Pub Date : 2025-02-10DOI: 10.1109/TIM.2025.3540132
Jin He;Fengmao Lv;Jun Liu;Min Wu;Badong Chen;Shiping Wang
The dropper plays a critical role in the overhead contact system (OCS) of high-speed railways, ensuring smooth power transmission and reducing vibration between the contact and messenger wires. However, adverse factors, such as temperature variations, inclement weather, and high-frequency vibrations can lead to dropper loosening and detachment, which deteriorates the collecting current through the pantograph. In severe cases, it can even result in pantograph breakage or contact wire damage, ultimately causing train malfunctions. Unfortunately, existing detection methods fall short in recognizing dropper defects in real-world scenarios. To address this challenge, we propose a novel cross-fusion of convolutional neural network and transformer for high-speed railway dropper defect detection (C2T-HR3D) network. Leveraging a cross-fusion of convolutional neural network (CNN) and transformers, this network accurately recognizes dropper defects in challenging scenarios, such as fog, rain, sun, and night-time conditions. Moreover, it can also accurately identify obscured and small dropper defects from a long distance, significantly improving recall and precision. Extensive experiments have demonstrated that our network outperforms CNN-based, transformer-based, and CNN-transformer state-of-the-art networks by 3.4%, 1.8%, and 2.1%, respectively. The C2T-HR3D network has been successfully deployed on over 300 high-speed trains, detecting more than 10000 dropper defects.
{"title":"C2T-HR3D: Cross-Fusion of CNN and Transformer for High-Speed Railway Dropper Defect Detection","authors":"Jin He;Fengmao Lv;Jun Liu;Min Wu;Badong Chen;Shiping Wang","doi":"10.1109/TIM.2025.3540132","DOIUrl":"https://doi.org/10.1109/TIM.2025.3540132","url":null,"abstract":"The dropper plays a critical role in the overhead contact system (OCS) of high-speed railways, ensuring smooth power transmission and reducing vibration between the contact and messenger wires. However, adverse factors, such as temperature variations, inclement weather, and high-frequency vibrations can lead to dropper loosening and detachment, which deteriorates the collecting current through the pantograph. In severe cases, it can even result in pantograph breakage or contact wire damage, ultimately causing train malfunctions. Unfortunately, existing detection methods fall short in recognizing dropper defects in real-world scenarios. To address this challenge, we propose a novel cross-fusion of convolutional neural network and transformer for high-speed railway dropper defect detection (C2T-HR3D) network. Leveraging a cross-fusion of convolutional neural network (CNN) and transformers, this network accurately recognizes dropper defects in challenging scenarios, such as fog, rain, sun, and night-time conditions. Moreover, it can also accurately identify obscured and small dropper defects from a long distance, significantly improving recall and precision. Extensive experiments have demonstrated that our network outperforms CNN-based, transformer-based, and CNN-transformer state-of-the-art networks by 3.4%, 1.8%, and 2.1%, respectively. The C2T-HR3D network has been successfully deployed on over 300 high-speed trains, detecting more than 10000 dropper defects.","PeriodicalId":13341,"journal":{"name":"IEEE Transactions on Instrumentation and Measurement","volume":"74 ","pages":"1-16"},"PeriodicalIF":5.6,"publicationDate":"2025-02-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143465718","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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 : 2025-02-10DOI: 10.1109/TIM.2025.3540129
Moirangthem James Singh;L. N. Sharma;Samarendra Dandapat
Effective screening for heart valve disease (HVD) is crucial for impeding its progression. However, current approaches lack transparency in classifying diverse HVDs. The seismocardiogram (SCG) signal provides comprehensive insights into cardiac activities across three axes, offering valuable information for detecting valvular abnormalities. To leverage this potential and address the aforementioned challenges, we propose HVDNet, an interpretable deep-learning framework for HVD classification using tri-axial SCG signals. The architecture integrates three modules: stacked 1-D convolutional neural networks with skip connections (sCNNs) to learn hierarchical features associated with morphological variations in SCG at different scales, long short-term memory (LSTM) layers to capture temporal variations within the feature maps, and self-attention (SA) layer to emphasize clinically relevant attributes. Evaluation on publicly available SCG databases demonstrate high accuracies: 99.35% on the validation set and 98.98% on the test set for HVD without co-existing diseases, and 99.21% on the validation set and 98.89% on the test set for aortic stenosis (AS) co-existing with other HVDs. Through an ablation study of different model variants, we found that integrating information from each axis component of the SCG signal yields optimal performance. Moreover, closely examining the learned attention weights reveals how the model emphasizes clinically relevant SCG attributes that characterize HVD. With its inherent transparency and superior performance compared to existing methods, the proposed model can become a reliable diagnostic tool for HVD, potentially improving patient care and treatment efficacy.
