The incremental hole-drilling technique requires calibration constants to relate the measured strain data to the residual stress distribution that existed in the material prior to drilling. For laminated composite materials, the calibration constants must be determined individually from finite element (FE) analyses for the specific combination of laminate configuration, material properties and experimental parameters used. Depending on the computational method and uncertainty sources considered, hundreds of FE analyses may be required per laminate configuration. This can become burdensome, especially if numerous laminate configurations are under investigation. A type of neural network known as a deep operator network (DeepONet) offers the potential to significantly reduce the amount of computation required. Investigate the use of a DeepONet to predict calibration constants for the integral method applied to balanced and symmetric fibre reinforced plastic (FRP) laminates of eight plies aligned at 0 ^∘ , 90^∘ , +45 ^∘ or -45 ^∘ . The DeepONet model was trained on the FE predicted top surface displacement fields, for every stress application depth and drilling increment, of 15 of the 70 possible E-glass/epoxy FRP laminate configurations. IHD was performed on two specimens, one each of [902/02]s and [02/+45/-45]s configuration, which were not part of the training set. The residual stress distributions obtained using DeepONet predicted and FE generated calibration constants showed good agreement, ≈ 1 – 1.5 MPa, over the measurement depth. The agreement in the calculated residual stress distributions demonstrates that the calibration constants predicted by the DeepONet model have acceptable accuracy for practical application, ≈ 10
Full-field deformation measurement in polycrystals at the mesoscale is challenging due to the inability to simultaneously observe grain boundaries (GBs) and large errors near GBs from deformation discontinuities. To develop a mesoscale segmentation-aided DIC (meso SA-DIC) method for accurate grain-scale deformation and to quantify deformation heterogeneity and discontinuities across GBs in a polycrystalline. A surface treatment method is proposed which enables the measurement of full-field deformation on the surface while simultaneously observing GBs. An adaptive subset generation method is developed, which confines correlation subsets within individual grains to avoid matching error induced by GBs. A method for characterizing the equivalent deformation of GBs is proposed, through which the weakening effect of GB to stiffness is explored and quantitatively characterized. The deformation of a typical triple-junction structure of a polycrystal is analyzed by the proposed method. The heterogeneous deformation within it is quantitatively quantified through stain concentration factors and rotation distribution in the grains. The displacement and strain continuity across GBs are analyzed to characterize their influence on deformation. The proposed meso SA-DIC method demonstrates superior accuracy in measuring the deformation of grains, and the findings provide significant insights into the mesoscale deformation mechanisms governed by GBs.
Standard bending tests usually give valid measurement results only in the linear region, rely on some beam model for evaluation, and transfer the load in some discrete points that localize the stress field. Pure bending tests can provide a solution to these issues and are therefore worth studying. To provide a deeper insight into the mechanics and possible evaluation methods of the pure bending tests. A pure bending test rig with a working principle similar to that of the Hill device [1] was developed. The mechanics of the device were analyzed, and possible issues were identified, including the effect of a parasitic force arising from friction. The bending moment is calculated from the loading force, while the beam curvature is measured by circle fitting to beam points obtained from digital photographs. Different evaluation methods of curvature measurement were tested, and the effect of measurement errors was analyzed with Monte Carlo simulations. The magnitude of the parasitic force was approximated and used for correction of the obtained raw moment-curvature relationship. The device and proposed methodology was validated on steel beam specimens of different lengths and a glass fiber laminate specimen for larger deformations. The study provides a validated measurement and evaluation method for pure bending tests capable of handling large deflections. It is demonstrated that while parasitic forces appear in these devices, the bending characteristic can be obtained accurately with the proposed correction procedure. In the elastic region, the measurements were in good agreement with the standard three-point bending results.
The expanding adoption of composite materials in aerospace, automotive, and rail increases the demand for reliable non-destructive testing (NDT), particularly for detecting internal delamination. Heatable composite laminates that embed carbon-based conductive fibers are prone to thermally induced damage under repeated thermal cycling, which degrades both structural integrity and heat transfer. To develop a practical, image-based NDT approach that quantitatively evaluates internal delamination in heatable composite laminates using a single infrared (IR) thermal image. A single IR thermal image was analyzed by extracting statistical descriptors—skewness (Sk) and kurtosis (Ku)—from the spatial temperature distribution to represent thermal anomalies caused by subsurface delamination. Experiments were performed on fabricated specimens with different delamination areas. Robustness was assessed by varying image resolution, temperature conditions, and the data-extraction region. Both Sk and Ku exhibited strong linear correlations with delamination area, enabling quantitative defect estimation. Sk provided higher accuracy, whereas Ku showed greater sensitivity to defect-related thermal anomalies. The method maintained performance across changes in resolution, thermal conditions, and extraction regions, supporting its robustness and potential field applicability. The proposed single-image, statistics-based thermographic method enables rapid, contactless, and interpretable delamination quantification without time-series measurements, complex equipment, or training datasets. It is promising for real-time NDT of advanced heatable composite structures where simple thermal inspection is required.
Tactile sensing is essential for intelligent manufacturing and embodied robotics. Vision-based tactile sensing provides high spatial resolution and immunity to electromagnetic interference compared with conventional electrical tactile sensors. However, existing vision-based tactile methods are still limited in accurate full-field deformation reconstructing, which degrades contact force estimation. In addition, real contact force supervision is still lacking for vision–tactile deep learning models. The objective is to establish an efficient and robust deep-learning pipeline integrated with DIC for vision-tactile sensing. In addition, a real paired dataset of deformation field and contact force is constructed for supervised training. Experimentally, a novel vision-based tactile sensor with speckles on the elastomer is fabricated. A robotic-arm-based experimental apparatus equipped with a distributed force sensor at the end-effector is developed to apply different contact forces to the elastomer. Images of the contact surface are captured by a stereo camera system, and the displacement field are extracted using stereo digital image correlation (DIC), thereby establishing a paired displacement field and contact force dataset. Meanwhile, a generative adversarial network (GAN) based on global and local architectures is designed. A two-stage workflow with spatial constraint is adopted to predict the contact force contour and reconstruct the detailed force. The trained model is evaluated on the test dataset and further validated in a practical contact force estimation experiment. It achieves an RMSE of 0.04 N over the point-wise distributed contact-force range of 0–0.72 N across all test samples, and an average relative error of 10.22
