Mechanical Systems and Signal Processing最新文献_第7页

CNC-VLM: An RLHF-optimized industrial large vision-language model with multimodal learning for imbalanced CNC fault detection CNC- vlm：一种基于rlhf优化的多模态学习工业大型视觉语言模型，用于不平衡数控故障检测

IF 8.9 1区工程技术 Q1 ENGINEERING, MECHANICAL

Mechanical Systems and Signal Processing

Pub Date : 2026-01-02 DOI: 10.1016/j.ymssp.2025.113838

Zisheng Wang , Junjie Chen , Chisen Wang , Cong Peng , Jianping Xuan , Tielin Shi , Mingjian Zuo

CNC fault detection in industrial scenarios still suffers from several important challenges, such as poor generalization to imbalanced samples, poor consistency between visual-textual semantics, and limited utilization of domain-specific multimodal knowledge. To address these issues, we propose an RLHF (Reinforcement Learning from Human Feedback)-optimized industrial large model (CNC-VLM) that leverages vision-language multimodal knowledge for autonomous fault detection. CNC-VLM consists of a visual encoder, a projector, and a textual decoder. Specifically, the visual encoder extracts discriminative visual landmarks from fault time-frequency images, which are then further mapped into a modality-aligned latent space via a projection layer. The decoder incorporates a vision-language cross-attention mechanism to combine visual semantics with textual prompts for precise reasoning and detection. We further design a direct preference optimization (DPO) algorithm for CNC fault detection based on the RLHF training paradigm, which combines the model output with human feedback (expert knowledge) to enhance the imbalanced detection ability for small sample faults. Through engineering samples obtained from CNC machines, the verification results demonstrate that CNC-VLM significantly outperforms existing state-of-the-art models in terms of accuracy and F1-score. Moreover, CNC-VLM can automatically generate detection descriptions and maintenance suggestions, showing good potential for practical deployment in the field of intelligent manufacturing.

工业场景中的CNC故障检测仍然面临着几个重要的挑战，例如对不平衡样本的泛化能力差，视觉文本语义之间的一致性差，以及特定于领域的多模态知识的利用有限。为了解决这些问题，我们提出了一种RLHF（从人类反馈中强化学习）优化的工业大型模型（CNC-VLM），该模型利用视觉语言多模态知识进行自主故障检测。CNC-VLM由一个视觉编码器、一个投影仪和一个文本解码器组成。具体而言，视觉编码器从故障时频图像中提取判别性视觉标志，然后通过投影层将其进一步映射到模态对齐的潜在空间中。该解码器采用视觉语言交叉注意机制，将视觉语义与文本提示相结合，以进行精确的推理和检测。我们进一步设计了一种基于RLHF训练范式的CNC故障检测直接偏好优化（DPO）算法，该算法将模型输出与人类反馈（专家知识）相结合，增强了对小样本故障的不平衡检测能力。通过从CNC机床上获得的工程样本，验证结果表明CNC- vlm在精度和f1分数方面显著优于现有的最先进模型。此外，CNC-VLM可以自动生成检测描述和维护建议，在智能制造领域具有良好的实际部署潜力。

{"title":"CNC-VLM: An RLHF-optimized industrial large vision-language model with multimodal learning for imbalanced CNC fault detection","authors":"Zisheng Wang , Junjie Chen , Chisen Wang , Cong Peng , Jianping Xuan , Tielin Shi , Mingjian Zuo","doi":"10.1016/j.ymssp.2025.113838","DOIUrl":"10.1016/j.ymssp.2025.113838","url":null,"abstract":"<div><div>CNC fault detection in industrial scenarios still suffers from several important challenges, such as poor generalization to imbalanced samples, poor consistency between visual-textual semantics, and limited utilization of domain-specific multimodal knowledge. To address these issues, we propose an RLHF (Reinforcement Learning from Human Feedback)-optimized industrial large model (CNC-VLM) that leverages vision-language multimodal knowledge for autonomous fault detection. CNC-VLM consists of a visual encoder, a projector, and a textual decoder. Specifically, the visual encoder extracts discriminative visual landmarks from fault time-frequency images, which are then further mapped into a modality-aligned latent space via a projection layer. The decoder incorporates a vision-language cross-attention mechanism to combine visual semantics with textual prompts for precise reasoning and detection. We further design a direct preference optimization (DPO) algorithm for CNC fault detection based on the RLHF training paradigm, which combines the model output with human feedback (expert knowledge) to enhance the imbalanced detection ability for small sample faults. Through engineering samples obtained from CNC machines, the verification results demonstrate that CNC-VLM significantly outperforms existing state-of-the-art models in terms of accuracy and F1-score. Moreover, CNC-VLM can automatically generate detection descriptions and maintenance suggestions, showing good potential for practical deployment in the field of intelligent manufacturing.</div></div>","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"245 ","pages":"Article 113838"},"PeriodicalIF":8.9,"publicationDate":"2026-01-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145876930","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Physics-guided periodic weighting sparse representation for fault feature extraction in rotating machinery 旋转机械故障特征提取的物理引导周期加权稀疏表示

IF 8.9 1区工程技术 Q1 ENGINEERING, MECHANICAL

Mechanical Systems and Signal Processing

Pub Date : 2026-01-02 DOI: 10.1016/j.ymssp.2025.113801

Lei Fu , Yuhao Jiang , Jia Liu , Fang Xu , Zepeng Ma

Extracting periodic fault components in complex interference environments remains a critical challenge in bearing fault diagnosis. Existing methods face issues such as low accuracy in periodic component extraction, insufficient interference suppression, and difficulty in separating multiple fault components under multi-source noise and compound fault conditions. To address these challenges, this paper proposes a Physics-Guided Periodic Weighting Sparse Representation (PG-PWSR) method. First, based on a bearing dynamic model, the method introduces a similarity-based fault subband value (SFSV) indicator to improve weak periodic component extraction and interference suppression under multi-source noise and compound faults. Then, a nonlinear exponentially weighted frequency energy operator spectrum is proposed to enhance periodic estimation and pulse localization, mitigating low-frequency disturbances and accurately identifying pulse positions. Finally, the method integrates adaptive sparse weighting, periodic pulse localization, and a sequential extraction framework to accurately extract weak fault features. Simulations and experiments validate that the PG-PWSR method significantly improves periodic fault feature extraction and effectively separates overlapping compound fault components.

