arXiv subjects

Hanfeng Zhai

Publications and source records attributed to Hanfeng Zhai.

15 recordsLinked to original sources

arXiv2026

Scalable dynamical inference of phase-field fracture from sparse and partial measurements

Evolving crack fields in structural health monitoring and fracture assessment must often be inferred from sparse mechanical measurements rather than dense full-field observations. We develop CNN2D--ConvGRU, a convolutional-recurrent framework for measurement-conditioned reconstruction of time-dependent phase-field brittle fracture. At each load step, the model maps a fixed-length history of phase and displacement fields and sparse current-step displacement measurements to the current full-field state. New measurements are assimilated during sequential deployment, making the framework a state-inference surrogate rather than an autonomous time integrator. It reproduces crack paths, damage evolution, and bulk displacement response, with the largest errors near propagating crack tips and steep displacement gradients and some drift at late stages. Without retraining, weights learned on a $256 \times 256$ raster are evaluated on a $512 \times 512$ raster of the same physical domain and finite-element discretization using a proportionally refined measurement grid. This empirical raster-and-sensing transfer preserves the principal damage topology and global damage evolution, although fine-scale displacement errors increase near crack tips. Comparisons with alternative spatial and temporal architectures show that CNN2D--ConvGRU offers a favorable balance between reconstruction accuracy and computational cost. Relative to repeated finite-element solutions, sequential reconstruction achieves mean speedups of $175\times$ on CPU and $253\times$ on GPU. These results demonstrate efficient full-field fracture-state reconstruction from sparse observations while retaining the spatial structure and history dependence of phase-field fracture.

physics.comp-ph↗

arXiv2026

A convolutional neural network surrogate for hierarchical homogenization: fast elastic moduli prediction of digital rocks

Digital rock physics (DRP) aims to estimate effective rock properties (e.g., elastic moduli) directly from 3D micro-CT images. However, direct numerical simulations (DNS) on high-resolution large 3D scans are often computationally prohibitive and severely limit the application of DRP. To address this bottleneck, we combine a lightweight 3D convolutional neural network (CNN) with hierarchical homogenization (HHM) and apply it to determine effective elastic moduli. In this scheme, a large rock image is divided into subcubes. The CNN replaces costly DNS by directly predicting subcube elastic moduli, while HHM upscales subcube-level predictions to the full rock. Using a shared convolutional backbone, we systematically compare three training targets: (i) full anisotropic $6\times6$ stiffness tensors, (ii) isotropic bulk and shear moduli $(K, G)$, and (iii) Hashin--Shtrikman (HS)-normalized factors. Across multiple rock types, all three models agree well with DNS results while substantially reducing the computational cost. Moreover, training from scratch on each rock type is fast enough that transfer learning is unnecessary. Across all three targets, the accuracy is comparable. In our comparative study, the HS-normalized factor offers the best overall speed--accuracy trade-off while guaranteeing physical consistency, making it a convenient default. The isotropic $(K, G)$ target is a slightly more accurate alternative.

physics.comp-ph↗

arXiv2026

Scalable Physics-Informed Neural Differential Equations and Data-Driven Algorithms for HVAC Systems

We present a scalable, data-driven simulation framework for large-scale heating, ventilation, and air conditioning (HVAC) systems that couples physics-informed neural ordinary differential equations (PINODEs) with differential-algebraic equation (DAE) solvers. At the component level, we learn heat-exchanger dynamics using an implicit PINODE formulation that predicts conserved quantities (refrigerant mass $M_r$ and internal energy $E_\text{hx}$) as outputs, enabling physics-informed training via automatic differentiation of mass/energy balances. Stable long-horizon prediction is achieved through gradient-stabilized latent evolution with gated architectures and layer normalization. At the system level, we integrate learned components with DAE solvers (IDA and DASSL) that explicitly enforce junction constraints (pressure equilibrium and mass-flow consistency), and we use Bayesian optimization to tune solver parameters for accuracy--efficiency trade-offs. To reduce residual system-level bias, we introduce a lightweight corrector network trained on short trajectory segments. Across dual-compressor and scaled network studies, the proposed approach attains multi-fold speedups over high-fidelity simulation while keeping errors low (MAPE below a few percent) and scales to systems with up to 16 compressor-condenser pairs.