{"title":"HVDNet: An Interpretable Deep Learning Framework for Heart Valve Disease Classification Using Tri-Axial Seismocardiogram Signals","authors":"Moirangthem James Singh;L. N. Sharma;Samarendra Dandapat","doi":"10.1109/TIM.2025.3540129","DOIUrl":"https://doi.org/10.1109/TIM.2025.3540129","url":null,"abstract":"Effective screening for heart valve disease (HVD) is crucial for impeding its progression. However, current approaches lack transparency in classifying diverse HVDs. The seismocardiogram (SCG) signal provides comprehensive insights into cardiac activities across three axes, offering valuable information for detecting valvular abnormalities. To leverage this potential and address the aforementioned challenges, we propose HVDNet, an interpretable deep-learning framework for HVD classification using tri-axial SCG signals. The architecture integrates three modules: stacked 1-D convolutional neural networks with skip connections (sCNNs) to learn hierarchical features associated with morphological variations in SCG at different scales, long short-term memory (LSTM) layers to capture temporal variations within the feature maps, and self-attention (SA) layer to emphasize clinically relevant attributes. Evaluation on publicly available SCG databases demonstrate high accuracies: 99.35% on the validation set and 98.98% on the test set for HVD without co-existing diseases, and 99.21% on the validation set and 98.89% on the test set for aortic stenosis (AS) co-existing with other HVDs. Through an ablation study of different model variants, we found that integrating information from each axis component of the SCG signal yields optimal performance. Moreover, closely examining the learned attention weights reveals how the model emphasizes clinically relevant SCG attributes that characterize HVD. With its inherent transparency and superior performance compared to existing methods, the proposed model can become a reliable diagnostic tool for HVD, potentially improving patient care and treatment efficacy.","PeriodicalId":13341,"journal":{"name":"IEEE Transactions on Instrumentation and Measurement","volume":"74 ","pages":"1-11"},"PeriodicalIF":5.6,"publicationDate":"2025-02-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143455204","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The practical applications of current PQD classification and identification algorithms are limited due to their often poor recognition of disturbance types outside the training sets. To deal with this issue, a novel classification and identification method for complex PQDs based on confidence-enhanced guided multilabel learning (CEGML) is proposed in this article. Initially, the fully convolutional network (FCN) is employed to extract deep and shallow features, which are subsequently fused. In addition, the concept of the simple parameter-free attention module (SimAM) is combined with the gated recurrent unit (GRU) to design the synchronized recurrent attention model (SRAM), enhancing the extraction of key features and the model’s ability to fit labels. At the same time, a linear layer is utilized to predict the confidence level of each label within the PQDs. Lastly, a confidence label is designed to differentiate between single disturbances and multiple disturbances, and a confidence enhancement factor is set to elevate the confidence levels of each label in the case of multiple disturbances, enabling the network to accurately identify more complex combinations of disturbance types beyond the training sets, even with limited and basic training data. Simulation and practical test results demonstrate that the PQD classification and identification method based on CEGML proposed in this article effectively recognizes complex disturbance types not included in the training set, achieving higher identification accuracy than existing methods.
