Accelerating composite sandwich plate analysis: A hybrid higher-order XFEM and ANN approach for natural frequency prediction

Free vibration analysis of laminated composites plays a critical role in design optimization, material selection, structural integrity assessment, vibration suppression, and fatigue life prediction. A novel coupled higher-order extended finite element method (HO-XFEM) and artificial neural network (ANN) approach is implemented for predicting the natural frequency of sandwich plates with a composite layout of 0°/θ°/0°/core/0°/θ°/0°, where θ varies from 15 to 90 degrees. A total of 557 data points are generated through HO-XFEM, which is then utilized to construct an optimized ANN model for predicting nondimensionalized natural frequency. The ANN model, designed specifically to predict the first mode of natural frequency, is optimized with a structure of 4-90-90-1 and softmax transfer functions in intermediate layers. It predicts the nondimensional natural frequency with the average, minimum, and maximum mean-squared error values of 0.0984 %, 0.00339 %, and 0.49 %, respectively. The optimized ANN model computes the natural frequency for different sandwich plate configurations a few million times faster compared to HO-XFEM and thus establishes a transformative framework for real-time structural integrity assessment.