No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Recent deep learning models for diagnosis and health monitoring: A review of research works and future challenges
Abstract
As an important branch of machine learning, deep learning (DL) models with multiple hidden layer structures have the ability to extract highly representative features from the input. At present, fault detection and diagnosis (FDD) and health monitoring solutions developed based on DL models have received extensive attention in academia and industry along with the rapid improvement of computing power. Therefore, this paper focuses on a comprehensive review of DL model–based FDD and health monitoring schemes in view of common problems of industrial systems. First, brief theoretical backgrounds of basic DL models are introduced. Then, related publications are discussed about the development of DL and graphical models in the industrial context. Afterwards, public data sets are summarized, which are associated with several research papers. More importantly, suggestions on DL model–based diagnosis and health monitoring solutions and future developments are given. Our work will have a positive impact on the selection and design of FDD solutions based on DL and graphical models in the future.
... Deep learning based methods [29] assume that the training set and the test set obey the same data distribution. The operating points of industrial systems vary with the environment, loads and production requirements, resulting in differences in the data distribution between the training and test sets [30]. ...
Domain generalization achieves fault diagnosis on unseen modes. In process industrial systems, fault samples are limited, and only single-mode fault data can be obtained. Extracting domain-invariant fault features from single-mode data for unseen mode fault diagnosis poses challenges. Existing methods utilize a generator module to simulate samples of unseen modes. However, multi-mode samples contain complex spatiotemporal information, which brings significant difficulties to accurate sample generation. Therefore, double gradient reversal network (DGRN) is proposed. First, the model is pre-trained to acquire fault knowledge from the single seen mode. Then, pseudo-fault feature generation strategy is designed by Adaptive instance normalization, to simulate fault features of unseen mode. The dual adversarial training strategy is created to enhance the diversity of pseudo-fault features, which models unseen modes with significant distribution differences. Subsequently, domain-invariant feature extraction strategy is constructed by contrastive learning and adversarial learning. This strategy extracts common features of faults and helps multi-mode fault diagnosis. Finally, the experiments were conducted on Tennessee Eastman process and continuous stirred-tank reactor. The experiments demonstrate that DGRN achieves high classification accuracy on unseen modes while maintaining a small model size.
... However, it is important to take into consideration the limitations and underlying bias associated with Sentiment Analysis techniques and methods. [6,7] The healthcare industry may use the Sentiment Analysis to get valuable information about patient experiences, public opinions of healthcare providers, and sentiment expressed in healthcare-related materials. ...
The book chapter titled “Application of Sentiment Analysis in Diverse Domains” basically covers the applications and uses in all important areas like business, finance, marketing, healthcare, education and many more. Being futuristic concept, its benefits apply across the various domains with same intensity. The capturing of learner’s feeling expressed in textual/audio/video/social media data and its understanding are the most important factors. It will improve further on advancement of the technology when more sentiment variables will be precisely detected and analysed by the latest tools described in the next chapter. This chapter illustrates a study on various applications of Sentiment Analysis that is from marketing and business to finance, commerce and trade, healthcare to pharma, aviation and beyond. It describes the categorisation of the tools, techniques and models employed in the study for extracting meaningful insights from the vast amount of textual/audio/video/social media information. Each domain is analysed in terms of lying opportunities and specific challenges. It is illustrated how the Sentiment Analysis adaptively predicting the requirements of the learners and recommending the best solutions for different industries. The applications emphasised further determining consumer preferences, assessing public opinion, improving decision-making processes and demonstrating the significant impact of Sentiment Analysis in future. This chapter can be used as a valuable resource by practitioners, researchers and enthusiasts to understand the wide scope of the Sentiment Analysis and how it can be used to improvised real-life experiences in diverse domains.
... The first step is to sample the original data to obtain a set containing samples and labels. The second step is to classify the samples using DL models integrating feature design, extraction, and classification [7], [8]. ...
