Pub Date : 2026-02-15DOI: 10.1016/j.ymssp.2026.113990
Jia-Yi Ding, Li Feng, Xu-Yang Cao, De-Cheng Feng, Michael Beer
Seismic events are characterized by abrupt onset, wide spatial impact, and substantial destructive potential, often leading to cascading socioeconomic consequences. At the regional scale, seismic assessment faces two persistent challenges: (i) building inventories typically lack complete, high-resolution structural information, and (ii) refined nonlinear models are computationally prohibitive for large portfolios. Against this background, this paper develops simplified multi-spring models for building portfolios in regional-scale seismic analysis. Specifically, the modeling framework consists of a lumped-shear multi-degree of freedom (MDOF) model for multi-story building, and a lumped flexural-shear-coupling MDOF model for high-rise buildings. Moreover, two types of parameters (i.e., coarse-scale and fine-scale) are compared in model generation, which are based on the building-level attributes and the component-level capacity characteristics, respectively. Both the parameter variability and seismic uncertainties (i.e., individual and combined parameter) are incorporated during the simulation to assess the demand variations. The results show that the simplified representation preserves essential seismic response characteristics while enabling high computational efficiency suitable for regional applications. To further enhance reliability, a lognormal-based probabilistic revision strategy is introduced to calibrate coarse-scale seismic demand statistics (median and dispersion) using fine-scale reference data. The resulting framework provides a practical and efficient solution for seismic assessments at a regional scale, particularly for diverse building portfolios.
{"title":"Quantification of model uncertainties in regional-scale seismic analysis of building portfolios","authors":"Jia-Yi Ding, Li Feng, Xu-Yang Cao, De-Cheng Feng, Michael Beer","doi":"10.1016/j.ymssp.2026.113990","DOIUrl":"https://doi.org/10.1016/j.ymssp.2026.113990","url":null,"abstract":"Seismic events are characterized by abrupt onset, wide spatial impact, and substantial destructive potential, often leading to cascading socioeconomic consequences. At the regional scale, seismic assessment faces two persistent challenges: (i) building inventories typically lack complete, high-resolution structural information, and (ii) refined nonlinear models are computationally prohibitive for large portfolios. Against this background, this paper develops simplified multi-spring models for building portfolios in regional-scale seismic analysis. Specifically, the modeling framework consists of a lumped-shear multi-degree of freedom (MDOF) model for multi-story building, and a lumped flexural-shear-coupling MDOF model for high-rise buildings. Moreover, two types of parameters (i.e., coarse-scale and fine-scale) are compared in model generation, which are based on the building-level attributes and the component-level capacity characteristics, respectively. Both the parameter variability and seismic uncertainties (i.e., individual and combined parameter) are incorporated during the simulation to assess the demand variations. The results show that the simplified representation preserves essential seismic response characteristics while enabling high computational efficiency suitable for regional applications. To further enhance reliability, a lognormal-based probabilistic revision strategy is introduced to calibrate coarse-scale seismic demand statistics (median and dispersion) using fine-scale reference data. The resulting framework provides a practical and efficient solution for seismic assessments at a regional scale, particularly for diverse building portfolios.","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"3 1","pages":""},"PeriodicalIF":8.4,"publicationDate":"2026-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146209038","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}
Pub Date : 2026-02-15DOI: 10.1016/j.ymssp.2026.113950
Tengjiao He , Jiancheng Liao , Kexi Liao , Huaixin Zhang , Xiaolong Shi , Feilong Zhou , Linxiang Wang , Guoqiang Xia , Yutong Jiang , Jing Tang
{"title":"Corrigendum to “Quantitative study on far-field magnetic signal response of steel pipe girth welds with weak magnetic excitation”. [Mech. Syst. Signal Process. 240 (2025) 113404]","authors":"Tengjiao He , Jiancheng Liao , Kexi Liao , Huaixin Zhang , Xiaolong Shi , Feilong Zhou , Linxiang Wang , Guoqiang Xia , Yutong Jiang , Jing Tang","doi":"10.1016/j.ymssp.2026.113950","DOIUrl":"10.1016/j.ymssp.2026.113950","url":null,"abstract":"","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"246 ","pages":"Article 113950"},"PeriodicalIF":8.9,"publicationDate":"2026-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146072597","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}
Ultra-low-frequency seismic vibration remains a major performance bottleneck for precision engineering platforms, such as metrology instruments, semiconductor equipment, and large-scale scientific facilities. To address this challenge, this work presents an enhanced H∞ loop-shaping framework integrated with a virtual sensor fusion architecture to meet these requirements in active seismic vibration isolation systems. The proposed formulation employs carefully constructed weighting functions to shape the loop for stringent low-frequency disturbance rejection while maintaining adequate robustness margins. To further mitigate the influence of unmodelled high-frequency dynamics—factors known to degrade H∞ controllers—a frequency-partitioned sensing scheme is introduced. Complementary filters allocate low-frequency feedback to physical sensors, ensuring accurate regulation, and high-frequency feedback to virtual sensor channels, effectively conditioning the control loop against adverse high-frequency dynamics. Experimental validation on a three-degree-of-freedom isolation platform demonstrates that the resulting controller achieves up to 65 dB reduction in transmitted motion within the target 0.1–10 Hz band. These results confirm that the proposed synthesis approach not only enhances practical seismic isolation performance but also extends the theoretical applicability of H∞ loop-shaping methods in systems where sensing constraints and broadband uncertainties are coupled.
