physics-constrained model tuning

This study challenges the prevailing assumption that higher surrogate model accuracy invariably yields better configuration tuning performance. Method: We conduct a large-scale empirical investigation—13,612 experiments over 13 months across 10 model families, 17 tuners, and 29 systems—introducing a fitness-landscape-based attribution framework. We employ multidimensional error metrics (e.g., RMSE, MAE), cross-system benchmarking, and statistical significance testing to rigorously quantify tuning outcomes. Contribution/Results: We are the first to empirically demonstrate a non-monotonic relationship between surrogate accuracy and tuning quality: up to 58% of cases show no improvement, and 24% exhibit statistically significant degradation. We identify widespread suboptimal surrogate selection and reveal that accuracy thresholds dynamically shift with error regimes. Our findings yield a reproducible tuner–surrogate matching guide and marginal accuracy-benefit analysis, shifting the surrogate selection paradigm from “higher accuracy is better” to “fitness-aware adaptivity.”

In automated performance tuning, optimizer hyperparameters have long been overlooked, and their impact on overall tuning efficacy remains systematically uninvestigated. Method: This paper introduces the novel paradigm of “tuner hyperparameter optimization” to address efficiency bottlenecks caused by suboptimal hyperparameter configurations. We design a robust cross-search-space evaluation protocol, construct a reproducible benchmark dataset and open-source toolkit, and incorporate a low-cost simulation replay mechanism to enable efficient meta-strategy optimization—fully adhering to FAIR (Findable, Accessible, Interoperable, Reusable) principles. Contribution/Results: Experiments demonstrate that lightweight hyperparameter tuning improves tuner performance by 94.8% on average; integrating meta-strategy optimization further boosts average gain to 204.7%. This work establishes both theoretical foundations and practical pathways for self-enhancement in automated tuning frameworks.

This study addresses the challenges of tool coordination and information consistency arising from downstream constraint backtracking in the multidisciplinary design of liquid oxygen/kerosene rocket thrust chambers. To this end, a long-range engineering agent framework is proposed. This framework employs a single coding agent to orchestrate performance, optimization, geometric, and multiphysics capabilities. It integrates a provenance-aware knowledge graph to assist method selection, constructs a typed intermediate representation to maintain shared parameters, and establishes revision-aware mechanisms for result invalidation and input blocking. Experimental results demonstrate that the proposed approach successfully handles operating point corrections following infeasible cooling searches, thereby validating both its sustained coordination capability across disciplinary workflows and the effectiveness of the dependency invalidation mechanism.

This study addresses the challenges of control design in complex industrial processes characterized by multivariable coupled dynamics by proposing an automated control strategy generation framework that integrates large language models (LLMs) with Bayesian optimization. The approach decomposes control design into structured code generation steps, ensuring physical consistency through execution-based validation and feedback-driven repair. It pioneers the automatic synthesis of decentralized PI controller architectures and their tuning environments directly from dynamic process models. Evaluated on a nonlinear gas preheater benchmark, the generated control schemes—subsequently refined via Bayesian optimization—achieve a 26.5% improvement in closed-loop performance and significantly enhance the transient response of pressure loops, thereby demonstrating the method’s effectiveness and novelty.

Model Predictive Control (MPC) suffers from a lack of performance guarantees during cost function parameter tuning. Method: This paper proposes a Bayesian optimization framework with explicit performance constraints. We introduce, for the first time, a Constrained Upper Confidence Bound (C-UCB) criterion that ensures—under Gaussian process modeling and posterior belief updates—the satisfaction of performance thresholds with high probability at any iteration. Coupled with a goal-directed optimistic exploration strategy, the framework enables performance-driven online learning. Contribution/Results: We theoretically prove finite-time convergence to the constrained optimal solution. Experiments on both autonomous racing simulation and real-world hardware platforms demonstrate that our method significantly reduces constraint violations and cumulative regret compared to classical and state-of-the-art Bayesian optimization approaches.

