Pub Date : 2026-02-15Epub Date: 2026-01-19DOI: 10.1016/j.ymssp.2026.113892
Fukang Xin , Lei Liu , Pan Wang , Huanhuan Hu , Huailiang Wang
In practical engineering, numerous dynamic uncertainties exist, and the input data or information is often inconsistent. It is important to consider the time-dependent reliability analysis under hybrid uncertainty. This work considers three types of uncertainty representation models to estimate the bounds of time-dependent failure probability. The conventional double-loop process results in an excessive number of simulator calls, which is often impractical in real-world applications. To overcome this challenge, a novel method, termed ‘Uncertainty Reduction guided Bayesian Optimization combined with Subset Simulation’ (URBO-SS), is proposed. It integrates both a double-loop strategy and a decoupling strategy to achieve Bayesian active learning by the proposed uncertainty reduction learning function and error-based stopping criterion. In addition, subset simulation is incorporated to reduce the size of the candidate sample pool. The decoupling strategy builds upon the double-loop strategy and adopts a sequential, collaborative updating way, thereby achieving high accuracy with significantly fewer simulator calls. Finally, the efficiency and accuracy of the URBO-SS method are demonstrated with test examples and two engineering examples.
{"title":"Uncertainty reduction guided Bayesian active learning method for hybrid time-dependent reliability analysis under three representation models","authors":"Fukang Xin , Lei Liu , Pan Wang , Huanhuan Hu , Huailiang Wang","doi":"10.1016/j.ymssp.2026.113892","DOIUrl":"10.1016/j.ymssp.2026.113892","url":null,"abstract":"<div><div>In practical engineering, numerous dynamic uncertainties exist, and the input data or information is often inconsistent. It is important to consider the time-dependent reliability analysis under hybrid uncertainty. This work considers three types of uncertainty representation models to estimate the bounds of time-dependent failure probability. The conventional double-loop process results in an excessive number of simulator calls, which is often impractical in real-world applications. To overcome this challenge, a novel method, termed ‘Uncertainty Reduction guided Bayesian Optimization combined with Subset Simulation’ (URBO-SS), is proposed. It integrates both a double-loop strategy and a decoupling strategy to achieve Bayesian active learning by the proposed uncertainty reduction learning function and error-based stopping criterion. In addition, subset simulation is incorporated to reduce the size of the candidate sample pool. The decoupling strategy builds upon the double-loop strategy and adopts a sequential, collaborative updating way, thereby achieving high accuracy with significantly fewer simulator calls. Finally, the efficiency and accuracy of the URBO-SS method are demonstrated with test examples and two engineering examples.</div></div>","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"246 ","pages":"Article 113892"},"PeriodicalIF":8.9,"publicationDate":"2026-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146001049","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 : 2026-02-15Epub Date: 2026-01-20DOI: 10.1016/j.ymssp.2026.113898
Lerui Chen , Zhendong Kang , Haiquan Wang , YukMing Tang , Yidan Ma , Shengjun Wen , Mohammed Woyeso Geda
Motors are pivotal in modern industry, especially as global demand for automation and smart manufacturing surges. Accurate fault diagnosis is crucial for stability maintenance, but existing approaches lack satisfactory accuracy and efficiency. This study integrates the multi-scale Convolution Neural Network (MSCNN),Bidirectional Mogrifier-Gated Recurrent Unit (BiMGRU), and Multi-head Attention Mechanism (MHAM) to propose a multimodal-based hybrid model of MSCNN-BiMGRU + MHAM for asynchronous motor fault diagnosis. The MSCNN channel is responsible for spatial feature extraction, and the BiMGRU channel is responsible for temporal feature extraction. While MHAM is responsible for efficient integration and extraction of multimodal features. Furthermore, to refine the model’s performance, an enhanced whale optimization algorithm (EWOA) is innovatively presented and embedded during model training, systematically optimizing hyperparameters to boost model generalization and training effectiveness. Numerous validations are conducted by the real vibration datasets of asynchronous motor gathered under noisy and various operating conditions. Compared to traditional approaches and the current mainstream deep learning models, the proposed hybrid model with EWOA optimization attains the impressive prediction accuracy. It delivers an effective and efficient approach to tackle the issues of motor fault diagnosis.