{"title":"Enhanced H∞ loop-shaping control with virtual sensor fusion for active seismic vibration isolation","authors":"Xiaoqi Yin, Xin Lin, Wenwu Feng, Shuyun Yang, Ziliang Zhang, Junxiang Lian, Guoying Zhao","doi":"10.1016/j.ymssp.2026.114016","DOIUrl":"https://doi.org/10.1016/j.ymssp.2026.114016","url":null,"abstract":"Ultra-low-frequency seismic vibration remains a major performance bottleneck for precision engineering platforms, such as metrology instruments, semiconductor equipment, and large-scale scientific facilities. To address this challenge, this work presents an enhanced <ce:italic>H<ce:inf loc=\"post\">∞</ce:inf></ce:italic> loop-shaping framework integrated with a virtual sensor fusion architecture to meet these requirements in active seismic vibration isolation systems. The proposed formulation employs carefully constructed weighting functions to shape the loop for stringent low-frequency disturbance rejection while maintaining adequate robustness margins. To further mitigate the influence of unmodelled high-frequency dynamics—factors known to degrade <ce:italic>H<ce:inf loc=\"post\">∞</ce:inf></ce:italic> controllers—a frequency-partitioned sensing scheme is introduced. Complementary filters allocate low-frequency feedback to physical sensors, ensuring accurate regulation, and high-frequency feedback to virtual sensor channels, effectively conditioning the control loop against adverse high-frequency dynamics. Experimental validation on a three-degree-of-freedom isolation platform demonstrates that the resulting controller achieves up to 65 dB reduction in transmitted motion within the target 0.1–10 Hz band. These results confirm that the proposed synthesis approach not only enhances practical seismic isolation performance but also extends the theoretical applicability of <ce:italic>H<ce:inf loc=\"post\">∞</ce:inf></ce:italic> loop-shaping methods in systems where sensing constraints and broadband uncertainties are coupled.","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"244 1","pages":""},"PeriodicalIF":8.4,"publicationDate":"2026-02-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146209039","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}
Pub Date : 2026-02-14DOI: 10.1016/j.ymssp.2026.113994
Li Wen, Yong Ruan, Hu Yang, Tao Tang
In astronomical observation, line-of-sight (LOS) accuracy is constrained by unknown disturbances and sensor-induced delays. This paper proposes an adaptive modular framework that incorporates frequency decomposition and estimation into parallel control channels, enabling structurally independent suppression of multiple narrow-band disturbances under sensor delays. Firstly, a cascaded and order-reduced parallel filtering framework is proposed to decouple multi-frequency disturbances into single-frequency components, thereby simplifying estimation and reducing computation. To address inherent image sensor delays, a frequency-based fractional time-delay compensation strategy is incorporated to ensure phase alignment with the delayed feedback, particularly in the high-frequency range. Finally, a small-gain stability analysis is performed to guarantee robust closed-loop performance under the proposed adaptive scheme. Experimental validation on an image-based piezoelectric tip-tilt stage demonstrates that the proposed method enhances multi-disturbance rejection up to the Nyquist frequency.
