Aug 2026· Applied Sciences· Vol 16, pp. 7891· 0 citations· 37 references
TL;DR
A Hybrid Knowledge-DL network (HKD-SVM) that utilizes Support Vector Machine (SVM) as the classifier, making the network well-suited for nonlinear, high-dimensional classification scenarios with limited training samples, which are common in power system applications.
Abstract
Accurate detection of forced oscillations is important for the stable operation of power systems. The method based on prior knowledge relies on manual feature extraction, which has limited ability to characterize non-stationary signals. While deep learning (DL) methods can automatically learn features, they may overlook the physical mechanisms of power systems, potentially leading to misjudgments. We propose a Hybrid Knowledge-DL network (HKD-SVM) that utilizes Support Vector Machine (SVM) as the classifier. In our method, Discrete Wavelet Transform (DWT) is used to represent the time–frequency structure of the input signals, and DL features are extracted by Convolutional Neural Network (CNN) from this time–frequency representation. These learned features are subsequently fused with prior knowledge features that carry explicit physical interpretations, thereby constructing a more discriminative feature representation space. Finally, SVM is adopted as the classifier, making the network well-suited for nonlinear, high-dimensional classification scenarios with limited training samples, which are common in power system applications. Experiments on both simulated and real-world phasor measurement unit (PMU) data demonstrate that HKD-SVM outperforms purely data-driven and purely knowledge-driven methods. The proposed method provides an effective solution for power system oscillation detection.
Brake fault diagnosis is critical for enhancing the safety and reliability of vehicles. This paper presents an approach of utilizing spectrogram deep learning, where the spectrogram is defined as the representation of the signal in terms of timefrequency plane. To diagnose faults in NVH signals, a novel framework combining deep learning models and unsupervised clustering methods is proposed, especially focusing on spectrograms. Raw multi-channel NVH data are divided into overlapping temporal segments and converted into timefrequency spectrogram images via Short-Time Fourier Transform (STFT). Afterward, intrinsic patterns of vibrations are extracted by K-means clustering, and a two-dimensional convolutional neural network (2D CNN) is trained based on the extracted clusters. Experimental datasets are formed by converting NVH signals into 37,962 spectrogram images through 38 sensor channels. Classification accuracy, precision, recall, and F1-score values reached up to 98.41% and more than 0.98 for all clusters, and the areas under ROC curves are close to 1.0 for all clusters.
Zakariya Abderrahmani, Ismail Anoiri, Mourad Kaddiri et al.· IEEE International Conferenc...· 0 citations
Motor fault identification is often constrained by scarce labeled samples and the limited representation capability of a single time–frequency transform. Conventional CNN–Softmax models may also produce unstable decision boundaries under small-sample conditions. To address these issues, this paper proposes a motor fault identification method based on the fusion of fixed-resolution and multiscale time–frequency features. Each vibration segment is transformed into short-time Fourier transform (STFT) and synchrosqueezed wavelet transform (SWT) maps. Two parallel convolutional branches extract complementary features, which are fused by element-wise addition and classified using a radial basis function support vector machine. Experiments on the HUST motor multimodal fault dataset show that the proposed method achieves 100% accuracy under the conventional 70%/30% train–test split. When the training proportion is reduced to 20%, 15%, 10%, and 5%, the corresponding accuracies remain at 99.46%, 99.10%, 98.78%, and 96.77%, respectively. Across operating speeds of 5, 10, 20, and 30 Hz, the average accuracies reach 98.75% and 94.61% under the 20% and 5% training conditions. The model also maintains 100% accuracy at signal-to-noise ratios of 15 dB and above. These results demonstrate that complementary time–frequency feature fusion combined with maximum-margin classification improves identification accuracy and decision-boundary stability under limited training data.
Jingyu Yang, Jikai Xu, Li Peng et al.· Machines· 0 citations
In this paper, a hybrid TransFAD system, which abides by voltage and current measurements is introduced and uses these measurements as a basis to detect and classify transmission line faults. Protection of a power system involves fast and reliable fault detection. But in the real world, the noise and transient behavior as well as class imbalance can impair the operation of traditional approaches. Current methods utilize either handcrafted characteristics or individual learning models, allowing them to be inadequate to allow characterization of both sub-instantaneous electrical attributes and temporal fault dynamics. To overcome it, TransFAD integrates classical machine-learning-based classifiers with a CNN-LSTM-based deep learning model. Representation of system imbalance is done using engineered electrical features, and the CNN-LSTM learns spatial and temporal patterns directly on sequence of signal signals. A weighted ensemble strategy is used to combine the predictions of the individual models in order to enhance robustness and accuracy. The efficacy of TransFAD is shown with reference to a publicly accessible electrical faults detection dataset, in which it is assessed by different faults and non-fault conditions and contrasted with standalone models.