Fault diagnosis is critical to maintaining the performance of rotating machinery and ensuring the safe operation of the equipment. Convolutional neural networks (CNNs) have recently shown great potential with excellent automatic feature learning and nonlinear mapping abilities in the field of rotating machinery fault diagnosis. However, the CNN-based methods still suffer from some defects, such as inadequate data utilization and uneconomical computational efficiency, which limit the further improvement of diagnosis performance. Therefore, this paper proposes a fault diagnosis method based on multi-sensor fusion and Convolutional Neural Network with Efficient Channel Attention (ECA-CNN). First, multi-sensor vibration signals are sampled, converted, and channel fused into multi-channel images with rich and comprehensive features. Then, the efficient channel attention mechanism is introduced into CNN to increase the feature learning ability by adaptively scoring and assigning weights to the channel features. The ECA-CNN is proposed to learn representative fault features from multi-sensor fusion data to achieve fault identification. Finally, two experimental cases on the bearing and gearbox datasets prove that the proposed method has excellent performance, strong generalization capability, and high computational efficiency.
... In recent years, deep learning (DL) -a subset of machine learning that employs layered artificial neural networks to model complex data abstractions -has gained popularity because of its excellent performance across different fields [14,15]. Despite its promising performance, DL models build the input-output knowledge by optimizing model parameters via the gradient backpropagation algorithm. ...
Deep learning (DL) has gained popularity in recent years as an effective tool for classifying the current health and predicting the future of industrial equipment. However, most DL models have black-box components with an underlying structure that is too complex to be interpreted and explained to human users. This presents significant challenges when deploying these models for safety-critical maintenance tasks, where non-technical personnel often need to have complete trust in the recommendations these models give. To address these challenges, we utilize a convolutional neural network (CNN) with Gradient-weighted Class Activation Mapping (Grad-CAM) activation map visualizations to form an interpretable DL method for classifying bearing faults. After the model training process, we apply Grad-CAM to identify a training sample's feature importance and to form a library of diagnosis knowledge (or health library) containing training samples with annotated feature maps. During the model evaluation process, the proposed approach retrieves prediction basis samples from the health library according to the similarity of the feature importance. The proposed method can be easily applied to any CNN model without modifying the model architecture, and our experimental results show that this method can select prediction basis samples that are intuitively and physically meaningful, improving the model's trustworthiness for human users.
The emerging trends of Sentiment Analysis has revolutionized the Educational Data Mining and transformed the other industries through its adaptive applications. It worked as a catalyst to speed up the further studies for further refinements. This book chapter analyses multifaceted applications of Sentiment Analysis in education, explores its significant impact on educational approaches, discusses about learner’s engagement and helps in making of institutional strategies. The chapter describes all about the through key themes, including the importance of the Sentiment Analysis in assessing learner’s feedback, the improvement of their personalized learning experiences and the overall development of positive learning spaces. It explains insights into the transformative potential of integration of Sentiment Analysis tools in unhiding the complex tapestry of human emotions within the digital world. It is necessary to ensure the unbiased consideration of the potential advantages by addressing challenges while taking leverage of Sentiment Analysis in education.
This paper presents a comprehensive review of the historical development, the current state of the art, and prospects of data-driven approaches for industrial process monitoring. The subject covers a vast and diverse range of works, which are compiled and critically evaluated based on the different perspectives they provide. Data-driven modeling techniques are surveyed and categorized into two main groups: multivariate statistics and machine learning. Representative models, namely principal component analysis, partial least squares and artificial neural networks, are detailed in a didactic manner. Topics not typically covered by other reviews, such as process data exploration and treatment, software and benchmarks availability, and real-world industrial implementations, are thoroughly analyzed. Finally, future research perspectives are discussed, covering aspects related to system performance, the significance and usefulness of the approaches, and the development environment. This work aims to be a reference for practitioners and researchers navigating the extensive literature on data-driven industrial process monitoring.
