Yancai Xiao

Deep learning-based fault diagnosis of induction motors often degrades under varying operating conditions because labeled fault data are limited and source and target domains may follow different distributions. To address this issue, this paper proposes a Residual Maximum Kernel Discrepancy Adaptation Network (RMKDAN) for cross-condition diagnosis of stator inter-turn short-circuit faults in a three-phase induction motor using the stator current signal measured from one phase. The proposed framework employs a lightweight one-dimensional residual network for hierarchical feature extraction and introduces multi-level Multi-Kernel Maximum Mean Discrepancy (MK-MMD) alignment to reduce distribution discrepancies between source and target domains. A dynamic decay strategy is further applied to the MK-MMD weights to emphasize domain alignment during early training while gradually strengthening discriminative learning. To evaluate the method, an induction-motor fault testbench was constructed, and current signals were collected under different motor frequencies and mechanical loads across eight health conditions. Across the five cross-condition transfer tasks, RMKDAN achieves an average classification accuracy of 77.5%. When the within-domain reference task is also included, the average accuracy across all six diagnostic tasks reaches 81.2%, exceeding DCTLN and FTNN by 4.6 and 5.0 percentage points, respectively. These results demonstrate the effectiveness of RMKDAN for fault diagnosis under changing operating conditions.

The arbitrary prescribed‐time attitude control problem for rigid spacecraft is addressed in this article. The definition of arbitrary prescribed‐time stabilization is introduced. Furthermore, by utilizing the properties of a non‐autonomous differential equation, the Lyapunov sufficient conditions are provided for the stability of nonlinear systems within an arbitrary prescribed time. An arbitrary prescribed‐time control method is developed for MIMO nonlinear systems via the backstepping technique. The designed control framework guarantees that the transformation errors and the states of the controlled system converge to zero within an arbitrary prescribed time, which is independent of the parameters and the initial conditions. Ultimately, the simulation results indicate that the methodology developed in this article is effective and practically useful.

Accurate coupling misalignment diagnosis is essential for reliable wind turbine drivetrain operation. However, vibration signals under variable-speed conditions exhibit strong nonstationarity, order-related feature drift, and coupling between speed-related and fault-related information, limiting diagnostic robustness. This paper proposes a dual-domain representation and speed-decoupled learning framework that addresses speed variation at three levels: signal alignment, complementary representation, and feature decoupling. First, vibration signals are angle-synchronously resampled using rotational speed information, and a one-dimensional (1D) order amplitude spectrum and a two-dimensional (2D) angle–order map are constructed. A dual-branch network is then developed, in which the 1D branch learns global order-related features and the 2D branch captures local angle–order textures. A hybrid attention module at the high-level stage of the 2D branch refines fault-sensitive regions and semantic channels. Next, a cross-domain attention fusion module integrates complementary features from both branches. Finally, a speed-decoupled multi-task learning strategy separates fault-related and speed-related features to reduce interference from speed variation. Experiments are conducted on a slf-built variable-speed coupling-misalignment test platform, with further evaluations on the KAIST and Ottawa variable-speed bearing datasets. The proposed method achieves diagnostic accuracies of 98.98%, 99.27%, and 99.66% on the three datasets, respectively. The public bearing datasets provide evidence of the framework’s applicability to other variable-speed rotating-machinery diagnosis tasks rather than direct validation of coupling-misalignment diagnosis. Results on the self-built dataset demonstrate the effectiveness of the proposed framework under the tested laboratory conditions and its potential for wind turbine drivetrain condition monitoring.