在复杂干扰环境中提取周期性故障分量是轴承故障诊断的关键问题。现有方法存在周期性分量提取精度低、干扰抑制不足、多源噪声和复合故障条件下多故障分量分离困难等问题。为了解决这些问题，本文提出了一种物理引导的周期加权稀疏表示（PG-PWSR）方法。首先，该方法在轴承动态模型的基础上，引入基于相似性的故障子带值（SFSV）指标，提高多源噪声和复合故障下的弱周期分量提取和干扰抑制能力；然后，提出了一种非线性指数加权频率能量算子谱，以增强周期估计和脉冲定位，减轻低频干扰，准确识别脉冲位置。最后，该方法结合自适应稀疏加权、周期脉冲定位和序列提取框架，精确提取弱故障特征。仿真和实验验证了PG-PWSR方法显著改善了周期性故障特征提取，有效分离了重叠复合故障分量。

{"title":"Physics-guided periodic weighting sparse representation for fault feature extraction in rotating machinery","authors":"Lei Fu , Yuhao Jiang , Jia Liu , Fang Xu , Zepeng Ma","doi":"10.1016/j.ymssp.2025.113801","DOIUrl":"10.1016/j.ymssp.2025.113801","url":null,"abstract":"<div><div>Extracting periodic fault components in complex interference environments remains a critical challenge in bearing fault diagnosis. Existing methods face issues such as low accuracy in periodic component extraction, insufficient interference suppression, and difficulty in separating multiple fault components under multi-source noise and compound fault conditions. To address these challenges, this paper proposes a Physics-Guided Periodic Weighting Sparse Representation (PG-PWSR) method. First, based on a bearing dynamic model, the method introduces a similarity-based fault subband value (SFSV) indicator to improve weak periodic component extraction and interference suppression under multi-source noise and compound faults. Then, a nonlinear exponentially weighted frequency energy operator spectrum is proposed to enhance periodic estimation and pulse localization, mitigating low-frequency disturbances and accurately identifying pulse positions. Finally, the method integrates adaptive sparse weighting, periodic pulse localization, and a sequential extraction framework to accurately extract weak fault features. Simulations and experiments validate that the PG-PWSR method significantly improves periodic fault feature extraction and effectively separates overlapping compound fault components.</div></div>","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"245 ","pages":"Article 113801"},"PeriodicalIF":8.9,"publicationDate":"2026-01-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145876929","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Optimality of the mean-square value, skewness, kurtosis, and synentropy for detection of weak additive fault signatures in (nearly) Gaussian noise 均方值、偏度、峰度和同熵的最优性在（近）高斯噪声中检测弱加性故障特征

IF 8.9 1区工程技术 Q1 ENGINEERING, MECHANICAL

Mechanical Systems and Signal Processing

Pub Date : 2026-01-02 DOI: 10.1016/j.ymssp.2025.113828

Jérôme Antoni , Pietro Borghesani , Wade A. Smith , Robert Bond Randall , Zhongxiao Peng

The mean-square value, the skewness, and the kurtosis are scalar indicators widely used in health monitoring systems. They are simple to use and have proven sensitive to fault occurrence. This paper theoretically proves that they are the optimal health indicators for the early detection of faults in additive Gaussian noise, in the sense of maximising the probability of detection given a constant false alarm rate. The mean-square value is optimal in scenarios where the scale parameter of the background noise is known; otherwise, it should be replaced by a new indicator, coined “synentropy”. Remarkably, these results hold true independently of the fault signature. The paper also shows how to correct these indicators when the background noise slightly departs from Gaussianity, provides their statistical thresholds, and evidences that their relative sensitivity depends on the data length.

均方值、偏度和峰度是健康监测系统中广泛使用的标量指标。它们使用简单，并已被证明对故障发生敏感。本文从理论上证明了它们是加性高斯噪声中故障早期检测的最佳健康指标，在给定恒定的虚警率的情况下，检测概率最大化。当背景噪声的尺度参数已知时，均方值是最优的；否则，它应该被一个新的指标所取代，即“同步熵”。值得注意的是，这些结果与断层特征无关。本文还说明了在背景噪声轻微偏离高斯性时如何校正这些指标，给出了它们的统计阈值，并证明了它们的相对灵敏度取决于数据长度。

引用次数: 0

A novel multi-domain dictionary learning for moving load identification using adaptive regularization path and logarithmic penalty function 一种基于自适应正则化路径和对数惩罚函数的多域字典学习移动载荷识别方法