{"title":"A Novel Classification and Identification Method for Complex Power Quality Disturbances Based on Confidence-Enhanced Guided Multilabel Learning","authors":"Dian Hong;Jianmin Li;Chengbin Liang;Haijun Lin;Wenxuan Yao;Jiaqi Yu","doi":"10.1109/TIM.2025.3540138","DOIUrl":"https://doi.org/10.1109/TIM.2025.3540138","url":null,"abstract":"The practical applications of current PQD classification and identification algorithms are limited due to their often poor recognition of disturbance types outside the training sets. To deal with this issue, a novel classification and identification method for complex PQDs based on confidence-enhanced guided multilabel learning (CEGML) is proposed in this article. Initially, the fully convolutional network (FCN) is employed to extract deep and shallow features, which are subsequently fused. In addition, the concept of the simple parameter-free attention module (SimAM) is combined with the gated recurrent unit (GRU) to design the synchronized recurrent attention model (SRAM), enhancing the extraction of key features and the model’s ability to fit labels. At the same time, a linear layer is utilized to predict the confidence level of each label within the PQDs. Lastly, a confidence label is designed to differentiate between single disturbances and multiple disturbances, and a confidence enhancement factor is set to elevate the confidence levels of each label in the case of multiple disturbances, enabling the network to accurately identify more complex combinations of disturbance types beyond the training sets, even with limited and basic training data. Simulation and practical test results demonstrate that the PQD classification and identification method based on CEGML proposed in this article effectively recognizes complex disturbance types not included in the training set, achieving higher identification accuracy than existing methods.","PeriodicalId":13341,"journal":{"name":"IEEE Transactions on Instrumentation and Measurement","volume":"74 ","pages":"1-11"},"PeriodicalIF":5.6,"publicationDate":"2025-02-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143480770","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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 : 2025-02-10DOI: 10.1109/TIM.2025.3540139
Yuwen Ruan;Long Meng;Xiaogang Hu
Surface electromyogram (EMG) signals have been a preferred modality for motor intent detections in the fields of robotic control, rehabilitation, and health monitoring. However, current EMG-based measurement techniques suffer a degradation in performance cross session over time due to factors such as shifts in electrode placement, changes in muscle states, and environmental noise. To address this challenge, we developed a novel neural-drive approach, capable of robust cross-day predictions of individual finger forces. Specifically, high-density EMG (HD-EMG) data were collected from flexor and extensor muscles during single-finger and multifinger tasks. The experimental procedure was repeated three times (sessions), with an average interval of 6.58 days between sessions. We first decomposed the EMG signals in a session to obtain separation matrices that contained motor unit (MU) information in the EMG signals. We then refined the separation matrices that accurately reflected individual fingers. The corresponding separation matrices were applied to EMG signals in the other two sessions to derive the neural drive for force predictions of individual fingers. Our results revealed that the cross-session performance was comparable with the within-session performance. In addition, the neural-drive approach can outperform the conventional EMG-amplitude approach, especially in the cross-session performance. Our developed approach can enhance the long-term reliability of finger force predictions and holds potential for various practical applications.
{"title":"Long-Term Finger Force Predictions Using Motoneuron Discharge Activities","authors":"Yuwen Ruan;Long Meng;Xiaogang Hu","doi":"10.1109/TIM.2025.3540139","DOIUrl":"https://doi.org/10.1109/TIM.2025.3540139","url":null,"abstract":"Surface electromyogram (EMG) signals have been a preferred modality for motor intent detections in the fields of robotic control, rehabilitation, and health monitoring. However, current EMG-based measurement techniques suffer a degradation in performance cross session over time due to factors such as shifts in electrode placement, changes in muscle states, and environmental noise. To address this challenge, we developed a novel neural-drive approach, capable of robust cross-day predictions of individual finger forces. Specifically, high-density EMG (HD-EMG) data were collected from flexor and extensor muscles during single-finger and multifinger tasks. The experimental procedure was repeated three times (sessions), with an average interval of 6.58 days between sessions. We first decomposed the EMG signals in a session to obtain separation matrices that contained motor unit (MU) information in the EMG signals. We then refined the separation matrices that accurately reflected individual fingers. The corresponding separation matrices were applied to EMG signals in the other two sessions to derive the neural drive for force predictions of individual fingers. Our results revealed that the cross-session performance was comparable with the within-session performance. In addition, the neural-drive approach can outperform the conventional EMG-amplitude approach, especially in the cross-session performance. Our developed approach can enhance the long-term reliability of finger force predictions and holds potential for various practical applications.","PeriodicalId":13341,"journal":{"name":"IEEE Transactions on Instrumentation and Measurement","volume":"74 ","pages":"1-10"},"PeriodicalIF":5.6,"publicationDate":"2025-02-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143465716","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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 : 2025-02-10DOI: 10.1109/TIM.2025.3538092