{"title":"Multimodal-based model for asynchronous motor fault diagnosis under noisy and variable operating conditions: a novel hybrid approach","authors":"Lerui Chen , Zhendong Kang , Haiquan Wang , YukMing Tang , Yidan Ma , Shengjun Wen , Mohammed Woyeso Geda","doi":"10.1016/j.ymssp.2026.113898","DOIUrl":"10.1016/j.ymssp.2026.113898","url":null,"abstract":"<div><div>Motors are pivotal in modern industry, especially as global demand for automation and smart manufacturing surges. Accurate fault diagnosis is crucial for stability maintenance, but existing approaches lack satisfactory accuracy and efficiency. This study integrates the multi-scale Convolution Neural Network (MSCNN),Bidirectional Mogrifier-Gated Recurrent Unit (BiMGRU), and Multi-head Attention Mechanism (MHAM) to propose a multimodal-based hybrid model of MSCNN-BiMGRU + MHAM for asynchronous motor fault diagnosis. The MSCNN channel is responsible for spatial feature extraction, and the BiMGRU channel is responsible for temporal feature extraction. While MHAM is responsible for efficient integration and extraction of multimodal features. Furthermore, to refine the model’s performance, an enhanced whale optimization algorithm (EWOA) is innovatively presented and embedded during model training, systematically optimizing hyperparameters to boost model generalization and training effectiveness. Numerous validations are conducted by the real vibration datasets of asynchronous motor gathered under noisy and various operating conditions. Compared to traditional approaches and the current mainstream deep learning models, the proposed hybrid model with EWOA optimization attains the impressive prediction accuracy. It delivers an effective and efficient approach to tackle the issues of motor fault diagnosis.</div></div>","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"246 ","pages":"Article 113898"},"PeriodicalIF":8.9,"publicationDate":"2026-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146014519","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 : 2026-02-15Epub Date: 2026-01-27DOI: 10.1016/j.ymssp.2026.113893
Jiasen Lin , Rongrong Hou , Yuequan Bao
Structural damage identification inevitably involves uncertainties, necessitating their explicit consideration to enhance the reliability and precision of detection frameworks. As a prominent sparse recovery technique, sparse Bayesian learning (SBL) has demonstrated effectiveness in damage identification by leveraging structural sparsity through automatic relevance determination (ARD) priors. However, conventional SBL implementations adopt an oversimplified probabilistic model that assumes mutual independence among damage parameters, thereby failing to account for inherent spatial correlations between adjacent structural elements. This study proposes a novel pattern-coupled SBL methodology that incorporates coupled Gaussian priors to simultaneously characterize and autonomously learn both sparsity patterns and parameter correlations. This dual-learning mechanism enables enhanced precision in quantifying damage severity through correlation-aware parameter estimation, and improved robustness against measurement noise and modeling errors. Furthermore, the proposed framework extends conventional sparse recovery capabilities by effectively resolving both distributed and block-sparse damage configurations—a crucial feature where traditional SBL approaches exhibit suboptimal performance. Numerical studies on a cable-stayed bridge model and experimental investigations of a space frame validate the method’s effectiveness in accurately identifying and quantifying single and multiple damage scenarios. Compared with the SBL method, the identification accuracy and robustness of the proposed method are significantly improved, especially for structural damage with block-sparse characteristics.