This paper presents a benchmark data set for condition monitoring of rolling bearings in combination with an extensive description of the corresponding bearing damage, the data set generation by experiments and results of data-driven classifications used as a diagnostic method. The diagnostic method uses the motor current signal of an electromechanical drive system for bearing diagnostic. The advantage of this approach in general is that no additional sensors are required, as current measurements can be performed in existing frequency inverters. This will help to reduce the cost of future condition monitoring systems. A particular novelty of the present approach is the monitoring of damage in external bearings which are installed in the drive system but outside the electric motor. Nevertheless, the motor current signal is used as input for the detection of the damage. Moreover, a wide distribution of bearing damage is considered for the benchmark data set. The results of the classifications show that the motor current signal can be used to identify and classify bearing damage within the drive system. However, the classification accuracy is still low compared to classifications based on vibration signals. Further, dependency on properties of those bearing damage that were used for the generation of training data are observed, because training with data of artificially generated and real bearing damages lead to different accuracies. Altogether a verified and systematically generated data set is presented and published online for further research.
Real-time and personalized lithium-ion battery health management is conducive to safety improvement for end-users. However, personalized prognostic of the battery health status is still challenging due to diverse usage interests, dynamic operational patterns and limited historical data. We generate a comprehensive dataset consisting of 77 commercial cells (77 discharge protocols) with over 140 000 charge-discharge cycles—the largest dataset to our knowledge of its kind, and develop a transfer learning framework to realize real-time personalized health status prediction for unseen battery discharge protocols, at any charge-discharge cycle. Our method can achieve mean testing errors of 0.176% and 8.72% for capacity estimation and remaining useful life (RUL) prediction, respectively. Additionally, the proposed framework can leverage the knowledge from two other well-known battery datasets, with a variety of charge configurations and a different battery chemistry respectively, to reliably estimate the capacity (0.328%/0.193%) and predict the RUL (9.80%/9.90%) of our cells. This study allows end users to tailor battery consumption plans and motivates manufacturers to improve battery designs.
With the wide deployment of rechargeable batteries, battery degradation prediction has emerged as a challenging issue. However, battery life defined by capacity loss provides limited information regarding battery degradation. In this article, we explore the prediction of voltage‐capacity curves over battery lifetime based on a sequence to sequence (seq2seq) model. We demonstrate that the data of one present voltage‐capacity curve can be used as the input of the seq2seq model to accurately predict the voltage‐capacity curves at 100, 200, and 300 cycles ahead. This offers an opportunity to update battery management strategies in response to the predicted consequences. Besides, the model avoids feature engineering and is flexible to incorporate different numbers of input and output cycles. Therefore, it can be easily transplanted to other battery systems or electrochemical components. Furthermore, the model features data generation, that is, we can use the data of only one cycle to generate a large spectrum of aging data at the future cycles for developing other battery diagnosis or prognosis methods. In this way, the time and energy consuming battery degradation tests can be sharply reduced. Accurate prediction of battery degradation is a prerequisite to advanced battery design and management. This article proposes a sequence to sequence deep‐learning model to predict future voltage‐capacity curves. The prediction results can more comprehensively reflect battery degradation than scalar degradation states, such as the capacity loss. The developed model can also be utilised to generate synthetic degradation data.
This paper presents a comprehensive review of the developments made in rotating bearing fault diagnosis, a crucial component of a rotatory machine, during the past decade. A data-driven fault diagnosis framework consists of data acquisition, feature extraction/feature learning, and decision making based on shallow/deep learning algorithms. In this review paper, various signal processing techniques, classical machine learning approaches, and deep learning algorithms used for bearing fault diagnosis have been discussed. Moreover, highlights of the available public datasets that have been widely used in bearing fault diagnosis experiments, such as Case Western Reserve University (CWRU), Paderborn University Bearing, PRONOSTIA, and Intelligent Maintenance Systems (IMS), are discussed in this paper. A comparison of machine learning techniques, such as support vector machines, k-nearest neighbors, artificial neural networks, etc., deep learning algorithms such as a deep convolutional network (CNN), auto-encoder-based deep neural network (AE-DNN), deep belief network (DBN), deep recurrent neural network (RNN), and other deep learning methods that have been utilized for the diagnosis of rotary machines bearing fault, is presented.