Hao Lu;Wenyu Wang;Jingyi Liu;Zhenzhan Wang;Jieying He;Shengwei Zhang;Xiaolin Tong;Te Wang
Hyperspectral microwave radiometer is a new type of passive microwave remote sensor for observing middle and upper atmospheric temperature, humidity, trace gas, and other parameters such as winds. The digital spectrometer, which allows the fine sampling of the spectral lines, is the core component of the radiometer. In this article, we propose the design and implementation of a new type of wideband, real-time channelized digital spectrometer, which realizes the core base-64 real-time complex fast Fourier transform (FFT) algorithm and channelization algorithm by improving the filter bank, 128-channel parallel processing of FFT and complex number processing. The digital spectrometer has a sampling rate of 20 Gsps, a quantization bit number of 8 bits, and an input bandwidth of 10 GHz, which realizes the spectrum analysis of 4096 channels. Then, an observation test was carried out using a V-band ground-based microwave radiometer equipped with the 10-GHz spectrometer, and the atmospheric temperature profile was successfully measured from the surface to the stratosphere. The retrieval results were compared with the ERA5 reanalysis data and the L2 temperature products of the FY-3-D/MWTS-MWHS and Aura/microwave limb sounder (MLS), with a better consistency, which proved the application and potential of the new wideband digital spectrometer in the atmospheric sounding.
{"title":"Development of a 10-GHz Wideband Digital Spectrometer for Hyperspectral Microwave Sounding","authors":"Hao Lu;Wenyu Wang;Jingyi Liu;Zhenzhan Wang;Jieying He;Shengwei Zhang;Xiaolin Tong;Te Wang","doi":"10.1109/TIM.2025.3538092","DOIUrl":"https://doi.org/10.1109/TIM.2025.3538092","url":null,"abstract":"Hyperspectral microwave radiometer is a new type of passive microwave remote sensor for observing middle and upper atmospheric temperature, humidity, trace gas, and other parameters such as winds. The digital spectrometer, which allows the fine sampling of the spectral lines, is the core component of the radiometer. In this article, we propose the design and implementation of a new type of wideband, real-time channelized digital spectrometer, which realizes the core base-64 real-time complex fast Fourier transform (FFT) algorithm and channelization algorithm by improving the filter bank, 128-channel parallel processing of FFT and complex number processing. The digital spectrometer has a sampling rate of 20 Gsps, a quantization bit number of 8 bits, and an input bandwidth of 10 GHz, which realizes the spectrum analysis of 4096 channels. Then, an observation test was carried out using a V-band ground-based microwave radiometer equipped with the 10-GHz spectrometer, and the atmospheric temperature profile was successfully measured from the surface to the stratosphere. The retrieval results were compared with the ERA5 reanalysis data and the L2 temperature products of the FY-3-D/MWTS-MWHS and Aura/microwave limb sounder (MLS), with a better consistency, which proved the application and potential of the new wideband digital spectrometer in the atmospheric sounding.","PeriodicalId":13341,"journal":{"name":"IEEE Transactions on Instrumentation and Measurement","volume":"74 ","pages":"1-9"},"PeriodicalIF":5.6,"publicationDate":"2025-02-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143379547","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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 : 2025-02-10DOI: 10.1109/TIM.2025.3534219
Zhibin Li;Yuxuan Zhang;Weirong Dong;Jing Yang;Jiansong Feng;Chengbi Zhang;Xiaolong Chen;Taihong Wang
Human beings can infer the shape and material characteristics of grasping objects based on multisensory information, which is still a technical challenge for modern robots. The cross-modal object perception mechanism holds promise to assist robots in effectively executing various operations or interactive tasks in complex applications, particularly in harsh visual scenes. Here, we present an associated learning architecture equipped with a parallel impedance sensing strategy, which enhances the perception of captured objects by integrating visual data with somatosensory data from frequency division multiplexing (FDM) parallel impedance and finger bending angles of the robotic hand. We design a cross-modal generative adversarial network (CGAN) in this architecture to achieve cross-modal feature learning for two types of sensory data, mimicking the psychological cognition of human senses. Additionally, the dynamic attention fusion mechanism is employed for feature transfer and fusion learning, enabling the network to adaptively adjust weights based on input cross-modal features, resulting in dynamic feature fusion. The architecture has undergone training and testing with ten categories of objects, successfully achieving cross-modal feature learning and fusion recognition of the two sensory data. Under low-quality image conditions, the recognition accuracy of attention fusion reaches up to 94.0%, significantly surpassing the accuracy of vision alone. This highlights the potential of our architecture to enhance robots to accurately perceive the outside world by integrating visual and somatosensory data, especially in challenging visual environments.
