Lin Cheng
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Lin Cheng has a total of 10 co-authors. Here are the top co-authors by number of shared publications:
- Ye Lu
Department of Mechanical Engineering University of Maryland Baltimore County Baltimore USA
2 shared publications - Christian Mathew
Department of Mechanical Engineering, Virginia Polytechnic Institute and State University, Blacksburg, VA 24061, United States of America
1 shared publication - Haozhe Yu
Department of Mechanical Engineering University of Maryland College Park Maryland USA
1 shared publication Jie Song
College of Engineering, SUNY Polytechnic Institute, Utica, NY 13502, United States of America
1 shared publication- Mohammad Madadi
Department of Mechanical Engineering Binghamton University Binghamton New York USA
1 shared publication - Pu Zhang
Department of Mechanical Engineering State University of New York at Binghamton 13902 Binghamton NY USA
1 shared publication
Latest Publications
- Open Access
- 29 August 2026
- Published by Elsevier in SSRN Electronic Journal
Execution-Verified Scientific Reasoning for Autonomous Finite-Element Solver Synthesis with Domain-Adapted Large Language Models
Partial differential equation (PDE)-based simulation is central to engineering analysis and design, yet constructing reliable finite element method (FEM) solvers from high-level physical descriptions remains labor intensive. This workflow requires physical interpretation, governing-equation formulation, weak-form derivation, numerical discretization, implementation, and runtime validation; an error at any stage can produce code that appears plausible but represents an incorrect or non-executable model. This paper presents an execution-verified framework for autonomous PDE-to-FEM solver synthesis using domain-adapted large language models (LLMs). The framework formulates solver synthesis as a structured reasoning process linking physical-semantic parsing, operator-level PDE assembly, boundary and initial-condition specification, variational formulation, finite-element discretization, executable FEniCSx/Gmsh code generation, and runtime verification. An execution-verified synthetic data factory generates reasoning-to-code trajectories and a physics-aware correction corpus, enabling a compact 2-billion-parameter open-source model to learn relationships among physical assumptions, numerical formulation, solver implementation, and execution outcome. Evaluation on 75 unseen PDE problems across canonical, intermediate, and complex tiers shows that the adapted model achieves 100.0%, 93.3%, and 60.0% execution success, respectively, and obtains the strongest weighted ReasoningScore, SpecScore, execution rate, and RuntimeQualityScore among all baselines under a single-pass, no-repair protocol. These results show that reliable AI-assisted physics-based simulation depends not only on model scale, but also on structured domain supervision and execution-based validation. This work contributes a structured, execution-verified reasoning-supervision method for LLMs, applied here to automate FEM solver generation for PDE-based engineering simulations.
- 29 August 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessExecution-Verified Scientific Reasoning for Autonomous Finite-Element Solver Synthesis with Domain-Adapted Large Language Models
Partial differential equation (PDE)-based simulation is central to engineering analysis and design, yet constructing reliable finite element method (FEM) solvers from high-level physical descriptions remains labor intensive. This workflow requires physical interpretation, governing-equation formulation, weak-form derivation, numerical discretization, implementation, and runtime validation; an error at any stage can produce code that appears plausible but represents an incorrect or non-executable model. This paper presents an execution-verified framework for autonomous PDE-to-FEM solver synthesis using domain-adapted large language models (LLMs). The framework formulates solver synthesis as a structured reasoning process linking physical-semantic parsing, operator-level PDE assembly, boundary and initial-condition specification, variational formulation, finite-element discretization, executable FEniCSx/Gmsh code generation, and runtime verification. An execution-verified synthetic data factory generates reasoning-to-code trajectories and a physics-aware correction corpus, enabling a compact 2-billion-parameter open-source model to learn relationships among physical assumptions, numerical formulation, solver implementation, and execution outcome. Evaluation on 75 unseen PDE problems across canonical, intermediate, and complex tiers shows that the adapted model achieves 100.0%, 93.3%, and 60.0% execution success, respectively, and obtains the strongest weighted ReasoningScore, SpecScore, execution rate, and RuntimeQualityScore among all baselines under a single-pass, no-repair protocol. These results show that reliable AI-assisted physics-based simulation depends not only on model scale, but also on structured domain supervision and execution-based validation. This work contributes a structured, execution-verified reasoning-supervision method for LLMs, applied here to automate FEM solver generation for PDE-based engineering simulations.
