To tackle the challenging task of extracting target signal features amidst significant interference from complex, time-varying non-stationary noise in industrial scenarios, we propose a novel time-domain all-neural beam- forming network tailored for non-stationary mechanical sound source separation, named TFANBNet. Leveraging time-frequency convolution and self-attention, TFANBNet promised robust performance in challenging acoustic environments. First, the proposed time-frequency convolution module employs parameterized complex-valued convolutional layers to simulate time-frequency transformations, thereby implicitly extracting time-frequency features. Then, the proposed adaptive attention transformer network enhances local attention interaction by integrating intermediate features, effectively enhancing long-range context modeling. Third, the beamforming module utilizes deep neural networks for beamforming weight estimation, replacing conventional noise covariance matrix computations, which have inherent performance limitations. Finally, the experimental results on the multi-channel spatial reverberation dataset synthesized from the MIMII dataset show that the proposed TFANBNet yields superior separation performance and generalization capabilities compared to the considered competitive methods.

Koopman spectral analysis plays a crucial role in understanding and modeling nonlinear dynamical systems as it reveals key system behaviors and long-term dynamics. However, the presence of measurement noise poses a significant challenge to accurately extracting spectral properties. In this work, we propose a robust method for identifying the Koopman operator and extracting its spectral characteristics in noisy environments. To address the impact of noise, our approach tackles an identification problem that accounts for both systematic errors from finite-dimensional approximations and measurement noise in the data. By incorporating Bayesian learning and Kalman smoothing, the method simultaneously identifies the Koopman operator and estimates system states, effectively decoupling these two error sources. The method's efficiency and robustness are demonstrated through extensive experiments, showcasing its accuracy across varying noise levels.

As a core component of aero-engines, the health condition of rolling bearings is crucial to the safety and stability of the aviation power system. In the fault diagnosis of aero-engine rolling bearings, the issue of insufficient fault data has long persisted. The training process of the widely introduced Generative Adversarial Networks (GANs) is difficult to control, and their generalization ability is relatively poor. Therefore, this paper proposes a data augmentation method based on denoising diffusion probabilistic model (DDPM) for aero-engine bearing fault diagnosis under few-shot. The proposed method is grounded in a well-defined probabilistic model and mathematical principles, allowing it to avoid instability and mode collapse during training. Specifically, we first propose using the synchro-squeezed wavelet transform (SST) to convert one-dimensional time-series signals into time-domain images as model input, addressing the insufficient feature extraction of traditional time-frequency analysis methods and clearly illustrating the fault characteristics of frequency variation with time. Next, we design a CRR-UNet based on residual connections to mitigate overfitting caused by insufficient data during the reverse denoising process of DDPM, thereby improving the quality of generated samples. Finally, a deep residual shrinkage network (DRSN) is employed to conduct fault diagnosis on the augmented fault dataset. The research results show that the samples generated by this method are highly similar to the original samples, outperforming existing data augmentation methods, and the fault diagnosis accuracy on the augmented dataset reaches 98%, significantly improving fault diagnosis performance.

Rolling bearings are critical components in machinery, directly affecting performance and lifespan. Traditional fault diagnosis methods rely on expert knowledge, whereas bearing data are often unlabeled and noisy. This paper proposes an unsupervised health state evaluation model based on Euclidean distance and Stacked denoising autoencoders (SDAE). The model directly extracts features and constructs a Health Indicator (HI) from unlabeled, noisy time-domain vibration signals of bearings throughout their entire lifecycle using relatively simple computations. First, preliminary feature extraction is performed using the SDAE. Second, the extracted features are utilized to construct the HI using Euclidean distance. Third, the proposed model is compared with four other models, experimental results demonstrate that the proposed method outperforms the comparison models in terms of smoothness and noise-handling capability. Furthermore, the superiority of the proposed approach is validated through a comprehensive evaluation using three metrics: monotonicity (Mon), correlation (Corr), and robustness (Rob). The proposed method effectively constructs the HI under unsupervised conditions while maintaining the advantages of simple modeling and efficient computation. However, the model cannot directly perform fault type diagnosis and remaining useful life prediction. Future research may consider integrating other classifiers or adopting advanced deep learning techniques such as Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks to enhance fault classification capabilities and improve the overall diagnostic performance.