{"title":"Structural damage identification based on pattern-coupled sparse Bayesian learning","authors":"Jiasen Lin , Rongrong Hou , Yuequan Bao","doi":"10.1016/j.ymssp.2026.113893","DOIUrl":"10.1016/j.ymssp.2026.113893","url":null,"abstract":"<div><div>Structural damage identification inevitably involves uncertainties, necessitating their explicit consideration to enhance the reliability and precision of detection frameworks. As a prominent sparse recovery technique, sparse Bayesian learning (SBL) has demonstrated effectiveness in damage identification by leveraging structural sparsity through automatic relevance determination (ARD) priors. However, conventional SBL implementations adopt an oversimplified probabilistic model that assumes mutual independence among damage parameters, thereby failing to account for inherent spatial correlations between adjacent structural elements. This study proposes a novel pattern-coupled SBL methodology that incorporates coupled Gaussian priors to simultaneously characterize and autonomously learn both sparsity patterns and parameter correlations. This dual-learning mechanism enables enhanced precision in quantifying damage severity through correlation-aware parameter estimation, and improved robustness against measurement noise and modeling errors. Furthermore, the proposed framework extends conventional sparse recovery capabilities by effectively resolving both distributed and block-sparse damage configurations—a crucial feature where traditional SBL approaches exhibit suboptimal performance. Numerical studies on a cable-stayed bridge model and experimental investigations of a space frame validate the method’s effectiveness in accurately identifying and quantifying single and multiple damage scenarios. Compared with the SBL method, the identification accuracy and robustness of the proposed method are significantly improved, especially for structural damage with block-sparse characteristics.</div></div>","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"246 ","pages":"Article 113893"},"PeriodicalIF":8.9,"publicationDate":"2026-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146072620","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 : 2026-02-15Epub Date: 2026-01-17DOI: 10.1016/j.ymssp.2026.113891
Tairan Wang, Sifeng Bi
This work aims to explore the integration of novel and powerful deep learning techniques and intractable engineering problems, especially by adopting deep generative models to tackle model updating problems under uncertainty. A conditional denoising diffusion probabilistic model-based updating framework is presented to extend the field of deep generative models-based model updating methods. The diffusion model is a representative generative AI technique that employs a Markov chain to progressively add noise to data (forward process), then train a deep neural network to reverse this corruption (reverse process), enabling high-quality data generation. The conditional denoising diffusion extends the standard diffusion model, which guides data synthesis by injecting conditional inputs into the diffusion process. The conditional diffusion-based model updating framework consists of two primary neural networks: a conditional network and a denoising network. The conditional network can summarise the synthetic/measured response data into an informative fixed-length vector, called a conditional embedding, for guiding the training and denoising process of the denoising network. The denoising network can learn to predict the noise added in the forward process and denoise to generate the posterior samples conditioned on the conditional embedding. Both networks are trained jointly, and their architectures are flexible and problem-dependent. The framework is applied to solve a simulation-based problem, which is a customised version of the NASA and DNV Uncertainty Quantification Challenge 2025, and an experimental case study, which is a recently designed benchmark testbed with both experiment uncertainty and controllable parameter uncertainty.
