Institution profile

University of Bath

Academic institutioneurope · gb
Official website
Research library196linked papers
Opportunities0open roles
Selected work

Representative Papers

A Novel Multi-Criteria Local Latin Hypercube Refinement System for Commutation Angle Improvement in IPMSMs

Mar 01, 2023IEEE transactions on industry applications

Optimizing the commutation angle γ across the wide-speed operating range of interior permanent magnet synchronous motors (IPMSMs) remains challenging, while simultaneously balancing permanent magnet (PM) volume reduction and high torque density. Method: This paper proposes a multi-criteria local Latin hypercube refinement (MLHR) sampling technique to construct a high-accuracy real-time γ mapping model, integrated with multi-objective optimization, vector diagram modeling, and coordinated maximum-torque-per-ampere (MTPA) and maximum-torque-per-volt (MTPV) control. Contribution/Results: The method achieves optimal γ trajectory planning without increasing phase current magnitude, significantly reducing PM volume while enhancing commutation accuracy and torque density. Experimental validation on the third-generation Toyota Prius IPMSM demonstrates an 18.7% reduction in PM mass, a 12.3% increase in torque density, and γ prediction error below 0.8°, confirming its engineering applicability.

6 citationsRead paper

A physics-informed Bayesian optimization method for rapid development of electrical machines

Feb 24, 2024Scientific Reports

To address the challenge of rapidly optimizing the stator slot fill factor (SFF) in electric vehicle traction motors, this paper proposes a mechanism-driven Bayesian optimization (BO) framework. The method explicitly incorporates electromagnetic physical constraints into the BO pipeline, integrating physics-informed modeling, an adaptive acquisition function, and multi-fidelity simulation to construct a high-fidelity, generalizable Gaussian process surrogate model. In permanent magnet synchronous motor (PMSM) design, the approach significantly enhances optimization reliability under limited data: it reduces the required number of iterations by 60% and improves optimization accuracy for key performance metrics—including efficiency and torque density—by 35%. This work establishes a new paradigm for efficient, interpretable, and physics-aware motor design.

2 citationsRead paper

Physics-constrained DeepONet for Surrogate CFD models: a curved backward-facing step case

Mar 14, 2025

This work addresses the insufficient accuracy of surrogate modeling for backward-facing curved step flows under sparse-data conditions. We propose a physics-constrained DeepONet (PC-DeepONet), which, for the first time, enforces mass conservation—i.e., zero divergence of the velocity field—as a hard constraint within the DeepONet architecture. The method integrates parameterized geometric mapping with CFD data-driven training. Compared to purely data-driven baselines, PC-DeepONet achieves convergence using only 50 training samples and 50 optimization iterations, significantly improving prediction accuracy and physical consistency in low-data regimes—particularly enhancing generalization capability for velocity and pressure fields. Our key contribution is the pioneering design of a divergence-free neural operator that simultaneously ensures high fidelity and strong physical interpretability, establishing a new paradigm for CFD surrogate modeling in geometrically complex, data-scarce scenarios.

1 citationsRead paper

Hierarchical Preference Optimization: Learning to achieve goals via feasible subgoals prediction

Nov 01, 2024arXiv.org

Hierarchical reinforcement learning (HRL) suffers from two key challenges: non-stationarity in high-level policies due to evolving low-level policies, and infeasible sub-goals generated by high-level policies that low-level policies cannot execute. To address these, we propose Hierarchical Preference Optimization (HPO), the first framework to integrate token-level direct preference optimization (DPO) into HRL—without requiring a pretrained reference policy. HPO jointly optimizes high-level goal generation and low-level action selection via a bilevel optimization formulation. We introduce a primitive-regularized DPO loss that mathematically enforces sub-goal feasibility and prevents degenerate solutions. Additionally, maximum entropy regularization is incorporated to enhance exploration robustness. Evaluated on robotic navigation and manipulation tasks, HPO achieves an average 35% performance gain over strong baselines, significantly mitigating both non-stationarity and sub-goal infeasibility. Ablation studies and quantitative analysis comprehensively validate its effectiveness.

1 citationsRead paper

LGR2: Language Guided Reward Relabeling for Accelerating Hierarchical Reinforcement Learning

Jun 09, 2024arXiv.org

Hierarchical reinforcement learning (HRL) for natural-language-instructed complex robotic control suffers from non-stationary high-level rewards due to low-level policy evolution, severely hindering high-level policy learning. Method: We propose a language-guided high-level reward relabeling mechanism that leverages large language models (LLMs) to semantically parse instructions and dynamically map sparse environmental feedback into decoupled, instruction-aligned high-level reward signals—thereby fundamentally mitigating reward non-stationarity in HRL. Our approach integrates LLM-based semantic understanding, hierarchical control architecture, and language-informed dynamic reward shaping. Contribution/Results: Evaluated on sparse-reward navigation and manipulation tasks, our method achieves over 70% success rate—substantially outperforming state-of-the-art baselines. Real-robot experiments further demonstrate robustness and generalization in complex, unstructured physical environments.

1 citationsRead paper
Recent publications

Latest Papers

Seeing Before Answering: Training-Free Visual Layer Profiling for Vision-Language Models

Aug 17, 2026

This study addresses the challenges of suboptimal visual layer selection and high search costs in vision-language models by proposing a training-free analytical framework grounded in representation geometry. Leveraging visual dataset entropy and Gromov-Wasserstein distance, this method accurately evaluates layer utility and identifies optimal candidates using only one hundred unlabeled samples, thereby eliminating inference overhead. Extensive experiments validate the strategy’s effectiveness, elucidate the impact of projectors on geometric signals, and demonstrate region-level guidance capabilities in tasks such as Video-LLaVA. Ultimately, this approach significantly reduces search costs while effectively enhancing performance across diverse multimodal tasks, offering a computationally efficient alternative to conventional layer selection paradigms.

0 citationsRead paper

Approximate Muon with low-rank adapters

Aug 14, 2026

This study addresses the incompatibility of the Muon optimizer with LoRA fine-tuning, which arises from the inability to orthogonalize low-rank parameters. To overcome this limitation, we propose sMuon, a novel method that achieves the first approximate solution to the Muon objective in low-rank settings. By leveraging linearization and least-squares formulations, sMuon transforms computationally intensive decomposition operations into efficient matrix multiplications. Extensive experiments demonstrate that sMuon successfully adapts Muon to parameter-efficient fine-tuning paradigms, yielding consistent performance improvements in both supervised fine-tuning and ReLoRA pretraining tasks. Consequently, this work presents an effective optimization strategy for low-rank adaptation, significantly enhancing computational efficiency while maintaining robust model performance across diverse training scenarios.

0 citationsRead paper