Based on the training with the large amounts of labeled data, intelligent data-driven approaches are able to assess the health status of rotating machinery. Traditional methods usually lack the multifunctionality required for health status assessment (HSA), resulting in incomplete assessment results. Furthermore, considering the challenge of machinery data silos, existing methods use collaborative model training solutions with multiple users, which place high demands on data privacy protection due to conflict of interests. To tackle this issue, a multifunctional HSA method based on decentralized federated temporal domain adaptation is proposed in this paper. First, a novel multifunctional HSA framework is designed for comprehensive assessment of rotating machinery, which performs synchronous health stage division and recognition, remaining useful life prediction, and reliability evaluation. Then, a temporal adversarial domain adaptation model is proposed to align temporal feature distributions through adversarial training between cross-domain encoders and a discriminator, both integrated with channel attention to enhance temporal feature extraction capability. Meanwhile, data privacy is ensured through decentralized federated learning involving multiple users. Experimental results show that the proposed method exhibits better generalization and stability, achieving an average error of 0.087 in dual prediction functions and an accuracy of 0.94 in recognition functionality.

In this article, a novel shared-database decentralized federated learning (SDeceFL) framework is developed for wafer defect pattern recognition (DPR). Specifically, a differential privacy shared-database strategy is proposed to overcome the interclass heterogeneity problem of different clients and enhance data privacy. A deformable convolutional autoencoder (DCAE) is designed for data augmentation for handling class imbalance. The vision transformer (ViT) is employed for wafer DPR. The proposed DCAE-ViT-SDeceFL framework is validated on three public datasets (e.g., WM-811K, NEU-CLS-64, and CIFAR-100). The experimental results show the superiority of the SDeceFL framework over Ratio Loss-FedAvg, MOON, FedNH, BalanceFL, federated averaging (FedAvg), DeceFL, and swarm learning (SL). Compared with some deep learning methods, experimental results exhibit the effectiveness of the proposed DCAE-ViT-SDeceFL method for wafer DPR on WM-811K.

In industrial applications, fault diagnosis of gear-boxes is critical for ensuring the reliability and safe operation of machinery and equipment. However, data-driven fault diagnosis techniques rely heavily on high-quality labeled data for training, which is often difficult to obtain in real industrial scenarios. To address this challenge, this paper presents a digital twin fault diagnosis framework. First, the digital twin technique is utilized to build a virtual model of a gearbox that generates sufficient labeled fault data. Second, we develop a hybrid adversarial domain discriminative network, which incorporates a spatial channel reconstruction convolution module instead of a conventional residual network to enhance the model's feature extraction capabilities. Additionally, we propose a dynamic weighted joint distribution adaptation method to improve domain confusion capability, minimizing the granularity of features between the two types of data. Finally, the effectiveness and reliability of the proposed method are validated through experiments.

Accurately assessing health status of rotating machinery is essential in reliable operation and predictive maintenance. Recently, health status assessment (HSA) of rotating machinery has acquired significant progress, particularly with the increasing popularity of deep learning. However, most methods regard HSA as a combination of isolated tasks, ignoring the hierarchical relationships between tasks in predictive maintenance. Also, they fail to learn multi-source features adequately. Besides, due to data silos and conflict of interests, the collaborative model training poses a strong demand for privacy protection. Thus, this paper proposes a data privacy protection HSA framework with dual-task feature fusion learning. First, we consider HSA of machinery as an integral dual-task process and synchronously obtain assessment results. The proposed framework constructs hierarchical health labels to ensure assessment accuracy, and utilizes adaptive information exchange channels to maintain balance between the dual tasks. Next, we design a feature fusion network based on channel independence and patch mechanisms to enhance the utilization of correlation information in multi-source signals. Finally, by employing decentralized swarm learning, achieve privacy-preserving joint training to address data silos. Experimental results validate the effectiveness and superiority of the proposed framework.

Fault diagnosis and prognosis (FDP) tries to recognize and locate the faults from the captured sensory data, and also predict their failures in advance, which can greatly help to take appropriate actions for maintenance and avoid serious consequences in industrial systems. In recent years, deep learning methods are being widely introduced into FDP due to the powerful feature representation ability, and its rapid development is bringing new opportunities to the promotion of FDP. In order to facilitate the related research, we give a summary of recent advances in deep learning techniques for industrial FDP in this paper. Related concepts and formulations of FDP are firstly given. Seven commonly used deep learning architectures, especially the emerging generative adversarial network, transformer, and graph neural network, are reviewed. Finally, we give insights into the challenges in current applications of deep learning-based methods from four different aspects of imbalanced data, compound fault types, multimodal data fusion, and edge device implementation, and provide possible solutions, respectively. This paper tries to give a comprehensive guideline for further research into the problem of intelligent industrial FDP for the community.