{"title":"Stochastic model updating using conditional diffusion-based probabilistic generative models","authors":"Tairan Wang, Sifeng Bi","doi":"10.1016/j.ymssp.2026.113891","DOIUrl":"10.1016/j.ymssp.2026.113891","url":null,"abstract":"<div><div>This work aims to explore the integration of novel and powerful deep learning techniques and intractable engineering problems, especially by adopting deep generative models to tackle model updating problems under uncertainty. A conditional denoising diffusion probabilistic model-based updating framework is presented to extend the field of deep generative models-based model updating methods. The diffusion model is a representative generative AI technique that employs a Markov chain to progressively add noise to data (forward process), then train a deep neural network to reverse this corruption (reverse process), enabling high-quality data generation. The conditional denoising diffusion extends the standard diffusion model, which guides data synthesis by injecting conditional inputs into the diffusion process. The conditional diffusion-based model updating framework consists of two primary neural networks: a conditional network and a denoising network. The conditional network can summarise the synthetic/measured response data into an informative fixed-length vector, called a conditional embedding, for guiding the training and denoising process of the denoising network. The denoising network can learn to predict the noise added in the forward process and denoise to generate the posterior samples conditioned on the conditional embedding. Both networks are trained jointly, and their architectures are flexible and problem-dependent. The framework is applied to solve a simulation-based problem, which is a customised version of the NASA and DNV Uncertainty Quantification Challenge 2025, and an experimental case study, which is a recently designed benchmark testbed with both experiment uncertainty and controllable parameter uncertainty.</div></div>","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"246 ","pages":"Article 113891"},"PeriodicalIF":8.9,"publicationDate":"2026-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145995655","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 : 2026-02-15Epub Date: 2026-01-23DOI: 10.1016/j.ymssp.2026.113902
Mengxiu Yang , Jie Wu , Cheng Shu
<div><div>The majority of structural health monitoring (SHM) studies focus primarily on measuring responses since access to inputs is highly restricted. This limitation is particularly pronounced when attempting to evaluate the operational performance of critical structures in complex environments. The inaccurate establishment of state-space models, caused by operational system parameters deviating from design specifications, is a significant challenge that fundamentally constrains the practical application of joint input-state identification techniques. To address the issue, this study proposes an augmented Kalman filtering algorithm with multi-scale Bayesian optimization (MSBO-AKF), which integrates time–frequency domain optimization functions across multiple time scales. The algorithm is specifically designed to mitigate the requirement of prior knowledge regarding the simplified mechanical system in real-world joint input-state estimation. Specifically, in the proposed algorithm, the time-domain function is constructed using the transferring difference, which efficiently extracts system error information while simultaneously avoiding the introduction of additional estimation errors. The frequency-domain function effectively constrains the convergence region of the optimization process, thereby ensuring the convergence toward the global optimum. Furthermore, the combination of multiple time scales enriches the information content for mechanical system. In the time domain, this is achieved by supplementing the information with more underdetermined matrices, and in the frequency domain, it mitigates the adverse effects of high-frequency noise. Additionally, a multi-model filtering strategy is employed to prevent the coupling errors of noise-induced parameters during the optimization step, which significantly enhances the robustness of the algorithm under varying noise conditions. In order to validate the effectiveness and robustness of the algorithm, a 5-degree-of-freedom (5-DOF) system is introduced, where comprehensive parameter studies and ablation study are conducted with limited measured responses. The results consistently demonstrate that the proposed innovations significantly enhance both the accuracy and stability of system identification. Additional comparison experiments have further proven the efficacy of the proposed algorithm in approximating the real values of system parameters, which facilitates the accurate joint identification of states and inputs. Finally, utilizing the SHM data collected from the 632-meter-high Shanghai Tower during an inland cyclone event, the wind load of the top, and three unknown displacements are successfully identified. The identification results are thoroughly analyzed in both the time and frequency domains. This demonstrates that the proposed MSBO-AKF algorithm, by incorporating inherent system identification capabilities, significantly contributes to advancing joint input-state estimation methods for