The early diagnosis of a motor is important. Many researchers have used deep learning to diagnose motor applications. This paper proposes a one-dimensional convolutional neural network for the diagnosis of permanent magnet synchronous motors. The one-dimensional convolutional neural network model is weakly supervised and consists of multiple convolutional feature-extraction modules. Through the analysis of the torque and current signals of the motors, the motors can be diagnosed under a wide range of speeds, variable loads, and eccentricity effects. The advantage of the proposed method is that the feature-extraction modules can extract multiscale features from complex conditions. The number of training parameters was reduced so as to solve the overfitting problem. Furthermore, the class feature map was proposed to automatically determine the frequency component that contributes to the classification using the weak learning method. The experimental results reveal that the proposed model can effectively diagnose three different motor states—healthy state, demagnetization fault state, and bearing fault state. In addition, the model can detect eccentric effects. By combining the current and torque features, the classification accuracy of the proposed model is up to 98.85%, which is higher than that of classical machine-learning methods such as the k-nearest neighbor and support vector machine.
Centrifugal pumps are the most vital part of any process industry. A fault in centrifugal pump can affect imperative industrial processes. To ensure reliable operation of the centrifugal pump, this paper proposes a novel automated health state diagnosis framework for centrifugal pump that combines a signal to time-frequency imaging technique and an Adaptive Deep Convolution Neural Network model (ADCNN). First, the vibration signals corresponding to different health conditions of the centrifugal pump are acquired. Vibration signals obtained from the centrifugal pump carry a great deal of information and generally, statistical features are extracted from the vibration signals to retain meaningful fault information. However, these features are either insensitive to weak incipient faults or unsuitable for tracking severe faults, thus, decreasing the fault classification accuracy. To tackle this problem, a signal to time-frequency imaging technique is applied to the pump vibration signals. For this purpose, Continuous Wavelet Transform (CWT) is applied to decompose the vibration signals over different time-frequency scales and extract the pump fault information in both the time and frequency domains. The CWT scales form two-dimensional time-frequency images commonly referred to as scalograms. The CWT scalograms are then converted into grayscale images (SGI). Over the past few decades, CNN models have been established as an effective practice to process images for classification and pattern recognition. Consequently, the extracted CWTSGIs are finally provided as inputs to the proposed ADCNN architecture to achieve feature extraction and classification for centrifugal pump faults. The performance of the proposed diagnostic framework (CWTSGI + ADCNN) is validated with a vibration dataset collected from a testbed specifically designed for centrifugal pump diagnosis. The experimental results suggest that the proposed technique based on CWTSGI and ADCNN outperformed existing methods with an average performance improvement of 4.7 – 15.6%.
Intelligent fault diagnosis can be related to applications of machine learning theories to machine fault diagnosis. Although there is a large number of successful examples, there is a gap in the optimization of the hyper-parameters of the machine learning model, which ultimately has a major impact on the performance of the model. Machine learning experts are required to configure a set of hyper-parameter values manually. This work presents a convolutional neural network based data-driven intelligent fault diagnosis technique for rotary machinery which uses model with optimized hyper-parameters and network structure. The proposed technique input raw three axes accelerometer signal as high definition 1-D data into deep learning layers with optimized hyper-parameters. Input is consisted of wide 12,800 × 1 × 3 vibration signal matrix. Model learning phase includes Bayesian optimization that optimizes hyper-parameters of the convolutional neural network. Finally, by using a Convolutional Neural Network (CNN) model with optimized hyper-parameters, classification in one of the 8 different machine states and 2 rotational speeds can be performed. This study accomplished the effective classification of different rotary machinery states in different rotational speeds using optimized convolutional artificial neural network for classification of raw three axis accelerometer signal input. Overall classification accuracy of 99.94% on evaluation set is obtained with the CNN model based on 19 layers. Additionally, more data are collected on the same machine with altered bearings to test the model for overfitting. Result of classification accuracy of 100% on second evaluation set has been achieved, proving the potential of using the proposed technique.