{"title":"Frequency estimation and parallel control for multi-disturbance rejection: Application to image-stabilized piezoelectric tip-tilt stages","authors":"Li Wen, Yong Ruan, Hu Yang, Tao Tang","doi":"10.1016/j.ymssp.2026.113994","DOIUrl":"https://doi.org/10.1016/j.ymssp.2026.113994","url":null,"abstract":"In astronomical observation, line-of-sight (LOS) accuracy is constrained by unknown disturbances and sensor-induced delays. This paper proposes an adaptive modular framework that incorporates frequency decomposition and estimation into parallel control channels, enabling structurally independent suppression of multiple narrow-band disturbances under sensor delays. Firstly, a cascaded and order-reduced parallel filtering framework is proposed to decouple multi-frequency disturbances into single-frequency components, thereby simplifying estimation and reducing computation. To address inherent image sensor delays, a frequency-based fractional time-delay compensation strategy is incorporated to ensure phase alignment with the delayed feedback, particularly in the high-frequency range. Finally, a small-gain stability analysis is performed to guarantee robust closed-loop performance under the proposed adaptive scheme. Experimental validation on an image-based piezoelectric tip-tilt stage demonstrates that the proposed method enhances multi-disturbance rejection up to the Nyquist frequency.","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"113 1","pages":""},"PeriodicalIF":8.4,"publicationDate":"2026-02-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146209042","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}
Pub Date : 2026-02-14DOI: 10.1016/j.ymssp.2026.113999
Jiaji Chen, Jie Hu, Pei Zhang, Wencai Xu, Yuxuan Tang, Donghao Yang
The autonomous driving system relies on time-varying states such as longitudinal velocity, lateral velocity, yaw rate, and centroid sideslip angle to achieve safe and stable control. However, the high cost of high-precision sensors hinders their widespread adoption in mass-produced vehicles, making state estimation techniques critical. To improve estimation accuracy and robustness, this paper proposes a hybrid LSTM-based cubature Kalman filter (HLCKF), which integrates a Long Short-Term Memory (LSTM) network into the Cubature Kalman Filter (CKF) framework. Firstly, a nonlinear system model is constructed to provide physical support for the estimation. Then, a Multi-output Head and Peeking-Coupled LSTM (MHPC-LSTM) is designed and embedded into the CKF to reduce the impact of model uncertainty and sensor noise on CKF prediction. Lastly, the predictive performance of MHPC-LSTM and the state estimation of HLCKF are validated by constructing a simulation platform. Experimental results show that the proposed MHPC-LSTM outperforms traditional model-based methods and standard LSTM networks in terms of both prediction accuracy and robustness, making it suitable for the CKF prediction phase. Further analysis indicates that the HLCKF surpasses both standalone CKF and LSTM methods in estimation accuracy and output stability, while maintaining an inference time that satisfies the real-time requirements of autonomous driving systems.
{"title":"A synchronized estimation method for time-varying vehicle states via hybrid of cubature Kalman filter and long short-term memory network","authors":"Jiaji Chen, Jie Hu, Pei Zhang, Wencai Xu, Yuxuan Tang, Donghao Yang","doi":"10.1016/j.ymssp.2026.113999","DOIUrl":"https://doi.org/10.1016/j.ymssp.2026.113999","url":null,"abstract":"The autonomous driving system relies on time-varying states such as longitudinal velocity, lateral velocity, yaw rate, and centroid sideslip angle to achieve safe and stable control. However, the high cost of high-precision sensors hinders their widespread adoption in mass-produced vehicles, making state estimation techniques critical. To improve estimation accuracy and robustness, this paper proposes a hybrid LSTM-based cubature Kalman filter (HLCKF), which integrates a Long Short-Term Memory (LSTM) network into the Cubature Kalman Filter (CKF) framework. Firstly, a nonlinear system model is constructed to provide physical support for the estimation. Then, a Multi-output Head and Peeking-Coupled LSTM (MHPC-LSTM) is designed and embedded into the CKF to reduce the impact of model uncertainty and sensor noise on CKF prediction. Lastly, the predictive performance of MHPC-LSTM and the state estimation of HLCKF are validated by constructing a simulation platform. Experimental results show that the proposed MHPC-LSTM outperforms traditional model-based methods and standard LSTM networks in terms of both prediction accuracy and robustness, making it suitable for the CKF prediction phase. Further analysis indicates that the HLCKF surpasses both standalone CKF and LSTM methods in estimation accuracy and output stability, while maintaining an inference time that satisfies the real-time requirements of autonomous driving systems.","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"328 1","pages":""},"PeriodicalIF":8.4,"publicationDate":"2026-02-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146209040","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}
Pub Date : 2026-02-14DOI: 10.1016/j.ymssp.2026.113995
Kai Yang, En-Guo Liu, Yu-Nan Zhu, Xiao-Ye Mao, Xiang-Ying Guo, Hu Ding, Li-Qun Chen
