{ "home": { "url": "/", "html": "

Hi! I’m Yanghao Chen, a final-year undergraduate at Tongji University, studying Engineering Mechanics in the School of Aerospace Engineering and Applied Mechanics. My interests center on scientific machine learning and computational mechanics.

\n\n

My research began with fine-tuning large language models using LoRA on Google Colab for question answering in mechanics of materials. I then developed ConvLSTM models to predict lithium concentration and von Mises stress image sequences in battery materials using MATLAB–COMSOL simulation data. I also studied shock propagation with physics-informed neural networks (PINNs), comparing standard and artificial-viscosity formulations against the analytical inviscid solution.

\n\n

Together, these experiences have shaped my interest in combining machine learning with numerical methods for scientific simulation. My current research focuses on neural operators, which learn mappings between function spaces, and hybrid FEM–neural operator solvers based on non-overlapping domain decomposition. I aim to combine AI-based models with GPU-accelerated numerical computing to make engineering simulations more efficient and scalable.

\n\n

At Tongji, I have been fortunate to work under the guidance of Prof. Ying Zhao and Prof. Xianyang (Tom) Chen. From January to October 2026, I was an undergraduate research intern in the Department of Civil and Systems Engineering at Johns Hopkins University, working with Prof. Somdatta Goswami in Centrum IntelliPhysics.

\n\n

News

\n\n
\n
\n
\n
Sep 2026
\n
3D Composite Extension: Extending the hybrid FEM–neural operator solver to a 100-fiber composite. A shared neural operator computes stress fields across all fiber subdomains, with component-specific weights shared across fibers. Periodic boundary conditions eliminate fiber-end effects. View the composite mesh.
\n
\n
\n
Aug 2026
\n
3D Bracket Studies: Extended the workflow to 3D bracket problems, using GRFs to sample interface displacement fields and tractions on internal surfaces. Trained neural operators to predict subdomain stress fields from these inputs.
\n
\n
\n
Jul 2026
\n
Cylinder Studies: Developed a hybrid FEM–neural operator solver for a cylinder problem, using GRFs to sample displacement fields on both inner and outer arcs. Coupled the FEM and neural-operator subdomains through iterative exchange of interface forces and displacements.
\n
\n
\n
Jul 2026
\n
Transitioned to an on-site undergraduate research internship at Johns Hopkins University to continue working with Prof. Somdatta Goswami.
\n
\n \n
\n
Nov 2025
\n
Developed standard and artificial-viscosity PINNs for one-dimensional Burgers shock problems, with analytical-reference comparisons of the predicted solution profiles.
\n
\n
\n
Jun 2025
\n
Began developing a MATLAB–COMSOL data-generation workflow and separate ConvLSTM models for concentration and stress image prediction in battery materials.
\n
\n \n
\n
Sep 2024
\n
Started curating instruction data and developing an LLM fine-tuning workflow for question answering in mechanics of materials.
\n
\n
\n
\n\n

Current Research

\n\n

I am developing GPU-accelerated hybrid FEM–neural operator solvers based on non-overlapping domain decomposition. FEM and neural-operator subdomains are connected through shared interfaces. Neural operators learn mappings from interface displacement fields to subdomain stress fields. Interface reaction forces computed from predicted stresses are transferred to the FEM solver. The FEM solver uses these forces to compute updated interface displacements and returns them to the neural operators for the next coupling iteration.

\n\n

View Research Projects in CV.

\n\n

Skills

\n\n\n\n

Quick Links

\n\n\n\n

Contact

\n\n\n" }, "cv": { "url": "/cv/", "pdf_url": "/files/Yanghao_Chen_CV.pdf?v=b183ba444d75cd6c20438899ccc2c33959c3c2af", "html": "\n

View CV (PDF) | Project repositories

\n\n

Education

\n\n
\n \"Tongji\n Tongji University, Shanghai, China\n
\n\n\n\n
\n \"Johns\n Johns Hopkins University, Baltimore, MD, USA\n
\n\n\n\n

Research Interests

\n\n\n\n

Current Research

\n\n\n\n

Research Projects

\n\n
\n
\n
1. GPU-Accelerated Hybrid FEM–Neural Operator Solver
\n
Jan. 2026 - Present
\n
\n
\n Lead Undergraduate Researcher|Advisor: Prof. Somdatta Goswami\n Johns Hopkins University\n
\n
\n\n\n\n

Ongoing 3D Composite Extension: We are extending the hybrid solver to a 100-fiber composite with non-overlapping fiber and matrix subdomains. FEM solves the matrix subdomain, while a shared neural operator computes stress fields in all fiber subdomains, using component-specific weights shared across all fibers. The two solvers exchange interface forces and displacements iteratively. We apply periodic boundary conditions to eliminate fiber-end effects.

