Xianhong Wu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Xianhong Wu has a total of 25 co-authors. Here are the top co-authors by number of shared publications:
- Juncai Song
Anhui University
10 shared publications - Siliang Lu
Anhui University
9 shared publications - Xiaoxian Wang
Anhui University
8 shared publications - Zhiyong Hu
Anhui University
3 shared publications - Fei Li
College of Internet, Anhui University, Hefei, China
2 shared publications - Jing Wu
School of Internet, Anhui University, Hefei, China
2 shared publications
Latest Publications
- 15 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Transportation Electrification
- Vol. 12 (1) , 811-824
Demagnetization Fault Diagnosis of a Permanent Magnet Synchronous Linear Motor Based on Noisy-KAN on an Edge Computing System
A noise-enhanced demagnetization fault (DF) diagnosis method for permanent magnet synchronous linear motors (PMSLMs) based on edge computing is proposed, aiming to address the limitations of the existing approaches, including poor real-time performance, low noise immunity, and high computational resource consumption. First, a simulation and experimental platform for PMSLM DFs is established. Subsequently, stray magnetic field signals in the external space are collected. Finally, a Noisy-Kolmogorov–Arnold network (Noisy-KAN) is proposed to extract the features for fault diagnosis. The proposed method is evaluated through comparative experiments with other baseline models. Training and testing are conducted on the Jetson Orin NX, while inference is performed on the Raspberry Pi (RPi) 4B. The Noisy-KAN model requires only 1.13 MB of memory occupation and 4.51-MB storage space, making it easily deployable on an ARM Cortex-A platform. The inference time per sample is 0.412 and 2.9 ms on Jetson and RPi, respectively. The average accuracy of the proposed method is 98.24% and 90.11% under noise-free and different noise levels, which is 2.36% and 7.86% higher than those of the baseline methods. Results demonstrate that the Noisy-KAN exhibits strong noise immunity and reduced computational overhead, enabling its efficient deployment on edge devices for accurate fault detection.
- 15 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Transportation Electrification
- Vol. 12 (1) , 811-824
Demagnetization Fault Diagnosis of a Permanent Magnet Synchronous Linear Motor Based on Noisy-KAN on an Edge Computing System
A noise-enhanced demagnetization fault (DF) diagnosis method for permanent magnet synchronous linear motors (PMSLMs) based on edge computing is proposed, aiming to address the limitations of the existing approaches, including poor real-time performance, low noise immunity, and high computational resource consumption. First, a simulation and experimental platform for PMSLM DFs is established. Subsequently, stray magnetic field signals in the external space are collected. Finally, a Noisy-Kolmogorov–Arnold network (Noisy-KAN) is proposed to extract the features for fault diagnosis. The proposed method is evaluated through comparative experiments with other baseline models. Training and testing are conducted on the Jetson Orin NX, while inference is performed on the Raspberry Pi (RPi) 4B. The Noisy-KAN model requires only 1.13 MB of memory occupation and 4.51-MB storage space, making it easily deployable on an ARM Cortex-A platform. The inference time per sample is 0.412 and 2.9 ms on Jetson and RPi, respectively. The average accuracy of the proposed method is 98.24% and 90.11% under noise-free and different noise levels, which is 2.36% and 7.86% higher than those of the baseline methods. Results demonstrate that the Noisy-KAN exhibits strong noise immunity and reduced computational overhead, enabling its efficient deployment on edge devices for accurate fault detection.
- 16 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-15
Fault Diagnosis of Linear Motion Guide System in PMSLM via STFC-RN Implemented on Edge Computing Node
A state monitoring and fault diagnosis method is proposed to achieve real-time state monitoring of the linear guide and balls of a permanent magnet synchronous linear motor (PMSLM) and ensure the reliable operation of the equipment. A PMSLM linear guide fault diagnosis platform is constructed to collect fault signals of the linear guide and balls. A novel deep learning framework, namely, Short-time Fourier convolution residual network (STFC-RN) is put forward, which integrates a Short-time Fourier convolution (STFC) layer in the first layer of the model to extract the time–frequency features of vibration signals. The main structure of STFC-RN is the residual structure containing backbone and branch. The backbone is mainly composed of residual blocks stacked to learn features extracted by the STFC layer. The branch consists of the depth-wise separable (DS) layer and the squeeze-and-excitation (SE) attention mechanism, realizing the reuse of primary features and reducing the computational burden. The model achieved a fault signal recognition rate of 97.3% at a speed of 600 mm/s, outperforming other models such as CNN, ResNet, DenseNet, and VGG. Finally, the model is deployed to an edge computing device for real-time condition monitoring of linear guide system.
