Cement mortar composite has a wide range of applications on construction sites, including masonry, plastering and concrete repair. In construction sites, scabbling process is a method to remove from a few millimeters to several centimeters of defect concrete surfaces. As a result, it is essential to investigate the scabbling characteristics for cement mortar with different silica sand proportion in laser scabbling process. In this study, 5 types of cement mortar with different silica sand proportions in mixing were fabricated and scabbled by using a high-density power laser beam. The effects of silica sand proportion in color changing and penetration depth of the samples after laser scabbling process were studied. Furthermore, the generation of micro-cracks and pores were observed by using scanning electron microscopy (SEM). In addition, chemical composition changes between processed zone and non-processed zone were also evaluated by Energy Dispersive X-ray (EDX) analysis. The results of this study are expected to provide valuable knowledge in understanding of the laser scabbling process for cement-based materials.
12 August 2025 in International Journal of Fatigue
Accurate state-of-charge (SOC) estimation is essential for ensuring the safe and reliable operation of lithium-ion batteries. Linear ultrasonic features such as time-of-flight and signal amplitude have been widely investigated as indicators for battery state monitoring. However, these features mainly depend on second-order elastic constants, which limit estimation accuracy. This study presents a nonlinear ultrasonic framework for accurate battery state estimation by combining bispectral analysis with a long short-term memory (LSTM) network. First, a theoretical relationship between the ultrasonic nonlinearity parameter and battery state is established based on elasticity theory. Next, ultrasonic measurements are performed using a through-transmission configuration. A modified ultrasonic nonlinearity parameter is extracted from reconstructed signals obtained through bispectral analysis and the LSTM network. The proposed framework achieves a SOC estimation error below 3% and improves estimation accuracy by a factor of 3.06 compared with linear ultrasonic features. Under C/2, 1C and 2C conditions, root mean squared errors of 3.1%, 2.4% and 2.9% are obtained, respectively. The results demonstrate that nonlinear ultrasonics combined with deep learning improves real-time battery state estimation accuracy.
Characterization of subsurface and internal defects in thin-walled metallic structures is important for maintaining structural integrity in safety-critical engineering applications. Shearography enables rapid full-field wavefield measurement and provides high spatial resolution for defect inspection. However, quantitatively characterizing defects from measured wavefield data remains a significant challenge. Physics-informed neural networks (PINNs) offer a unified framework combining data and physics for solving wavefield inversion problems. Nevertheless, conventional time-domain formulations suffer from substantial computational cost due to the high-order spatial and temporal derivatives, particularly when analyzing narrowband signals from piezoelectric excitation. This study presents a frequency-domain defect-parameterized PINN framework for rapid monochromatic wavefield inversion and defect characterization. The governing equations are formulated in the frequency domain to reduce problem dimensionality. A mapping function is integrated into the governing equations to enable direct inversion of defect location and dimensions through a compact parameter set. Finite difference discretization is introduced to compute second-order spatial derivatives within the physics loss, replacing automatic differentiation to improve computational efficiency. Additionally, an SVD-based motion magnification with spatial filtering is applied to enhance the signal-to-noise ratio of shearography measurements prior to inversion. The framework is validated using simulated data and experimental shearography measurements acquired from aluminum plates containing rectangular and circular defects. Results under varying excitation conditions demonstrate accurate reconstruction of wavefields and reliable defect localization, with an average normalized mean squared error of approximately 0.083% for the predicted wavefield and a mean intersection-over-union of about 0.523 for defect characterization. In addition, the finite difference-based implementation reduces training time by approximately 1.95-fold while maintaining comparable inversion accuracy.
14 November 2023 in Journal of Building Engineering