{"title":"Associated Learning Architecture Equipped With Parallel Impedance Sensing Strategy to Enhance Cross-Modal Object Perception","authors":"Zhibin Li;Yuxuan Zhang;Weirong Dong;Jing Yang;Jiansong Feng;Chengbi Zhang;Xiaolong Chen;Taihong Wang","doi":"10.1109/TIM.2025.3534219","DOIUrl":"https://doi.org/10.1109/TIM.2025.3534219","url":null,"abstract":"Human beings can infer the shape and material characteristics of grasping objects based on multisensory information, which is still a technical challenge for modern robots. The cross-modal object perception mechanism holds promise to assist robots in effectively executing various operations or interactive tasks in complex applications, particularly in harsh visual scenes. Here, we present an associated learning architecture equipped with a parallel impedance sensing strategy, which enhances the perception of captured objects by integrating visual data with somatosensory data from frequency division multiplexing (FDM) parallel impedance and finger bending angles of the robotic hand. We design a cross-modal generative adversarial network (CGAN) in this architecture to achieve cross-modal feature learning for two types of sensory data, mimicking the psychological cognition of human senses. Additionally, the dynamic attention fusion mechanism is employed for feature transfer and fusion learning, enabling the network to adaptively adjust weights based on input cross-modal features, resulting in dynamic feature fusion. The architecture has undergone training and testing with ten categories of objects, successfully achieving cross-modal feature learning and fusion recognition of the two sensory data. Under low-quality image conditions, the recognition accuracy of attention fusion reaches up to 94.0%, significantly surpassing the accuracy of vision alone. This highlights the potential of our architecture to enhance robots to accurately perceive the outside world by integrating visual and somatosensory data, especially in challenging visual environments.","PeriodicalId":13341,"journal":{"name":"IEEE Transactions on Instrumentation and Measurement","volume":"74 ","pages":"1-11"},"PeriodicalIF":5.6,"publicationDate":"2025-02-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143388571","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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 : 2025-02-10DOI: 10.1109/TIM.2025.3540144
Zijian Wang;Linxia Zhang;Zhe Li;Shukai Duan;Jia Yan
The theory and technology of electronic nose (E-nose) systems have been vigorously developed, and these systems have achieved success in many practical applications, such as medical diagnosis, food quality inspection, and environmental detection. However, the drift problem of a sensor array affects the industrialization and commercialization of E-nose systems. In this article, a novel ELM-based dual-level joint domain adaptation method (JDAELM) is proposed to effectively suppress drift and address the distribution discrepancy issue. Specifically, the proposed method implements joint domain adaptation (DA) at the feature level and label level. For source domain data without drift, the information of the data could be preserved as much as possible. Considering the inconsistent distribution caused by drift, the marginal and conditional distribution discrepancies are reduced to a minimum at the feature level to achieve domain alignment. To reduce the impact of pseudolabels on the model, we align the label space to achieve DA at the label level. By maximizing the Hilbert–Schmidt independence criterion, the relationship between the feature projection space and label projection space is strengthened in this model. The joint learning model is effectively solved by an efficient alternative optimization strategy. The average accuracy of the proposed method is 88.30% and 87.21% under long-term drift and short-term drift, respectively, and 96.41% on the instrument variation dataset, which is superior to that of other comparison methods in terms of accuracy. This proves that the JDAELM can be well adapted to long-term and short-term drift scenarios, and can effectively compensate for instrument variation drift caused by inherent differences.
