Lingyun You

Microscopic deformation stage recognition from molecular dynamics (MD) trajectories is crucial for understanding the evolution of material damage; however, traditional empirical analysis and black-box single deep-learning models lack both high-throughput spatiotemporal modeling and transparent physical interpretability. This work develops a multiscale interpretable deep-learning framework to automatically classify elastic, plastic, and fracture stages from uniaxial tensile MD trajectories and physically decode model decision logic. First, a 3D atomic trajectory is converted to 2D gray-scale image sequences; a CNN-LSTM hybrid architecture is built to jointly extract spatial atomic textures and long-time deformation dynamics, reaching 98.8% test accuracy and far surpassing spatial-only CNN baselines in both supervised classification and unsupervised clustering. More importantly, a hierarchical multiscale interpretability toolkit, including multilayer feature heatmaps, smoothed gradient saliency maps, gradient-weighted class activation maps, and regularized SHAP attribution, is integrated to quantify positive/negative feature contributions and localize model focus regions. The visualized attention zones perfectly match core physical fields (von Mises strain, atomic displacement, and nonaffine deformation), resolving the explainability gap of conventional MD data mining pipelines. This work establishes a generalizable, trustworthy AI paradigm that connects data-driven prediction to intrinsic microscale mechanical mechanisms for computational material research.

1. In recent years, the resource utilisation of solid waste in road engineering has become a research hotspot. Its secondary use not only achieves the transformation of waste into valuable resource...

Alkali-activated slag (AAS) systems offer a sustainable approach for valorizing industrial solid wastes, thereby lessening the environmental impact of both cement production and landfilling. Although granite (GN) and dolomite (DM) wastes are abundant by-products, their use in AAS materials for performance enhancement has not been systematically investigated. This study therefore examined the utilization of GN and DM wastes to improve mechanical strength development in AAS systems. The compressive strength and microstructural evolution were systematically analyzed after replacing 50 wt.% of ground granulated blast furnace slag with GN and DM at varying ratios. Results demonstrate that a GN-to-DM ratio of 2:3 maximizes compressive strength, achieving a 28 d value of 37.21 MPa. The underlying mechanism involves a two-stage process: DM contributes to early strength gain by rapidly releasing calcium (Ca2+) and magnesium (Mg2+) ions, thereby accelerating the formation of dense reaction products. Subsequently, the aluminosilicate network of GN optimizes the gel composition, promoting mid- and long-term stability. At a GN substitution level of 20 wt.%, the synergy between GN’s slowly reactive aluminosilicate properties and DM’s ionic contribution achieves balanced reactivity and enhanced durability. In contrast, systems relying solely on GN substitution form a poorly developed, Ca-deficient calcium-aluminosilicate hydrate (C-A-S-H) gel because of insufficient calcium availability, resulting in reduced mechanical performance. These findings provide mechanistic insights into the eco-friendly recycling of GN–DM wastes for producing high-performance AAS materials.

For complex multicomponent condensed-phase systems, coarse-grained (CG) molecular dynamics often struggles to simultaneously achieve structural fidelity and thermodynamic consistency. Using base asphalt as a representative system, this study proposes and validates a hierarchical hybrid, closed-loop parametrization strategy. Bonded interactions are derived via iterative Boltzmann inversion (IBI) and further stabilized through robust statistical treatment and reduced representations, improving the parameter identifiability and numerical stability. Non-bonded interactions are built on the Martini 3 framework and are calibrated in a low-dimensional manner using only two global scaling factors, followed by geometric fine-tuning for the overall correction. The resulting model reproduces the density-temperature response within the experimentally relevant window and remains consistent with all-atom (AA) references in key structural statistics. Dynamically, it preserves the relative diffusion ranking among components and enables consistent time rescaling to the AA reference. Arrhenius-like diffusion behavior and the temperature dependence of the viscosity are also captured. At the mesoscale, the model reproduces the evolution of asphaltene self-aggregation, representative stacking-like local arrangements of aromatic-core regions, and bee-like morphological features consistent with experimental observations. Overall, this work provides a reproducible and parsimonious paradigm for CG force field construction in complex mixtures, enabling coupled convergence of structure and thermodynamics.