In high-cost simulation-driven design, automatically translating natural language requirements into executable mathematical optimization models remains a critical bottleneck. Method: This paper proposes a solver-agnostic Automated Problem Formulation (APF) framework that integrates large language model (LLM) supervised fine-tuning, formal natural-language-to-optimization-model mapping, synthetic data generation, and semantic consistency verification. Crucially, APF introduces the first pipeline for generating and annotating training data without requiring real simulation feedback. Contribution/Results: Evaluated on antenna design tasks, APF achieves significantly higher requirement formalization accuracy and radiation efficiency curve compliance rates than state-of-the-art methods, demonstrating strong practicality, generalizability, and robustness. To our knowledge, this is the first work enabling end-to-end, fully automated translation from ambiguous engineering requirements to executable optimization models—eliminating reliance on domain experts and manual modeling.

This work addresses the lack of systematic methodologies in model optimization, which often relies on heuristic choices and struggles to accommodate diverse deployment constraints. It formalizes model compression and acceleration as a constraint-aware multi-objective engineering decision problem, establishing a unified and actionable framework grounded in five key dimensions: data availability, latency, memory footprint, accuracy tolerance, and retraining budget. By integrating techniques such as quantization, pruning, knowledge distillation, parameter-efficient fine-tuning (PEFT), and inference optimization, the study proposes tailored optimization pipelines for four representative industrial scenarios, delivering a reproducible and quantifiable guide for technology selection.

This work addresses the challenge that traditional PID tuning relies heavily on model identification and fails to capture the empirical expertise of engineers who iteratively adjust parameters based on observed system responses. To bridge this gap, the authors propose a novel tuning framework that integrates control-domain knowledge with the reasoning capabilities of large and small language models. The approach formalizes the engineer’s tuning process into an executable task by leveraging closed-loop response characteristics, diagnostic cues, tuning preferences, and IMC-based examples to guide parameter generation and refinement. Physical constraints and reinforcement learning are incorporated to enhance performance, with the method employing supervised fine-tuning (SFT) and a physics-informed group relative policy optimization (PI-GRPO). Evaluated on 200 FOPDT/SOPDT processes, the cloud-based large model achieves a success rate of 75–89%, while a locally deployed Qwen3-0.6B model, after optimization, attains a first-recommendation success rate of 94.0%.

This work addresses the challenge that existing automatic code generation methods often produce structurally invalid or physically inconsistent models, which are unsuitable for engineering simulation. To ensure physical consistency and simulatability, the authors propose a procedural modeling framework that integrates domain knowledge injection, constraint-guided fine-tuning, and closed-loop simulation validation. Key contributions include CivilInstruct—the first instruction-following dataset tailored for structural engineering—along with a two-stage fine-tuning strategy and MBEval, a validation-driven evaluation benchmark. Experimental results demonstrate that the proposed approach significantly outperforms baseline methods across multiple rigorous metrics, effectively suppressing hallucinations and constraint violations, and enabling the direct use of generated models in structural dynamics simulations.

This work addresses the key bottleneck in high-fidelity aerodynamic drag prediction for vehicles—namely, geometric preprocessing, mesh generation, resource contention, and reproducibility challenges—rather than solver runtime. The authors propose a contract-centered, self-evolving coded agent framework that formulates drag coefficient prediction as a constrained optimization problem in program space. By leveraging multi-objective selection and structured mutation operators spanning data, models, loss functions, and partitioning strategies, the framework generates executable and auditable surrogate pipelines. A key innovation lies in evolving complete, executable programs while enforcing hard evaluation contracts to guarantee leak-free, replayable, and robust evaluations. A “filter-and-upgrade” deployment strategy balances efficiency and reliability. The best-performing system achieves a Combined Score of 0.9335 and a symbolic accuracy of 0.9180, demonstrating the critical role of adaptive sampling and island migration in enhancing convergence quality.