No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Richly connected spatial–temporal graph neural network for rotating machinery fault diagnosis with multi-sensor information fusion
Multi-sensor fault diagnosis, especially when using heterogeneous sensors, substantially enhances the accuracy of fault detection in asynchronous motors operating under high-interference conditions. A critical challenge in multi-sensor fault diagnosis lies in effectively fusing data from different sensors. Deep learning offers a promising solution by transforming multi-sensor data into a unified representation, thereby facilitating robust data fusion. However, existing approaches often fail to fully exploit inter-sensor correlations and inherent prior physical knowledge. To address this limitation, we propose a novel graph neural network-based model that emphasizes graph structure construction for heterogeneous multi-sensor information fusion. Our framework includes (1) a multi-task enhanced autoencoder for node feature extraction, enabling discriminative representation learning, particularly with heterogeneous sensor data; (2) an adjacency matrix builder integrated with physical prior constraints to improve the generalization and robustness of the model; and (3) a graph isomorphism network to derive graph-level representations for fault classification. Our experimental results demonstrate the model’s effectiveness in diagnosing faults, as it achieves superior performance compared to conventional methods on two heterogeneous asynchronous motor datasets.
The vibration signal acquired by a single sensor contains limited information and is easily interfered by noise signals, resulting in the inability to fully express the operating characteristics and state of a gearbox. To address this problem, our study proposes a gearbox fault diagnosis method based on multi-sensor deep spatiotemporal feature representation. This method utilizes two vibration sensors to obtain the vibration information of the gearbox. A fault diagnosis model (PCNN–GRU) combined with a parallel convolutional neural network (PCNN) and gated recurrent unit (GRU) was used to fuse the gearbox vibration information. The parallel convolutional neural network was used to extract the spatial information of the vibration signals collected by different position sensors, and the timing information was mined through the gated recurrent unit. The deep spatiotemporal features that fuse the multi-sensor spatial and temporal information were composed. The collected multi-sensor vibration signals were directly input into the PCNN–GRU model, and an end-to-end intelligent diagnosis of the gearbox faults was realized. Finally, through experimental verification, the accuracy rate of this model can reach up to 99.92%. Compared with other models, this model has a higher diagnostic accuracy and stability.
Representation learning is vital for the performance of Multivariate Time Series (MTS) related tasks. Given high-dimensional MTS data, researchers generally rely on deep learning (DL) models to learn representative features. Among them, the methods that can capture the spatial-temporal dependencies within MTS data generally achieve better performance. However, they ignored hierarchical relations and the dynamic property within MTS data, hindering their performance. To address these problems, we propose a Hierarchical Correlation Pooling boosted graph neural network (HierCorrPool) for MTS data representation learning. First, we propose a novel correlation pooling scheme to learn and capture hierarchical correlations between sensors. Meanwhile, a new assignment matrix is designed to ensure the effective learning of hierarchical correlations by adaptively combining both sensor features and correlations. Second, we learn sequential graphs to represent the dynamic property within MTS data, so that this property can be captured for learning decent representations. We conducted extensive experiments to test our model on various MTS tasks, including remaining useful life prediction, human activity recognition, and sleep stage classification. Experimental results have shown the effectiveness of our proposed model.
With the rapid development of industry, fault diagnosis plays a more and more important role in maintaining the health of equipment and ensuring the safe operation of equipment. Due to large-size monitoring data of equipment conditions, deep learning (DL) has been widely used in the fault diagnosis of rotating machinery. In the past few years, a large number of related solutions have been proposed. Although many related survey papers have been published, they lack a generalization of the issues and methods raised in existing research and applications. Therefore, this paper reviews recent research on DL-based intelligent fault diagnosis for rotating machinery. Based on deep learning models, this paper divides existing research into five categories: deep belief networks (DBN), autoencoders (AE), convolutional neural networks (CNN), recurrent neural networks (RNN), and generative adversarial networks (GAN). This paper introduces the basic principles of these mainstream solutions, discusses related applications, and summarizes the application features of various solutions. The main problems of existing DL-based intelligent fault diagnosis (IFD) research are summarized as small-size sample imbalance and transfer fault diagnosis. The future research trends and hotspots are pointed out. It is expected that this survey paper can help readers understand the current problems and existing solutions in DL-based rotating machinery fault diagnosis, and effectively carry out related research.
Artificial intelligence fields have been using deep learning in recent years. Due to its powerful data mining capabilities, deep learning has a wide-ranging impact on the diagnosis of motor faults. A method for diagnosing motor faults based on the multi-feature fusion of convolutional neural network (CNN) is presented in this paper. As far as the method is concerned, CNN is used as the basic framework, and the CNN model has been improved. First, the collected vibration and current signals are preprocessed. Second, segmented multi-time window synchronous input is performed on the processed data. In addition, a multi-scale feature extraction process and time series fusion of vibration and current signals subject to synchronous input in the same time window can be performed, which ultimately enables the identification of motor faults with a high degree of accuracy. In order to verify the validity of the proposed fault diagnosis model, an experimental platform for fault simulation was built for the motor, and vibration and current signals of different motor states were collected and verified by experimentation. According to the results of the experiment, the method can effectively combine motor vibration and current signal fault features, and thus motor fault diagnosis can be improved. In comparison with a single signal input, a multi-signal input provides greater accuracy and stability. As compared to other multi-signal feature fusion methods, such a deep learning model is able to extract fault features in a more comprehensive manner, which helps to improve the accuracy of motor fault diagnosis.