IF 8.9 1区工程技术 Q1 ENGINEERING, MECHANICAL

Mechanical Systems and Signal Processing

Pub Date : 2025-12-31 DOI: 10.1016/j.ymssp.2025.113824

Zhilong Hou , Chudong Pan , Hong Hao , Ling Yu

Dictionary learning methods offer superior capabilities in modeling complex vehicle loads compared with traditional fixed-dictionary approaches. However, their performance remains constrained by inherent limitations, including the empirical selection of sparsity levels, insufficient inter-domain heterogeneity, and limited generalization ability. To address these issues, a multi-domain dictionary learning model driven by vehicle-induced bridge responses is proposed in this study, aiming to enhance the accuracy and robustness of moving load identification (MLI). Specifically, the vehicle loads are firstly decomposed into zero-frequency, narrowband, and broadband components based on their spectral distribution characteristics. For each component, corresponding frequency and time domain dictionaries are constructed to match their respective physical attributes. Subsequently, a dual-sparsity coding technique is introduced to jointly optimize the frequency-domain dictionary atoms and their sparse coefficients. To mitigate the sensitivity of traditional regularization strategies for parameter selection, a parameter-free logarithmic penalty function is introduced to update the time-domain dictionary. Furthermore, a sparsity selection strategy based on feature contributions in the regularization path is proposed to adaptively adjust the sparsity level. Finally, the effectiveness of the proposed method is verified through a series of vehicle-bridge coupling simulations and model experiments. The results show that the proposed MLI framework outperforms both fixed-dictionary and single-domain dictionary learning methods in capturing the intrinsic structural characteristics of vehicle loads, particularly under conditions of higher road surface roughness. Besides, comparative studies further show that the adaptive sparsity effectively balances identification accuracy and efficiency in various models without prior information of sparsity, which provides an innovative and efficient MLI tool in bridge structural health monitoring.

与传统的固定字典方法相比，字典学习方法在复杂车辆负载建模方面具有优越的能力。然而，它们的性能仍然受到固有局限性的制约，包括稀疏度水平的经验选择、域间异质性不足和有限的泛化能力。为了解决这些问题，本研究提出了一种基于车辆诱导桥梁响应的多域字典学习模型，旨在提高移动荷载识别的准确性和鲁棒性。具体而言，首先根据车辆载荷的频谱分布特征将其分解为零频、窄带和宽带分量。对于每个组件，构建相应的频域和时域字典以匹配其各自的物理属性。随后，引入双稀疏编码技术，对频域字典原子及其稀疏系数进行联合优化。为了降低传统正则化策略对参数选择的敏感性，引入无参数对数惩罚函数来更新时域字典。在此基础上，提出了一种基于正则化路径中特征贡献的稀疏度选择策略，自适应调整稀疏度水平。最后，通过一系列车桥耦合仿真和模型实验验证了所提方法的有效性。结果表明，所提出的MLI框架在捕获车辆载荷的内在结构特征方面优于固定字典和单域字典学习方法，特别是在较高的路面粗糙度条件下。此外，对比研究进一步表明，自适应稀疏度在没有稀疏度先验信息的情况下，有效地平衡了各种模型的识别精度和效率，为桥梁结构健康监测提供了一种创新而高效的MLI工具。

{"title":"A novel multi-domain dictionary learning for moving load identification using adaptive regularization path and logarithmic penalty function","authors":"Zhilong Hou , Chudong Pan , Hong Hao , Ling Yu","doi":"10.1016/j.ymssp.2025.113824","DOIUrl":"10.1016/j.ymssp.2025.113824","url":null,"abstract":"<div><div>Dictionary learning methods offer superior capabilities in modeling complex vehicle loads compared with traditional fixed-dictionary approaches. However, their performance remains constrained by inherent limitations, including the empirical selection of sparsity levels, insufficient inter-domain heterogeneity, and limited generalization ability. To address these issues, a multi-domain dictionary learning model driven by vehicle-induced bridge responses is proposed in this study, aiming to enhance the accuracy and robustness of moving load identification (MLI). Specifically, the vehicle loads are firstly decomposed into zero-frequency, narrowband, and broadband components based on their spectral distribution characteristics. For each component, corresponding frequency and time domain dictionaries are constructed to match their respective physical attributes. Subsequently, a dual-sparsity coding technique is introduced to jointly optimize the frequency-domain dictionary atoms and their sparse coefficients. To mitigate the sensitivity of traditional regularization strategies for parameter selection, a parameter-free logarithmic penalty function is introduced to update the time-domain dictionary. Furthermore, a sparsity selection strategy based on feature contributions in the regularization path is proposed to adaptively adjust the sparsity level. Finally, the effectiveness of the proposed method is verified through a series of vehicle-bridge coupling simulations and model experiments. The results show that the proposed MLI framework outperforms both fixed-dictionary and single-domain dictionary learning methods in capturing the intrinsic structural characteristics of vehicle loads, particularly under conditions of higher road surface roughness. Besides, comparative studies further show that the adaptive sparsity effectively balances identification accuracy and efficiency in various models without prior information of sparsity, which provides an innovative and efficient MLI tool in bridge structural health monitoring.</div></div>","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"244 ","pages":"Article 113824"},"PeriodicalIF":8.9,"publicationDate":"2025-12-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145884634","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Development of a stall-flutter-based piezoelectric wind energy harvester with a diamond-shaped airfoil and endplates for high performance 基于失速颤振的高性能菱形翼型和端板压电风能收集器的研制

IF 8.9 1区工程技术 Q1 ENGINEERING, MECHANICAL

Mechanical Systems and Signal Processing

Pub Date : 2025-12-31 DOI: 10.1016/j.ymssp.2025.113816

Yonghao Liu , Ji-Heon Baek , Kai Xue , Jongwon Seok

Flutter-based wind energy harvesters often suffer from limited efficiency under low-to-moderate wind conditions, hindering their application in distributed power systems. To address this challenge, we propose a stall flutter based piezoelectric harvester featuring a diamond-shaped airfoil with endplates. This design introduces a static angle of attack to induce stall flutter and leverages 1:1 internal resonance in conjunction with multi-point flow separation to enhance energy conversion efficiency.