{"title":"A system-input-state joint estimation algorithm with multi-scale Bayesian optimization","authors":"Mengxiu Yang , Jie Wu , Cheng Shu","doi":"10.1016/j.ymssp.2026.113902","DOIUrl":"10.1016/j.ymssp.2026.113902","url":null,"abstract":"<div><div>The majority of structural health monitoring (SHM) studies focus primarily on measuring responses since access to inputs is highly restricted. This limitation is particularly pronounced when attempting to evaluate the operational performance of critical structures in complex environments. The inaccurate establishment of state-space models, caused by operational system parameters deviating from design specifications, is a significant challenge that fundamentally constrains the practical application of joint input-state identification techniques. To address the issue, this study proposes an augmented Kalman filtering algorithm with multi-scale Bayesian optimization (MSBO-AKF), which integrates time–frequency domain optimization functions across multiple time scales. The algorithm is specifically designed to mitigate the requirement of prior knowledge regarding the simplified mechanical system in real-world joint input-state estimation. Specifically, in the proposed algorithm, the time-domain function is constructed using the transferring difference, which efficiently extracts system error information while simultaneously avoiding the introduction of additional estimation errors. The frequency-domain function effectively constrains the convergence region of the optimization process, thereby ensuring the convergence toward the global optimum. Furthermore, the combination of multiple time scales enriches the information content for mechanical system. In the time domain, this is achieved by supplementing the information with more underdetermined matrices, and in the frequency domain, it mitigates the adverse effects of high-frequency noise. Additionally, a multi-model filtering strategy is employed to prevent the coupling errors of noise-induced parameters during the optimization step, which significantly enhances the robustness of the algorithm under varying noise conditions. In order to validate the effectiveness and robustness of the algorithm, a 5-degree-of-freedom (5-DOF) system is introduced, where comprehensive parameter studies and ablation study are conducted with limited measured responses. The results consistently demonstrate that the proposed innovations significantly enhance both the accuracy and stability of system identification. Additional comparison experiments have further proven the efficacy of the proposed algorithm in approximating the real values of system parameters, which facilitates the accurate joint identification of states and inputs. Finally, utilizing the SHM data collected from the 632-meter-high Shanghai Tower during an inland cyclone event, the wind load of the top, and three unknown displacements are successfully identified. The identification results are thoroughly analyzed in both the time and frequency domains. This demonstrates that the proposed MSBO-AKF algorithm, by incorporating inherent system identification capabilities, significantly contributes to advancing joint input-state estimation methods for ","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"246 ","pages":"Article 113902"},"PeriodicalIF":8.9,"publicationDate":"2026-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146033434","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}
Simultaneous low-frequency and broadband absorption is still difficult to achieve in compact acoustic metamaterials, as most existing designs address only one aspect. To address this limitation, a hybrid acoustic metastructure combining a sonic black hole with a membrane-type acoustic metamaterial is proposed to realize efficient broadband absorption at low frequencies within a compact configuration. A transfer matrix model, validated by finite element simulations, confirms that the sonic black hole provides broadband dissipation by guiding and attenuating acoustic energy, while the membrane-type acoustic metamaterial introduces tunable low-frequency resonances. Parametric studies further reveal the critical influence of the coupling cavity and back cavity dimensions in shaping the absorption peaks. Comparative analyses with conventional sonic black hole-based designs demonstrate that the proposed acoustic metastructure achieves superior low-frequency control and compactness. Finally, impedance tube experiments corroborate the numerical predictions, underscoring the strong potential of the acoustic metastructure for practical broadband low-frequency noise control applications.