Deep-neural network-based fault diagnosis methods have been widely used according to the state of the art. However, a few of them consider the prior knowledge of the system of interest, which is beneficial for fault diagnosis. To this end, a new fault diagnosis method based on the graph convolutional network (GCN) using a hybrid of the available measurement and the prior knowledge is proposed. Specifically, this method first uses the structural analysis (SA) method to prediagnose the fault and then converts the prediagnosis results into the association graph. Then, the graph and measurements are sent into the GCN model, in which a weight coefficient is introduced to adjust the influence of measurements and the prior knowledge. In this method, the graph structure of GCN is used as a joint point to connect SA based on the model and GCN based on data. In order to verify the effectiveness of the proposed method, an experiment is carried out. The results show that the proposed method, which combines the advantages of both SA and GCN, has better diagnosis results than the existing methods based on common evaluation indicators.
To find out the hidden danger in the industrial production process in time, it is necessary to monitor the health condition of the key components of the mechanical system in operation. However, traditional fault diagnosis methods usually adopt manual feature extraction, which not only costs expensively and depends on prior knowledge. Therefore, it is of great significance to study the application of automatic fault identification based on the original vibration signals. Recently, existing studies have shown that the most fault diagnosis technology is implemented by using deep neural network. Although these methods have achieved satisfactory performances, there are obvious limitations in real application, that is, the complexity of deep neural network requires a lot of hardware computing resources. This hinders the development of online fault diagnosis tools. To solve this problem, this paper proposes a fault diagnosis model based on lightweight convolutional neural network MobileNet, and realizes an end-to-end intelligent fault classification and diagnosis application. We evaluated the proposed method with the rolling bearing dataset from Western Reserve University. The best average precision, recall and F1 score of 10 different bearing health conditions are about 96%, 82% and 88%, respectively. In addition, we also compare the accuracy of the rolling bearing fault diagnosis classification model under the traditional ReLU and the improved ReLU. Experimental results show that both of them obtain good performances, but the improved ReLU reach the over 96% accuracy more rapidly.
Fault diagnosis is critical in industrial systems since early detection of problems can not only save valuable time but also reduce maintenance costs. The feature extraction process of traditional fault diagnosis is time-consuming and laborious work. Recently, with the rapid development of the deep learning (DL) method, it has shown its superiority with an end-to-end process and has been applied to classification and other fields. To a certain extent, it solves the disadvantages of manual feature extraction in the traditional fault diagnosis method. However, the available training data is often limited, and it will degrade the performance of DL methods. A new DL method that combines deep convolutional neural network (DCNN) and transfers learning (TL) for fault diagnosis is proposed in this paper to handle different fault types. A signal processing method that converts one-dimensional time-series signals into grayscale images is firstly applied, and it can eliminate the effect of handcrafted features. Secondly, an optimal DCNN is designed and trained with the ImageNet datasets, which can extract the high-level features of massive images. Finally, TL is further developed to apply the knowledge learned in the source data distribution to the target data distribution, which greatly reduces the dependence on training data and improves the generalization performance of DCNN. Three well-known datasets, including the bearing vibration dataset from the Case Western Reserve University (CWRU), self-priming centrifugal pump dataset (SPCP), and bearing force dataset from the University of Paderborn, are utilized to the performance of the proposed method. Some popular classification methods are also added to the comparison. Results show that the proposed method can precisely identify different fault types and have the highest classification accuracy among other methods.
Intelligent operation and maintenance is an important part of Industry 4.0. In order to realize the intelligent of plant equipment, it will make full use of artificial intelligence methods to evaluate the operating status of the equipment. Fault diagnosis of industrial equipment represented by bearings is critical in smart manufacturing. Early, online and accurate diagnostics can save the plant a lot of time and expense. With the development of sensor technology and deep learning technology, multi-sensor information fusion and convolutional neural network (CNN) provide a solution to the above problems. In this paper, based on the characteristics of mechanical vibration signal propagation in space, a new multi-sensor information fusion method is proposed to implement fault classification. This method constructs the time-domain vibration signals of multiple sensors from different position into a rectangular two-dimensional matrix, and then uses an improved 2D CNN to realize signal classification. The method is validated on the open dataset Case Western Reserve University, the University of Cincinnati IMS bearing database and the dataset form designed bearing fault test rig, has achieved prediction of 99.92%, 99.68%, and 99.25% respectively. Compared with the traditional 1D, 2D CNN and other fault classification methods, the model can utilize less data and computational complexity, achieve higher fault prediction accuracy.