Bearing is one of the most important parts of rotating machinery with high failure rate, and its working state directly affects the performance of the entire equipment. Hence, it is of great significance to diagnose bearing faults, which can contribute to guaranteeing running stability and maintenance, thus promoting production efficiency and economic benefits. Usually, the bearing fault features are difficult to extract effectively, which results in low diagnosis performance. To solve the problem, this paper proposes a bearing fault feature extraction method and it establishes a bearing fault diagnosis method that is based on feature fusion. The basic idea of the method is as follows: firstly, the time-frequency feature of the bearing signal is extracted through Wavelet Packet Transform (WPT) to form the time-frequency characteristic matrix of the signal; secondly, the Multi-Weight Singular Value Decomposition (MWSVD) is constructed by singular value contribution rate and entropy weight. The features of the time-frequency feature matrix obtained by WPT are further extracted, and the features that are sensitive to fault in the time-frequency feature matrix are retained while the insensitive features are removed; finally, the extracted feature matrix is used as the input of the Support Vector Machine (SVM) classifier for bearing fault diagnosis. The proposed method is validated by data sets from the time-varying bearing data from the University of Ottawa and Case Western Reserve University Bearing Data Center. The results show that the algorithm can effectively diagnose the bearing under the steady-state and unsteady state. This paper proposes that the algorithm has better fault diagnosis capabilities and feature extraction capabilities when compared with methods that aree based on traditional feature technology.
Intelligent fault diagnosis methods based on deep learning becomes a research hotspot in the fault diagnosis field. Automatically and accurately identifying the incipient micro-fault of rotating machinery, especially for fault orientations and severity degree, is still a major challenge in the field of intelligent fault diagnosis. The traditional fault diagnosis methods rely on the manual feature extraction of engineers with prior knowledge. To effectively identify an incipient fault in rotating machinery, this paper proposes a novel method, namely improved the convolutional neural network-support vector machine (CNN-SVM) method. This method improves the traditional convolutional neural network (CNN) model structure by introducing the global average pooling technology and SVM. Firstly, the temporal and spatial multichannel raw data from multiple sensors is directly input into the improved CNN-Softmax model for the training of the CNN model. Secondly, the improved CNN are used for extracting representative features from the raw fault data. Finally, the extracted sparse representative feature vectors are input into SVM for fault classification. The proposed method is applied to the diagnosis multichannel vibration signal monitoring data of a rolling bearing. The results confirm that the proposed method is more effective than other existing intelligence diagnosis methods including SVM, K-nearest neighbor, back-propagation neural network, deep BP neural network, and traditional CNN.
The temporal-spatial distribution of photovoltaic (PV) array output in large-scale PV plant has following features: (i) In the DC (direct current) side of PV plant, the outputs from different arrays display a substantial correlation with each other. (ii) The dynamic difference for a PV array is covered by the random fluctuation of PV output. (iii) It is difficult to explain the occurrence and evolution process of PV faults with operation data only. Currently, most fault diagnosis methods for the PV arrays do not take advantage of the temporal-spatial distribution information contained in the operation data. To solve these above problem, a new fault diagnosis method using the spatio-temporal distribution characteristics of photovoltaic array output is proposed. Here, the temporal fluctuation and spatial distribution characteristics of PV array output under different fault conditions were analyzed. The spatio-temporal composite function is constructed, and then used to set fuzzy rules for fault diagnosis of the PV array. Finally, an example is used to verify the effectiveness of the proposed method, and the results show that the method can describe the characters of spatial and temporal distribution of PV output under faults conditions, and it can effectively classify different faults.
Rotating machines are extensively utilized in diverse industries, and their malfunctions can result in significant financial consequences and safety risks. Consequently, there has been growing research interest in the intelligent fault diagnosis of rotating machines, particularly through the utilization of multi-sensor condition monitoring data. However, a comprehensive review focusing on multi-sensor data fusion methods is lacking. To bridge this gap, this paper provides a comprehensive analysis of the existing literature on the application of multi-sensor data fusion techniques to diagnose faults in rotating machines. Basic concepts of multi-sensor data fusion are first provided, establishing a robust foundation for subsequent discussions. The review then provides an in-depth analysis of the applications of multi-sensor data fusion in intelligent diagnosis for rotating machines. Furthermore, this review paper highlights the current challenges encountered in multi-sensor data fusion for intelligent fault diagnosis of rotating machines. By considering these challenges and consolidating knowledge from various sources, this paper proposes future research directions in this field. This review article serves as a valuable resource for researchers, practitioners, and decision-makers in the domain of intelligent fault diagnosis of rotating machines. The review provides comprehensive insights into the latest advancements of multi-sensor data fusion techniques and guiding future research directions in the measurement sciences.