Printed circuit boards (PCBs) operating in harsh vibrational environments are prone to cracking and mission-critical failures. This study proposes a reconfigurable, partition-enabled particle damper (PD) whose cavity can be rapidly switched between single- and multi-unit configurations without redesign. Mode‑resolved sine-sweep tests on a PCB, combined with discrete element method (DEM) simulations, quantify vibration attenuation at the first two bending modes and clarify the underlying damping mechanisms. Energy‑budget analysis shows that inelastic normal collisions and tangential friction dominate dissipation, whereas rotational effects are negligible; their relative contributions vary strongly with filling ratio. DEM further reveals distinct mode-dependent particle‑motion regimes: a collect-and-collide state at the first resonance, and a gas-like state at the second, consistent with the measured frequency-dependent performance. Parametric studies demonstrate that damping efficiency is highly sensitive to particle material, size, and filling ratio, while cavity geometry in the present design space plays a secondary role. An empty‑cavity control confirms that the observed vibration reduction arises primarily from granular dissipation rather than added mass. Compared with existing PCB vibration studies that use fixed particle dampers or address single-mode response, this work (i) introduces a reconfigurable, partition-enabled PD that can be rapidly switched between single- and multi-unit layouts without redesign and (ii) systematically links DEM-resolved energy dissipation mechanisms to mode-resolved PCB experiments. The resulting framework provides physics-based, practically oriented design guidelines for implementing granular damping in high-reliability electronic systems.
{"title":"Mode-resolved, reconfigurable particle damping for printed circuit boards vibration control","authors":"Kai Yang, En-Guo Liu, Yu-Nan Zhu, Xiao-Ye Mao, Xiang-Ying Guo, Hu Ding, Li-Qun Chen","doi":"10.1016/j.ymssp.2026.113995","DOIUrl":"https://doi.org/10.1016/j.ymssp.2026.113995","url":null,"abstract":"Printed circuit boards (PCBs) operating in harsh vibrational environments are prone to cracking and mission-critical failures. This study proposes a reconfigurable, partition-enabled particle damper (PD) whose cavity can be rapidly switched between single- and multi-unit configurations without redesign. Mode‑resolved sine-sweep tests on a PCB, combined with discrete element method (DEM) simulations, quantify vibration attenuation at the first two bending modes and clarify the underlying damping mechanisms. Energy‑budget analysis shows that inelastic normal collisions and tangential friction dominate dissipation, whereas rotational effects are negligible; their relative contributions vary strongly with filling ratio. DEM further reveals distinct mode-dependent particle‑motion regimes: a collect-and-collide state at the first resonance, and a gas-like state at the second, consistent with the measured frequency-dependent performance. Parametric studies demonstrate that damping efficiency is highly sensitive to particle material, size, and filling ratio, while cavity geometry in the present design space plays a secondary role. An empty‑cavity control confirms that the observed vibration reduction arises primarily from granular dissipation rather than added mass. Compared with existing PCB vibration studies that use fixed particle dampers or address single-mode response, this work (i) introduces a reconfigurable, partition-enabled PD that can be rapidly switched between single- and multi-unit layouts without redesign and (ii) systematically links DEM-resolved energy dissipation mechanisms to mode-resolved PCB experiments. The resulting framework provides physics-based, practically oriented design guidelines for implementing granular damping in high-reliability electronic systems.","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"48 1","pages":""},"PeriodicalIF":8.4,"publicationDate":"2026-02-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146209041","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}
Pub Date : 2026-02-14DOI: 10.1016/j.ymssp.2026.113983
Qingbin Tong, Jilong Zhao, Xuedong Jiang, Baohua Wang, Feiyu Lu, Shouxin Du, Xin Du, Jianjun Xu, Jingyi Huo
In recent years, single-source generalization (SDG), which involves training with only one source domain and generalizing to multiple target domains, has garnered widespread attention. However, the current research focus is primarily concentrated on the generation of pseudo-domains. Methods used for generating pseudo-domains inevitably introduce unnecessary noise or produce unreliable features. Therefore, this paper proposes a single-domain generalization framework integrating Zernike-based detail-blurred domain-invariant feature extraction and the Mamba global attention mechanism, focusing on extracting domain-invariant features from a single source domain. Firstly, Zernike moments are utilized to conduct further feature extraction on the features transformed by Short-Time Fourier Transform (STFT). In combination with the selective mixup method, the high-order Zernike moments that depict noise and other details are discarded, while the integrative expression of the