Recently, various deep learning models, which are mainly based on data-driven algorithms, have received more and more attention in the field of intelligent fault diagnosis and prognostics. However, there are two major assumptions accepted by default in the existing studies: 1) The training (source domain) and testing (target domain) data sets obey the same feature distribution; 2) Sufficient labeled data with fault information is available for model training. In real industrial scenarios, especially for different machines, these assumptions are mostly invalid, which makes it a huge challenge to build reliable diagnostic model. Motivated by transfer learning, we present a novel intelligent method named deep transfer network (DTN) with multi-kernel dynamic distribution adaptation (MDDA) to address the problem of cross-machine fault diagnosis. In the proposed approach, the DTN has wide first-layer convolutional kernel and several small convolutional layers, which is utilized to extract transferable features across different machines and suppress high frequency noise. Then, the MDDA method constructs a weighted mixed kernel function to map different transferable features to a unified feature space, and the relative importance of the marginal and conditional distributions are also evaluated dynamically. The proposed method is verified by three transfer learning tasks of bearings, in which the health states of wind turbine bearings in real scenario are identified by using diagnosis knowledge from two different bearings in laboratories. The results show that the proposed method can achieve higher diagnosis accuracy and better transfer performance even under different noisy environment conditions than many other state-of-the-art methods. The presented framework offers a promising approach for cross-machine fault diagnosis.
Hydraulic piston pump is the heart of hydraulic transmission system. On account of the limitations of traditional fault diagnosis in the dependence on expert experience knowledge and the extraction of fault features, it is of great meaning to explore the intelligent diagnosis methods of hydraulic piston pump. Motivated by deep learning theory, a novel intelligent fault diagnosis method for hydraulic piston pump is proposed via combining wavelet analysis with improved convolutional neural network (CNN). Compared with the classic AlexNet, the proposed method decreases the number of parameters and computational complexity by means of modifying the structure of network. The constructed model fully integrates the ability of wavelet analysis in feature extraction and the ability of CNN in deep learning. The proposed method is employed to extract the fault features from the measured vibration signals of the piston pump and realize the fault classification. The fault data are mainly from five different health states: central spring failure, sliding slipper wear, swash plate wear, loose slipper, and normal state, respectively. The results show that the proposed method can extract the characteristics of the vibration signals of the piston pump in multiple states, and effectively realize intelligent fault recognition. To further demonstrate the recognition property of the proposed model, different CNN models are used for comparisons, involving standard LeNet-5, improved 2D LeNet-5, and standard AlexNet. Compared with the models for contrastive analysis, the proposed method has the highest recognition accuracy, and the proposed model is more robust.
Papers from the 2006 flagship meeting on neural computation, with contributions from physicists, neuroscientists, mathematicians, statisticians, and computer scientists.
The annual Neural Information Processing Systems (NIPS) conference is the flagship meeting on neural computation and machine learning. It draws a diverse group of attendees—physicists, neuroscientists, mathematicians, statisticians, and computer scientists—interested in theoretical and applied aspects of modeling, simulating, and building neural-like or intelligent systems. The presentations are interdisciplinary, with contributions in algorithms, learning theory, cognitive science, neuroscience, brain imaging, vision, speech and signal processing, reinforcement learning, and applications. Only twenty-five percent of the papers submitted are accepted for presentation at NIPS, so the quality is exceptionally high. This volume contains the papers presented at the December 2006 meeting, held in Vancouver.