cs.LG↗

arXiv2026

Learning Interatomic Force Coefficients from X-ray Thermal Diffuse Scattering Data

We present a fully automated framework for extracting interatomic force constants (IFCs) directly from X-ray thermal diffuse scattering (TDS) data. By formulating scattering intensity as a differentiable function of a symmetry-reduced IFC parameterization, we enable gradient-based optimization via direct, Cholesky-based sampling of correlated atomic displacements at thermal equilibrium. This approach bypasses the computational bottleneck of repeated Hessian matrix diagonalizations, significantly accelerating the inversion process. Benchmark tests demonstrate that the framework accurately recovers ground-truth IFCs and phonon dispersion relations, providing a robust, high-throughput pathway for studying lattice dynamics across diverse crystalline materials. This method bridges the gap between experimental observations and computational modeling, enabling the direct integration of TDS data into the refinement of high-fidelity inter-atomic potentials.

physics.comp-ph↗

arXiv2026

Atomistic and data-driven insights into the local slip resistances in random refractory multi-principal element alloys

Refractory multi-principal element alloys (RMPEAs) have attracted growing interest for their exceptional high-temperature strength, yet their complex compositions hinder a mechanistic understanding of plastic deformation. Here, we perform atomistic simulations to determine local slip resistances (LSRs) of edge and screw dislocations on primary BCC slip planes in 12 equal-molar RMPEAs. Machine learning is employed to uncover relationships between LSR and underlying material properties, enabling systematic assessment of compositional effects on dislocation behavior. Based on these insights, we develop a thermally activated, dislocation-based model to predict macroscopic yield stress. We find that increasing the fraction of hexagonal close-packed elements above 50% significantly reduces unstable stacking fault energy, ideal shear strength, and screw LSR across all slip planes. Higher elastic anisotropy further lowers these quantities, while lattice distortion modifies relative slip resistances between dislocation characters and slip systems. By combining an autoencoder with a random forest model, we identify elastic constants and lattice distortion as the dominant factors controlling LSR. The resulting framework accurately predicts tensile yield stress in BCC RMPEAs and provides guidance for alloy design.

cond-mat.mtrl-sci↗

arXiv2025

Link Statistics of Dislocation Network during Strain Hardening

Dislocations are line defects in crystals that multiply and self-organize into a complex network during strain hardening. The length of dislocation links, connecting neighboring nodes within this network, contains crucial information about the evolving dislocation microstructure. By analyzing data from Discrete Dislocation Dynamics (DDD) simulations in face-centered cubic (fcc) Cu, we characterize the statistical distribution of link lengths of dislocation networks during strain hardening on individual slip systems. Our analysis reveals that link lengths on active slip systems follow a double-exponential distribution, while those on inactive slip systems conform to a single-exponential distribution. The distinctive long tail observed in the double-exponential distribution is attributed to the stress-induced bowing out of long links on active slip systems, a feature that disappears upon removal of the applied stress. We further demonstrate that both observed link length distributions can be explained by extending a one-dimensional Poisson process to include different growth functions. Specifically, the double-exponential distribution emerges when the growth rate for links exceeding a critical length becomes super-linear, which aligns with the physical phenomenon of long links bowing out under stress. This work advances our understanding of dislocation microstructure evolution during strain hardening and elucidates the underlying physical mechanisms governing its formation.

cond-mat.mtrl-sci↗

arXiv2024

Stress Predictions in Polycrystal Plasticity using Graph Neural Networks with Subgraph Training