Gearbox fault diagnosis is a critical aspect of machinery maintenance and reliability, as it ensures the safe and efficient operation of various industrial systems. The cross-domain fault diagnosis method based on transfer learning has been extensively researched to enhance the engineering applications of data-driven methods. Currently, the state-of-the-art gearbox cross-domain fault diagnosis primarily relies on single-modal data, which may not capture the full information needed for robust fault diagnosis under varying conditions. To address this issue, we propose a novel multi-modal data cross-domain fusion network that utilizes vibration signals and thermal images to capture comprehensive information about the gearbox's health conditions. First, one-dimensional and two-dimensional convolutional neural networks are constructed for feature extraction and fusion of multi-modal data. Then, the Maximum Mean Discrepancy loss is introduced to achieve cross-domain feature alignments within the modal. Finally, the cross-modal consistency learning strategy is constructed to enhance the cross-domain diagnosis performance of the model. To validate the effectiveness of the proposed method, we conducted experiments on a real-world gearbox test rig. Experimental results demonstrate that the proposed method is superior to single-modal methods and existing fusion methods in terms of diagnosis performance, proving that the proposed method offers a promising solution for gearbox fault diagnosis in industrial settings.
A gearbox is an indispensable component of a mechanical transmission system, and in the event of an anomaly, it can have a serious impact on the stability of the system. Graph convolutional networks have achieved some results in the study of fault diagnosis, but the way of topology construction from a single metric space is somewhat inexact and one-sided. Meanwhile, graph convolutional networks can only aggregate node features in the spatial dimension, which does not guarantee a comprehensive expression of fault information. Therefore, we propose a gearbox fault diagnosis method based on refined topology and spatio-temporal graph convolutional networks to accurately identify the gearbox health state. First, two topologies are constructed based on the distance metric and the similarity metric, respectively, and the edge structure is pruned to obtain a more lightweight topology based on the analysis of the structural similarity between the two. Then, depending on the number of common neighbors of two nodes, the weights between them are enhanced or weakened to further refine the topological information. Finally, a network model with spatio-temporal feature co-extraction is developed to mine fault features in depth in two dimensions to improve network performance. Case studies, ablation experiments, and robustness validation of the proposed method are carried out, and the results show that the accuracy of the proposed method is as high as 99.41% and 99.93% on the two datasets, respectively, which is much higher than that of other comparison methods, and at the same time, it has good noise robustness.
The key of mobile robot fault diagnosis is to model the spatial-temporal correlations from multi-sensor data. However, the majority of existing deep learning (DL)-based studies only focus on extracting either temporal or spatial information. Additionally, data imbalance problem, which seriously affects models’ generalization performance, cannot be ignored in robot fault diagnosis. To address these issues, a novel spatial-temporal graph attention network (STGATN) is proposed, which has three primary characteristics: 1) the feature-enhanced spatial-temporal graph is constructed based on robot multi-sensors data; 2) an attention-based spatial-temporal feature extraction module (A-STFEM) is designed to mine spatial-temporal correlation information among multi-sensors; 3) a regulatory cross-entropy loss function is developed to enhance the model robustness under imbalanced data scenario. The effectiveness of STGATN is verified by fault diagnosis experiments for a wheeled mobile robot and results show that STGATN can achieve outstanding diagnosis performance and robustness.
Recently, rotating machinery fault diagnosis studies based on graph neural networks (GNN) have received some satisfactory achievements. But most of them are based on the analysis of the single sensor signals, which cannot capture the comprehensive fault information, especially aiming at large rotating machineries. A few research using GNN for multi-sensor fault diagnosis only fuse multi-source features in the construction of the input graph, and the fusion effect largely depends on the manual feature selection. Graph attention network (GAT), as an emerging GNN, can give trainable weights to vertices based on the self-attention mechanism to improve the effectiveness of feature learning. And it has not yet been used in the field of multi-sensor fault diagnosis. To fill this gap and utilize GAT’s advantages, this paper presents a multi-sensor multi-head GAT (MMHGAT) model for large rotating machinery fault diagnosis. With the input of several subgraphs, the designed MMHGAT model consisting of two graph attention layers (GAL), a feature fusion process and a Softmax classifier, can dynamically fuse and mine the high-level fault characteristics during the training process. By employing the experiment on the axial flow pump, the effectiveness and superiority of the proposed method are validated.
This dataset includes vibration sensor data from accelerometers located on the support bearings on a rotary machine designed as a fault simulator. Data collection for known faulty components include: bearing inner and outer raceway faults and bent shaft. 38 singles and double fault scenarios and a one no fault scenario were collected at three different operating frequencies (shaft rpm). Data was collected for approximately 10 seconds per scenario at a rate of 6400 hertz. Data can be used for machine learning classification.
Traditional deep learning (DL)-based rolling bearing fault diagnosis methods usually use signals collected under specific working condition to train the diagnosis models. This may lead to the lack of domain adaptive ability of these trained models, thus making it difficult to obtain satisfactory diagnosis accuracy when working conditions fluctuate. To address it, a novel fault diagnosis framework based on the graph neural network (GNN) and dynamic graph embedding mechanism (DGE) was proposed in this paper. Firstly, convolutional neural network (CNN) is used to extract the hidden fault features from raw bearing vibration signals. Secondly, DGE module is designed with edge dropout mechanism to transform the features exacted by CNN into higher-level graph-structured features dynamically. Then, GNN is applied to further mine the fault features sensitivity to the fluctuating bearing working conditions. Finally, a novel mechanism named node voters is proposed to replace traditional graph-level attribute update function in GNN to obtain optimal fault pattern recognition results. Experiment results shows that the proposed framework can not only realize the end-to-end fault diagnosis of rolling bearings, but also has excellent domain adaptive ability to obtain better stability and diagnosis accuracy under variable working conditions compared to traditional DL-based methods.