{"title":"Sonic black hole coupled with membrane-type acoustic metamaterial for broadband and low-frequency sound absorption","authors":"Wei-Qin Wu , Yong-Bin Zhang , Liu-Xian Zhao , Ting-Gui Chen , Yi-Feng Wang","doi":"10.1016/j.ymssp.2026.113932","DOIUrl":"10.1016/j.ymssp.2026.113932","url":null,"abstract":"<div><div>Simultaneous low-frequency and broadband absorption is still difficult to achieve in compact acoustic metamaterials, as most existing designs address only one aspect. To address this limitation, a hybrid acoustic metastructure combining a sonic black hole with a membrane-type acoustic metamaterial is proposed to realize efficient broadband absorption at low frequencies within a compact configuration. A transfer matrix model, validated by finite element simulations, confirms that the sonic black hole provides broadband dissipation by guiding and attenuating acoustic energy, while the membrane-type acoustic metamaterial introduces tunable low-frequency resonances. Parametric studies further reveal the critical influence of the coupling cavity and back cavity dimensions in shaping the absorption peaks. Comparative analyses with conventional sonic black hole-based designs demonstrate that the proposed acoustic metastructure achieves superior low-frequency control and compactness. Finally, impedance tube experiments corroborate the numerical predictions, underscoring the strong potential of the acoustic metastructure for practical broadband low-frequency noise control applications.</div></div>","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"246 ","pages":"Article 113932"},"PeriodicalIF":8.9,"publicationDate":"2026-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146047934","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 : 2026-02-15Epub Date: 2026-01-26DOI: 10.1016/j.ymssp.2026.113905
Khairul Jauhari, Achmad Zaki Rahman, Fitriana Nur Hasanah Aji Pramesti, Sri Kliwati, Wahyu Widada, Mahfudz Al Huda
Chatter detection plays a critical role in modern milling operations, as regenerative vibrations can severely degrade surface quality, accelerate tool wear, and destabilize the cutting process. Although existing techniques such as Fast Fourier Transform (FFT), Short-Time Fourier Transform (STFT), and wavelet-based analysis have been widely adopted, their performance is often limited by high computational demand and reduced effectiveness when dealing with rapidly changing or non-stationary signals. To overcome these limitations, this study introduces a novel time-domain chatter detection approach based on the Natural Observation Filter (NOF). The method decomposes vibration signals using lightweight recursive filters and employs a Mean Filter Index (MFI) to capture energy shifts associated with transitions from stable cutting to chatter. The proposed framework is validated through both numerical simulations and controlled milling experiments using a 3-axis CNC machine. Results show that the method can accurately distinguish stable, transition, and chatter states even under varying dynamic conditions. With a computational complexity of O(M·N), the NOF algorithm achieves ultra-low processing latency, enabling real-time deployment on low-power embedded platforms such as microcontrollers. These advantages highlight its potential for practical, industry-scale chatter monitoring and integration into intelligent machining systems.
{"title":"A novel online milling chatter detection using natural observation filters and mean filter index","authors":"Khairul Jauhari, Achmad Zaki Rahman, Fitriana Nur Hasanah Aji Pramesti, Sri Kliwati, Wahyu Widada, Mahfudz Al Huda","doi":"10.1016/j.ymssp.2026.113905","DOIUrl":"10.1016/j.ymssp.2026.113905","url":null,"abstract":"<div><div>Chatter detection plays a critical role in modern milling operations, as regenerative vibrations can severely degrade surface quality, accelerate tool wear, and destabilize the cutting process. Although existing techniques such as Fast Fourier Transform (FFT), Short-Time Fourier Transform (STFT), and wavelet-based analysis have been widely adopted, their performance is often limited by high computational demand and reduced effectiveness when dealing with rapidly changing or non-stationary signals. To overcome these limitations, this study introduces a novel time-domain chatter detection approach based on the Natural Observation Filter (NOF). The method decomposes vibration signals using lightweight recursive filters and employs a Mean Filter Index (MFI) to capture energy shifts associated with transitions from stable cutting to chatter. The proposed framework is validated through both numerical simulations and controlled milling experiments using a 3-axis CNC machine. Results show that the method can accurately distinguish stable, transition, and chatter states even under varying dynamic conditions. With a computational complexity of <em>O</em>(<em>M</em>·<em>N</em>), the NOF algorithm achieves ultra-low processing latency, enabling real-time deployment on low-power embedded platforms such as microcontrollers. These advantages highlight its potential for practical, industry-scale chatter monitoring and integration into intelligent machining systems.</div></div>","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"246 ","pages":"Article 113905"},"PeriodicalIF":8.9,"publicationDate":"2026-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146047937","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 : 2026-02-15Epub Date: 2026-01-20DOI: 10.1016/j.ymssp.2026.113900
Shuai Zhao , Guangming Li , Chengyou Wang , Kaixuan Hui
Bridge cables are critical load-bearing components that transfer the weight of the main beam and deck in cable-stayed bridges. To ensure safe operation and effective maintenance, real-time monitoring of cable integrity is essential. Acoustic emission (AE) technology provides an effective means of real-time damage monitoring; however, distinguishing broken wire signals from noise remains a challenge, particularly in complex operational environments. To address this issue, this paper proposes a deep learning framework that integrates conventional manual features temporal feature with a long short-term memory (LSTM) autoencoder, and spatial feature extraction using graph convolutional networks (GCN): MLG-Net (manual LSTM GCN-Net). First, AE signals are collected from full-scale bridge cable breakage experiments, and conventional manual features are extracted. Next, a long short-term memory autoencoder captures the temporal evolution of AE signals, while a graph convolutional network leverages spatial correlations among multi-sensor AE data. Experimental results demonstrate that the proposed method achieves 99.4% accuracy, 99.0% recall, and a 99.2% F1 score, significantly outperforming conventional classifiers. This study highlights the potential of integrating manual and deep learning-based feature extraction for bridge cable health monitoring and provides a foundation for future research on real-world AE-based structural health assessment.