A theoretical framework is established using the extended Hamilton’s principle, the Galerkin method, and the expansion theorem. A two-step modal analysis is applied to discretize the continuous system and analyze its dynamic and energetic responses. Nonlinear dynamic behaviors are examined through the perturbation method and harmonic balance technique, revealing the presence of limit cycle oscillations and hysteretic phenomena. To elucidate the aerodynamic enhancement mechanisms, computational fluid dynamics and finite-time Lyapunov exponent analyses are conducted. These demonstrate that multi-point flow separation facilitates dynamic stall vortex formation, which strengthens the interaction between the elastic structure and aerodynamic forces. Experimental validation is performed using an optimized prototype, showing strong agreement with theoretical predictions. The proposed harvester achieves a power density of 167.1 W/m³ at 9 m/s, significantly outperforming conventional designs. This work not only presents a unified theoretical–experimental framework for analyzing stall-flutter-based harvesters but also underscores the crucial role of nonlinear flow–structure interactions in optimizing energy harvesting efficiency. The methodologies and findings contribute valuable insights for the design of compact, self-powered devices and advance the development of next-generation sustainable energy harvesting systems.

基于颤振的风能采集器在低至中等风力条件下效率有限，阻碍了其在分布式电力系统中的应用。为了解决这一挑战，我们提出了一种基于失速颤振的压电收割机，该收割机具有带端板的菱形翼型。该设计引入了一个静态攻角来诱导失速颤振，并利用1:1的内部共振与多点流动分离相结合来提高能量转换效率。利用扩展的哈密顿原理、伽辽金方法和展开定理建立了理论框架。采用两步模态分析方法对连续系统进行离散化，分析其动力响应和能量响应。通过摄动法和谐波平衡技术研究了非线性动力学行为，揭示了极限环振荡和滞回现象的存在。为了阐明气动增强机理，进行了计算流体力学和有限时间李雅普诺夫指数分析。这表明多点流动分离有利于动态失速涡的形成，加强了弹性结构与气动力之间的相互作用。利用优化后的原型进行了实验验证，结果与理论预测非常吻合。该收割机在9米/秒的速度下达到167.1 W/m3的功率密度，显著优于传统设计。这项工作不仅提出了一个统一的理论-实验框架来分析基于失速颤振的集热器，而且强调了非线性流-结构相互作用在优化能量收集效率中的关键作用。这些方法和发现为紧凑、自供电设备的设计提供了宝贵的见解，并推动了下一代可持续能源收集系统的发展。

{"title":"Development of a stall-flutter-based piezoelectric wind energy harvester with a diamond-shaped airfoil and endplates for high performance","authors":"Yonghao Liu , Ji-Heon Baek , Kai Xue , Jongwon Seok","doi":"10.1016/j.ymssp.2025.113816","DOIUrl":"10.1016/j.ymssp.2025.113816","url":null,"abstract":"<div><div>Flutter-based wind energy harvesters often suffer from limited efficiency under low-to-moderate wind conditions, hindering their application in distributed power systems. To address this challenge, we propose a stall flutter based piezoelectric harvester featuring a diamond-shaped airfoil with endplates. This design introduces a static angle of attack to induce stall flutter and leverages 1:1 internal resonance in conjunction with multi-point flow separation to enhance energy conversion efficiency.</div><div>A theoretical framework is established using the extended Hamilton’s principle, the Galerkin method, and the expansion theorem. A two-step modal analysis is applied to discretize the continuous system and analyze its dynamic and energetic responses. Nonlinear dynamic behaviors are examined through the perturbation method and harmonic balance technique, revealing the presence of limit cycle oscillations and hysteretic phenomena. To elucidate the aerodynamic enhancement mechanisms, computational fluid dynamics and finite-time Lyapunov exponent analyses are conducted. These demonstrate that multi-point flow separation facilitates dynamic stall vortex formation, which strengthens the interaction between the elastic structure and aerodynamic forces. Experimental validation is performed using an optimized prototype, showing strong agreement with theoretical predictions. The proposed harvester achieves a power density of 167.1 W/m<sup>3</sup> at 9 m/s, significantly outperforming conventional designs. This work not only presents a unified theoretical–experimental framework for analyzing stall-flutter-based harvesters but also underscores the crucial role of nonlinear flow–structure interactions in optimizing energy harvesting efficiency. The methodologies and findings contribute valuable insights for the design of compact, self-powered devices and advance the development of next-generation sustainable energy harvesting systems.</div></div>","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"244 ","pages":"Article 113816"},"PeriodicalIF":8.9,"publicationDate":"2025-12-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145884639","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Evaluation of zero-group-velocity Lamb waves induced by local feature variation in directed energy deposition 定向能沉积中局部特征变化诱导的零群速兰姆波的评价

IF 8.9 1区工程技术 Q1 ENGINEERING, MECHANICAL

Mechanical Systems and Signal Processing

Pub Date : 2025-12-31 DOI: 10.1016/j.ymssp.2025.113821

Peipei Liu , Junhao Chen , Yang Yang , Menglong Liu , Kiyoon Yi , Hoon Sohn , Zhao-Dong Xu