The effective fault diagnosis of bearing can guarantee the safety of rotating machinery and is very important for its stable operation. The information fusion of multi-sensor data has been a feasible method to enhance the performance of fault diagnosis. However, how to fuse the joint information from different channels or even different kinds of sensors is still an important challenge. The present study proposes a novel multi-sensor information coupling network (MICN) for bearing fault diagnosis, which handles the signals from the same or different types of sensors, and the deeper features can be extracted from multi-sensors independently and simultaneously fused layer by layer. Especially, during the multi-layer feature fusion process, a novel feature-level information coupling model is developed based on the mutual attention mechanism. Finally, to validate the efficiency of the proposed method, several different experiments are designed, and the results show validity and superiority.
The bearing is the key component of rotating mechanical equipment, so the fault diagnosis of bearings is important to improve the reliability and safety of equipment operation. In recent years, feature fusion method has been extensively explored in the health monitoring and fault diagnosis of bearings. However, almost all the existing feature-fusion-based fault diagnosis methods extract features from different signals independently and concatenating them simply. It will lead to the failure of achieving the expected diagnostic accuracy because the complementary fault information is not fully mined and fused. This article proposes a novel bearing fault diagnosis approach based on mutual attention and bilinear model to address these issues. The features extracted from different input are interactive through mutual attention and are fused by the bilinear model, so the complementary fault features are effectively extracted and fine-grained fused. Experiments are conducted on the Paderborn bearing dataset to verify the effectiveness of the proposed method. Results show that the proposed method can effectively extract complementary fault features from different signals and deeply fuse them, and its diagnosis accuracy is up to 99.86%. Its performance is much better than that of simple concatenation and the feature fusion methods proposed in the reference.
The critical issue of wheeled robot fault diagnosis is to comprehensively evaluate its health condition using multisensor data, but traditional deep learning-based methods are hard to model the relationships among sensor measurements. Unlike these methods, the graph convolutional network (GCN), which uses the graph-structured data along with the association graph as input, is more efficient for relationship modeling. However, existing GCN-based fault diagnosis methods suffer from the following weaknesses: the association graphs are obtained according to the similarity of data samples or their features, which cannot guarantee accuracy; and these models are focused on spatial correlations and neglect temporal correlations. To address these problems, we propose to construct the association graph based on prior knowledge,
i.e
., a simplified mathematical model of the wheeled robot. Moreover, we develop a spatial-temporal difference graph convolutional network (STDGCN) for wheeled robot fault diagnosis. This network contains a difference layer that utilizes localized difference properties for feature enhancement, and the spatial-temporal graph convolutional modules are introduced to jointly capture the spatial-temporal correlations. To verify the effectiveness of the STDGCN for fault diagnosis, experiments are carried out, and the results show that the STDGCN achieves superior performance.
Deep learning (DL)-based methods have advanced the field of Prognostics and Health Management (PHM) in recent years, because of their powerful feature representation ability. The data in PHM are typically regular data represented in the Euclidean space. Nevertheless, there are an increasing number of applications that consider the relationships and interdependencies of data and represent the data in the form of graphs. Such kind of irregular data in non-Euclidean space pose a huge challenge to the existing DL-based methods, making some important operations (e.g., convolutions) easily applied to Euclidean space but difficult to model graph data in non-Euclidean space. Recently, graph neural networks (GNNs), as the emerging neural networks, have been utilized to model and analyze the graph data. However, there still lacks a guideline on leveraging GNNs for realizing intelligent fault diagnostics and prognostics. To fill this research gap, a practical guideline is proposed in this paper, and a novel intelligent fault diagnostics and prognostics framework based on GNN is established to illustrate how the proposed guideline works. In this framework, three types of graph construction methods are provided, and seven kinds of graph convolutional networks (GCNs) with four different graph pooling methods are investigated. To afford benchmark results for helping further study, a comprehensive evaluation of these models is performed on eight datasets, including six fault diagnosis datasets and two prognosis datasets. Finally, four issues related to the performance of GCNs are discussed and potential research directions are provided. The code library is available at: https://github.com/HazeDT/PHMGNNBenchmark.
The condition monitoring(CM) of rotating machinery(RM) is an essential operation for improving the reliability of mechanical systems. For this purpose, an efficient CM method that possesses simple and intuitive attributes is required for industrial applications. For condition monitoring that connects fault detection, degradation assessment, and prognosis applications, health indicators(HIs) have been developed in the past few decades. The construction of a HI is the decisive procedure for extracting informative fault information from the monitoring signal. From the initial statistical parameter-based construction methods to the introduction of data-oriented intelligent methods such as deep learning in recent years, HIs construction methods have ranged from fault mechanism-based approaches to a data-based approach, which involve two different technologies regardless of superiority or inferiority. This paper provides a systematic review of the HIs construction methods for rotating machinery proposed in the literature. It emphasizes the classical technical approaches and recent interesting research trends and analyzes the benefits and potential of efficient HIs for condition monitoring. The current challenges and future research opportunities are also presented in this paper. The Engineers and researchers interested in this research can be informed of current research ideas and directions in the field by reading this paper, as well as inspiring potentially excellent research work in the future.