- 16 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-15
Fault Diagnosis of Linear Motion Guide System in PMSLM via STFC-RN Implemented on Edge Computing Node
A state monitoring and fault diagnosis method is proposed to achieve real-time state monitoring of the linear guide and balls of a permanent magnet synchronous linear motor (PMSLM) and ensure the reliable operation of the equipment. A PMSLM linear guide fault diagnosis platform is constructed to collect fault signals of the linear guide and balls. A novel deep learning framework, namely, Short-time Fourier convolution residual network (STFC-RN) is put forward, which integrates a Short-time Fourier convolution (STFC) layer in the first layer of the model to extract the time–frequency features of vibration signals. The main structure of STFC-RN is the residual structure containing backbone and branch. The backbone is mainly composed of residual blocks stacked to learn features extracted by the STFC layer. The branch consists of the depth-wise separable (DS) layer and the squeeze-and-excitation (SE) attention mechanism, realizing the reuse of primary features and reducing the computational burden. The model achieved a fault signal recognition rate of 97.3% at a speed of 600 mm/s, outperforming other models such as CNN, ResNet, DenseNet, and VGG. Finally, the model is deployed to an edge computing device for real-time condition monitoring of linear guide system.
- 5 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Informatics
- Vol. 21 (3) , 2214-2223
DPMSLM Demagnetization Fault Diagnosis Based on Deep Feature Fusion of External Stray Flux Signal
To detect the demagnetization fault (DF) of a dual-sided permanent magnet synchronous linear motor, a new method based on deep feature fusion of external stray flux signal (ESFS) is proposed. First, finite element models under ideal materials and assembly conditions are established to extract ESFS to reflect DF information. Second, Markov transition field and recurrence plot transform 1-D signals into 2-D images, to realize DF feature visual enhancement. Low-rank representation networks can merge the advantages of both methods by image fusion. Then, an accurate diagnosis framework, as efficient channel attention-MobileNetV3, is proposed to conduct deep feature extraction and realize diagnosis in both qualitative fault type classification and fault degree evaluation aspects. The classification accuracy reaches 98.50%, and the evaluation index R2 reaches 0.96, superior to other frameworks. Finally, a tunnel magnetoresistance sensor is applied to realize ESFS noninvasive online measurement, and an experimental platform is built to certify the superiority.
- 5 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Informatics
- Vol. 21 (3) , 2214-2223
DPMSLM Demagnetization Fault Diagnosis Based on Deep Feature Fusion of External Stray Flux Signal
To detect the demagnetization fault (DF) of a dual-sided permanent magnet synchronous linear motor, a new method based on deep feature fusion of external stray flux signal (ESFS) is proposed. First, finite element models under ideal materials and assembly conditions are established to extract ESFS to reflect DF information. Second, Markov transition field and recurrence plot transform 1-D signals into 2-D images, to realize DF feature visual enhancement. Low-rank representation networks can merge the advantages of both methods by image fusion. Then, an accurate diagnosis framework, as efficient channel attention-MobileNetV3, is proposed to conduct deep feature extraction and realize diagnosis in both qualitative fault type classification and fault degree evaluation aspects. The classification accuracy reaches 98.50%, and the evaluation index R2 reaches 0.96, superior to other frameworks. Finally, a tunnel magnetoresistance sensor is applied to realize ESFS noninvasive online measurement, and an experimental platform is built to certify the superiority.
- 11 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Transportation Electrification
- Vol. 11 (2) , 5999-6011
Accurate Detection and Evaluation of the Airgap Asymmetry Fault in DS-PMSLM Based on OSVT and ECA-ENet
A new method based on magnetic signal analysis and artificial intelligence framework is proposed to detect airgap asymmetry fault (AAF) of double-sided permanent magnet synchronous linear motor (DS-PMSLM). First, 3D finite element model is established to extract external stray magnetic field signals (ESMFS) under different AAFs as efficient fault signals. Second, an optimized snowflake view transformer (OSVT) signal processing method is proposed to convert 1D ESMFS into 2D feature-enhanced images, which can realize signal visual display and characteristics enhancement. Then, a novel deep learning framework named efficient channel attention-EfficientNet (ECA-ENet) is proposed to conduct AAF feature extraction and realize precise diagnosis of AAF types and severity degrees. The AAF type classification accuracy and F1 score are as high as 99.50% and 99.74%, respectively. Furthermore, the AAF degree diagnostic index MAE and RMSE are remarkably low at 0.0720 mm and 0.0842 mm, respectively. These results indicate the proposed method is better than compared methods, such as ResNet50, ShuffleNet, EfficientNetV1, and EfficientNetV2. Finally, tunnel magnetoresistance (TMR) sensor circuit hardware is integrated into design of the motor mover module to realize ESMFS data non-invasive online measurement, and DS-PMSLM experimental platform is built to validate the superiority and robustness of the proposed method.
