国际机械系统动力学学报(英文)最新文献

Cover Caption: Scoliosis Rehabilitation with a Robotic Brace Powered by RL-based Impedance Control and Digital Twin: Adolescent Idiopathic Scoliosis (AIS) is commonly treated with traditional braces that rely solely on passive strap tensioning, lacking intelligent control strategies. This study proposes a reinforcement learning-based position-based impedance control (RLPIC) method for robotic braces to enable active human–robot interaction. To safely simulate and train the control system, a novel five-dimensional, three-layer digital twin (DT) model is developed, integrating physical modeling, digital modeling, bidirectional interaction, and optimization, enhanced by a neural network-based parameter estimator. Both numerical simulations and real-time experiments validate the DT and RLPIC framework, demonstrating improved tracking and interaction performance in AIS treatment.

Marine environment is a harsh environment that can cause major issues for marine structures while operating in this environment, including fatigue cracking and corrosion damage, which can yield catastrophic consequences, such as human life losses, financial losses, environmental pollution, and so forth. Therefore, it is critical to take necessary actions before undesired situations happen. One potential solution is to install structural health monitoring systems on marine structures. Structural health monitoring is a technology to enhance the safety, stability, and functionality of large engineering structures. The inverse Finite Element Method (iFEM) is a promising technique for this purpose. In this study, the corrosion damage detection capability of iFEM is presented by introducing two new damage parameters for plates under tension and bending loading conditions. The contribution of newly introduced parameters to the accuracy of iFEM on damage detection is demonstrated for multiple corrosion scenarios and sensor configurations.

Front Cover Caption: Control of a lambda-robot based on machine learning surrogates for inverse kinematics and kinetics: Tracking control of multibody systems with closed-loop mechanisms presents significant computational challenges due to the complexity of inverse kinematics and dynamics. This study introduces an innovative approach that replaces traditional model-based methods with artificial intelligence by training surrogate models on simulation data. Using the λ-robot, a parallel mechanism, as a case study, the workspace is analyzed to ensure comprehensive data coverage for training. The trained surrogates provide control inputs that enable the use of a linear quadratic regulator (LQR) for trajectory tracking. An additional feedback loop addresses model uncertainties. Simulation results validate the effectiveness of this AI-enhanced, data-driven control framework.

Back Cover Caption: Transfer learning in Physics-informed Neural Networks: This study explores the generalization capabilities of physics-informed neural networks (PINNs) through transfer learning techniques applied to partial differential equation (PDE) problems. Traditional PINNs require retraining when problem conditions change, whereas this approach leverages full finetuning, lightweight finetuning, and low-rank adaptation (LoRA) to enhance efficiency across varying boundary conditions, materials, and geometries. Benchmark cases include the Taylor-Green Vortex, functionally graded elastic materials, and structural problems such as a square plate with a circular hole. The results demonstrate that full finetuning and LoRA significantly improve convergence and accuracy, highlighting their potential in developing more adaptable and efficient PINN-based solvers.

A robust control method for the uncertain vertical electric stabilization system (VESS) with flexible nonlinearity is proposed, and the mismatched uncertainty is considered and compensated based on the backstepping idea. First, based on evaluating the coupling effects of the flexible nonlinearity, the analytical dynamics model of the VESS is established. Second, the tracking error is defined as the evaluation of the system's pitch-pointing tracking control, and then the mismatched state space model with two interconnected subsystems is established as the controlled system. Third, the original mismatched system is converted to the locally matched system using the backstepping design to transform the system state variables. The robust control is proposed to handle the flexible nonlinearity and mismatched uncertainty, which can make both the original system and the reconfigured system present practical stability. Finally, the effectiveness of the proposed control is verified by numerical simulation experiments. This study should be the first to consider flexible nonlinearity coupling and two different uncertainties (matched and mismatched uncertainty) in the design of pitch-pointing tracking control for the vertical electric stabilization system (VESS).

AI for PDEs has garnered significant attention, particularly physics-informed neural networks (PINNs). However, PINNs are typically limited to solving specific problems, and any changes in problem conditions necessitate retraining. Therefore, we explore the generalization capability of transfer learning in the strong and energy forms of PINNs across different boundary conditions, materials, and geometries. The transfer learning methods we employ include full finetuning, lightweight finetuning, and low-rank adaptation (LoRA). Numerical experiments include the Taylor-Green Vortex in fluid mechanics and functionally graded materials with elastic properties, as well as a square plate with a circular hole in solid mechanics. The results demonstrate that full finetuning and LoRA can significantly improve convergence speed while providing a slight enhancement in accuracy. However, the overall performance of lightweight finetuning is suboptimal, as its accuracy and convergence speed are inferior to those of full finetuning and LoRA.