Remote fault-diagnosis services in dynamic edge–cloud environments depend on timely monitoring-data upload, remote inference, and result delivery, making their communication layer sensitive to variations in available bandwidth, link latency, and packet loss rate. This study addresses the network-state forecasting layer that supports such services rather than the fault-classification model itself. We propose Horizon-Aware Transformer–Ridge Fusion (HATR-Fusion), which combines a nonlinear Transformer expert with a low-variance ridge-regression expert for joint short- and long-horizon forecasting. Historical available bandwidth, link latency, packet loss rate, mobility, and offered load are used as inputs. Preprocessing statistics are estimated using the training set only, and validation-calibrated convex fusion weights are frozen before test inference. Experiments on controlled synthetic trajectories from eight links sampled at 1-min intervals, using 10 neural-network initialization seeds and ten baselines including DLinear and iTransformer, show that HATR-Fusion reduces mean absolute error (MAE) relative to the standalone Transformer by 6.94–8.31% over the 10-min horizon and by 3.02–4.40% over the 60-min horizon, with all six paired improvements remaining significant after Holm correction. Against iTransformer, HATR-Fusion is significantly more accurate for short-horizon bandwidth and latency, whereas iTransformer is significantly more accurate for long-horizon latency and packet loss; short-horizon packet loss and long-horizon bandwidth are not significantly different after Holm correction. The six-task mean normalized mean absolute error (NMAE) is 0.07106 for HATR-Fusion and 0.07047 for iTransformer, indicating comparable overall accuracy with task-dependent differences between the two methods. Ablation results show complementary short- and long-range contributions from ridge regression and Transformer, while input-quality sensitivity analysis identifies a limitation of the fixed fusion weights under corrupted or missing history. The conclusions are therefore restricted to scenarios with relatively stable input quality and distribution shifts comparable to those evaluated in this study.

With the increasing digitization and intelligence of next-generation spacecraft, telemetry data exhibit increasingly complex characteristics, including dynamic inter-variable coupling, multivariate heterogeneity, and non-stationarity. Conventional anomaly detection approaches often struggle to effectively model the high-dimensional temporal-spatial dependencies inherent in such data. To overcome the limitations of existing deep learning methods in modeling dynamic topological evolution and heterogeneous telemetry, this paper proposes a dynamic graph temporal-spatial interaction multivariate anomaly detection (DGTSI-MAD) method. A novel dynamic graph construction module is designed to automatically learn the topology among telemetry variables evolving over time based on segmented data without prior knowledge, thus capturing the dynamic spatial dependencies. Furthermore, a heterogeneous feature differentiation strategy is introduced to map variables with different attributes into a unified high-dimensional latent space while preserving the semantic and temporal logic of discrete status variables. By integrating the temporal-spatial interaction mechanism and a dual-branch decision network, the model can simultaneously capture long-range temporal dependence and fine-grained spatial correlation, while maintaining high sensitivity to abrupt and distributed anomalies. Experimental results on a launch-vehicle Inertial Measurement Unit dataset and two public spacecraft telemetry benchmarks demonstrate the effectiveness of DGTSI-MAD. This method provides a more precise data-driven anomaly detection solution for multivariate telemetry data.

Wind turbine (WT) misalignment fault diagnosis is constrained by critical signal processing challenges: weak fault features, intense background noise, and poor generalization. This study proposes a lightweight method for high-precision fault diagnosis. A fixed-threshold wavelet denoising method with the scene-specific pre-optimized parameter a (0 < a ≤ 1.3) is proposed: the parameter a is determined via offline grid search using the feature retention rate (FRR) as the objective function for typical wind farm operating scenarios. A multi-scale depthwise separable CNN (MS-DSCNN) captures multi-scale spatial features via 3 × 1 and 5 × 1 kernels, reducing computational complexity by 73.4% versus standard CNNs. An attention-based minimal peephole LSTM (AttMPLSTM) enhances temporal feature measurement, using minimal peephole connections for long-term dependencies and channel attention to weight fault-relevant signals. Joint L1–L2 regularization mitigates overfitting and environmental interference, improving model robustness. Validated on a WT test bench, the Adams simulation dataset, and the CWRU benchmark, the model achieves a 90.2 ± 1.4% feature retention rate (FRR) in signal processing, an over 98% F1-score for fault classification, and over 99% accuracy. With 2.5 s single-epoch training and a 12.8 ± 0.5 ms single-sample inference time, the reduced parameters enable real-time deployment in embedded systems, advancing signal processing for rotating machinery fault diagnosis.