Directed energy deposition (DED) process often introduces porosity defects that compromise structural integrity. Zero‐group‐velocity (ZGV) Lamb waves, generated by laser ultrasonics, offer a non-contact and in-situ approach for defect detection in DED thin structures. However, the response of ZGV modes to local feature variations in DED remains underexplored. In this paper, we present our research investigating the evaluation of ZGV Lamb waves induced by local feature variation produced during DED processes. Finite element models were first built to investigate the ZGV Lamb wave behavior in DED thin structures when encountering isolated geometric features, discrete porosity defects, and their coupled effects. The results demonstrate that during the DED process, in a defect-free structure, typical geometric variations lead to frequency shift of the S₁-ZGV mode while maintaining mode purity. In contrast, porosity defects generate distinct feature guided waves (FGW) and associated FGW-ZGV mode, while also altering the S₁-ZGV frequency. When both geometric features and porosity defects are present, their coupled influence produces measurable and distinguishable modifications to both modes, enabling effective defect discrimination. Experimental validation using laser ultrasonic testing on specimens with controlled defects confirms the high sensitivity of ZGV modes to both individual and coupled features. Leveraging the non-contact, in-situ capabilities of laser ultrasonics, this approach demonstrates strong potential for real-time porosity monitoring and quality assurance in DED process.

定向能沉积（DED）工艺通常会引入气孔缺陷，从而影响结构的完整性。由激光超声产生的零群速度兰姆波（ZGV）为DED薄结构的缺陷检测提供了一种非接触的原位方法。然而，ZGV模式对DED局部特征变化的响应仍未得到充分研究。本文主要研究了由局部特征变化引起的ZGV Lamb波的评价问题。首先建立了有限元模型，研究了当遇到孤立的几何特征、离散的孔隙缺陷以及它们的耦合效应时，在DED薄结构中的ZGV Lamb波行为。结果表明，在DED过程中，在无缺陷结构中，典型的几何变化导致S1-ZGV模态频移，同时保持模态纯度。相反，孔隙缺陷会产生明显的特征导波（FGW）和相关的FGW- zgv模式，同时也会改变S1-ZGV频率。当几何特征和孔隙缺陷同时存在时，它们的耦合影响会对两种模式产生可测量和可区分的修改，从而实现有效的缺陷识别。采用激光超声检测控制缺陷试样的实验验证证实了ZGV模式对单个和耦合特征的高灵敏度。利用激光超声的非接触式原位功能，该方法在DED过程中显示了实时孔隙度监测和质量保证的强大潜力。

{"title":"Evaluation of zero-group-velocity Lamb waves induced by local feature variation in directed energy deposition","authors":"Peipei Liu , Junhao Chen , Yang Yang , Menglong Liu , Kiyoon Yi , Hoon Sohn , Zhao-Dong Xu","doi":"10.1016/j.ymssp.2025.113821","DOIUrl":"10.1016/j.ymssp.2025.113821","url":null,"abstract":"<div><div>Directed energy deposition (DED) process often introduces porosity defects that compromise structural integrity. Zero‐group‐velocity (ZGV) Lamb waves, generated by laser ultrasonics, offer a non-contact and in-situ approach for defect detection in DED thin structures. However, the response of ZGV modes to local feature variations in DED remains underexplored. In this paper, we present our research investigating the evaluation of ZGV Lamb waves induced by local feature variation produced during DED processes. Finite element models were first built to investigate the ZGV Lamb wave behavior in DED thin structures when encountering isolated geometric features, discrete porosity defects, and their coupled effects. The results demonstrate that during the DED process, in a defect-free structure, typical geometric variations lead to frequency shift of the S<sub>1</sub>-ZGV mode while maintaining mode purity. In contrast, porosity defects generate distinct feature guided waves (FGW) and associated FGW-ZGV mode, while also altering the S<sub>1</sub>-ZGV frequency. When both geometric features and porosity defects are present, their coupled influence produces measurable and distinguishable modifications to both modes, enabling effective defect discrimination. Experimental validation using laser ultrasonic testing on specimens with controlled defects confirms the high sensitivity of ZGV modes to both individual and coupled features. Leveraging the non-contact, in-situ capabilities of laser ultrasonics, this approach demonstrates strong potential for real-time porosity monitoring and quality assurance in DED process.</div></div>","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"244 ","pages":"Article 113821"},"PeriodicalIF":8.9,"publicationDate":"2025-12-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145884640","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Active vibration control enabled by intelligent excitation adaptability for frequency-variable excitations 主动振动控制，通过智能励磁自适应实现变频励磁

IF 8.9 1区工程技术 Q1 ENGINEERING, MECHANICAL

Mechanical Systems and Signal Processing

Pub Date : 2025-12-31 DOI: 10.1016/j.ymssp.2025.113831

Shuai Chen , Yilong Wang , Jian Zhang , Qianjing Wu , Xiaoyun Zhang , Dengqing Cao , Biao Wang

Active vibration control (AVC) is a well-validated solution for vibration suppression in stable environments with specific frequency ranges. However, maintaining its high performance under variable and unpredictable excitation conditions remains a significant challenge. This article develops an intelligent excitation-adaptive active vibration control (IEA-AVC) method to achieve optimal control of structures under frequency-variable excitations. The method integrates conventional AVC techniques with real-time identification of excitation frequency and autonomous adaptation of feedback gains. A typical IEA-AVC system employs multiple complementary control gain settings featuring staggered resonance characteristics and switches among them as required. A general linear multi-degree-of-freedom dynamic model is established, and a model-based algorithm is developed for real-time identification of variable harmonic excitation frequencies using only acceleration measurements. Experiments confirm that the proposed algorithm rapidly (min. to 95 ms) and accurately tracks time-varying excitation frequencies, enabling real-time autonomous switching between preset control gain settings. Based on this validated architecture, theoretical analyses and simulations are conducted to evaluate system performance under different gain settings, and corresponding customized control schemes are consequently proposed. The results show that the resonance at each modal order is effectively avoided, leading to significantly enhanced vibration suppression.