- Open Access
- 13 August 2026
- Published by Elsevier in SSRN Electronic Journal
Execution-Verified Scientific Reasoning for Autonomous Finite-Element Solver Synthesis with Domain-Adapted Large Language Models
Partial differential equation (PDE)-based simulation is central to engineering analysis and design, yet constructing reliable finite element method (FEM) solvers from high-level physical descriptions remains labor intensive. This workflow requires physical interpretation, governing-equation formulation, weak-form derivation, numerical discretization, implementation, and runtime validation; an error at any stage can produce code that appears plausible but represents an incorrect or non-executable model. This paper presents an execution-verified framework for autonomous PDE-to-FEM solver synthesis using domain-adapted large language models (LLMs). The framework formulates solver synthesis as a structured reasoning process linking physical-semantic parsing, operator-level PDE assembly, boundary and initialcondition specification, variational formulation, finite-element discretization, executable FEniCSx/Gmsh code generation, and runtime verification. An execution-verified synthetic data factory generates reasoningto-code trajectories and a physics-aware correction corpus, enabling a compact 2-billion-parameter opensource model to learn relationships among physical assumptions, numerical formulation, solver implementation, and execution outcome. Evaluation on 75 unseen PDE problems across canonical, intermediate, and complex tiers shows that the adapted model achieves 100.0%, 93.3%, and 60.0% execution success, respectively, and obtains the strongest weighted ReasoningScore, SpecScore, execution rate, and RuntimeQualityScore among all baselines under a single-pass, no-repair protocol. These results show that reliable AI-assisted physics-based simulation depends not only on model scale, but also on structured domain supervision and execution-based validation. This work contributes a structured, execution-verified reasoning-supervision method for LLMs, applied here to automate FEM solver generation for PDE-based engineering simulations.
- 13 August 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessExecution-Verified Scientific Reasoning for Autonomous Finite-Element Solver Synthesis with Domain-Adapted Large Language Models
Partial differential equation (PDE)-based simulation is central to engineering analysis and design, yet constructing reliable finite element method (FEM) solvers from high-level physical descriptions remains labor intensive. This workflow requires physical interpretation, governing-equation formulation, weak-form derivation, numerical discretization, implementation, and runtime validation; an error at any stage can produce code that appears plausible but represents an incorrect or non-executable model. This paper presents an execution-verified framework for autonomous PDE-to-FEM solver synthesis using domain-adapted large language models (LLMs). The framework formulates solver synthesis as a structured reasoning process linking physical-semantic parsing, operator-level PDE assembly, boundary and initialcondition specification, variational formulation, finite-element discretization, executable FEniCSx/Gmsh code generation, and runtime verification. An execution-verified synthetic data factory generates reasoningto-code trajectories and a physics-aware correction corpus, enabling a compact 2-billion-parameter opensource model to learn relationships among physical assumptions, numerical formulation, solver implementation, and execution outcome. Evaluation on 75 unseen PDE problems across canonical, intermediate, and complex tiers shows that the adapted model achieves 100.0%, 93.3%, and 60.0% execution success, respectively, and obtains the strongest weighted ReasoningScore, SpecScore, execution rate, and RuntimeQualityScore among all baselines under a single-pass, no-repair protocol. These results show that reliable AI-assisted physics-based simulation depends not only on model scale, but also on structured domain supervision and execution-based validation. This work contributes a structured, execution-verified reasoning-supervision method for LLMs, applied here to automate FEM solver generation for PDE-based engineering simulations.
- Open Access
- 22 July 2026
- Published by Wiley in International Journal for Numerical Methods in Engineering
- Vol. 127 (14) , e70387
A Hierarchical Deep Symbolic AI Method for Closed‐Form, Interpretable Constitutive Modeling
Recent advances in advanced manufacturing have enabled the development of multifunctional materials such as soft composites, whose complex nonlinear behaviors demand accurate and interpretable constitutive modeling. Traditional phenomenological formulations rely heavily on assumed functional forms and expert intuition, requiring extensive trial and error and offering limited generality. In contrast, data‐driven approaches based on neural networks provide strong predictive capability but often require large datasets and lack interpretability, stability, and physical consistency. To address these challenges, we propose a hierarchical deep symbolic artificial intelligence (HDSAI) method for discovering constitutive laws directly from data while providing closed‐form analytical expressions. The method constructs constitutive relations hierarchically in a layer‐by‐layer manner using symbolic mathematical operators as building blocks. Each layer combines and transforms the outputs from the previous one, progressively enriching functional complexity while maintaining analytical transparency. A layer‐wise evolutionary and selection algorithm is introduced to optimize the symbolic operators within and across layers, efficiently balancing model accuracy and parsimony. The HDSAI method offers three key advantages: (1) deeper symbolic composition while avoiding uncontrolled symbolic‐tree explosion; (2) flexible control of model complexity via progressive symbolic evolution; and (3) enhanced stability and generalization compared with existing symbolic regression and neural‐network‐based models. Benchmark studies against state‐of‐the‐art symbolic regression including GPTIPS, PySR, and LoHM methods demonstrate that the proposed HDSAI method not only achieves higher numerical accuracy and stability but also successfully rediscovers the exact ground‐truth constitutive expressions that existing methods fail to capture. This capability underscores the effectiveness of the hierarchical architecture in navigating complex search spaces and preserving physically interpretable functional forms. Systematic investigations of key parameters, including population size, layer depth, data availability, and noisy level of the data, further reveal that the hierarchical design enhances convergence robustness and expressiveness, enabling HDSAI to efficiently identify compact, physically meaningful constitutive laws with superior interpretability and generalization.