Tejinder Kaur, Abirami.R, Anitha Rani Palakayala et al.· 2026 International Conferenc...· 0 citations
To address the insufficient feature representation of conventional data-driven transient stability assessment (TSA) models and their limited adaptability to changes in power-system operating conditions, which result in inadequate assessment accuracy, this paper proposes a TSA method based on high-level sample feature extraction and model updating. First, a self-supervised contrastive random feature perturbation model for transient stability assessment, called TSA-SCRF, is developed. By introducing random perturbations into steady-state power flow features and employing contrastive learning, the proposed model extracts robust deep feature representations while preserving fault-type information. Second, a boundary-aware ensemble support vector machine (BAESVM) is constructed, which exploits multiple kernel functions to learn complementary discriminative information and dynamically assigns classifier weights according to both classifier performance and the samples’ decision distances. Finally, high-value newly acquired samples are selected based on sample uncertainty and, together with the support vectors of the original model, are utilized for model updating. Case studies conducted on a provincial power grid in China demonstrate that the proposed method improves the accuracy of data-driven transient stability assessment and enhances the model’s adaptability to changes in power-system operating conditions.
Shuolin Zhang, Yue Yu, Ye Tao et al.· Processes· 0 citations
Bearing monitoring is typically based on signals acquired from accelerometers, where the operational status is inferred by analyzing potential signal features. In practical situations, fault features are often weak and obscured by background noise, which significantly increases the difficulty of weak feature extraction and condition diagnosis. Sparse representation (SR) has been widely adopted for feature extraction from vibration signals. However, classical algorithms generally exhibit limited robustness under low signal-to-noise ratio (SNR) conditions. To address this issue, a sparse impulse feature learning (SIFL) method is proposed in this article. First, a shift kurtosis spectrum (SKS) method is developed to automatically identify the potential number of impulse components and their initial center frequencies. Second, a composite convolutional constraint is constructed and incorporated into convolutional dictionary learning (CDL). Bandwidth and sparse nonconvex constraints are imposed simultaneously during optimization. Furthermore, an optimization strategy for the constrained bandwidth and center frequency is proposed. During iteration, SIFL adaptively updates the bandwidths and spectrum locations of different atoms to mine hidden features. Meanwhile, a differential envelope energy (DEE) is proposed to effectively evaluate the performance of different models. Compared with SR and its variants, SIFL achieves superior performance in both fault feature frequency identification and amplitude integrity preservation.
Wei Lu, Changkun Han, Li Qiu et al.· IEEE Transactions on Instrum...· 0 citations
In this study, a hybrid deep learning architecture is proposed for robust vibration-based fault diagnosis in industrial machinery by jointly modeling time-domain, frequency-domain, and temporal dynamics. In the time domain, convolutional layers combined with an Enhanced Gated Attention (EGA) mechanism emphasize informative signal components while suppressing noise. Temporal evolution is modeled using Neural Ordinary Differential Equations (Neural ODEs), enabling smooth and stable continuous-time feature representations. In parallel, a Fourier Neural Operator (FNO) extracts frequency-domain characteristics, augmented with gated attention to focus on fault-related spectral patterns. Long Short-Term Memory (LSTM) layers capture long-range dependencies, while Squeeze-and-Excitation (SE) blocks adaptively recalibrate channel-wise feature responses. A multi-scale attention-based fusion module integrates domain-specific representations and auxiliary features to enhance discrimination under varying operating conditions. The proposed model is evaluated on the SUBFv1.0 dataset through extensive ablation studies and experiments under multiple noise levels, achieving 98.41% accuracy in noise-free conditions and maintaining performance above 91% even at 5 dB SNR. Unlike existing multi-path approaches that combine heterogeneous features in a loosely coupled or discrete manner, the proposed architecture uniquely integrates multi-domain feature learning with bidirectional attention mechanisms and continuous-time temporal dynamics, enabling coherent cross-domain interaction and robust fault characterization.
Canan Taştimur· Information Technology and C...· 0 citations