- 11 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Transportation Electrification
- Vol. 11 (2) , 5999-6011
Accurate Detection and Evaluation of the Airgap Asymmetry Fault in DS-PMSLM Based on OSVT and ECA-ENet
A new method based on magnetic signal analysis and artificial intelligence framework is proposed to detect airgap asymmetry fault (AAF) of double-sided permanent magnet synchronous linear motor (DS-PMSLM). First, 3D finite element model is established to extract external stray magnetic field signals (ESMFS) under different AAFs as efficient fault signals. Second, an optimized snowflake view transformer (OSVT) signal processing method is proposed to convert 1D ESMFS into 2D feature-enhanced images, which can realize signal visual display and characteristics enhancement. Then, a novel deep learning framework named efficient channel attention-EfficientNet (ECA-ENet) is proposed to conduct AAF feature extraction and realize precise diagnosis of AAF types and severity degrees. The AAF type classification accuracy and F1 score are as high as 99.50% and 99.74%, respectively. Furthermore, the AAF degree diagnostic index MAE and RMSE are remarkably low at 0.0720 mm and 0.0842 mm, respectively. These results indicate the proposed method is better than compared methods, such as ResNet50, ShuffleNet, EfficientNetV1, and EfficientNetV2. Finally, tunnel magnetoresistance (TMR) sensor circuit hardware is integrated into design of the motor mover module to realize ESMFS data non-invasive online measurement, and DS-PMSLM experimental platform is built to validate the superiority and robustness of the proposed method.
- 6 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 73 (00189456) , 1-13
Multisource Deep Feature Fusion of Optimized Symmetrized Dot Patterns for SRM Fault Diagnosis
This work proposes a new diagnostic method for high-resistance contact (HRC) and bearing faults in a switched reluctance motor (SRM) on the basis of multisource signal visual fusion and an intelligent classifier. First, an electric vehicle experimental platform with an SRM prototype is built, and the stator winding current and vibration signals are collected in a noninvasive manner. Second, an optimizing multisource signal symmetrized dot pattern (OMSSDP) method is proposed to characterize fault signals and realize 2-D visualization enhancement under different fault states. This method can also process multisource signal and fuse them into one image. Third, a MobileViT-ECA model is built to extract the global and local fault features of SRM. The model can effectively distinguish the type and severity degree of compound faults in SRM. Finally, comparison experiments with different time-series signal processing methods and intelligent classifiers are performed. Results prove that the proposed method can accurately identify the HRC fault phase location and bearing fault type, and its classification accuracy can reach 98.71%.
- 6 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 73 (00189456) , 1-13
Multisource Deep Feature Fusion of Optimized Symmetrized Dot Patterns for SRM Fault Diagnosis
This work proposes a new diagnostic method for high-resistance contact (HRC) and bearing faults in a switched reluctance motor (SRM) on the basis of multisource signal visual fusion and an intelligent classifier. First, an electric vehicle experimental platform with an SRM prototype is built, and the stator winding current and vibration signals are collected in a noninvasive manner. Second, an optimizing multisource signal symmetrized dot pattern (OMSSDP) method is proposed to characterize fault signals and realize 2-D visualization enhancement under different fault states. This method can also process multisource signal and fuse them into one image. Third, a MobileViT-ECA model is built to extract the global and local fault features of SRM. The model can effectively distinguish the type and severity degree of compound faults in SRM. Finally, comparison experiments with different time-series signal processing methods and intelligent classifiers are performed. Results prove that the proposed method can accurately identify the HRC fault phase location and bearing fault type, and its classification accuracy can reach 98.71%.
