Inverse design of phononic crystal filter for higher harmonics modulation in nonlinear ultrasonic testing: Numerical...

Signal modulation of higher harmonics can reduce the complexity of signal processing in nonlinear ultrasonic testing (NUT), which are effectively achieved by utilizing the bandgap characteristics of phononic crystals. However, proposing effective configuration of phononic crystals for NUT could be a significant challenge based on the complex conventional design process. Therefore, machine learning technique provides a feasible approach for inverse design of phononic crystal structures. This paper establishes an inverse design framework for phononic crystal configurations based on multi-branch convolutional neural networks (CNN) and generative adversarial networks (GAN). Then, the phononic crystal filter (PCF) with a truncated conical configuration is proposed. Meanwhile, numerical simulations are employed to verify the filtering ability of the designed filter, and the influences of the spatial arrangement and geometric parameters of PCF's unit-cell are also numerically investigated. Finally, the 3D-printed truncated conical PCF is manufactured to experimentally validate the efficient filtering ability for higher harmonics modulation. This study provides numerical and experimental basis for the application of machine learning-based inverse design of PCF in non-destructive testing.