{"title":"MLG-Net: A hybrid framework for bridge cable damage identification using acoustic emission technology","authors":"Shuai Zhao , Guangming Li , Chengyou Wang , Kaixuan Hui","doi":"10.1016/j.ymssp.2026.113900","DOIUrl":"10.1016/j.ymssp.2026.113900","url":null,"abstract":"<div><div>Bridge cables are critical load-bearing components that transfer the weight of the main beam and deck in cable-stayed bridges. To ensure safe operation and effective maintenance, real-time monitoring of cable integrity is essential. Acoustic emission (AE) technology provides an effective means of real-time damage monitoring; however, distinguishing broken wire signals from noise remains a challenge, particularly in complex operational environments. To address this issue, this paper proposes a deep learning framework that integrates conventional manual features temporal feature with a long short-term memory (LSTM) autoencoder, and spatial feature extraction using graph convolutional networks (GCN): MLG-Net (manual LSTM GCN-Net). First, AE signals are collected from full-scale bridge cable breakage experiments, and conventional manual features are extracted. Next, a long short-term memory autoencoder captures the temporal evolution of AE signals, while a graph convolutional network leverages spatial correlations among multi-sensor AE data. Experimental results demonstrate that the proposed method achieves 99.4% accuracy, 99.0% recall, and a 99.2% F1 score, significantly outperforming conventional classifiers. This study highlights the potential of integrating manual and deep learning-based feature extraction for bridge cable health monitoring and provides a foundation for future research on real-world AE-based structural health assessment.</div></div>","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"246 ","pages":"Article 113900"},"PeriodicalIF":8.9,"publicationDate":"2026-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146014518","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 : 2026-02-15Epub Date: 2026-01-28DOI: 10.1016/j.ymssp.2026.113923
Keshun You , Yingkui Gu , Haidong Shao , Yajun Wang
For the problems of dynamic feature attenuation, low efficiency of multimodal fusion and insufficient diagnostic interpretability in rotating machinery fault diagnosis, this paper proposes an interpretable multimodal heterogeneous fusion liquid impulse neural network (LINN) model. First, a liquid state coding layer based on differential equations is constructed to model the time-series dynamic evolution features in non-stationary signals via a chunked feedback mechanism. Moreover, a multi-channel leaky integrate-and-fire (MC-LIF) impulse neurons are introduced to enhance the extraction of transient shock features by combining alternative gradient and membrane potential attenuation strategies. Finally, an attention-guided multimodal fusion mechanism is designed to realize adaptive integration and contribution interpretability quantification of time–frequency features. In the high-noise and variable-load condition tests, LINN achieves more than 98.7% accuracy with only 4.1 M parameters and 88.64% cross-condition generalization accuracy. The ablation experiments verify the key role of liquid layer and impulse mechanism in enhancing dynamic modelling and noise immunity, and the interpretability analysis based on time–frequency domain attention (TFDA) further reveals the sensitive response of the model to key time–frequency modal contributions. The method provides an effective solution with high accuracy, strong generalization and interpretability for intelligent diagnosis under complex working conditions.