主动振动控制（AVC）是一种经过验证的解决方案，用于在特定频率范围的稳定环境中抑制振动。然而，在可变和不可预测的激励条件下保持其高性能仍然是一个重大挑战。本文提出了一种智能激励自适应主动振动控制（IEA-AVC）方法，以实现变频激励下结构的最优控制。该方法将传统的AVC技术与激励频率的实时识别和反馈增益的自适应相结合。典型的IEA-AVC系统采用具有交错共振特性的多个互补控制增益设置，并根据需要在它们之间切换。建立了通用的线性多自由度动力学模型，并开发了一种基于模型的算法，用于仅利用加速度测量实时识别变谐波激励频率。实验验证了该算法的有效性。到95毫秒)，并准确地跟踪时变的激励频率，使实时自主切换之间的预设控制增益设置。在此基础上，进行了理论分析和仿真，评估了系统在不同增益设置下的性能，并提出了相应的定制控制方案。结果表明，有效地避免了各模态阶的共振，显著增强了对振动的抑制。

{"title":"Active vibration control enabled by intelligent excitation adaptability for frequency-variable excitations","authors":"Shuai Chen , Yilong Wang , Jian Zhang , Qianjing Wu , Xiaoyun Zhang , Dengqing Cao , Biao Wang","doi":"10.1016/j.ymssp.2025.113831","DOIUrl":"10.1016/j.ymssp.2025.113831","url":null,"abstract":"<div><div>Active vibration control (AVC) is a well-validated solution for vibration suppression in stable environments with specific frequency ranges. However, maintaining its high performance under variable and unpredictable excitation conditions remains a significant challenge. This article develops an intelligent excitation-adaptive active vibration control (IEA-AVC) method to achieve optimal control of structures under frequency-variable excitations. The method integrates conventional AVC techniques with real-time identification of excitation frequency and autonomous adaptation of feedback gains. A typical IEA-AVC system employs multiple complementary control gain settings featuring staggered resonance characteristics and switches among them as required. A general linear multi-degree-of-freedom dynamic model is established, and a model-based algorithm is developed for real-time identification of variable harmonic excitation frequencies using only acceleration measurements. Experiments confirm that the proposed algorithm rapidly (min. to 95 ms) and accurately tracks time-varying excitation frequencies, enabling real-time autonomous switching between preset control gain settings. Based on this validated architecture, theoretical analyses and simulations are conducted to evaluate system performance under different gain settings, and corresponding customized control schemes are consequently proposed. The results show that the resonance at each modal order is effectively avoided, leading to significantly enhanced vibration suppression.</div></div>","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"244 ","pages":"Article 113831"},"PeriodicalIF":8.9,"publicationDate":"2025-12-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145884726","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Fault-tolerant path planning for cable-driven manipulators based on a diffusion strategy 基于扩散策略的缆索驱动机器人容错路径规划

IF 8.9 1区工程技术 Q1 ENGINEERING, MECHANICAL

Mechanical Systems and Signal Processing

Pub Date : 2025-12-31 DOI: 10.1016/j.ymssp.2025.113840

Dongyang Hu , Dawei Xu , Hamid Reza Karimi , Qianming Wang , Yanfeng Wang , Huanlong Zhang , Quan Li , Yongjie Zhai

Cable-driven manipulators often perform tasks in radiation, toxic, high-temperature and high-pressure environments. In such conditions, cable wear or failure cannot be addressed on-site. To address operational safety issues arising from cable abrasion in these scenarios, this study proposes a physics-informed diffusion policy framework for fault-tolerant path planning of cable-driven manipulators. Given the hyper-redundant degrees of freedom in cable-driven manipulators, the system can redistribute loads and continue functioning despite cable damage. First, a virtual-work-based cable tension estimation network embedding physical constraints is developed to achieve high-accuracy tension prediction corresponding to joint postures. Subsequently, tension guidance is integrated into the denoising diffusion model, leveraging collision-free reference trajectories as priors, thus iteratively generating fault-tolerant trajectories in high-dimensional joint spaces that satisfy both low-tension and obstacle avoidance constraints. Multidimensional experiments evaluating tension prediction accuracy, path feasibility, and algorithm scalability demonstrate that the proposed method effectively reduces average tension in damaged cables while ensuring path safety, achieving a planning time of only 9.7 s. Furthermore, the method maintains stable convergence and low jitter even with up to three damaged cables. Experimental results validate the algorithm’s effectiveness in fault-tolerant path planning for cable-driven manipulators, presenting a promising new paradigm for continuous and reliable operations in extreme environments.