Numerical modeling of polycrystal plasticity is computationally intensive. We employ Graph Neural Networks (GNN) to predict stresses on complex geometries for polycrystal plasticity from Finite Element Method (FEM) simulations. We present a novel message-passing GNN that encodes nodal strain and edge distances between FEM mesh cells, and aggregates to obtain embeddings and combines the decoded embeddings with the nodal strains to predict stress tensors on graph nodes. The GNN is trained on subgraphs generated from FEM mesh graphs, in which the mesh cells are converted to nodes and edges are created between adjacent cells. We apply the trained GNN to periodic polycrystals with complex geometries and learn the strain-stress maps based on crystal plasticity theory. The GNN is accurately trained on FEM graphs, in which the $R^2$ for both training and testing sets are larger than 0.99. The proposed GNN approach speeds up more than 150 times compared with FEM on stress predictions. We also apply the trained GNN to unseen simulations for validations and the GNN generalizes well with an overall $R^2$ of 0.992. The GNN accurately predicts the von Mises stress on polycrystals. The proposed model does not overfit and generalizes well beyond the training data, as the error distributions demonstrate. This work outlooks surrogating crystal plasticity simulations using graph data.

cond-mat.mtrl-sci↗

arXiv2023

Benchmarking Inverse Optimization Algorithms for Materials Design

Machine learning-based inverse materials discovery has attracted enormous attention recently due to its flexibility in dealing with black box models. Yet, many metaheuristic algorithms are not as widely applied to materials discovery applications as machine learning methods. There are ongoing challenges in applying different optimization algorithms to discover crystals with single- or multi-elemental compositions and how these algorithms differ in mining the ideal materials. We comprehensively compare 11 different optimization algorithms for the design of single- and multi-elemental crystals with targeted properties. By maximizing the bulk modulus and minimizing the Fermi energy through perturbing the parameterized elemental composition representations, we estimated the unique counts of elemental compositions, mean density scan of the objectives space, mean objectives and frequency distributed over the materials' representations and objectives. We found that nature-inspired algorithms contain more uncertainties in the defined elemental composition design tasks, which correspond to their dependency on multiple hyperparameters. Runge Kutta optimization (RUN) exhibits higher mean objectives whereas Bayesian optimization (BO) displayed low mean objectives compared with other methods. Combined with materials count and density scan, we propose that BO strives to approximate a more accurate surrogate of the design space by sampling more elemental compositions and hence have lower mean objectives, yet RUN will repeatedly sample the targeted elemental compositions with higher objective values. Our work shed light on the automated digital design of crystals with single- and multi-elemental compositions and is expected to elicit future studies on materials optimization such as composite and alloy design based on specific desired properties.

cond-mat.mtrl-sci↗

arXiv2023

Controlling Biofilm Transport with Porous Metamaterials Designed with Bayesian Learning

Biofilm growth and transport in confined systems frequently occur in natural and engineered systems. Designing customizable engineered porous materials for controllable biofilm transportation properties could significantly improve the rapid utilization of biofilms as engineered living materials for applications in pollution alleviation, material self-healing, energy production, and many more. We combine Bayesian optimization (BO) and individual-based modeling to conduct design optimizations for maximizing different porous materials' (PM) biofilm transportation capability. We first characterize the acquisition function in BO for designing 2-dimensional porous membranes. We use the expected improvement acquisition function for designing lattice metamaterials (LM) and 3-dimensional porous media (3DPM). We find that BO is 92.89% more efficient than the uniform grid search method for LM and 223.04% more efficient for 3DPM. For all three types of structures, the selected characterization simulation tests are in good agreement with the design spaces approximated with Gaussian process regression. All the extracted optimal designs exhibit better biofilm growth and transportability than unconfined space without substrates. Our comparison study shows that PM stimulates biofilm growth by taking up volumetric space and pushing biofilms' upward growth, as evidenced by a 20% increase in bacteria cell numbers in unconfined space compared to porous materials, and 128% more bacteria cells in the target growth region for PM-induced biofilm growth compared with unconfined growth. Our work provides deeper insights into the design of substrates to tune biofilm growth, analyzing the optimization process and characterizing the design space, and understanding biophysical mechanisms governing the growth of biofilms.