middle and low-order moment features is enhanced. Consequently, preliminary detail-blurred domain-invariant features with summarizing properties are obtained. To enhance the expression of single-domain features while maintaining their domain invariance, a novel Global-Local Feature Fusion Model (GLFM) is constructed. The Mamba Attention-Guided Global Feature Extraction module (MGFE) aims to extract global features, while the Convolutional Local Feature Extraction module assisted by spatial attention (CLFE) aims to extract local features. These two types of features are fused and selected through the Fusion Selection Module (FSM) to enhance useful features and suppress less useful ones. Experiments were conducted on 12 single-domain generalization tasks across the CWRU and BJTU datasets. The proposed method demonstrated excellent diagnostic performance, with average accuracies of 98.48% and 95.38%, respectively.
{"title":"A single-domain generalization framework integrating Zernike-Based detail blur feature extraction and Mamba global attention mechanism","authors":"Qingbin Tong, Jilong Zhao, Xuedong Jiang, Baohua Wang, Feiyu Lu, Shouxin Du, Xin Du, Jianjun Xu, Jingyi Huo","doi":"10.1016/j.ymssp.2026.113983","DOIUrl":"https://doi.org/10.1016/j.ymssp.2026.113983","url":null,"abstract":"In recent years, single-source generalization (SDG), which involves training with only one source domain and generalizing to multiple target domains, has garnered widespread attention. However, the current research focus is primarily concentrated on the generation of pseudo-domains. Methods used for generating pseudo-domains inevitably introduce unnecessary noise or produce unreliable features. Therefore, this paper proposes a single-domain generalization framework integrating Zernike-based detail-blurred domain-invariant feature extraction and the Mamba global attention mechanism, focusing on extracting domain-invariant features from a single source domain. Firstly, Zernike moments are utilized to conduct further feature extraction on the features transformed by Short-Time Fourier Transform (STFT). In combination with the selective mixup method, the high-order Zernike moments that depict noise and other details are discarded, while the integrative expression of the middle and low-order moment features is enhanced. Consequently, preliminary detail-blurred domain-invariant features with summarizing properties are obtained. To enhance the expression of single-domain features while maintaining their domain invariance, a novel Global-Local Feature Fusion Model (GLFM) is constructed. The Mamba Attention-Guided Global Feature Extraction module (MGFE) aims to extract global features, while the Convolutional Local Feature Extraction module assisted by spatial attention (CLFE) aims to extract local features. These two types of features are fused and selected through the Fusion Selection Module (FSM) to enhance useful features and suppress less useful ones. Experiments were conducted on 12 single-domain generalization tasks across the CWRU and BJTU datasets. The proposed method demonstrated excellent diagnostic performance, with average accuracies of 98.48% and 95.38%, respectively.","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"59 1","pages":""},"PeriodicalIF":8.4,"publicationDate":"2026-02-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146209043","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}
Terahertz (THz) technology, a promising alternative to traditional non-destructive testing (NDT) methods, emerges great potential for quantitative characterization of debonding defect in Glass Fiber Reinforced Polymer (GFRP) composites. However, during the intelligent identification of debonding defects inside composite, there are challenges with class imbalance in THz datasets leading to reduced accuracy, and traditional deep learning models struggle to balance classification precision with inference speed. Therefore, a lightweight THz three-dimensional characterization system based on Ghost Bottleneck Attention Module Network (GBAMNet) for THz class-imbalanced dataset is proposed to ultimately achieve the automatic characterization of hidden debonding defect inside GFRP. Ghost layer convolution and Bottleneck Attention Module (BAM) modules are specially designed to capture essential features from the raw THz signals with a low computational cost, and thus realizing real-time, high-accuracy categorization of THz signals. The ghost feature factor s is introduced to control the number of ghost feature mappings generated by linear transformations within the Ghost layer, increasing the quantity of feature mappings in the network to enhance representational capabilities with minimal computational and parameter overhead. Moreover, the Gradient Harmonizing Mechanism (GHM) Loss function addresses the class-imbalance issue by dynamically adjusting the gradient weights of each sample to achieve a balanced gradient distribution. The comprehensive performance of GBAMNet with respect to classification accuracy and inference velocity are validated by a series of experiments on the class-imbalanced dataset. Overall, our proposed system will promote the actual applications of Terahertz non-destructive testing (THz NDT) scenarios for automatic and real-time defect characterization.
