Motor failure can result in damage to resources and property. Real-time motor fault diagnosis technology can detect faults and diagnosis in time to prevent serious consequences caused by the continued operation of the machine. Neural network models can easily and accurately fault diagnose from vibration signals. However, they cannot notice faults in time. In this study, a deep learning model based on a temporal convolutional network (TCN) and attention is proposed for real-time motor fault diagnosis. TCN can extract features from shorter vibration signal sequences to allow the system to detect and diagnose faults faster. In addition, attention allows the model to have higher diagnostic accuracy. The experiments demonstrate that the proposed model is able to detect faults in time when they occur and has an excellent diagnostic accuracy.
This paper addresses the high-frequency oscillations issue commonly associated with the sliding mode observer (SMO) algorithm in sensorless control of surface-mounted permanent magnet synchronous motor (SPMSM) and the problem of parameter variation caused by magnetic saturation and temperature rise during motor operation. A novel information acquisition optimized recursive least squares (IAORLS) algorithm, which incorporates an IAO optimized forgetting factor, is proposed to identify motor parameters. The identified parameters are utilized in conjunction with the Super Twisting Algorithm Sliding Mode Observer (STA-SMO) for motor control. By discretizing the stator voltage equations of the SPMSM via the forward Euler method, IAORLS enables efficient parameter identification. These parameters are then integrated into the STA-SMO, allowing the observer to receive real-time updates, thereby preventing performance degradation due to parameter mismatches. Additionally, the method reduces high-frequency oscillations when the motor’s extended back electromotive force (back-EMF) is observed, improving the accuracy of rotor position and speed estimation. Simulations and experimental results confirm the effectiveness and reliability of the proposed approach.
Due to its high infectivity, COVID-19 has rapidly spread worldwide, emerging as one of the most severe and urgent diseases faced by the global community in recent years. Currently, deep learning-based diagnostic methods can automatically detect COVID-19 cases from chest X-ray images. However, these methods often rely on large-scale labeled datasets. To address this limitation, we propose a novel neural network model called CN2A-CapsNet, aiming to enhance the automatic diagnosis of COVID-19 in chest X-ray images through efficient feature extraction techniques. Specifically, we combine CNN with an attention mechanism to form the CN2A model, which efficiently mines relevant information from chest X-ray images. Additionally, we incorporate capsule networks to leverage their ability to understand spatial information, ultimately achieving efficient feature extraction. Through validation on a publicly available chest X-ray image dataset, our model achieved a 98.54% accuracy and a 99.01% recall rate in the binary classification task (COVID-19/Normal) on a six-fold cross-validation dataset. In the three-class classification task (COVID-19/Pneumonia/Normal), it attained a 96.71% accuracy and a 98.34% recall rate. Compared to the previous state-of-the-art models, CN2A-CapsNet exhibits notable advantages in diagnosing COVID-19 cases, specifically achieving a high recall rate even with small-scale datasets.
1 January 2024 in Journal of Electrotechnology, Electrical Engineering and Management
In view of the existing variable flux permanent magnet synchronous motor with limited range of magnetization and no automatic control of magnetization accuracy, an intelligent variable flux permanent magnet synchronous motor with auto-rotating permanent magnets and rotor built-in drive controller (IVFPMSM) is proposed. This paper analyzes the internal structure design and operation principle of this intelligent variable flux permanent magnet synchronous motor (IVFPMSM), and numerically analyzes the electromagnetic characteristics such as induced electric potential, winding magnetic chain, air gap flux density and weak magnetic characteristics when the permanent magnets (PMs) rotate at 0°, 30°, 45° and 90°, respectively. The results show that the PM can be rotated at any angle under the composite drive, which greatly widens the range of motor magnetization. The validity of the theory and design is verified by making a prototype and building a test platform.
This paper addresses the chattering issue commonly associated with the sliding mode observer (SMO) algorithm in sensorless control of surface-mounted permanent magnet synchronous motor (SPMSM) and the problem of parameter variation caused by magnetic saturation and temperature rise during motor operation. A novel information acquisition optimized recursive least squares (IAORLS) algorithm, which incorporates an IAO-optimized forgetting factor, is proposed to identify motor parameters. The identified parameters are utilized in conjunction with the Super Twisting Algorithm Sliding Mode Observer (STA-SMO) for motor control. By discretizing the stator voltage equations of the PMSM via the forward Euler method, IAORLS enables efficient parameter identification. These parameters are then integrated into the STA-SMO, allowing the observer to receive real-time updates, thereby preventing performance degradation due to parameter mismatches. Additionally, the method reduces chattering when the motor’s back electromotive force (EMF) is observed, improving the accuracy of rotor position and speed estimation. Simulations and experimental results confirm the effectiveness and reliability of the proposed approach.
To address the problems of high model complexity and scarce labeled data in rolling bearing fault diagnosis under industrial field conditions, this paper proposes a limited-data rolling bearing fault diagnosis method based on reptile-optimized decoupled knowledge distillation (Reptile-DKD). First, under a limited-data task setting, a teacher–student KD framework is constructed. A deep network with strong feature representation capability is adopted as the teacher model, while a lightweight network is used as the student model, thereby reducing the number of model parameters and computational complexity while maintaining the capability of fault feature extraction. Second, a DKD strategy is introduced to divide the distillation objective into target class knowledge and non-target class knowledge. This strategy enhances the effective supervision of the teacher model over the student model and reduces the interference of irrelevant class information during limited-data learning. Finally, a reptile-based update strategy is incorporated into the distillation process to dynamically update the parameters of the teacher model according to the feedback performance of the student model during task learning. As a result, the teacher model can generate supervisory information that is better adapted to the current learning state of the student model, thereby improving the quality of knowledge transfer. Experimental results show that, with 200 total training samples, the proposed method achieves mean accuracies of 96.92% and 95.74% on the 5-shot tasks of the CWRU and PU datasets, respectively, while containing only 4.15 K parameters and 0.46 M FLOPs. Comparative experiments with multiple models verify the effectiveness of the proposed method. The proposed method can be effectively applied to limited-data bearing fault diagnosis, achieving high diagnostic accuracy while maintaining a lightweight model structure. It provides a theoretical reference for fault diagnosis in resource-constrained environments.