physics.bio-ph↗

arXiv2022

Multiscale mechanics of thermal gradient coupled graphene fracture: A molecular dynamics study

The thermo-mechanical coupling mechanism of graphene fracture under thermal gradients possesses rich applications whereas is hard to study due to its coupled non-equilibrium nature. We employ non-equilibrium molecular dynamics to study the fracture of graphene by applying a fixed strain rate under different thermal gradients by employing different potential fields. It is found that for AIREBO and AIREBO-M, the fracture stresses do not strictly follow the positive correlations with the initial crack length. Strain-hardening effects are observed for "REBO-based" potential models of small initial defects, which is interpreted as blunting effect observed for porous graphene. The temperature gradients are observed to not show clear relations with the fracture stresses and crack propagation dynamics. Quantized fracture mechanics verifies our molecular dynamics calculations. We provide a unique perspective that the transverse bond forces share the loading to account for the nonlinear increase of fracture stress with shorter crack length. Anomalous kinetic energy transportation along crack tips is observed for "REBO-based" potential models, which we attribute to the high interatomic attractions in the potential models. The fractures are honored to be more "brittle-liked" carried out using machine learning interatomic potential (MLIP), yet incapable of simulating post-fracture dynamical behaviors. The mechanical responses using MLIP are observed to be not related to temperature gradients. The temperature configuration of equilibration simulation employing the dropout uncertainty neural network potential with a dropout rate of 0.1 is reported to be the most accurate compared with the rest. This work is expected to inspire further investigation of non-equilibrium dynamics in graphene with practical applications in various engineering fields.

cond-mat.mtrl-sci↗

arXiv2022

Computational design of antimicrobial active surfaces via automated Bayesian optimization

Biofilms pose significant problems for engineers in diverse fields, such as marine science, bioenergy, and biomedicine, where effective biofilm control is a long-term goal. The adhesion and surface mechanics of biofilms play crucial roles in generating and removing biofilm. Designing customized nano-surfaces with different surface topologies can alter the adhesive properties to remove biofilms more easily and greatly improve long-term biofilm control. To rapidly design such topologies, we employ individual-based modeling and Bayesian optimization to automate the design process and generate different active surfaces for effective biofilm removal. Our framework successfully generated ideal nano-surfaces for biofilm removal through applied shear and vibration. Densely distributed short pillar topography is the optimal geometry to prevent biofilm formation. Under fluidic shearing, the optimal topography is to sparsely distribute tall, slim, pillar-like structures. When subjected to either vertical or lateral vibrations, thick trapezoidal cones are found to be optimal. Optimizing the vibrational loading indicates a small vibration magnitude with relatively low frequencies is more efficient in removing biofilm. Our results provide insights into various engineering fields that require surface-mediated biofilm control. Our framework can also be applied to more general materials design and optimization.

physics.bio-ph↗

arXiv2022

Comparison of Deep Learning and Deterministic Algorithms for Control Modeling

Controlling nonlinear dynamics arise in various engineering fields. We present efforts to model the forced van der Pol system control using physics-informed neural networks (PINN) compared to benchmark methods, including idealized nonlinear feedforward (FF) control, linearized feedback control (FB), and feedforward-plus-feedback combined (C). The aim is to implement circular trajectories in the state space of the van der Pol system. A designed benchmark problem is used for testing the behavioral differences of the disparate controllers and then investigating controlled schemes and systems of various extents of nonlinearities. All methods exhibit a short initialization accompanying arbitrary initialization points. FF control successfully converges to the desired trajectory, and PINN executes good controls with higher stochasticity observed for higher-order terms based on the phase portraits. In contrast, combined control failed. Varying trajectory amplitudes revealed that FF, FB, and combined control all fail for unity nonlinear damping gain. Traditional control methods display a robust fluctuation for higher-order terms. For some various nonlinearities, PINN failed to implement the desired trajectory instead of becoming "trapped" in the phase of a small radius, yet FB successfully implemented controls. PINN generally exhibits lower relative errors for varying targeted trajectories. However, PINN also shows evidently higher computational burden compared with traditional control theory methods, with at least more than 30 times longer control time compared with benchmark idealized nonlinear feed-forward control. This manuscript proposes a comprehensive comparative study for future controller employment considering deterministic and machine learning approaches.