电缆驱动的机械手经常在辐射、有毒、高温和高压环境中执行任务。在这种情况下，电缆磨损或故障无法现场解决。为了解决这些情况下由电缆磨损引起的操作安全问题，本研究提出了一种物理信息扩散策略框架，用于电缆驱动机械手的容错路径规划。考虑到电缆驱动机械手的超冗余自由度，系统可以重新分配负载并在电缆损坏的情况下继续工作。首先，建立了嵌入物理约束的基于虚拟工作的缆索张力估计网络，实现了与关节姿态相对应的高精度张力预测；随后，将张力导引集成到去噪扩散模型中，利用无碰撞参考轨迹作为先验，在高维关节空间中迭代生成同时满足低张力和避障约束的容错轨迹。多维度实验评估了张力预测精度、路径可行性和算法可扩展性，结果表明，该方法在保证路径安全的同时有效降低了受损电缆的平均张力，规划时间仅为9.7 s。此外，该方法即使在多达三条电缆损坏的情况下也能保持稳定的收敛性和低抖动。实验结果验证了该算法在缆索驱动机械臂容错路径规划中的有效性，为极端环境下的连续可靠操作提供了一种有希望的新范式。

{"title":"Fault-tolerant path planning for cable-driven manipulators based on a diffusion strategy","authors":"Dongyang Hu , Dawei Xu , Hamid Reza Karimi , Qianming Wang , Yanfeng Wang , Huanlong Zhang , Quan Li , Yongjie Zhai","doi":"10.1016/j.ymssp.2025.113840","DOIUrl":"10.1016/j.ymssp.2025.113840","url":null,"abstract":"<div><div>Cable-driven manipulators often perform tasks in radiation, toxic, high-temperature and high-pressure environments. In such conditions, cable wear or failure cannot be addressed on-site. To address operational safety issues arising from cable abrasion in these scenarios, this study proposes a physics-informed diffusion policy framework for fault-tolerant path planning of cable-driven manipulators. Given the hyper-redundant degrees of freedom in cable-driven manipulators, the system can redistribute loads and continue functioning despite cable damage. First, a virtual-work-based cable tension estimation network embedding physical constraints is developed to achieve high-accuracy tension prediction corresponding to joint postures. Subsequently, tension guidance is integrated into the denoising diffusion model, leveraging collision-free reference trajectories as priors, thus iteratively generating fault-tolerant trajectories in high-dimensional joint spaces that satisfy both low-tension and obstacle avoidance constraints. Multidimensional experiments evaluating tension prediction accuracy, path feasibility, and algorithm scalability demonstrate that the proposed method effectively reduces average tension in damaged cables while ensuring path safety, achieving a planning time of only 9.7 s. Furthermore, the method maintains stable convergence and low jitter even with up to three damaged cables. Experimental results validate the algorithm’s effectiveness in fault-tolerant path planning for cable-driven manipulators, presenting a promising new paradigm for continuous and reliable operations in extreme environments.</div></div>","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"244 ","pages":"Article 113840"},"PeriodicalIF":8.9,"publicationDate":"2025-12-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145884733","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A projection basis adaptation scheme for the update of reduced order models using vibration measurements 基于振动测量的降阶模型更新的投影基自适应方案

IF 8.9 1区工程技术 Q1 ENGINEERING, MECHANICAL

Mechanical Systems and Signal Processing

Pub Date : 2025-12-31 DOI: 10.1016/j.ymssp.2025.113796

Jake Hollins , Tinkle Chugh , Evangelos Papatheou , Keith Worden , Konstantinos Tatsis , Konstantinos Agathos

A methodology is proposed for updating Reduced-Order Models (ROMs), constructed from a linear, high-fidelity Finite Element (FE) model, based on experimental vibration measurements. The ability of ROMs to represent experimentally measured responses is maximised by updating the Reduced-Order Bases (ROBs) employed in their construction. To render this update efficient, the updated ROBs are formed by adapting initial bases of higher dimensionality via a linear transformation, which is parametrised in a way that preserves desirable qualities of the original bases, such as mass orthogonality. It is also shown that only updating ROBs can limit the range of natural frequencies of the updated models, thus additional parameters, in the form of substructure-stiffness modifiers, are introduced in the updating process. The resulting methodology is demonstrated in two experimentally-tested structures, one where a model with good initial correlation is available, and one where the available model is very poorly correlated, but a large quantity of experimental measurements is available. In both cases, the proposed approach produces models with excellent correlation to the measurements, while in the second one it is also shown to accurately expand mode shapes in sensor locations that are not used for updating.

提出了一种基于实验振动测量的线性高保真有限元模型的降阶模型（ROMs）更新方法。rom表示实验测量响应的能力通过更新其构建中使用的降阶基（ROBs）来最大化。为了提高更新效率，更新后的罗布是通过线性变换来适应高维的初始基来形成的，该变换以一种保留原始基的理想品质的方式进行参数化，例如质量正交性。研究还表明，只有更新rob才能限制更新模型的固有频率范围，因此在更新过程中引入额外的参数，以子结构-刚度修正器的形式。所得到的方法在两个实验测试的结构中得到了证明，其中一个是具有良好初始相关性的模型，另一个是可用的模型相关性非常差，但有大量的实验测量可用。在这两种情况下，所提出的方法产生的模型与测量结果具有良好的相关性，而在第二种情况下，它也被证明可以准确地扩展不用于更新的传感器位置的模态振型。

{"title":"A projection basis adaptation scheme for the update of reduced order models using vibration measurements","authors":"Jake Hollins , Tinkle Chugh , Evangelos Papatheou , Keith Worden , Konstantinos Tatsis , Konstantinos Agathos","doi":"10.1016/j.ymssp.2025.113796","DOIUrl":"10.1016/j.ymssp.2025.113796","url":null,"abstract":"<div><div>A methodology is proposed for updating Reduced-Order Models (ROMs), constructed from a linear, high-fidelity Finite Element (FE) model, based on experimental vibration measurements. The ability of ROMs to represent experimentally measured responses is maximised by updating the Reduced-Order Bases (ROBs) employed in their construction. To render this update efficient, the updated ROBs are formed by adapting initial bases of higher dimensionality via a linear transformation, which is parametrised in a way that preserves desirable qualities of the original bases, such as mass orthogonality. It is also shown that only updating ROBs can limit the range of natural frequencies of the updated models, thus additional parameters, in the form of substructure-stiffness modifiers, are introduced in the updating process. The resulting methodology is demonstrated in two experimentally-tested structures, one where a model with good initial correlation is available, and one where the available model is very poorly correlated, but a large quantity of experimental measurements is available. In both cases, the proposed approach produces models with excellent correlation to the measurements, while in the second one it is also shown to accurately expand mode shapes in sensor locations that are not used for updating.</div></div>","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"244 ","pages":"Article 113796"},"PeriodicalIF":8.9,"publicationDate":"2025-12-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145884734","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Knowledge-driven compound fault detection enhancement based on multi-modal feature generative adversarial collaborative learning for industrial rotating machinery 基于多模态特征生成对抗协同学习的工业旋转机械知识驱动复合故障检测增强

