Wenhai Wang

Modelica is an industry-standard language for modeling and simulating complex cyber-physical systems, playing a critical role in digital twin development. Driven by the escalating demand for complex system modeling, there is a growing interest in leveraging the advanced code generation capabilities of Large Language Models (LLMs). To explore this potential, we establish a comprehensive benchmark to systematically evaluate LLMs in generating Modelica component models. Our evaluation reveals that current LLMs struggle to construct simulation-ready components due to strict acausal semantics, with failures primarily stemming from: (1) syntactic violations, (2) semantic hallucinations, and (3) physical inconsistencies. To overcome these challenges, we propose ModiGen, a framework designed to promote syntactic correctness along with structural and physical consistency in Modelica generation. ModiGen features supervised fine-tuning to facilitate the internalization of domain knowledge. Crucially, it employs Modelica-specific structural retrieval-augmented generation to inject precise topological constraints and physical laws. Furthermore, it incorporates a fully automated, feedback-driven refinement mechanism that transforms simulation diagnostics into fine-grained instructions, guiding LLMs to resolve specific syntactic and semantic errors. Experimental results demonstrate that ModiGen achieves up to 18.65% improvement in the functional validation pass rate over baselines, significantly enhancing the reliability of component generation.

Gas–liquid flow in millimeter-scale tubes exhibits transient dynamics at both bulk and individual-bubble scales, which are difficult to measure simultaneously under low-frame-rate imaging. This study develops a computer-vision framework for synchronized measurement of bulk flow parameters and Lagrangian bubble motion in bubbly and slug flows using single-view 30 fps backlit image sequences. Attention-UNet segmentation is combined with a kinematically constrained matching strategy based on foreground-normalized mask overlap and an inertial continuity constraint. The framework extracts a relative image-derived gas-occupancy proxy, global bubble velocity Uglobal, individual-bubble trajectories, projected-area variation, image-derived apparent near-wall clearance, and in-plane nose curvature from the same video stream. Cross-validation yielded a segmentation Dice coefficient of 0.9736. Synthetic-displacement tests gave a mean relative offset error of 0.95%, and the representative type-B uncertainty in individual-bubble velocity was 0.47%. The global bubble-velocity signal contained a dominant low-frequency component whose frequency increased nearly linearly with U¯global (R2=0.984); the fitted slope corresponded to an apparent advective length of 1.52 m, close to the apparatus length and supporting a facility-scale interpretation. At the bubble scale, the higher Weber-number end-member group exhibited lower mean in-plane nose curvature. The framework therefore enables synchronized characterization of millimeter-scale gas–liquid flow at global and bubble-resolved scales using a single low-frame-rate image sequence.

This study proposes a hybrid flow-field modeling framework that integrates physics-based simulation, parameter-space interpolation, and deep learning for rapid and high-accuracy prediction of wind farm wakes under uniform yaw conditions. First, a steady-state CFD yaw-wake simulation is developed to produce physically consistent and high-quality dataset. Second, this paper rigorously proves the Lipschitz continuity of steady yaw wake solutions and their image representations with respect to the input parameters. Based on this theoretical basis, we propose the Parameter-Space Intermediate Flow Estimation method (PS-IFE) for steady wake image augmentation, which recasts video frame interpolation into the parameter space to generate intermediate-condition steady wakes. Finally, a lightweight deep neural network (DNN) is trained on the augmented dataset for data-driven uniform-yaw wake modeling. In a 55 wind farm with 7D spacing using the NREL 5 MW turbine, PS-IFE achieves augmentation errors as low as RMSE: 0.063 m/s and MAPE: 0.584%, outperforming conventional augmentation algorithms with tailored superiority. Sensitivity analysis further recommends m/s for reliable interpolation path design. The resulting data-driven yaw wake model attains RMSE: 0.127 m/s and MAPE: 1.041%, about 75% lower than the Bastankhah model, while maintaining millisecond-level inference time. The proposed framework maintains physical consistency while achieving both high accuracy and high efficiency, with strong scalability to other complex flow-field modeling scenarios.