- 22 July 2026
- Published by Wiley in International Journal for Numerical Methods in Engineering
- Vol. 127 (14) , e70387
Open AccessA Hierarchical Deep Symbolic AI Method for Closed‐Form, Interpretable Constitutive Modeling
Recent advances in advanced manufacturing have enabled the development of multifunctional materials such as soft composites, whose complex nonlinear behaviors demand accurate and interpretable constitutive modeling. Traditional phenomenological formulations rely heavily on assumed functional forms and expert intuition, requiring extensive trial and error and offering limited generality. In contrast, data‐driven approaches based on neural networks provide strong predictive capability but often require large datasets and lack interpretability, stability, and physical consistency. To address these challenges, we propose a hierarchical deep symbolic artificial intelligence (HDSAI) method for discovering constitutive laws directly from data while providing closed‐form analytical expressions. The method constructs constitutive relations hierarchically in a layer‐by‐layer manner using symbolic mathematical operators as building blocks. Each layer combines and transforms the outputs from the previous one, progressively enriching functional complexity while maintaining analytical transparency. A layer‐wise evolutionary and selection algorithm is introduced to optimize the symbolic operators within and across layers, efficiently balancing model accuracy and parsimony. The HDSAI method offers three key advantages: (1) deeper symbolic composition while avoiding uncontrolled symbolic‐tree explosion; (2) flexible control of model complexity via progressive symbolic evolution; and (3) enhanced stability and generalization compared with existing symbolic regression and neural‐network‐based models. Benchmark studies against state‐of‐the‐art symbolic regression including GPTIPS, PySR, and LoHM methods demonstrate that the proposed HDSAI method not only achieves higher numerical accuracy and stability but also successfully rediscovers the exact ground‐truth constitutive expressions that existing methods fail to capture. This capability underscores the effectiveness of the hierarchical architecture in navigating complex search spaces and preserving physically interpretable functional forms. Systematic investigations of key parameters, including population size, layer depth, data availability, and noisy level of the data, further reveal that the hierarchical design enhances convergence robustness and expressiveness, enabling HDSAI to efficiently identify compact, physically meaningful constitutive laws with superior interpretability and generalization.
- Open Access
- 31 March 2026
- Published by IOP Publishing in Modelling and Simulation in Materials Science and Engineering
- Vol. 34 (3)
Crystallographic Orientation Effects on Pitting Corrosion and Pit-to-Crack Transition in Austenitic Stainless Steel
This study investigates the influence of crystallographic orientation and grain boundary sensitivity on pitting corrosion and pit-to-crack transition in austenitic stainless steels, including conventionally processed and laser powder bed fusion (LPBF) conditions. A multiphysics, microstructure-sensitive phase-field framework is developed to model orientation-dependent corrosion kinetics by incorporating crystallographic Euler angles into electrochemical corrosion potentials and overpotentials, together with a grain boundary sensitivity parameter to capture boundary-mediated corrosion effects. By integrating Electron Backscatter Diffraction (EBSD)-derived microstructures, the model reproduces anisotropic corrosion morphologies, grain-selective dissolution, and texture-dependent corrosion fronts consistent with experimental observations. Coupled electrochemical–mechanical simulations reveal that the interaction between grain boundary sensitivity and local orientation contrast governs stress localization, crack initiation location, and crack-path tortuosity. In particular, the simulations show that strongly textured LPBF microstructures develop smoother corrosion fronts than conventionally wrought materials while exhibiting distinct crack-initiation behavior, demonstrating the important role of additive-manufacturing texture in governing localized corrosion and crack-path evolution. The proposed framework provides a mechanistic basis for linking crystallographic texture, grain boundary–controlled microstructural heterogeneity, and localized corrosion-driven cracking in stainless steels.
- 31 March 2026
- Published by IOP Publishing in Modelling and Simulation in Materials Science and Engineering
- Vol. 34 (3)
Open AccessCrystallographic Orientation Effects on Pitting Corrosion and Pit-to-Crack Transition in Austenitic Stainless Steel
This study investigates the influence of crystallographic orientation and grain boundary sensitivity on pitting corrosion and pit-to-crack transition in austenitic stainless steels, including conventionally processed and laser powder bed fusion (LPBF) conditions. A multiphysics, microstructure-sensitive phase-field framework is developed to model orientation-dependent corrosion kinetics by incorporating crystallographic Euler angles into electrochemical corrosion potentials and overpotentials, together with a grain boundary sensitivity parameter to capture boundary-mediated corrosion effects. By integrating Electron Backscatter Diffraction (EBSD)-derived microstructures, the model reproduces anisotropic corrosion morphologies, grain-selective dissolution, and texture-dependent corrosion fronts consistent with experimental observations. Coupled electrochemical–mechanical simulations reveal that the interaction between grain boundary sensitivity and local orientation contrast governs stress localization, crack initiation location, and crack-path tortuosity. In particular, the simulations show that strongly textured LPBF microstructures develop smoother corrosion fronts than conventionally wrought materials while exhibiting distinct crack-initiation behavior, demonstrating the important role of additive-manufacturing texture in governing localized corrosion and crack-path evolution. The proposed framework provides a mechanistic basis for linking crystallographic texture, grain boundary–controlled microstructural heterogeneity, and localized corrosion-driven cracking in stainless steels.