{"title":"Improving E-Nose Performance: A Novel ELM-Based Dual-Level Joint Domain Adaptation Method for Sensor Drift Data","authors":"Zijian Wang;Linxia Zhang;Zhe Li;Shukai Duan;Jia Yan","doi":"10.1109/TIM.2025.3540144","DOIUrl":"https://doi.org/10.1109/TIM.2025.3540144","url":null,"abstract":"The theory and technology of electronic nose (E-nose) systems have been vigorously developed, and these systems have achieved success in many practical applications, such as medical diagnosis, food quality inspection, and environmental detection. However, the drift problem of a sensor array affects the industrialization and commercialization of E-nose systems. In this article, a novel ELM-based dual-level joint domain adaptation method (JDAELM) is proposed to effectively suppress drift and address the distribution discrepancy issue. Specifically, the proposed method implements joint domain adaptation (DA) at the feature level and label level. For source domain data without drift, the information of the data could be preserved as much as possible. Considering the inconsistent distribution caused by drift, the marginal and conditional distribution discrepancies are reduced to a minimum at the feature level to achieve domain alignment. To reduce the impact of pseudolabels on the model, we align the label space to achieve DA at the label level. By maximizing the Hilbert–Schmidt independence criterion, the relationship between the feature projection space and label projection space is strengthened in this model. The joint learning model is effectively solved by an efficient alternative optimization strategy. The average accuracy of the proposed method is 88.30% and 87.21% under long-term drift and short-term drift, respectively, and 96.41% on the instrument variation dataset, which is superior to that of other comparison methods in terms of accuracy. This proves that the JDAELM can be well adapted to long-term and short-term drift scenarios, and can effectively compensate for instrument variation drift caused by inherent differences.","PeriodicalId":13341,"journal":{"name":"IEEE Transactions on Instrumentation and Measurement","volume":"74 ","pages":"1-11"},"PeriodicalIF":5.6,"publicationDate":"2025-02-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143480829","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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 : 2025-02-10DOI: 10.1109/TIM.2025.3540131
Rourou Li;Tangbin Xia;Yimin Jiang;Jianhua Wu;Xiaolei Fang;Nagi Gebraeel;Lifeng Xi
Fault diagnosis (FD) of industrial robots (IRs) plays an increasingly indispensable role in modern manufacturing. Fault-related component obscurity by strong noise, feature exploitation insufficiency with scarce fault samples, and limited physical interpretation hinder existing diagnostic models’ application to IRs. A deep, complex wavelet denoising network (DCWDN) is, thus, proposed to achieve high-performance and interpretable FD with robustness against noise and class-imbalanced data. Hereinto, a dual-tree cascade autoencoder with trainable convolutional filters is constructed. Significantly, complex wavelet conditions such as orthogonality, approximate analyticity, and sparsity are imposed on the filters to structure their optimization. Meanwhile, shrinkage-based denoising with learnable thresholds is integrated to suppress noise-related components. The proposed DCWDN organically combines the data adaptivity of deep learning (DL) and wavelets’ time-frequency representation ability. Its interpretability is embodied through the explainable structure, learned scientifically meaningful filters, and extracted coefficients with explicit fault indications. Case studies on real IR datasets and experimental drivetrain benchmarks are conducted to demonstrate the effectiveness and superiority of the proposed method.