Large Language Models have demonstrated remarkable progress in general-purpose capabilities and can achieve strong performance in specific domains through fine-tuning on domain-specific data. However, acquiring high-quality data for target domains remains a significant challenge. Existing data synthesis approaches follow a deductive paradigm, heavily relying on explicit domain descriptions expressed in natural language and careful prompt engineering, limiting their applicability in real-world scenarios where domains are difficult to describe or formally articulate. In this work, we tackle the underexplored problem of domain-specific data synthesis through an inductive paradigm, where the target domain is defined only through a set of reference examples, particularly when domain characteristics are difficult to articulate in natural language. We propose a novel framework, DOMINO, that learns a minimal sufficient domain representation from reference samples and leverages it to guide the generation of domain-aligned synthetic data. DOMINO integrates prompt tuning with a contrastive disentanglement objective to separate domain-level patterns from sample-specific noise, mitigating overfitting while preserving core domain characteristics. Theoretically, we prove that DOMINO expands the support of the synthetic data distribution, ensuring greater diversity. Empirically, on challenging coding benchmarks where domain definitions are implicit, fine-tuning on data synthesized by DOMINO improves Pass@1 accuracy by up to 4.63% over strong, instruction-tuned backbones, demonstrating its effectiveness and robustness. This work establishes a new paradigm for domain-specific data synthesis, enabling practical and scalable domain adaptation without manual prompt design or natural language domain specifications. The full version is available at https://arxiv.org/abs/2605.30039.

Modern software libraries continually evolve to maintain activeness, improve performance, and enhance security. This evolution frequently introduces major version updates that include incompatible modifications (breaking changes), which often cause downstream projects to fail and significantly increase the effort required for maintenance. To reduce this burden on developers, recent studies propose automated approaches that leverage API characterization techniques and compilation error reports to guide LLM-based updates. However, these methods rely on static typing and compilation feedback, which are not available in dynamic languages such as JavaScript. As a result, when applied to JavaScript, they degrade to a knowledge-free setting and cannot effectively support automated dependency updates. In this paper, we propose a knowledge-based approach that mines explicit update knowledge, referred to as breaking change records, from library repositories to support automated dependency updates in the presence of breaking changes (i.e., breaking dependency updates). We first conduct an empirical investigation of how breaking change records are maintained in top-ranked JavaScript libraries. Our study reveals that 83.8% of libraries maintain such records within their repositories, often spread across multiple locations. However, the quality of these records varies considerably, with issues such as implicit references to changed objects and vague adaptation instructions. To address this issue, we systematically characterize breaking change records by identifying recurring patterns in both their locations and the types of information they convey. Building on these insights, we develop BDUpdater, an agent based framework that aggregates repository sources and refines raw records into structured breaking change lists, thereby mitigating the absence of API differencing tools. Leveraging this mined knowledge, BDUpdater pinpoints changed objects and statically identifies the affected client code, while supplying fine-grained change information to LLMs to guide accurate client code migration in the absence of compilation error messages. Our evaluation on 13 popular JavaScript libraries and 84 clients shows that BDUpdater recovers 90.5% of the breaking dependency updates with high semantic equivalence to developer-authored adaptations, while incurring an average cost of only $0.05 per client.

As objective dimensionality increases, the environmental selection of NSGA-III becomes increasingly vulnerable to ambiguous directional association, weakened local discrimination, and imbalanced survivor allocation, which jointly reduce convergence quality and deteriorate solution-set structure in many-objective optimization. To address this issue, this paper proposes Directional Adaptive Ranking and Matching (DARM), a direction-consistent environmental selection strategy for NSGA-III. DARM combines adaptive local ranking with directional filtering that restricts cosine-based association to compact candidate direction sets, thereby improving within-direction discrimination while preserving balanced survivor allocation across directions. Experimental results on benchmark suites show that NSGA-III DARM remains competitive and stable under fixed and increasing objective dimensionalities. At fixed dimensionality, it improves convergence quality, solution-set quality, and search stability across PF characteristics; as dimensionality increases, it maintains a robust degradation pattern across structured and heterogeneous families. Qualitative, sensitivity, and ablation results further support the framework’s regularity, robustness, and component complementarity. These results indicate that the advantage of DARM lies in strengthening coordination and structural robustness in environmental selection, rather than improving an isolated selection component.