Quantitative Damage Prediction for Composite Laminates Based on Wave Propagation and Artificial Neural Networks

Targeted at an online health monitoring technique for in-service composite structures, a Lamb wave propagation-based deterioration assessment approach is developed using an artificial neural network (ANN) algorithm and a PZT transducer network. Structural dynamic responses are numerically simulated using three-dimensional FEM analyses, and signal characteristics are then extracted with a Signal Processing and Interpretation Package (SPIP) in terms of the wavelet transform technique. A damage parameters database (DPD) is constructed to accommodate the extracted wave spectrographic characteristics, and adopted for ANN training under the supervision of an error-backpropagation neural algorithm. The validity of this methodology is evaluated by identifying through-hole-type damages in [45/45/0/90]s quasi-isotropic CF/EP (T650/F584) laminates. The results exhibit excellent quantitative prediction for damage in the CF/EP composites, including position, geometric identity, and orientation. Additionally, the dependence of ANN performance on inherent network configurations is also evaluated.