IF 8.9 1区工程技术 Q1 ENGINEERING, MECHANICAL

Mechanical Systems and Signal Processing

Pub Date : 2025-12-30 DOI: 10.1016/j.ymssp.2025.113822

Yan Liu , Zuhua Xu , Fei Gao , Jun Zhao , Chunyue Song

Compound faults of industrial rotating machinery are characterized by temporal-frequency multisource signals coupling, existing methods struggle to cope with the intrinsic correlations and distribution gap between the temporal signal and the frequency signal, leading to suboptimal detection performance. To tackle it, we proposed a multi-modal feature collaborative generative adversarial learning method for industrial rotating machinery compound fault detection enhancement. In this method, the feature representations of inter-modal complementarity (modal-shared) and the intra-modal specificity (modal-specific) are collaboratively learnt. First, we design two parallel knowledge-informed modal-specific feature extraction network models and a multi-modal transfer model. Two modal-specific feature extraction models utilize the prior knowledge to learn modal-specific feature from the temporal and frequency modals. The multi-modal transfer model aggregates and transfer modal-shared feature by calculating modal-specific feature neighbors of each modal. Then, a generative adversarial-based multi-modal collaborative training strategy is developed for feature collaborative learning. In this strategy, three feature learning models acts as multi-modal feature generator, which intends to generate and fuse modal-specific and modal shared features to form the cross-modal propagated feature, and a feature modal discriminator is introduced to identify the source belonging modal of the cross-modal propagated feature. Based on such feature generative adversarial training, the cross-modal propagated feature no longer clearly belongs to one certain modal, thus achieving effective multi-modal feature transfer. Finally, two industrial rotating machineries cases demonstrate the compound fault performance of the proposed method.

工业旋转机械复合故障具有时频多源信号耦合的特点，现有方法难以处理时频信号之间的内在相关性和分布间隙，导致检测性能欠佳。针对这一问题，提出了一种多模态特征协同生成对抗学习方法，用于工业旋转机械复合故障检测增强。在该方法中，协同学习了模态间互补性（模态共享）和模态内特异性（模态特异性）的特征表示。首先，我们设计了两个并行的基于知识的特定模式特征提取网络模型和一个多模式转移模型。两种模态特征提取模型利用先验知识从时间模态和频率模态中学习模态特征。多模态迁移模型通过计算每个模态的模态特征邻域来聚合和迁移模态共享特征。然后，针对特征协同学习，提出了一种基于生成对抗的多模态协同训练策略。在该策略中，三个特征学习模型作为多模态特征生成器，旨在生成和融合模态特定特征和模态共享特征以形成跨模态传播特征，并引入特征模态鉴别器来识别跨模态传播特征的源所属模态。基于这种特征生成对抗训练，跨模态传播的特征不再明确地属于某一模态，从而实现了有效的多模态特征转移。最后，以两个工业旋转机械为例，验证了该方法的复合故障性能。

{"title":"Knowledge-driven compound fault detection enhancement based on multi-modal feature generative adversarial collaborative learning for industrial rotating machinery","authors":"Yan Liu , Zuhua Xu , Fei Gao , Jun Zhao , Chunyue Song","doi":"10.1016/j.ymssp.2025.113822","DOIUrl":"10.1016/j.ymssp.2025.113822","url":null,"abstract":"<div><div>Compound faults of industrial rotating machinery are characterized by temporal-frequency multisource signals coupling, existing methods struggle to cope with the intrinsic correlations and distribution gap between the temporal signal and the frequency signal, leading to suboptimal detection performance. To tackle it, we proposed a multi-modal feature collaborative generative adversarial learning method for industrial rotating machinery compound fault detection enhancement. In this method, the feature representations of inter-modal complementarity (modal-shared) and the intra-modal specificity (modal-specific) are collaboratively learnt. First, we design two parallel knowledge-informed modal-specific feature extraction network models and a multi-modal transfer model. Two modal-specific feature extraction models utilize the prior knowledge to learn modal-specific feature from the temporal and frequency modals. The multi-modal transfer model aggregates and transfer modal-shared feature by calculating modal-specific feature neighbors of each modal. Then, a generative adversarial-based multi-modal collaborative training strategy is developed for feature collaborative learning. In this strategy, three feature learning models acts as multi-modal feature generator, which intends to generate and fuse modal-specific and modal shared features to form the cross-modal propagated feature, and a feature modal discriminator is introduced to identify the source belonging modal of the cross-modal propagated feature. Based on such feature generative adversarial training, the cross-modal propagated feature no longer clearly belongs to one certain modal, thus achieving effective multi-modal feature transfer. Finally, two industrial rotating machineries cases demonstrate the compound fault performance of the proposed method.</div></div>","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"244 ","pages":"Article 113822"},"PeriodicalIF":8.9,"publicationDate":"2025-12-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145885231","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0