In this article, the physics informed neural networks (PINNs) is employed for the numerical simulation of heat transfer involving a moving source under mixed boundary conditions. To reduce computational effort and increase accuracy, a new training method is proposed that uses a continuous time-stepping through transfer learning. A single network is initialized and used as a sliding window function across the time domain. On this single network each time interval is trained with the initial condition for $$ iteration as the solution obtained at $$ iteration. Thus, this framework enables the computation of large temporal intervals without increasing the complexity of the network itself. The proposed framework is used to estimate the temperature distribution in a homogeneous medium with a moving heat source. The results from the proposed framework is compared with traditional finite element method and a good agreement is seen.

Nonlinearity in parallel compliance can be exploited to improve the performance of locomotion systems in terms of (1) energy efficiency, (2) control robustness, and (3) gait optimality; that is, attaining energy efficiency across a set of motions. Thus far, the literature has investigated and validated only the first two benefits. In this study, we present a new mathematical framework for designing nonlinear compliances in cyclic tasks encompassing all three benefits. We present an optimization-based formulation for each benefit to obtain the desired compliance profile. Furthermore, we analytically prove that, compared to linear compliance, using nonlinear compliance leads to (1) lower energy consumption, (2) better closed-loop performance, specifically in terms of tracking error, and (3) a higher diversity of natural frequencies. To compare the performance of linear and nonlinear compliance, we apply the proposed methods to a diverse set of robotic systems performing cyclic tasks, including a 2-DOF manipulator, a 3-DOF bipedal walker, and a hopper model. Compared to linear compliance, the nonlinear compliance leads to better performance in all aspects; for example, a 70% reduction in energy consumption and tracking error for the manipulator simulation. Regarding gait optimality, for all robotic simulation models, compared to linear compliance, the nonlinear compliance has lower energy consumption and tracking error over the considered set of motions. The proposed analytical studies and simulation results strongly support the idea that using nonlinear compliance significantly improves robotic system performance in terms of energy efficiency, control robustness, and gait optimality.

This article investigates the nonlinear vibration behavior of porous multidirectional piezoelectric functionally graded nonuniform (PFGN) plates resting on orthotropic variable elastic foundations and subjected to hygrothermal loading. The sigmoidal law is employed to define the multidirectional gradation properties, incorporating porosity along both the axial and thickness directions. The governing equations for the porous multidirectional PFGN plate are derived using the modified first-order shear deformation theory (FSDT) with nonlinear von Kármán strain and Hamilton's principle. A higher-order finite element (FE) approach, combined with a modified Newton-Raphson method, is utilized to solve the resulting equations. The study reveals that orthotropic variable elastic foundations significantly influence the vibration behavior of multidirectional PFGN porous plates compared to conventional elastic foundations under hygrothermal loading. Additionally, the effects of various parameters such as thickness ratio, tapered ratio, material exponent, boundary conditions, porosity distribution, electrical loading, temperature variation, and moisture change on the vibration behavior are comprehensively analyzed. The results of this study have direct applications in energy harvesting and structural health monitoring, offering a novel approach to designing and optimizing smart materials for engineering systems operating under hygrothermal and thermoelectrical conditions.

The advancement of artificial intelligence (AI) in material design and engineering has led to significant improvements in predictive modeling of material properties. However, the lack of interpretability in machine learning (ML)-based material informatics presents a major barrier to its practical adoption. This study proposes a novel quantitative computational framework that integrates ML models with explainable artificial intelligence (XAI) techniques to enhance both predictive accuracy and interpretability in material property prediction. The framework systematically incorporates a structured pipeline, including data processing, feature selection, model training, performance evaluation, explainability analysis, and real-world deployment. It is validated through a representative case study on the prediction of high-performance concrete (HPC) compressive strength, utilizing a comparative analysis of ML models such as Random Forest, XGBoost, Support Vector Regression (SVR), and Deep Neural Networks (DNNs). The results demonstrate that XGBoost achieves the highest predictive performance ($$), while SHAP (Shapley Additive Explanations) and LIME (Local Interpretable Model-Agnostic Explanations) provide detailed insights into feature importance and material interactions. Additionally, the deployment of the trained model as a cloud-based Flask-Gunicorn API enables real-time inference, ensuring its scalability and accessibility for industrial and research applications. The proposed framework addresses key limitations of existing ML approaches by integrating advanced explainability techniques, systematically handling nonlinear feature interactions, and providing a scalable deployment strategy. This study contributes to the development of interpretable and deployable AI-driven material informatics, bridging the gap between data-driven predictions and fundamental material science principles.