{"title":"Terahertz characterization of internal debonding defects of composites under class imbalance condition using lightweight network","authors":"Xingyu Wang, Yafei Xu, Yuqing Cui, Wenkang Li, Rong Wang, Liuyang Zhang, Ruqiang Yan, Xuefeng Chen","doi":"10.1016/j.ymssp.2026.114018","DOIUrl":"https://doi.org/10.1016/j.ymssp.2026.114018","url":null,"abstract":"Terahertz (THz) technology, a promising alternative to traditional non-destructive testing (NDT) methods, emerges great potential for quantitative characterization of debonding defect in Glass Fiber Reinforced Polymer (GFRP) composites. However, during the intelligent identification of debonding defects inside composite, there are challenges with class imbalance in THz datasets leading to reduced accuracy, and traditional deep learning models struggle to balance classification precision with inference speed. Therefore, a lightweight THz three-dimensional characterization system based on Ghost Bottleneck Attention Module Network (GBAMNet) for THz class-imbalanced dataset is proposed to ultimately achieve the automatic characterization of hidden debonding defect inside GFRP. Ghost layer convolution and Bottleneck Attention Module (BAM) modules are specially designed to capture essential features from the raw THz signals with a low computational cost, and thus realizing real-time, high-accuracy categorization of THz signals. The ghost feature factor s is introduced to control the number of ghost feature mappings generated by linear transformations within the Ghost layer, increasing the quantity of feature mappings in the network to enhance representational capabilities with minimal computational and parameter overhead. Moreover, the Gradient Harmonizing Mechanism (GHM) Loss function addresses the class-imbalance issue by dynamically adjusting the gradient weights of each sample to achieve a balanced gradient distribution. The comprehensive performance of GBAMNet with respect to classification accuracy and inference velocity are validated by a series of experiments on the class-imbalanced dataset. Overall, our proposed system will promote the actual applications of Terahertz non-destructive testing (THz NDT) scenarios for automatic and real-time defect characterization.","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"PAS-103 1","pages":""},"PeriodicalIF":8.4,"publicationDate":"2026-02-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146208636","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}
Pub Date : 2026-02-13DOI: 10.1016/j.ymssp.2026.113985
Niki Tsivouraki, Spilios D. Fassois, Konstantinos Tserpes
The problem of purely random vibration based monitoring of progressive fatigue damage for a population of nominally identical thermoplastic coupons is considered in the absence of material/structural, loading, damage accumulation information or other assumptions. The study aims at: (a) Exploring the basis for effective monitoring by modeling the evolution of the population structural dynamics under progressive damage and examining their monotonicity with Fatigue Cycles, and (b) postulating a novel and holistic, purely random vibration-based, Functional Multiple Model (F-MM) framework for damage monitoring encompassing damage detection, Fatigue Cycles characterization, and, for the first time, damage level estimation. The postulated framework is based on stochastic Functionally Pooled Multiple Model data-based representations of the dynamics and is generally applicable to any type of coupons. The Functionally Pooled aspect allows for explicitly modeling the dynamics under any fatigue loading cycles, while the Multiple Model aspect allows for accounting for significant uncertainty.
