate real-time SHM conditions, static strain fields are augmented with time-varying synthetic noise and fault models, producing spatio-temporal sequences for robust learning. These inputs are processed through a hybrid deep learning model that combines Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks to perform three core tasks: (i) classify the current sensor state, (ii) correct corrupted signals through regression-based recovery, and (iii) dynamically update sensor reliability metrics based on prediction confidence and fault recurrence. The framework is demonstrated on a composite wing panel with fibre-cut damage, showcasing its ability to distinguish between structural degradation and sensor anomalies.