Aiming at the wind power characteristics of temporality, periodicity and complexity, the periodic law of short-term and long-term repetitive patterns is studied, and an integrated dual-channel prediction model is proposed. A practical periodic characteristic extracting strategy is designed to show the hidden periodic law of the original signal. Combining the grid search algorithm with the variation trend of amplitude/period, the optimal periodic step is determined. Based on the above analysis, the original signal is decomposed into temporal and periodic components. Then the temporal attention network and the encoder-decoder attention network are schemed out to dispose the two components respectively. Finally, the linear regression attention network is adopted to realize data fitting. The integrated forecasting framework can deal with the long-term and short-term dependencies of the original data at the same time, and ensure the rapid convergence of training process, thereby improve the prediction accuracy and stability. The multi-dimensional experimental verification is carried out through the comparison of evaluation indicators, prediction trends, scatter plots and box plots.

Due to the scarcity of labeled faulty data in industrial practice, fault diagnosis models often face challenges related to overfitting and limited accuracy. This article introduces a novel solution to tackle the problem—a fault diagnosis method that leverages a combination of a semi-supervised prototype network and contrastive learning. Firstly, a limited number of labeled samples is employed to construct pairs of positive and negative samples. And the training method of contrastive learning is used to provide appropriate initial parameters for the autoencoder. Secondly, the autoencoder is employed as a feature mapping function within the prototype network and obtains the prototype using a limited number of labeled samples. Finally, a prototype refinement method fine-tunes the prototype through unlabeled data, reducing the impact of anomalous data and obtaining a more stable and accurate prototype. The effectiveness of the proposed method is evaluated using gearbox data as a validation dataset. Through comparative analysis with other methods, the results demonstrate that the proposed approach achieves superior diagnostic accuracy.

As one of the core equipment in the gas industry, gas flow meter guides the operation and maintenance of gas companies and reflects the gas consumption habits of downstream users. Therefore, the anomaly detection of gas flow meter has important theoretical value and practical significance. However, the sensors of the gas flow meter are coupled with each other, and the collected data has the characteristics of difficult to collect negative samples. For the solution of the above challenges, an approach to unsupervised anomaly detection based on generative adversarial networks (GAN) is introduced in this paper. Firstly, after the flow meter signal is processed with the data screening, the Savitzky Golay (S-G) filter can filter out the noise from the raw data. The information between channels is captured by the feature attention module for improving the accuracy of the potential representation of samples, in addition, residual blocks are used to prevent the network degradation phenomenon in the deep network. Finally, the proposed model was demonstrated to be valid by using the actual user gas flow meter data. Compared with the current popular depth learning methods, the experimental results reveal that the proposed method can be very accurate in terms of ACC and F1 scores.

In order to resolve the problem that the traditional fault diagnosis methods can not accurately diagnose under complex working conditions and few fault data, a fault diagnosis method based on meta-learning and time convolution network (TCN) is proposed in this paper. Firstly, Ensemble Empirical Mode Decomposition-Independent Component Analysis (EEMDICA) is used to denoise the original signals under different working conditions, and the acquired signals are randomly sampled to form meta tasks according to the meta-learning protocol. Through the TCN model training, the priori knowledge is obtained by optimizing the initialization parameters. The knowledge is used to improve accuracy in unforeseen condition tasks. Finally, simulation experiments are carried out on the gearbox dataset. The results indicate that the method has advantages in solving the fault classification problem of few shot gearbox under complex working conditions.

In integrated energy systems, load forecasting has become increasingly significant on the background of carbon peaking and carbon neutrality. This study aims to apply a probabilistic learning method, i.e. sparse Bayesian learning, to achieve day-ahead and week-ahead load forecasting. Different from previous studies, the proposed method integrates multiple regression with SBL method by using it as a weather forecasting step while the whole learning and prediction process is based on feature engineering and moving window techniques. Test cases are conducted on historical load data from New York Independent System Operator (NYISO). In terms of prediction performance, the SBL-based hybrid method outperforms single relevance vector machine (RVM), artificial neural network and multiple regression method on load forecasting.

Summary In this article, we propose a deep Koopman model predictive control (MPC) strategy to improve the transient stability of power grids in a fully data‐driven manner. Due to the high‐dimensionality and the nonlinearity of the transient process, we use the Koopman operator to map the original nonlinear dynamics into an infinite dimensional linear system. To facilitate the control design, we first utilize the deep neural network method to efficiently train observable functions to approximate the Koopman operator so that the obtained dynamics in the high dimensional space is a linear system. We then propose an MPC strategy for the obtained high dimensional linear system. The proposed control strategy utilizes energy storage units, which inject or absorb real power at the synchronous generator buses to enhance the transient stability. Simulation studies implemented on the IEEE 9‐bus 3‐machine test system and the IEEE 39‐bus 10‐machine test system illustrate the performance of the proposed deep Koopman MPC strategy. The results demonstrate that the proposed control strategy effectively enhances the transient stability of the system even in the presence of severe faults.