As wind power generation technology continues to advance, issues related to misalignment faults in wind turbine drive systems have garnered increasing attention. Under actual operating conditions, vibration signals obtained from wind turbine often exhibit low-frequency, nonlinear, and non-steady-state characteristics, and are easily influenced by factors such as fluctuations in wind speed, mechanical friction, and changes in environmental temperature. This complex background significantly reduces the stability and accuracy of traditional fault diagnosis methods. To address these challenges, this paper proposes a novel fault diagnosis model that combines the residual hybrid convolutional attention module with the stacked denoising autoencoder (SDAE), and designs a multi-stage denoising training strategy (MSD). The RHCBAM integrates multi-scale convolutional structures, multi-channel attention mechanisms, and residual connection mechanisms, significantly enhancing the ability to capture weak and complex fault features and improving training stability. By enhancing feature extraction depth, it addresses issues such as gradient vanishing and insufficient fusion capabilities that traditional DAE models encounter when processing complex non-stationary signals. The proposed MSD achieves robustness enhancement during the unsupervised training phase by constructing typical noise scenarios such as Gaussian white noise, periodic interference, and pulse noise, combined with noise samples of different intensity levels, significantly improving its generalization ability in real complex environments and effectively addressing multi-source, multi-level noise pollution issues in wind turbine operating environments. Extensive experiments on wind turbine drive system fault test benches and the CWRU bearing fault dataset demonstrate that the proposed method achieves fault identification accuracy rates of 99.64% and 99.70%, respectively, significantly outperforming existing mainstream deep learning models, thereby validating its diagnostic performance and key feature extraction capabilities in complex noise environments.

Aiming at the noise interference problem in wing fatigue tests, this paper improves the traditional LMS algorithm using the variational Versoria function and the variational Gaussian function. Additionally, this paper proposes a variable step‐size LMS (VSS‐LMS) filtering algorithm based on the composite function (CVSS‐LMS). The composite function combines the variational Versoria function and the variational Gaussian function to describe the nonlinear relationship between the iteration step size and the error. To adapt to environments with different signal‐to‐noise ratios, the algorithm replaces the fixed parameters with a combination of current and previous errors, thus enabling adaptive adjustment of the parameters. Moreover, a step‐size dynamic constraint rule is proposed to further improve the stability of the algorithm. The algorithm is normalized using a combination of the cumulative sum of error squares, the mean square error (MSE), and the power of the input signal, which reduces the sensitivity to the input signal amplitude. The above parts finally constitute the adaptive CVSS‐LMS (ACVSS‐LMS) filtering algorithm. The convergence of the ACVSS‐LMS algorithm is verified through theoretical derivation. The ACVSS‐LMS algorithm is experimentally analyzed by using the simulation data generated by MATLAB and the actual data collected from the wing fatigue test, and the results show that the ACVSS‐LMS algorithm proposed in this paper has a faster convergence speed and lower steady‐state error compared to other algorithms.

The key circuits of servo channel loading card with high‐precision data acquisition and output functions for wing fatigue testing are studied and improved in this paper. To solve the problem of large offset voltage in some op‐amps, non‐inverting combined amplifier circuit (NCAC) and inverting combination amplifier circuit (ICAC) are proposed by utilizing precision op‐amps combined with wide voltage and high‐power output op‐amps. In the force signal acquisition section, a non‐inverting combined zeroing filter circuit (NCZFC) is proposed to improve the precision of signal acquisition. In the excitation source output section, a non‐inverting combined excitation output circuit (NCEOC) is proposed to improve the precision of the positive excitation output; an inverting combined excitation output circuit (ICEOC) is also proposed, which improves the precision of the negative excitation output and also realizes the inversion of the input and output signals. In the servo drive output section, a non‐inverting combined voltage and current conversion circuit is proposed to improve the precision of voltage and current outputs, which has two modes of non‐inverting combined voltage output (NCVO) and non‐inverting combined current output (NCIO) modes. Simulation experiments in Multisim and actual tests have been carried out on the above improved circuits to verify the stability and precision.