Bradford Books imprint
Deep Transfer Learning (DTL) is a new paradigm of machine learning, which can not only leverage the advantages of Deep Learning (DL) in feature representation, but also benefit from the superiority of Transfer Learning (TL) in knowledge transfer. As a result, DTL techniques can make DL-based fault diagnosis methods more reliable, robust and applicable, and they have been widely developed and investigated in the field of Intelligent Fault Diagnosis (IFD). Although several systematic and valuable review articles have been published on the topic of IFD, they summarized relevant research only from an algorithm perspective and overlooked practical applications in industry scenarios. Furthermore, a comprehensive review on DTL-based IFD methods is still lacking. From this insight, it is particularly important and more necessary to comprehensively survey the relevant publications of DTL-based IFD, which will help readers to conveniently understand the current state-of-the-art techniques and to quickly design an effective solution for solving IFD problems in practice. First, theoretical backgrounds of DTL are briefly introduced to present how the transfer learning techniques can be integrated with deep learning models. Then, major applications of DTL and their recent developments in the field of IFD are detailed and discussed. More importantly, suggestions on how to select DTL algorithms in practical applications, and some future challenges are shared. Finally, conclusions of this survey are given. At last, we have reason to believe that the works done in this article can provide convenience and inspiration for the researchers who want to devote their efforts in the progress and advance of IFD.
Due to its ability to learn the relationship among nodes from graph data, graph convolution network (GCN) has received extensive attention. In machine fault diagnosis field, it needs to construct input graphs reflecting features and relationships of the monitoring signals. Thus, the quality of the input graph affects the diagnostic performance. But it still has two limitations: 1) the constructed input graph usually has redundant edges, consuming excessive computational costs; 2) the constructed input graph cannot reflect the relationship between the noisy signals well. In order to overcome them, a dynamic graph-based feature learning with few edges considering noisy samples is proposed for rotating machinery fault diagnosis in this paper. Noisy vibration signals are converted into one spectrum features-based static graph, where redundant edges are simplified by the distance metric function. Edge connections of input static graph are updated according to the relationship among high-level features extracted by the GCN. Based on this, dynamic input graphs are reconstructed as new graph representation for noisy samples. To verify effectiveness of the proposed method, validation experiments were conducted on practical platforms, and results show that the dynamic input graph with few edges can effectively improve the diagnostic performance under different SNRs.
Electromagnetic radiation (EMR) waveforms contain rich information on coal fracture law, which is of great value to the refined study of their characteristics. In this paper, the Hilbert-Huang transform (HHT) method is used to analyze and process EMR waveform of coal failure process. The results show that HHT method can adapt well to the nonstationary and nonlinear characteristics of EMR waveform. It adaptively decomposes EMR waveforms by empirical mode decomposition (EMD), extracts the intrinsic mode function (IMF) characteristics of EMR waveforms, and obtains the law of instantaneous energy changes of EMR waveforms. At the same time, EMR waveform energy is quantitatively expressed on the three-dimensional Hilbert energy spectrum (time-frequency-energy distribution), and the refined features of EMR waveform signal are effectively extracted. The instantaneous energy of EMR and the three-dimensional Hilbert energy spectrum processed by HHT method more easily describe the fracture law of coal body. Compared with conventional characteristic parameters, such as EMR amplitude and dominant frequency, the identification degree of the precursor characteristics of coal body instability is greatly improved. Among the short-term Fourier transform (STFT), wavelet transform (WT) and HHT nonstationary signal processing methods, HHT method has an improved adaptability and superiority and can reveal the characteristics of coal deformation and failure more carefully and accurately, which provides a new means for better using and improving EMR method to monitor coal instability and failure.
Contactless fault diagnosis is one of the most important technique for fault identification of equipment. Based on the idea of contactless fault diagnosis, this paper presents a sound-based diagnosis method for railway point machines (RPMs). First, the sound signals are preprocessed using empirical mode decomposition (EMD). Entropy, time-domain and frequency-domain statistical parameters of the first 15 intrinsic mode functions (IMFs) are then extracted. Second, a two-stage feature selection strategy blending Filter method and Wrapper method is proposed, which can significantly reduce the dimension of features and select the optimal features. The superiority and effectiveness of the proposed feature selection strategy are verified by comparing with other feature selection methods. Third, a weighted majority voting (WMV)-based ensemble classifier optimized using particle swarm optimization (PSO) is developed and compared with single classifiers. And the ensemble patterns are discussed to select the most optimal ensemble pattern. The average diagnosis accuracies of 10 repeated trails of reverse-normal and normal-reverse switching processes reach 99% and 99.93%, respectively, which indicates the effectiveness and feasibility of the proposed method.