- 29 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 24 (11) , 18219-18231
DPMSLM Eccentricity Fault Detection Based on Multiview of Mystery Curve Transformation and Deep Feature Extraction
To detect the eccentricity fault of dual-sided permanent magnet synchronous linear motor (DPMSLM) and ensure the stable operation of the equipment, a new method based on multi-view of mystery curve transformation (MCT) and deep feature extraction is proposed to detect complex eccentricity faults of DPMSLM in this work. First, finite element analysis (FEA) calculation models of DPMSLM under different static eccentricity and dynamic eccentricity fault conditions are established to extract the external magnetic leakage signal (EMLS) as the efficient fault diagnostic signals. Second, a MCT signal processing method is proposed to convert a 1D EMLS into 3D curve and obtain 2D projections of multiple views (top, front, and side). This method achieves eccentricity fault signal visual display in 2D multi-view fusion image and realize complementary enhancement of fault characteristics. Thereafter, a novel classification deep learning framework, named SA-ConvNeXt, is proposed to conduct deep fault feature extraction and realize eccentricity faults accurate classification in fault types and severity levels. The diagnostic accuracy of SA-ConvNeXt is as high as 99.5%, which is better than those of comparison models, such as CNN, ResNet-34, ShuffleNet, and ConvNeXt. Finally, tunnel magnetoresistance (TMR) sensor circuit hardware is integrated designation with motor mover module to realize EMLS data non-invasive online measurement, and the DPMSLM experimental platform under several eccentricity faults is built to verify the superiority and robustness of the proposed method.
- 29 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 24 (11) , 18219-18231
DPMSLM Eccentricity Fault Detection Based on Multiview of Mystery Curve Transformation and Deep Feature Extraction
To detect the eccentricity fault of dual-sided permanent magnet synchronous linear motor (DPMSLM) and ensure the stable operation of the equipment, a new method based on multi-view of mystery curve transformation (MCT) and deep feature extraction is proposed to detect complex eccentricity faults of DPMSLM in this work. First, finite element analysis (FEA) calculation models of DPMSLM under different static eccentricity and dynamic eccentricity fault conditions are established to extract the external magnetic leakage signal (EMLS) as the efficient fault diagnostic signals. Second, a MCT signal processing method is proposed to convert a 1D EMLS into 3D curve and obtain 2D projections of multiple views (top, front, and side). This method achieves eccentricity fault signal visual display in 2D multi-view fusion image and realize complementary enhancement of fault characteristics. Thereafter, a novel classification deep learning framework, named SA-ConvNeXt, is proposed to conduct deep fault feature extraction and realize eccentricity faults accurate classification in fault types and severity levels. The diagnostic accuracy of SA-ConvNeXt is as high as 99.5%, which is better than those of comparison models, such as CNN, ResNet-34, ShuffleNet, and ConvNeXt. Finally, tunnel magnetoresistance (TMR) sensor circuit hardware is integrated designation with motor mover module to realize EMLS data non-invasive online measurement, and the DPMSLM experimental platform under several eccentricity faults is built to verify the superiority and robustness of the proposed method.
- 29 March 2024
- Published by Elsevier in Measurement
- Vol. 231
Air gap flux density measurement of PMSLM based on TMR sensing external stray magnetic field and CNN-LSTM
- 29 March 2024
- Published by Elsevier in Measurement
- Vol. 231
Air gap flux density measurement of PMSLM based on TMR sensing external stray magnetic field and CNN-LSTM
- 25 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 73 (00189456) , 1-12
PMSLM Eccentricity Fault Diagnosis Based on Deep Feature Fusion of Stray Magnetic Field Signals
In order to detect the eccentric fault of permanent magnet synchronous linear motor (PMSLM) safely, efficiently, and accurately, this paper proposes a diagnosis method based on fusion feature extraction and a new hybrid network model. First, according to motor topology structure characteristics, tunneling magnetoresistance (TMR) sensor integrated design scheme is put forward to measure external stray magnetic field density signal of PMSLM, which can realize non-invasive fault diagnosis. Second, 2D image coding methods, namely, Gramian Angular Field (GAF) and Markov Transition Field (MTF) are introduced to convert 1D fault signal into 2D images, which can realize fault feature enhanced display. Visual saliency map and weighted least square (VSMWLS) image fusion technology is applied to fuse GAF and MTF images, combining the advantages of both methods. It can also enhance the complementary information of fault features. Then, a hybrid multi-feature fusion classification framework called ConvNeXt-ResNet-CNN-GCNet (CRCG) is proposed to integrate different deep feature extraction capabilities of the three models. In addition, GRCG incorporates GCNet attention mechanism to capture multi channels’ dependencies to realize quantitative and refined diagnosis of eccentricity faults. Finally, PMSLM prototype experimental platform is established. The accuracy and F1 score are up to 97.4% and 97.9%, respectively. All these results are higher than other popular methods that can certify the superiority and effectiveness of this method.