{"title":"A liquid-impulse neural network model based on heterogeneous fusion of multimodal information for interpretable rotating machinery fault diagnosis","authors":"Keshun You , Yingkui Gu , Haidong Shao , Yajun Wang","doi":"10.1016/j.ymssp.2026.113923","DOIUrl":"10.1016/j.ymssp.2026.113923","url":null,"abstract":"<div><div>For the problems of dynamic feature attenuation, low efficiency of multimodal fusion and insufficient diagnostic interpretability in rotating machinery fault diagnosis, this paper proposes an interpretable multimodal heterogeneous fusion liquid impulse neural network (LINN) model. First, a liquid state coding layer based on differential equations is constructed to model the time-series dynamic evolution features in non-stationary signals via a chunked feedback mechanism. Moreover, a multi-channel leaky integrate-and-fire (MC-LIF) impulse neurons are introduced to enhance the extraction of transient shock features by combining alternative gradient and membrane potential attenuation strategies. Finally, an attention-guided multimodal fusion mechanism is designed to realize adaptive integration and contribution interpretability quantification of time–frequency features. In the high-noise and variable-load condition tests, LINN achieves more than 98.7% accuracy with only 4.1 M parameters and 88.64% cross-condition generalization accuracy. The ablation experiments verify the key role of liquid layer and impulse mechanism in enhancing dynamic modelling and noise immunity, and the interpretability analysis based on time–frequency domain attention (TFDA) further reveals the sensitive response of the model to key time–frequency modal contributions. The method provides an effective solution with high accuracy, strong generalization and interpretability for intelligent diagnosis under complex working conditions.</div></div>","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"246 ","pages":"Article 113923"},"PeriodicalIF":8.9,"publicationDate":"2026-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146072613","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 : 2026-02-15Epub Date: 2026-01-19DOI: 10.1016/j.ymssp.2026.113870
Agustín Hernández Rocha , Damián H. Zanette , Marian Wiercigroch
This article presents a generic methodology to investigate dynamics and bifurcation scenarios of multi degrees-of-freedom piecewise linear systems. The method takes advantage of the analytical solutions for linear regimes to define mapping transformations, which in turn allow to determine all periodic orbits. The methodology is applied to analyse dynamic interactions between two oscillators connected via an elastic link. A rich variety and complexity of solutions are obtained close to the first grazing frequency. Zones with isolated solutions and co-existence of three to five orbits were found. The in-phase and out-phase modes are sensitive to the phase shift between the forces and the ratio between the natural frequencies of the individual oscillators.
{"title":"Semi-analytical method to analyse periodic orbits of piecewise linear oscillators","authors":"Agustín Hernández Rocha , Damián H. Zanette , Marian Wiercigroch","doi":"10.1016/j.ymssp.2026.113870","DOIUrl":"10.1016/j.ymssp.2026.113870","url":null,"abstract":"<div><div>This article presents a generic methodology to investigate dynamics and bifurcation scenarios of multi degrees-of-freedom piecewise linear systems. The method takes advantage of the analytical solutions for linear regimes to define mapping transformations, which in turn allow to determine all periodic orbits. The methodology is applied to analyse dynamic interactions between two oscillators connected via an elastic link. A rich variety and complexity of solutions are obtained close to the first grazing frequency. Zones with isolated solutions and co-existence of three to five orbits were found. The in-phase and out-phase modes are sensitive to the phase shift between the forces and the ratio between the natural frequencies of the individual oscillators.</div></div>","PeriodicalId":51124,"journal":{"name":"Mechanical Systems and Signal Processing","volume":"246 ","pages":"Article 113870"},"PeriodicalIF":8.9,"publicationDate":"2026-02-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146001047","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}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号