{"title":"Deep Complex Wavelet Denoising Network for Interpretable Fault Diagnosis of Industrial Robots With Noise Interference and Imbalanced Data","authors":"Rourou Li;Tangbin Xia;Yimin Jiang;Jianhua Wu;Xiaolei Fang;Nagi Gebraeel;Lifeng Xi","doi":"10.1109/TIM.2025.3540131","DOIUrl":"https://doi.org/10.1109/TIM.2025.3540131","url":null,"abstract":"Fault diagnosis (FD) of industrial robots (IRs) plays an increasingly indispensable role in modern manufacturing. Fault-related component obscurity by strong noise, feature exploitation insufficiency with scarce fault samples, and limited physical interpretation hinder existing diagnostic models’ application to IRs. A deep, complex wavelet denoising network (DCWDN) is, thus, proposed to achieve high-performance and interpretable FD with robustness against noise and class-imbalanced data. Hereinto, a dual-tree cascade autoencoder with trainable convolutional filters is constructed. Significantly, complex wavelet conditions such as orthogonality, approximate analyticity, and sparsity are imposed on the filters to structure their optimization. Meanwhile, shrinkage-based denoising with learnable thresholds is integrated to suppress noise-related components. The proposed DCWDN organically combines the data adaptivity of deep learning (DL) and wavelets’ time-frequency representation ability. Its interpretability is embodied through the explainable structure, learned scientifically meaningful filters, and extracted coefficients with explicit fault indications. Case studies on real IR datasets and experimental drivetrain benchmarks are conducted to demonstrate the effectiveness and superiority of the proposed method.","PeriodicalId":13341,"journal":{"name":"IEEE Transactions on Instrumentation and Measurement","volume":"74 ","pages":"1-11"},"PeriodicalIF":5.6,"publicationDate":"2025-02-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143465760","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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 : 2025-02-10DOI: 10.1109/TIM.2025.3540141
Jian Xiao;Chang Liu;Xi Wang;Zhenya Wang;Xing Wu
The foundation of contemporary industry, mechanical equipment is essential to many critical fields, including industrial production, transportation, energy generation, and utilization. Residual neural networks (ResNet) have made considerable progress in the identification of mechanical equipment faults in the past few years. However, due to its relatively high model complexity and large number of parameters, ResNet is challenging to implement in industrial settings and deploy on embedded platforms with limited resources. As a result, this research suggests a knowledge-transfer-based approach for diagnosing equipment faults in variable conditions that use enhanced ResNets. This method adopts a knowledge distillation architecture, where the teacher network uses an improved ResNet50 network to enhance feature information mining capability; the student network uses a simplified depthwise separable convolutional neural network (DSCNN) to achieve lightweight deployment by reducing network size. First, the short-time Fourier transform (STFT) is used to convert the gathered variable condition data into 2-D time-frequency pictures, which are then fed into the neural network model. Then, the ResNet50 model is utilized as the teacher network model and its design is gotten to the next level. Next, a simplified DSCNN and knowledge distillation method are used to train a more lightweight and efficient student network, transferring the complex knowledge from the teacher network to the lightweight depthwise separable (DS) convolutional network. Finally, utilizing the rolling bearing experimental dataset under varied conditions, the suggested method is experimentally validated. The findings demonstrate that with a 96.14% accuracy rate, the computational and parameter complexity was reduced by approximately 238 times, and the runtime was shortened nearly three times. In addition, experimental validation is conducted on a homemade RV gearbox fault simulation test bench. The experimental results demonstrate that the method can achieve robust and efficient fault diagnosis results in different conditions and practical application scenarios.
{"title":"A Fault Diagnosis Method for Variable Condition Equipment Based on Knowledge Transfer and Improved Residual Neural Networks","authors":"Jian Xiao;Chang Liu;Xi Wang;Zhenya Wang;Xing Wu","doi":"10.1109/TIM.2025.3540141","DOIUrl":"https://doi.org/10.1109/TIM.2025.3540141","url":null,"abstract":"The foundation of contemporary industry, mechanical equipment is essential to many critical fields, including industrial production, transportation, energy generation, and utilization. Residual neural networks (ResNet) have made considerable progress in the identification of mechanical equipment faults in the past few years. However, due to its relatively high model complexity and large number of parameters, ResNet is challenging to implement in industrial settings and deploy on embedded platforms with limited resources. As a result, this research suggests a knowledge-transfer-based approach for diagnosing equipment faults in variable conditions that use enhanced ResNets. This method adopts a knowledge distillation architecture, where the teacher network uses an improved ResNet50 network to enhance feature information mining capability; the student network uses a simplified depthwise