Graph neural networks provide a powerful toolkit for embedding real-world graphs into low-dimensional spaces according to specific tasks. Up to now, there have been several surveys on this topic. However, they usually lay emphasis on different angles so that the readers cannot see a panorama of the graph neural networks. This survey aims to overcome this limitation and provide a systematic and comprehensive review on the graph neural networks. First of all, we provide a novel taxonomy for the graph neural networks, and then refer to up to 327 relevant literatures to show the panorama of the graph neural networks. All of them are classified into the corresponding categories. In order to drive the graph neural networks into a new stage, we summarize four future research directions so as to overcome the challenges faced. It is expected that more and more scholars can understand and exploit the graph neural networks and use them in their research community.
Lots of learning tasks require dealing with graph data which contains rich relation information among elements. Modeling physics systems, learning molecular fingerprints, predicting protein interface, and classifying diseases demand a model to learn from graph inputs. In other domains such as learning from non-structural data like texts and images, reasoning on extracted structures (like the dependency trees of sentences and the scene graphs of images) is an important research topic which also needs graph reasoning models. Graph neural networks (GNNs) are neural models that capture the dependence of graphs via message passing between the nodes of graphs. In recent years, variants of GNNs such as graph convolutional network (GCN), graph attention network (GAT), graph recurrent network (GRN) have demonstrated ground-breaking performances on many deep learning tasks. In this survey, we propose a general design pipeline for GNN models and discuss the variants of each component, systematically categorize the applications, and propose four open problems for future research.
It is difficult to find the rolling bearing fault and its detailed information in the early stage due to its complex structure, heavy load, harsh working environment, and long-term high-speed working condition. This paper proposes a fault diagnosis method based on Transient-extracting Transform (TET) and Linear Discriminant Analysis (LDA). Firstly, the transient component of vibration signal in the time-frequency domain is extracted using the TET method. And then, 10 statistical features of the original signal and its transient component are obtained respectively by computing the mean, standard deviation, skewness, etc. Next, LDA is used to extract their fisher features, which can reduce the dimensionality of features and improve the fault recognition accuracy and efficiency. Finally, the diagnosis is accomplished by the Nearest Neighbor Classifier (NNC). The experimental results demonstrate that the proposed method can effectively identify rolling bearing faults with different positions, damage levels and damage types.
It is a challenging problem to explore the capability of multi-sensor networks due to the identity of the underlying information space across modalities. In this paper, the information space for multi-sensor networks is developed from information geometry. The relationship between information space and the performance of multi-sensor networks is investigated. Different sensor information obtained by multi-sensor networks is represented, analyzed and fused concisely. The structure of the information space is studied such as geodesic, Ricci tensor and the information metric matrix. The structural properties of the information space are introduced: i) the symmetry; ii) the connection between information space’s curvature and Einstein’s field equation; iii) noise essence conjecture. The proposed analysis techniques are validated in many scenarios. The theoretical demonstration and numerical results indicate that the information described in different coordinate systems is equivalent.
Bearing fault diagnosis is an important part of rotating machinery maintenance. Existing diagnosis methods based on single-modal signals not only have unsatisfactory accuracy, but also bear the inherent risk of being misguided by single-modal signal noise. A new method is put forward that fuses multi-modal sensor signals, i.e. the data collected by an accelerometer and a microphone, to realize more accurate and robust bearing-fault diagnosis. The proposed method extracts features from raw vibration signals and acoustic signals, and fuses them using the 1D-CNN-based networks. Extensive experimental results obtained on ten groups of bearings are used to evaluate the performance of the proposed method. By analyzing the loss function and accuracy rate under different SNRs, it is empirically found that the proposed method achieves higher rate of diagnosis accuracy than the algorithms based on a single-modal sensor. Moreover, a visualization analysis is also conducted to investigate the inner mechanism of the proposed 1D-CNN-based method.
The supervisory control and data acquisition (SCADA) systems are widely installed on wind turbines (WTs) in major wind farms, which produce a large amount of sensory data that can be used for fault diagnosis of WTs. However, these SCADA data are naturally multivariate time series which represent complex temporal correlations between each sensor variable and spatial correlations among different sensor variables. To effective capture spatio-temporal correlations in SCADA data, we propose a new spatio-temporal multiscale neural network (STMNN). The proposed STMNN model contains two parallel feature extraction modules: 1) a multiscale deep echo state network (MSDeepESN) module to extract temporal multi-scale features; 2) a multiscale residual network (MSResNet) module to extract spatial multi-scale features. Additionally, to address the data imbalance problem of SCADA data and enhance fault diagnosis performance, instead of cross entropy loss, the STMNN model adopts focal loss (FL) as loss function. Our proposed STMNN method can provide end-to-end fault diagnosis solution with imbalanced SCADA data, and evaluated through experiments on an SCADA dataset from a real wind farm. The experimental results and comparative analysis have proved the effectiveness of our proposed STMNN model in practical applications.
Deep learning (DL) based methods have swept the field of mechanical fault diagnosis, because of the powerful ability of feature representation. However, many of existing DL methods fail in relationship mining between signals explicitly. Unlike those deep neural networks, graph convolutional networks (GCNs) taking graph data with topological structure as input is more efficient for data relationship mining, making GCN to be powerful for feature representation from graph data in non-Euclidean space. Nevertheless, existing GCNs have two limitations. Firstly, most GCNs are constructed on unweighted graphs, considering importance of neighbors as the same, which is not in line with reality. Secondly, the receptive field of GCNs is fixed, which limits the effectiveness of GCNs for feature representation. To address these issues, a multi-receptive field graph convolutional network (MRF-GCN) is proposed for effective intelligent fault diagnosis. In MRF-GCN, data samples are converted into weighted graphs to indicate differences in relationship of data samples. Moreover, MRF-GCN learns not only features from different receptive field, but also fuses learned features as an enhanced feature representation. To verify the efficacy of MRF-GCN for machine fault diagnosis, case studies are implemented, and the results show that MRF-GCN can achieve superior performance even under imbalanced dataset. Index Terms-Deep learning, graph convolutional networks, multi-receptive field, mechanical fault diagnosis.
