Autoencoder enhanced Bayesian fusion method for damage imaging of composite materials under variable temperature using ultrasonic guided waves

Abstract

Carbon Fiber Reinforced Polymer (CFRP) structures are prone to damage under variable temperature environments. Temperature fluctuations not only accelerate damage evolution but also adversely affect guided wave–based techniques commonly used for CFRP damage detection. To address this issue, this study proposes an Autoencoder (AE) -based temperature compensation method combined with a Bayesian fusion framework using ultrasonic guided wave data. The goal is to mitigate the effects of environmental temperature fluctuations and enhance the accuracy of defect localization. The approach is proposed by training the AE with baseline signals collected under a subset of temperature conditions and then reconstructing baseline signals for other temperature by processing damage signals. Experimental validation shows that, with baseline data from only 39 temperature points, the proposed method can accurately reconstruct baseline signals at an additional 117 temperature points. Subsequently, wavelet transform is employed to extract the Time of Flight (TOF) of scattered signals, and a Bayesian data fusion framework is utilized to integrate the Reconstruction Algorithm for Probabilistic Inspection of Damage (RAPID) method with TOF information for precise defect localization. The reconstructed baselines closely align with actual measured baselines, confirming the efficacy of the proposed temperature compensation strategy. In two representative damage scenarios, the proposed Bayesian fusion method reduces the average localization errors by 38.52% and 26.43%, respectively, compared with the conventional RAPID method, while significantly suppressing imaging artifacts.