Sparse wavefield reconstruction based on Physics-Informed neural networks

In recent years, the widespread application of laser ultrasonic (LU) devices for obtaining internal material information has been observed. However, this approach demands a significant amount of time to acquire complete wavefield data. Hence, there is a necessity to reduce the acquisition time. In this work, we propose a method based on physics-informed neural networks to decrease the required sampling measurements. We utilize sparse sampling of full experimental data as input data to reconstruct complete wavefield data. Specifically, we employ physics-informed neural networks to learn the propagation characteristics from the sparsely sampled data and partition the complete grid to reconstruct the full wavefield. We achieved 95% reconstruction accuracy using four hundredth of the total measurements. The proposed method can be utilized not only for sparse wavefield reconstruction in LU testing but also for other wavefield reconstructions, such as those required in online monitoring systems.