Centrifugal pumps are widely used in modern industry, and blades are the key parts of it. The cracks on blades may result in a very serious consequence. In this paper, a fault diagnosis method based on principal component analysis (PCA) and Gaussian mixed model (GMM) was proposed, which combined signal processing and knowledge. Also, the theory model of proposed methods was established, and Expectation Maximization (EM) algorithm was used to make the model converge. PCA was used to reduce the data dimensionalities and increase the feature resolution, and GMM was used as classifier for crack fault. In order to verify the diagnostic effect of the model, the experimental bench of centrifugal pump was established, and various working conditions of the centrifugal pump were simulated. Experimental results showed that the classifier based on these parameters performed very well in data testing.
The existing intelligent fault diagnosis techniques of bevel gear focus on single-sensor signal analysis under the steady operation condition. In this study, a new method is proposed based on ensemble deep transfer learning and multisensor signals to enhance the fault diagnosis adaptability and reliability of bevel gear under various operation conditions. First, a novel stacked autoencoder (NSAE) is constructed using a denoising autoencoder, batch normalization, and the Swish activation function. Second, a series of source-domain NSAEs with multisensor vibration signals is pretrained. Third, the good model parameters provided by the source-domain NSAEs are transferred to initialize the corresponding target-domain NSAEs. Finally, a modified voting fusion strategy is designed to obtain a comprehensive result. The multisensor signals collected under the different operation conditions of bevel gear are used to verify the proposed method. The comparison results show that the proposed method can diagnose different faults in an accurate and stable manner using only one target-domain sample, thereby outperforming the existing methods.
Numerous sensors have been deployed in different locations in components of wind turbines to continuously monitor the health status of the turbine system and accordingly, generate a large volume of operation data by the supervisory control and data acquisition (SCADA) system. Naturally, these sensory data are multivariate time series with high spatio-temporal correlations. It is still challenging to effectively model such correlations and then enable an accurate fault diagnosis. To this end, we proposed a new spatio-temporal fusion neural network (STFNN) for wind turbine fault diagnosis. Specifically, a multi-kernel fusion convolution neural network (MKFCNN) with multiple convolution kernels of different sizes is first designed to extract multi-scale spatial correlations among different variables. Then, we adopt the long short-term memory (LSTM) to further learn the temporal dependence of the learned spatial features. The proposed STFNN model provides an end-to-end fault diagnosis way, which can directly learn spatio-temporal dependency from the raw SCADA data and give the fault diagnosis result. The effectiveness and superiority of the proposed method are evaluated on a generic wind turbine benchmark simulation dataset and a SCADA dataset from a real wind farm. Both experimental results have indicated that the proposed method outperformed several compared methods.
Intelligent diagnosis algorithms can monitor faults with industrial production of a timely manner via their powerful learning ability. Multi-sensor diagnosis systems can more comprehensively describe the state of equipment and avoid the influence of incorrect data acquisition locations, which is beneficial to fault diagnosis. The fusion of the original data is a difficult problem, and it is hard to express effective information via traditional algorithms. This paper presents an adaptive data fusion strategy based on deep learning called the convolutional neural network with atrous convolution for the adaptive fusion of multiple source data (FAC-CNN). Specifically, an adaptive-sized convolution kernel that matches the channel of data sources is constructed to capture multi-source data without tedious preprocessing, and the channel of data sources is not limited. The atrous convolution kernel is introduced to expand the field of view of the FAC-CNN and extracts fusion sequence features without repeated computation, resulting in improved stability. The 1D-CNN is added to extract features after atrous convolution. In addition, batch normalization optimizes the distribution of fusion data and the structure of the model. The parametric rectified linear unit activation function and global average pooling are also introduced to improve network performance. The proposed method is validated on an industrial fan system with non-manufacturing faults and a centrifugal pump. Compared with other fusion methods and diagnosis algorithms based on feature engineering, namely CNN, ANN, and SVM, the FAC-CNN model is found to exhibit superior performance.
Gearboxes are widely used in rotating machinery and various industrial applications for transmission of power and torque. They operate for prolong hours and under different working conditions which may increase their probability of failure. Sudden failure of a gearbox may lead to significant downtime and increase maintenance costs. In industrial applications, usually fault detection and diagnosis techniques based on vibration signal are used for monitoring the health condition of gearboxes. In most of these techniques, time and frequency domain features are manually extracted from a vibration sensor and used for fault detection and diagnosis. In this research, a fault diagnosis methodology based on motor current signature analysis is proposed. The acquired data from multiple current sensors are fused by a novel 2-D convolutional neural network architecture and used for classification purpose directly without any need for manual feature extraction. Performance of the proposed method has been evaluated on the motor current data obtained from an industrial gearbox test rig in various health condition and with different working speeds. In comparison with classical machine learning (ML) algorithms, the presented methodology exhibits the best classification performance for gearbox fault detection and diagnosis.