{"title":"Progressive fatigue damage monitoring for a population of thermoplastic coupons via a holistic functional multiple model random vibration data-based framework","authors":"Niki Tsivouraki, Spilios D. Fassois, Konstantinos Tserpes","doi":"10.1016/j.ymssp.2026.113985","DOIUrl":"https://doi.org/10.1016/j.ymssp.2026.113985","url":null,"abstract":"The problem of purely random vibration based monitoring of progressive fatigue damage for a population of nominally identical thermoplastic coupons is considered in the absence of material/structural, loading, damage accumulation information or other assumptions. The study aims at: (a) Exploring the basis for effective monitoring by modeling the evolution of the population structural dynamics under progressive damage and examining their monotonicity with Fatigue Cycles, and (b) postulating a novel and holistic, purely random vibration-based, Functional Multiple Model (F-MM) framework for damage monitoring encompassing damage detection, Fatigue Cycles characterization, and, for the first time, damage level estimation. The postulated framework is based on stochastic Functionally Pooled Multiple Model data-based representations of the dynamics and is generally applicable to any type of coupons. The Functionally Pooled aspect allows for explicitly modeling the dynamics under any fatigue loading cycles, while the Multiple Model aspect allows for accounting for significant uncertainty.","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"24 1","pages":""},"PeriodicalIF":8.4,"publicationDate":"2026-02-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146208638","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}
Pub Date : 2026-02-13DOI: 10.1016/j.ymssp.2026.113989
Xiaocen Wang, Zhongwen Xu, Die Su, Xueyan Ma, Xiumin Jiang, Yang An, Zhigang Qu
Hydrate blockage and pipeline leak are two urgent problems that must be addressed to ensure the secure and reliable transportation of natural gas. In this paper, a 16-bit orthogonal complementary Golay (A, B) code signal with a sinusoidal carrier is applied and discussed with a view to improving the signal-to-noise ratio (SNR) of location monitoring for these two abnormal events in natural gas pipelines. Theoretically, the measurement principle and feasibility of orthogonal complementary Golay (A, B) code excitation signal are analyzed and obtained by simulation. The corresponding experiments based on these theories are then carried out. Experimental results demonstrate that the excitation signal has a significant effect on improving the SNR of the matched filtered (MF) signal. When the standard deviation of Gaussian white noise is increased to 1, the absolute positioning error (APE) and SNR of hydrate blockage are 0.11 m and 11.09 dB, respectively, and the APE and SNR of pipeline leak are 0.106 m and 14.83 dB. This method also has strong ability to resist color noise and flow-induced noise. Furthermore, sidelobe suppression leads to enhanced spatial resolution. Consequently, the proposed method has considerable application prospects in natural gas pipeline safety monitoring under the condition of high-noise level environment.
{"title":"Natural gas pipeline safety monitoring technology: Golay code excitation improves anti-noise performance","authors":"Xiaocen Wang, Zhongwen Xu, Die Su, Xueyan Ma, Xiumin Jiang, Yang An, Zhigang Qu","doi":"10.1016/j.ymssp.2026.113989","DOIUrl":"https://doi.org/10.1016/j.ymssp.2026.113989","url":null,"abstract":"Hydrate blockage and pipeline leak are two urgent problems that must be addressed to ensure the secure and reliable transportation of natural gas. In this paper, a 16-bit orthogonal complementary Golay (A, B) code signal with a sinusoidal carrier is applied and discussed with a view to improving the signal-to-noise ratio (SNR) of location monitoring for these two abnormal events in natural gas pipelines. Theoretically, the measurement principle and feasibility of orthogonal complementary Golay (A, B) code excitation signal are analyzed and obtained by simulation. The corresponding experiments based on these theories are then carried out. Experimental results demonstrate that the excitation signal has a significant effect on improving the SNR of the matched filtered (MF) signal. When the standard deviation of Gaussian white noise is increased to 1, the absolute positioning error (APE) and SNR of hydrate blockage are 0.11 m and 11.09 dB, respectively, and the APE and SNR of pipeline leak are 0.106 m and 14.83 dB. This method also has strong ability to resist color noise and flow-induced noise. Furthermore, sidelobe suppression leads to enhanced spatial resolution. Consequently, the proposed method has considerable application prospects in natural gas pipeline safety monitoring under the condition of high-noise level environment.","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"37 1","pages":""},"PeriodicalIF":8.4,"publicationDate":"2026-02-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146208646","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}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：