math.DS↗

arXiv2022

Multiscale Computational Modeling of Biofilm

We review the computation models for biofilm and bacteria cells, providing perspectives on biofilm's various properties and potential serving as engineering living materials (ELMs), considering the omnipresence of such biological matter. The minireview starts from the molecular regime, bottom-up to the mesoscale, to continuum, with an emphasis on the mesoscale algorithms such as dissipative particles dynamics (DPD) and individual-based modeling (IbM). Some representative works are highlighted considering different modeling methods on each scale. The advantages and limitations of each algorithm for different scales are elaborated given the existed research works. Specifically, the potential for IbM, also known as the discrete element method (DEM) is targeted for its accurate description of both biological and mechanical properties.

physics.bio-ph↗

arXiv2021

Controlling Chaos in Van Der Pol Dynamics Using Signal-Encoded Deep Learning

Controlling nonlinear dynamics is a long-standing problem in engineering. Harnessing known physical information to accelerate or constrain stochastic learning pursues a new paradigm of scientific machine learning. By linearizing nonlinear systems, traditional control methods cannot learn nonlinear features from chaotic data for use in control. Here, we introduce Physics-Informed Deep Operator Control (PIDOC), and by encoding the control signal and initial position into the losses of a physics-informed neural network (PINN), the nonlinear system is forced to exhibit the desired trajectory given the control signal. PIDOC receives signals as physics commands and learns from the chaotic data output from the nonlinear van der Pol system, where the output of the PINN is the control. Applied to a benchmark problem, PIDOC successfully implements control with higher stochasticity for higher-order terms. PIDOC has also been proven to be capable of converging to different desired trajectories based on case studies. Initial positions slightly affect the control accuracy at the beginning stage yet do not change the overall control quality. For highly nonlinear systems, PIDOC is not able to execute control with high accuracy compared with the benchmark problem. The depth and width of the neural network structure do not greatly change the convergence of PIDOC based on case studies of van der Pol systems with low and high nonlinearities. Surprisingly, enlarging the control signal does not help to improve the control quality. The proposed framework can potentially be applied to many nonlinear systems for nonlinear controls.

eess.SY↗

arXiv2021

BubbleNet: Inferring micro-bubble dynamics with semi-physics-informed deep learning

Micro-bubbles and bubbly flows are widely observed and applied in chemical engineering, medicine, involves deformation, rupture, and collision of bubbles, phase mixture, etc. We study bubble dynamics by setting up two numerical simulation cases: bubbly flow with a single bubble and multiple bubbles, both confined in the microchannel, with parameters corresponding to their medical backgrounds. Both the cases have their medical background applications. Multiphase flow simulation requires high computation accuracy due to possible component losses that may be caused by sparse meshing during the computation. Hence, data-driven methods can be adopted as an useful tool. Based on physics-informed neural networks (PINNs), we propose a novel deep learning framework BubbleNet, which entails three main parts: deep neural networks (DNN) with sub nets for predicting different physics fields; the semi-physics-informed part, with only the fluid continuum condition and the pressure Poisson equation $\mathcal{P}$ encoded within; the time discretized normalizer (TDN), an algorithm to normalize field data per time step before training. We apply the traditional DNN and our BubbleNet to train the coarsened simulation data and predict the physics fields of both the two bubbly flow cases. The BubbleNets are trained for both with and without $\mathcal{P}$, from which we conclude that the 'physics-informed' part can serve as inner supervision. Results indicate our framework can predict the physics fields more accurately, estimating the prediction absolute errors. Our deep learning predictions outperform traditional numerical methods computed with similar data density meshing. The proposed network can potentially be applied to many other engineering fields.

physics.flu-dyn↗