- 25 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 73 (00189456) , 1-12
PMSLM Eccentricity Fault Diagnosis Based on Deep Feature Fusion of Stray Magnetic Field Signals
In order to detect the eccentric fault of permanent magnet synchronous linear motor (PMSLM) safely, efficiently, and accurately, this paper proposes a diagnosis method based on fusion feature extraction and a new hybrid network model. First, according to motor topology structure characteristics, tunneling magnetoresistance (TMR) sensor integrated design scheme is put forward to measure external stray magnetic field density signal of PMSLM, which can realize non-invasive fault diagnosis. Second, 2D image coding methods, namely, Gramian Angular Field (GAF) and Markov Transition Field (MTF) are introduced to convert 1D fault signal into 2D images, which can realize fault feature enhanced display. Visual saliency map and weighted least square (VSMWLS) image fusion technology is applied to fuse GAF and MTF images, combining the advantages of both methods. It can also enhance the complementary information of fault features. Then, a hybrid multi-feature fusion classification framework called ConvNeXt-ResNet-CNN-GCNet (CRCG) is proposed to integrate different deep feature extraction capabilities of the three models. In addition, GRCG incorporates GCNet attention mechanism to capture multi channels’ dependencies to realize quantitative and refined diagnosis of eccentricity faults. Finally, PMSLM prototype experimental platform is established. The accuracy and F1 score are up to 97.4% and 97.9%, respectively. All these results are higher than other popular methods that can certify the superiority and effectiveness of this method.
- 20 March 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 72 (00189456) , 1-12
DPMSLM Demagnetization Fault Detection Based on Texture Feature Analysis of Grayscale Fusion Image
This study investigates a novel image morphology texture feature extraction method to realize the demagnetization fault location and severity detection of double-sided permanent magnet synchronous linear motor (DPMSLM). Initially, according to the constraints of DPMSLMs topology structure, the three lines magnetic density signal in motor air gap is extracted by finite element analysis as effective fault signal. Then, the grayscale fusion image (GFI) method is introduced to transform 1D data signal to 2D fused grayscale image which can better describe the demagnetization fault information. The unique features are visualized using the image enhancement techniques, and the image morphology texture features such as the area, Euler number, perimeter operator, correlation of binary image and so on can be extracted to constitute the demagnetization fault indexes. In addition, fisher score (FS) is used for feature optimization which can reduce the feature dimension. Furthermore, the two-level multiverse optimization support vector machine (MVO-SVM) algorithm is established to conduct demagnetization fault classification. Comparison experiments with other classifiers show that the MVO-SVM has a high fault identification accuracy of more than 98.3% and low running time less than 2.57s. Finally, the motor prototype experiment results show that the proposed method can accurately identify the location and severity of DPMSLM demagnetization faults, and it is an effective and feasible method which can be applied in DPMLM batch demagnetization inspection before delivery.
- 20 March 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 72 (00189456) , 1-12
DPMSLM Demagnetization Fault Detection Based on Texture Feature Analysis of Grayscale Fusion Image
This study investigates a novel image morphology texture feature extraction method to realize the demagnetization fault location and severity detection of double-sided permanent magnet synchronous linear motor (DPMSLM). Initially, according to the constraints of DPMSLMs topology structure, the three lines magnetic density signal in motor air gap is extracted by finite element analysis as effective fault signal. Then, the grayscale fusion image (GFI) method is introduced to transform 1D data signal to 2D fused grayscale image which can better describe the demagnetization fault information. The unique features are visualized using the image enhancement techniques, and the image morphology texture features such as the area, Euler number, perimeter operator, correlation of binary image and so on can be extracted to constitute the demagnetization fault indexes. In addition, fisher score (FS) is used for feature optimization which can reduce the feature dimension. Furthermore, the two-level multiverse optimization support vector machine (MVO-SVM) algorithm is established to conduct demagnetization fault classification. Comparison experiments with other classifiers show that the MVO-SVM has a high fault identification accuracy of more than 98.3% and low running time less than 2.57s. Finally, the motor prototype experiment results show that the proposed method can accurately identify the location and severity of DPMSLM demagnetization faults, and it is an effective and feasible method which can be applied in DPMLM batch demagnetization inspection before delivery.