Fast and robust strain signal processing for aircraft structural health monitoring

This work elaborates a fast and robust structural health monitoring scheme for copying with aircraft structural fatigue. The type of noise in structural strain signals is determined by using a statistical analysis method, which can be regarded as a mixture of Gaussian-like (tiny hairy signals) and impulse-like noise (single signals with anomalous movements in peak and valley areas). Based on this, a least squares filtering method is employed to preprocess strain signals. To precisely eliminate noise or outliers in strain signals, we propose a novel variational model to generate step signals instead of strain ones. Expert judgments are employed to classify the generated signals. Based on the classification labels, whether the aircraft is structurally healthy is accurately judged. By taking the generated step count vectors and labels as an input, a discriminative neural network is proposed to realize automatic signal discrimination. The network output means whether the aircraft structure is healthy or not. Experimental results demonstrate that the proposed scheme is effective and efficient, as well as achieves more satisfactory results than other peers.