\n\n
\n
\n \n \"Cylinder\n \n
\n Cylinder Subdomain Layout
\n The FEM subdomain (gray) and neural-operator subdomain (green) are connected through a shared interface (blue).\n
\n
\n\n
\n \n \"Full\n \n
\n Non-overlapping Domain Decomposition
\n The mesh is partitioned into non-overlapping FEM and neural-operator subdomains with a shared interface.\n
\n
\n\n
\n \n \"Neural-operator\n \n
\n GRF-Sampled Boundary Displacements
\n Sampled displacement fields on the inner and outer arcs using Gaussian random fields (GRFs). Set horizontal displacement to zero on the left edge and vertical displacement to zero on the bottom edge.\n
\n
\n\n
\n \n \"Subdomain\n \n
\n Subdomain Mesh and Point Cloud
\n Constructed the point cloud from 4,102 finite element nodes, including 3,847 interior nodes (blue) and 255 boundary nodes (red), for neural-operator training.\n
\n
\n\n
\n \n \"Predicted\n \n
\n σxx Stress Prediction and Error
\n Compared the predicted σxx stress field with the FEM reference. The relative L2 error is 0.6262%, and the absolute error map shows the spatial error distribution.\n
\n
\n\n
\n \n \"Cylinder\n \n
\n Coupled FEM–Neural Operator Results
\n Compared the σxx, σyy, and τxy components of the coupled FEM–neural operator solution with the full FEM reference and plotted the absolute error for each component.\n
\n
\n
\n\n
\n
\n \n \"Bottom\n \n
\n 3D Composite Mesh with 100 Fibers
\n Cross-sectional view of the 3D composite mesh containing 100 fibers, with a close-up of the fiber–matrix interfaces.\n
\n
\n
\n\n
\n
\n
2. ConvLSTM Prediction of Concentration and Stress Fields in Battery Materials
\n
Jun. 2025 - Oct. 2025
\n
\n
\n Undergraduate Researcher|Advisor: Prof. Ying Zhao\n Tongji University\n
\n
\n\n\n\n
\n
\n \n \"Animated\n \n
\n Lithium Concentration Evolution
\n Simulated lithium concentration evolution in a polycrystalline NMC particle using MATLAB–COMSOL.\n
\n
\n\n
\n \n \"Animated\n \n
\n von Mises Stress Evolution
\n Simulated von Mises stress evolution in the same particle under the same loading conditions.\n
\n
\n\n
\n \n \"Grain\n \n
\n Grain Orientation Map
\n Encoded grain orientations relative to the global x-axis as a color map for the conditional ConvLSTM models.\n
\n
\n\n
\n \n \"ConvLSTM\n \n
\n Conditional ConvLSTM Prediction
\n Used separate conditional ConvLSTM models to predict future concentration and stress image sequences from past image sequences, grain-orientation maps, and C-rate inputs.\n
\n
\n\n
\n \n \"Internal\n \n
\n ConvLSTM Cell Structure
\n Used convolutional input, forget, and output gates to update the cell and hidden states across the image sequence.\n
\n
\n\n
\n \n \"Scheduled\n \n
\n Scheduled Sampling
\n Selected reference or model-predicted frames as inputs during training to prepare the models for autoregressive prediction.\n
\n
\n\n
\n\n
\n
\n \n \"Prediction\n \n
\n Concentration Sequence Prediction
\n Compared autoregressive concentration predictions with COMSOL reference images and analyzed error accumulation across successive prediction steps using error maps.\n
\n
\n
\n\n
\n
\n
3. PINNs for Shock Capturing in the Burgers Equation
\n
Nov. 2025 - Mar. 2026
\n
\n
\n Undergraduate Researcher|Advisor: Prof. Xianyang (Tom) Chen\n Tongji University\n
\n
\n\n\n\n
\n
\n \n \"Burgers\n \n
\n Burgers Problem and PINN Training
\n Defined the inviscid Burgers shock problem with initial and boundary conditions. Trained the PINN using PDE-residual, initial-condition, and boundary-condition losses.\n
\n
\n\n
\n \n \"Animation\n \n
\n Analytical Shock Evolution
\n Illustrated the analytical solution of the inviscid Burgers equation, with a discontinuity moving at a constant speed of 0.5.\n
\n
\n\n
\n \n \"Animation\n \n
\n Analytical Shock Trajectory
\n Tracked the analytical shock position along xs(t) = 0.5t in the space–time plane.\n
\n
\n\n
\n
\n \n \"Animation\n \n
\n Standard PINN Prediction
\n Compared the standard PINN prediction with the analytical inviscid solution over time, examining the shock location and transition width.\n
\n
\n\n
\n \n \"Animation\n \n
\n Artificial-Viscosity PINN Prediction
\n Compared the PINN prediction with fixed artificial viscosity against the analytical inviscid solution, examining the shock location and transition width.\n
\n
\n
\n
\n\n
\n
\n
4. Domain-Specific LLM Fine-Tuning for Mechanics of Materials
\n
Sep. 2024 - Apr. 2025
\n
\n
\n Undergraduate Researcher|Advisor: Prof. Ying Zhao\n Tongji University\n
\n
\n\n\n\n
\n
\n \n \"Workflow\n \n
\n LLM Fine-Tuning and Release Workflow
\n Built instruction–response datasets for mechanics of materials, fine-tuned Qwen2.5-7B using LoRA on Google Colab, and released the datasets and model checkpoints on Hugging Face.\n
\n
\n
\n\n

