Latin American Journal of Solids and Structures

The purpose of this study is to propose a new understanding of the influence of plastic softening of strength parameters on the mechanical properties of the coal surrounding borehole. Firstly, a thorough analysis of experimental data from previous studies reveals that the plastic softening behavior of strength parameters essentially manifests as a reduction in cohesion and a variation in internal friction angle. Unlike the traditional viewpoint, the change in the internal friction angle after the stress peak is not fixed, manifesting as an increase, a decrease, or remaining unchanged. Subsequently, a theoretical model for deformation and stress distribution of coal surrounding borehole is established, which considers the softening effect of strength parameters and covers different regions. The effect of residual strength parameters on the shape and size of the plastic region is examined in depth, and it is found that the plastic region shape is primarily dominated by the lateral pressure coefficient. Additionally, this study demonstrates that as residual strength parameters decrease, the tangential peak stress transfers to deeper areas, and the plastic deformation region expands progressively. Next, a numerical model considering the plastic softening behavior of strength parameters is developed, and a systematic comparison is conducted between the numerical and the theoretical results. The maximum error between the numerical and theoretical results is found to be within 12%, which validates the reliability of the proposed analytical model. Finally, field verification results confirm that the residual internal friction angle constitutes another critical factor affecting the stress and deformation characteristics of the coal around borehole. Notably, the mechanical response results derived from the model considering only the plastic softening effect of cohesion are smaller than the field-measured data, indicating that the influence of the internal friction angle must also be adequately accounted for in related analyses.

Residual vertical velocity during the late stage of airbag cushioning may produce localized hard contact between an airdrop-equipment subframe and the ground. Conventional whole-vehicle overload measurements cannot resolve response differences among individual load-bearing regions. This study aims to characterize region-specific hard-impact responses of the subframe and identify the regions most vulnerable to landing-induced shock. An equivalent contact framework based on residual-velocity equivalence was established by considering contact reaction force, load-transfer paths, and regional dynamic responses. Free-drop tests from 100, 150, and 200 mm were performed on a subframe divided into four functional regions, with six uniaxial accelerometers sampled at 50 kHz. The 0.4 s acceleration records were processed by wavelet detrending and 10 kHz low-pass filtering, and peak acceleration, root-mean-square (RMS) acceleration, and the pseudo-velocity shock response spectrum (PV-SRS) were used for multi-index evaluation. With increasing drop height, the response progressively concentrated in the wheel-arch beam and central circular load-bearing regions. At 200 mm, measurement point C6 exhibited a peak acceleration of 707 g, an RMS acceleration of 86 g, and a PV-SRS peak of 6.9 m/s, while the corresponding peak frequency remained between 322 and 335 Hz. The dominant spectral peak at C5 shifted from approximately 1 kHz to 276 Hz. The proposed region-specific evaluation framework reveals hard-impact differences that are obscured by whole-vehicle overload measurements. It provides an experimental basis for assessing landing-impact safety and structural reliability of airdrop equipment.

Crashworthiness surrogate models are usually trained only on completed finite element simulations and therefore assume that every candidate can be evaluated numerically. Automated full-vehicle workflows violate this assumption because some mixed-variable combinations prevent model reconstruction, meshing, solution, or response extraction. This study treats such failures as a hidden numerical constraint and develops a simulation-admissibility-aware framework for electric-vehicle side-pole design. A battery-electric SUV was parameterized with 32 continuous geometry variables and 36 discrete material/thickness variables. Among 600 SFE-LS-DYNA designs, 528 were admissible, and nested cross-validation yielded an accepted-set validity of 0.96 for the calibrated classifier. Within the learned admissible domain, MS-GIN modeled interactions among structural modules and achieved mean R²=0.91 and NRMSE=6.5% on an independent test set. Bootstrap SHAP reduced the search space from 68 variables to 18. High-fidelity FE verification of a compromise design reduced mass, B-pillar chest intrusion, and battery-neighborhood intrusion by 3.5%, 8.6%, and 12.4%, respectively. The remaining constraints were satisfied, showing that explicit failure-risk control can support reliable surrogate-based crashworthiness optimization within the investigated design domain.

A physics-informed machine learning framework is developed to predict temporal force–displacement responses and three-dimensional deformation histories of hexagonal crash boxes under multi-angle loading. To reduce the computational cost of nonlinear explicit finite element simulations, a three-dimensional convolutional autoencoder (3D-ConvAE) encodes crash box geometries into a low-dimensional latent representation, while a long short-term memory (LSTM) network predicts their temporal deformation evolution. A physics-informed loss function incorporating kinematic admissibility and energy conservation is introduced to suppress non-physical responses and improve mechanical consistency. Validation against benchmark numerical data demonstrates high predictive accuracy, with mean errors of 7.24% for peak force and 6.31% for energy absorption. The framework also maintains geometric fidelity across complex failure transitions, including the shift from progressive folding to global bending, achieving a mean Chamfer distance of 4 mm. The proposed approach provides a robust and computationally efficient tool for predicting nonlinear crash responses and supporting crashworthiness optimization.

The design of occupant restraint system (ORS) parameters is directly related to the crash safety performance of vehicles. However, traditional iterative simulation and manual parameter tuning methods suffer from long design cycles and high computational costs, while existing deep learning methods mostly focus on indirect modeling or injury prediction, lacking direct parameter design capability for multimodal data. To address this issue, this paper proposes a CNN-based ORS parameter inverse design model (ORS-PDM) that achieves unified input and joint representation of three types of multimodal parameters: time-series, image, and numerical data. Specifically, the crash pulse (time-series modality), the grayscale image of the human-machine layout (image modality), and the HIC metric (numerical modality) are taken as heterogeneous inputs. After unified preprocessing, these data are encoded into tensor channels and concatenated at the channel level, and then fed into the network to directly predict the optimal ORS parameter combination, thereby realizing inverse parameter design.