Uncertainty quantification for damage detection in 3D printed pure PLA auxetic lattice structure using ultrasonic guided waves and Flipout probabilistic convolutional neural network

This paper presents a novel framework for health diagnosis and uncertainty quantification in 3D-printed auxetic structures made of polylactic acid. The approach integrates compression with simultaneous ultrasonic testing to capture ultrasonic signals across various deformation states. Critical damage deformation is identified through observed patterns and signal energy variations. Damage-sensitive features, extracted via Hilbert transform, serve as inputs for a Flipout probabilistic convolutional neural network (FPCNN). The FPCNN, incorporating pseudo-independent weight perturbations and a Gaussian probabilistic layer within a modified VGG-13 architecture, predicts structural deformations and associated uncertainties. A warm-up algorithm has been used to optimize the learning rate. The framework, based on variational inference and conditional covariance law, effectively quantifies aleatoric and epistemic uncertainties in damage detection. This framework's feasibility is demonstrated through compression, ultrasonic tests and the FPCNN.