Deep learning has revolutionized many machine learning tasks in recent years, ranging from image classification and video processing to speech recognition and natural language understanding. The data in these tasks are typically represented in the Euclidean space. However, there is an increasing number of applications, where data are generated from non-Euclidean domains and are represented as graphs with complex relationships and interdependency between objects. The complexity of graph data has imposed significant challenges on the existing machine learning algorithms. Recently, many studies on extending deep learning approaches for graph data have emerged. In this article, we provide a comprehensive overview of graph neural networks (GNNs) in data mining and machine learning fields. We propose a new taxonomy to divide the state-of-the-art GNNs into four categories, namely, recurrent GNNs, convolutional GNNs, graph autoencoders, and spatial-temporal GNNs. We further discuss the applications of GNNs across various domains and summarize the open-source codes, benchmark data sets, and model evaluation of GNNs. Finally, we propose potential research directions in this rapidly growing field.
Intelligent fault diagnosis (IFD) refers to applications of machine learning theories to machine fault diagnosis. This is a promising way to release the contribution from human labor and automatically recognize the health states of machines, thus it has attracted much attention in the last two or three decades. Although IFD has achieved a considerable number of successes, a review still leaves a blank space to systematically cover the development of IFD from the cradle to the bloom, and rarely provides potential guidelines for the future development. To bridge the gap, this article presents a review and roadmap to systematically cover the development of IFD following the progress of machine learning theories and offer a future perspective. In the past, traditional machine learning theories began to weak the contribution of human labor and brought the era of artificial intelligence to machine fault diagnosis. Over the recent years, the advent of deep learning theories has reformed IFD in further releasing the artificial assistance since the 2010s, which encourages to construct an end-to-end diagnosis procedure. It means to directly bridge the relationship between the increasingly-grown monitoring data and the health states of machines. In the future, transfer learning theories attempt to use the diagnosis knowledge from one or multiple diagnosis tasks to other related ones, which prospectively overcomes the obstacles in applications of IFD to engineering scenarios. Finally, the roadmap of IFD is pictured to show potential research trends when combined with the challenges in this field.
In recent years, graph neural networks (GNNs) have emerged as a powerful neural architecture to learn vector representations of nodes and graphs in a supervised, end-to-end fashion. Up to now, GNNs have only been evaluated empirically—showing promising results. The following work investigates GNNs from a theoretical point of view and relates them to the 1-dimensional Weisfeiler-Leman graph isomorphism heuristic (1-WL). We show that GNNs have the same expressiveness as the 1-WL in terms of distinguishing non-isomorphic (sub-)graphs. Hence, both algorithms also have the same shortcomings. Based on this, we propose a generalization of GNNs, so-called k-dimensional GNNs (k-GNNs), which can take higher-order graph structures at multiple scales into account. These higher-order structures play an essential role in the characterization of social networks and molecule graphs. Our experimental evaluation confirms our theoretical findings as well as confirms that higher-order information is useful in the task of graph classification and regression.
Bearing fault diagnosis has extensively exploited vibration signals because of their rich information about bearing health conditions. However, this approach is expensive because the measurement of vibration signals requires external accelerometers. Moreover, in machine systems that are inaccessible or unable to be installed external sensors, the vibration signal-based approach is impracticable. Otherwise, motor current signals are easily measured by the inverters which are available components of those systems. Therefore, the motor current-signal-based bearing fault diagnosis approach has attracted considerable attention from researchers. However, the performance of this approach is still not good as the vibration signal-based approach, especially in the case of fault diagnosis for external bearings (the bearings which installed outside of the electric motors). Accordingly, this paper proposes a motor current-signal-based fault diagnosis method utilizing deep learning and information fusion that can be applied to external bearings in rotary machine systems. The proposed method uses raw signals from multiple phases of the motor current as direct input, features are extracted from the current signals of each phase. Then each feature set is classified separately by a convolutional neural network. To enhance the classification accuracy, a novel decision-level information fusion technique is introduced to fuse information from all of the utilized convolutional neural networks. The problem of decision-level information fusion is transformed into a simple pattern classification task, which can be solved effectively by familiar supervised learning algorithms. The effectiveness of the proposed fault diagnosis method is verified through experiments carried out with actual bearing fault signals.
Aimed at identifying the health state of wind turbines accurately by comprehensively using the change information in spatial and temporal scale of the supervisory control and data acquisition (SCADA) data, a novel condition monitoring method of wind turbines based on spatio-temporal features fusion of SCADA data by convolutional neural networks (CNN) and gated recurrent unit (GRU) was proposed in this paper. First, missing value complement and selection of variables with Pearson prod-moment correlation coefficient were applied to improve the effectiveness of SCADA data. Second, a deep learning model was constructed by the structures of CNN and GRU. The spatial features in SCADA data were extracted by CNN at every step, and the temporal features in the sequence of spatial features were extracted and fused by GRU. Third, the historical healthy SCADA data was used to train the normal behavior model. At last, the trained model received measured data and output the predicted values. The entire residual between the actual data and the predicted output was calculated to put into the exponential weighted moving average control chart for recognizing the condition of the wind turbine. The effectiveness and availability of the proposed method were proved in measured SCADA data experiments.