separable convolutional neural network (DSCNN) to achieve lightweight deployment by reducing network size. First, the short-time Fourier transform (STFT) is used to convert the gathered variable condition data into 2-D time-frequency pictures, which are then fed into the neural network model. Then, the ResNet50 model is utilized as the teacher network model and its design is gotten to the next level. Next, a simplified DSCNN and knowledge distillation method are used to train a more lightweight and efficient student network, transferring the complex knowledge from the teacher network to the lightweight depthwise separable (DS) convolutional network. Finally, utilizing the rolling bearing experimental dataset under varied conditions, the suggested method is experimentally validated. The findings demonstrate that with a 96.14% accuracy rate, the computational and parameter complexity was reduced by approximately 238 times, and the runtime was shortened nearly three times. In addition, experimental validation is conducted on a homemade RV gearbox fault simulation test bench. The experimental results demonstrate that the method can achieve robust and efficient fault diagnosis results in different conditions and practical application scenarios.","PeriodicalId":13341,"journal":{"name":"IEEE Transactions on Instrumentation and Measurement","volume":"74 ","pages":"1-16"},"PeriodicalIF":5.6,"publicationDate":"2025-02-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143465675","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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 : 2025-02-10DOI: 10.1109/TIM.2025.3540142
Qiangsheng Gao;Ka Ho Cheng;Li Qiu;Zijun Gong
Relative localization is an essential part of autonomous multiagent systems. In this study, drawing inspiration from animals that achieve collective behaviors solely through individual perception of relative information, we propose an infrastructure-free 2-D distributed relative localization framework utilizing onboard ranging sensors. We start with system modeling, based on which optimal sensor configuration and algorithm design are conducted. Subsequently, we perform a thorough performance analysis and validate the overall system design through field tests using unmanned ground vehicles (UGVs) equipped with ultrawideband (UWB) ranging sensors and microcontroller units (MCUs) onboard. Contributions include the following: the geometric dilution of precision (GDOP) and Cramér-Rao lower bound (CRLB) are derived; a novel Euclidean distance matrix (EDM)-based trilateration (EDMT) algorithm and a maximum likelihood estimation (MLE) algorithm are proposed; the computational complexities of the proposed algorithms are compared with the state-of-the-art methods; and comprehensive simulation and field tests are conducted to validate the viability of the proposed framework. Two use cases are considered: to localize a target sensor and to localize an agent. The theoretical, numerical, and experimental results will shed light on the design and optimization of infrastructure-free relative localization systems, and our proposed framework holds potential for future extensions to 3-D scenarios, different unmanned vehicle platforms, and multirobot cooperative systems.
{"title":"Infrastructure-Free Relative Localization: System Modeling, Algorithm Design, Performance Analysis, and Field Tests","authors":"Qiangsheng Gao;Ka Ho Cheng;Li Qiu;Zijun Gong","doi":"10.1109/TIM.2025.3540142","DOIUrl":"https://doi.org/10.1109/TIM.2025.3540142","url":null,"abstract":"Relative localization is an essential part of autonomous multiagent systems. In this study, drawing inspiration from animals that achieve collective behaviors solely through individual perception of relative information, we propose an infrastructure-free 2-D distributed relative localization framework utilizing onboard ranging sensors. We start with system modeling, based on which optimal sensor configuration and algorithm design are conducted. Subsequently, we perform a thorough performance analysis and validate the overall system design through field tests using unmanned ground vehicles (UGVs) equipped with ultrawideband (UWB) ranging sensors and microcontroller units (MCUs) onboard. Contributions include the following: the geometric dilution of precision (GDOP) and Cramér-Rao lower bound (CRLB) are derived; a novel Euclidean distance matrix (EDM)-based trilateration (EDMT) algorithm and a maximum likelihood estimation (MLE) algorithm are proposed; the computational complexities of the proposed algorithms are compared with the state-of-the-art methods; and comprehensive simulation and field tests are conducted to validate the viability of the proposed framework. Two use cases are considered: to localize a target sensor and to localize an agent. The theoretical, numerical, and experimental results will shed light on the design and optimization of infrastructure-free relative localization systems, and our proposed framework holds potential for future extensions to 3-D scenarios, different unmanned vehicle platforms, and multirobot cooperative systems.","PeriodicalId":13341,"journal":{"name":"IEEE Transactions on Instrumentation and Measurement","volume":"74 ","pages":"1-14"},"PeriodicalIF":5.6,"publicationDate":"2025-02-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143489067","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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：