Additive manufacturing is becoming increasingly popular because of its unique advantages, especially fused deposition modelling (FDM) which has been widely used due to its simplicity and comparatively low price. However, in current FDM processes, it is difficult to fabricate parts with highly accurate dimensions. One of the reasons is due to the slicing process of 3D models. Current slicing software divides the parts into layers and then lines (paths) based on a fixed value. However, in a real printing process, the printed line width will change when the process parameters are set in different values. The various printed widths may result in inaccuracy of printed dimensions of parts if using a fixed value for slicing. In this paper, a mathematical model is proposed to predict the printed line width in different layer heights. Based on this model, a method is proposed for calculating the optimal width value for slicing 3D parts. In the future, the proposed mathematical model can be integrated into slicing software to slice 3D models for precision additive manufacturing.

Short-term electricity load forecasting has become increasingly important due to the privatization and deregulation in the energy market. This study proposes a probabilistic learning method to predict hourahead and day-ahead load demand. Unlike methods in previous studies, the proposed method integrates wavelet transform and feature selection as key preprocessing steps. Features are divided into current state related features and historical information related features. Current state related features are forecasted by the regression model before being added into the load prediction model. The entire learning and prediction process is based on the relevance vector machine (RVM) that utilizes load data characteristics. A number of test cases are presented using benchmark datasets from the New York Independent System Operator (NYISO) and ISO New England. Based on the detailed empirical comparison, the proposed RVM-based integrated method outperforms classical time series approaches and state-of-the-art artificial intelligence methods on short-term load forecasting. (C) 2020 Elsevier B.V. All rights reserved.

微电网由负载、储能系统和分布式电源互联集成到能源系统中,微电网系统可以作为一个整体系统与电网并行运行或以孤岛模式运行.负载建模是微电网运行和管理中的一个基本问题.本文着重解决以下两个关键问题:1)协调负载模型结构的合理性和简洁性;2)负载模型参数的校准.与常规负载建模方法不同,本文提出了一类数据驱动建模方法以同时实现负载模型结构选择和参数校准.具体地,该方法从量测数据中稀疏学习静态负载模型和动态负载模型,其关键方法分别来自于稀疏贝叶斯学习方法和交替方向方法,即从一组备选非线性字典函数中稀疏学习最主要的非线性项以平衡数据拟合度并实现模型学习.所提出的方法将机器学习与稀疏表示相结合,旨在对负载模型从物理角度提供机理解释并向配电网系统操作员提供有关负载的动态信息.在孤岛微电网测试系统中验证并评估了所提出的算法.研究测例表明所提出算法从量测数据中实现负载稀疏学习的合理性和对于噪声的鲁棒性.

With increasing penetration intermittent renewable energy and the interactions between different power sources in an interconnected system, low frequency oscillations may occur and potentially threaten the security of power systems if the grid cannot support adequate damping. Locating sources of low frequency oscillations is of great importance, which needs finding the mechanism of the damping of low frequency oscillations for interpretation. The difficulty of this problem lies in the fact that the parameters associated with the power system model can range from slightly uncertain to entirely unknown. Therefore, we firstly focus on identifying the network-reduced model to characterize low frequency oscillation, and then utilizing Hamilton analysis to reveal mechanism of oscillation. Accordingly, the problem of locating low frequency oscillation is equivalent to identify equivalent negative damping coefficient. In this paper, we propose a novel data-driven method that estimates equivalent damping coefficients and topological parameters of the network-reduced model simultaneously. More specifically, the proposed method utilizes the sparse representation to select the most dominant nonlinear terms from a set of dictionary functions, which finally balances the data fitness and achieves dynamics learning. We validate and evaluate the proposed method on IEEE 9-bus test system and IEEE 39-bus test system. The results demonstrate the effectiveness of the proposed method in achieving dynamics learning and locating sources of low frequency oscillations from measurement data.

It is crucial to maintain the system-wide stability for the function of distribution systems in the form of interconnected microgrids with meshed topology. In this paper, we study the system-wide stability of coupled microgrids under the practical situation where the transmission line is lossy. The stability of coupled microgrids are analyzed through small-signal stability approach considering both frequency and voltage dynamics. The condition of system-wide stability is then provided with the designed Lyapunov function, and the criterion solved by the linear matrix inequality (LMI) is given based on Lur’e system theory. Simulations performed on IEEE 123 node feeder demonstrate the effectiveness of the proposed stability criterion. Results also show that the droop gains have an impact on the system-wide stability.