Inverse Problem for Health Monitoring of Functionally Graded Plates Using Deep Learning

In this work, damage detection of functionally graded (FG) plates is studied by deep learning (DL). A dataset produced from isogeometric analysis (IGA) is utilized to feed into learning models. The displacement in the plate’s plane and the displacement across the plate thickness are respectively presented by non-uniform rational B-spline (NURBS) functions and the generalized shear deformation theory (GSDT). The rule of mixture scheme is employed to homogenize FG material. Then, several learning models are constructed to recognize the damage of FG plates based on inverse problems. More concretely, the input data of these models are time-series acceleration signals, while the output ones are the parameters for representing thickness change and two gradient indexes for representing the material distribution in the plate plane. Moreover, due to the limitation of measurement sensors in practice, only a few signals recorded at key degrees of freedom (DOFs) are used as the input data for learning models. Acquired results have proven the reliability of the proposed paradigm. Deep neural network (DNN) and Extreme Gradient Boosting (XGBoost) are utilized to test a square FG plate to confirm the validation of the approach.