Contact

\n\n\n\n" }, "projects": { "url": "/projects/", "items": [{"id":"fem-neural-operator","title":"GPU-Accelerated Hybrid FEM–Neural Operator Solver","short_title":"Hybrid FEM–NO","preview":"/images/research/cylinder_subdomain_supports_1234.png","preview_alt":"Cylinder FEM and neural-operator subdomains connected through a shared interface.","preview_caption":"Cylinder subdomain layout","status":"Ongoing","description":"A hybrid solver based on non-overlapping domain decomposition, with FEM and neural-operator subdomains connected through a shared interface. The two solvers exchange interface displacements and reaction forces iteratively across this interface. Uses GPU-accelerated JAX-FEM simulations to generate training data and trains Transolver in JAX to learn mappings from interface displacement fields to local stress fields.","research_url":"/cv/#fem-neural-operator","repository":null},{"id":"convlstm-battery","title":"ConvLSTM Prediction of Concentration and Stress Fields in Battery Materials","short_title":"ConvLSTM","preview":"/images/research/battery_concentration.gif","preview_alt":"Lithium concentration evolution in the COMSOL simulation dataset.","preview_caption":"Concentration evolution · COMSOL","description":"Trains conditional ConvLSTM models in PyTorch to predict future sequences of concentration and stress fields in polycrystalline battery materials. Separate models for each field use past field images, grain-orientation maps, and C-rate inputs, with MATLAB–COMSOL simulation data and a combined MSE and SSIM loss.","research_url":"/cv/#convlstm-battery","repository":"https://github.com/cyhcyh070126-bot/convlstm-battery-field-prediction#readme"},{"id":"burgers-pinn","title":"PINNs for Shock Capturing in the Burgers Equation","short_title":"PINNs","preview":"/images/research/burgers_standard_pinn_prediction.gif","preview_alt":"Standard PINN prediction compared with the analytical Burgers shock solution.","preview_caption":"Standard PINN prediction","description":"Trains PINNs in PyTorch to predict shock propagation in the one-dimensional Burgers equation, using PDE-residual, initial-condition, and boundary-condition losses. Compares standard and constant-artificial-viscosity PINNs against the analytical inviscid solution to assess the predicted shock profiles.","research_url":"/cv/#burgers-pinn","repository":"https://github.com/cyhcyh070126-bot/burgers-pinn#readme"},{"id":"mechanics-llm","title":"Domain-Specific LLM Fine-Tuning for Mechanics of Materials","short_title":"LLM Fine-Tuning","preview":"/images/research/llm_finetuning_workflow_v2.png","preview_alt":"Mechanics dataset preparation, LoRA fine-tuning, and model release workflow.","preview_caption":"Fine-tuning and release workflow","description":"Adapts Qwen2.5-7B for question answering and concept explanation in mechanics of materials through LoRA fine-tuning on Google Colab. Uses curated instruction–response examples from textbooks and technical literature, with datasets and fine-tuned models released on Hugging Face.","research_url":"/cv/#mechanics-llm","repository":"https://huggingface.co/CYHcyh66","repository_label":"Hugging Face"}] }, "contact": { "email": "2350083@tongji.edu.cn", "linkedin": "yanghao-chen-830677399", "github": "cyhcyh070126-bot", "huggingface": "CYHcyh66" } }