A hybrid Chirplet Transform and neural network approach for compensating environmental effects in guided wave structural health monitoring

Environmental effects often interfere with the accurate diagnosis of structural damage in guided wave-based structural health monitoring (SHM) systems. Distinguishing these environmental influences, termed environment-sensitive features, from damage-sensitive features is critical, as factors such as temperature can mask indicators of structural damage such as cracks. This study proposes a hybrid methodology that combines Chirplet Transform (CT) for feature extraction and artificial neural networks (ANNs) for environmental compensation. Guided elastic wave data were collected from an undamaged structure under varying environmental and operational conditions. Preliminary analysis under controlled conditions justified the feature selection, showing a strong linear correlation between the CT’s time-shift coefficient and temperature ( R =0.99), and between its scaling modulus and damage size ( R =0.98). An ANN was then trained on data from an undamaged structure to model the baseline relationship between temperature and these CT coefficients, achieving a validation root mean squared error of 14%. For damage assessment, the trained ANN generates pseudo-reference coefficients for the current temperature, allowing for the damage quantification by comparing them with the coefficients from the measured signal. The approach circumvents the extensive data requirements of optimal baseline selection and the oversimplified models of baseline signal stretching, enabling a direct framework for damage quantification with low computational cost. By successfully separating environmental effects from damage indicators in a simplified scenario, this work presents a promising and computationally efficient methodology for damage assessment in SHM.