Fault diagnosis of rotating machinery plays a significant role in the reliability and safety of modern industrial systems. The traditional fault diagnosis methods usually need manually extracting the features from raw sensor data before classifying them with pattern recognition models. This requires much professional knowledge and complex feature extraction, only to cause results in a poor flexibility of the model, which only applies to the diagnosis of a fault in particular equipment. In recent years, deep learning has developed rapidly, and great achievements have been made in image analysis, speech recognition and natural language processing. However, its application in fault diagnosis of rotating machinery is still at the initial stage. In order to solve the problem of end-to-end fault diagnosis, this paper focuses on developing a convolutional neural network to learn features directly from the original vibration signals and then diagnose faults. The effectiveness of the proposed method is validated through PHM (Prognostics and Health Management) 2009 gearbox challenge data and a planetary gearbox test rig. Compared with the other three traditional methods, the results show that the one-dimensional convolutional neural network (1-DCNN) model has higher accuracy for fixed-shaft gearbox and planetary gearbox fault diagnosis than that of the traditional diagnostic ones.
We develop a novel deep learning framework to achieve highly-accurate machine fault diagnosis using transfer learning to enable and accelerate the training of deep neural network. Compared with existing methods, the proposed method is faster to train and more accurate. First, original sensor data are converted to images by conducting a Wavelet transformation to obtain time-frequency distributions. Next, a pre-trained network is used to extract lower level features. The labeled time-frequency images are then used to fine-tune the higher-levels of the neural network architecture. This paper creates a machine fault diagnosis pipeline and experiments are carried out to verify the effectiveness and generalization of the pipeline on three main mechanical datasets including induction motors, gearboxes, and bearings with sizes of 6,000, 9,000, and 5,000 time series samples, respectively. We achieve state-of-the-art results on each dataset, with most datasets showing test accuracy near 100%, and in the gearbox dataset, we achieve significant improvement from 94.8% to 99.64%. We created a repository including these datasets located at mlmechanics.ics.uci.edu.
Fault diagnosis of rotating machinery plays a significant role for the reliability and safety of modern industrial systems. As an emerging field in industrial applications and an effective solution for fault recognition, artificial intelligence (AI) techniques have been receiving increasing attention from academia and industry. However, great challenges are met by the AI methods under the different real operating conditions. This paper attempts to present a comprehensive review of AI algorithms in rotating machinery fault diagnosis, from both the views of theory background and industrial applications. A brief introduction of different AI algorithms is presented first, including the following methods: k-nearest neighbour, naive Bayes, support vector machine, artificial neural network and deep learning. Then, a broad literature survey of these AI algorithms in industrial applications is given. Finally, the advantages, limitations, practical implications of different AI algorithms, as well as some new research trends, are discussed.
This paper proposes a novel intelligent fault diagnosis method to automatically identify different health conditions of wind turbine gearbox. Different from traditional approaches where feature extraction and classification are separately designed and performed, this study aims to automatically learns effective fault features directly from raw vibration signals and meanwhile classifies the type of faults in a single learning framework, thus leading to an end-to-end learning based fault diagnosis system for wind turbine gearbox without additional signal processing techniques and diagnostic expertise. Considering multi-scale characteristics inherent in vibration signals of a gearbox, in this paper, a new multiscale convolutional neural network (MSCNN) framework is proposed to perform multi-scale feature extraction and classification simultaneously. A key advantage of the proposed MSCNN is the ability to learn complementary and rich fault pattern features at different time scales by incorporating multiscale learning into the CNN architecture, which greatly improves feature learning to enable better diagnosis performance. The proposed MSCNN framework is evaluated through experiments on a wind turbine test rig. The results demonstrate that the proposed MSCNN framework outperforms state-of-the-art methods and exhibits superior robustness against noise.
This paper presents a convolutional neural network (CNN)-based approach for fault diagnosis of rotating machinery. The proposed approach incorporates sensor fusion by taking advantages of the CNN structure to achieve higher and more robust diagnosis accuracy. Both temporal and spatial information of the raw data from multiple sensors is considered during the training process of the CNN. Representative features can be extracted automatically from the raw signals. It avoids manual feature extraction or selection, which rely heavily on prior knowledge of specific machinery and fault types. The effectiveness of the developed method is evaluated by using datasets from two types of typical rotating machinery, roller bearings and gearboxes. Compared with traditional approaches using manual feature extraction, the results show the superior diagnosis performance of the proposed method. The present approach can be extended to fault diagnosis of other machinery with various types of sensors, due to its end to end feature learning capability.
The dominant sequence transduction models are based on complex recurrent or convolutional neural networks in an encoder-decoder configuration. The best performing models also connect the encoder and decoder through an attention mechanism. We propose a new simple network architecture, the Transformer, based solely on attention mechanisms, dispensing with recurrence and convolutions entirely. Experiments on two machine translation tasks show these models to be superior in quality while being more parallelizable and requiring significantly less time to train. Our model achieves 28.4 BLEU on the WMT 2014 English-to-German translation task, improving over the existing best results, including ensembles by over 2 BLEU. On the WMT 2014 English-to-French translation task, our model establishes a new single-model state-of-the-art BLEU score of 41.0 after training for 3.5 days on eight GPUs, a small fraction of the training costs of the best models from the literature. We show that the Transformer generalizes well to other tasks by applying it successfully to English constituency parsing both with large and limited training data.