Accurate degradation monitoring over battery life is indispensable for the safe and durable operation of battery-powered applications. In this work, we extend conventional capacity degradation estimation to the estimation of entire constant-current charging curves. A deep neural network (DNN) is developed to estimate complete charging curves by featuring small portions of the charging curves to form the input. We demonstrate that the charging curves can be accurately captured with an error of less than 16.9 mAh for 0.74 Ah batteries with 30 points collected in less than 10 min. Validation based on batteries working at different current rates and temperatures further demonstrates the effectiveness of the proposed method. This method also enjoys the advantage of transfer learning; that is, a DNN trained on one battery dataset can be used to improve the curve estimation of other batteries operating under different scenarios by using few training data.
The requirements for accuracy of piezoresistive pressure sensor in modern society are becoming increasingly high. Besides, widely application range of sensor is required. However, due to the influence of material properties, many piezoresistive pressure sensors have high temperature coefficient, which limits their application temperature range. With the change of ambient temperature, the response characteristic of the sensor has strong nonlinearity. To solve above crucial problems, a dynamic chaos quantum-behaved particle swarm optimization optimized multiple kernel relevance vector machine (DCQPSO-MKRVM) algorithm is presented in this paper. First, a basic theory of temperature effect is given and a new idea of temperature compensation is proposed. Second, the multi-kernel relevance vector machine (MKRVM) is adopted to estimate the bias values of input pressure. By using heterogeneous kernel learning method, the kernels of MKRVM maintain diversity to obtain higher estimation accuracy. Third, DCQPSO is employed to optimize the optimal sparse weights of kernel functions in MKRVM. Moreover, the dynamic parameter is applied for the boundary of chaos search between original QPSO swarm and the chaos swarm. The experimental result indicates the complexity nonlinear relationship of temperature effect, and the method proposed in this paper can effectively and accurately estimate the bias of input pressure fast to achieve temperature compensation goal. The mean relative accuracy (MRA) of estimation results achieves 99.5proves that the method proposed in this paper is applicable and effective for industrial application.
As the power system develops from informatization to intelligence. Research on data services based on the Internet of Things (IoT) focuses more on application functions, but the research on the data quality of the IoT itself is insufficient. Long-term continuous operation of the big data IoT system has the risk of performance degradation or even partial fault, which leads to a decrease in the availability of collected data for intelligent analysis. In this article, based on the power IoT message data, the characteristics are established through a variety of improved detection methods, and then the abnormal data type is obtained through
Q
learning and fusion of the random forest (RF) identification features. Finally, the topology of the specific power user IoT system is combined with kernel principal component analysis (KPCA) + improved RF algorithm getting the abnormal location of the IoT. The results show that the research method has a significantly higher positioning accuracy (from 61% to 97%) than the traditional RF method, and the combination method has more advantages in parameter adjustment and classification accuracy than directly using a multilayer perceptron (MLP).
In recent years, advances in computer technology and the emergence of big data have enabled deep learning to achieve impressive successes in bearing condition monitoring and fault detection. While existing deep learning approaches are able to efficiently detect and classify bearing faults, most of these approaches depend exclusively on data and do not incorporate physical knowledge into the learning and prediction processes—or more importantly, embed the physical knowledge of bearing faults into the model training process, which makes the model physically meaningful. To address this challenge, we propose a physics-informed deep learning approach that consists of a simple threshold model and a deep convolutional neural network (CNN) model for bearing fault detection. In the proposed physics-informed deep learning approach, the threshold model first assesses the health classes of bearings based on known physics of bearing faults. Then, the CNN model automatically extracts high-level characteristic features from the input data and makes full use of these features to predict the health class of a bearing. We designed a loss function for training and validating the CNN model that selectively amplifies the effect of the physical knowledge assimilated by the threshold model when embedding this knowledge into the CNN model. The proposed physics-informed deep learning approach was validated using (1) data from 18 bearings on an agricultural machine operating in the field, and (2) data from bearings on a laboratory test stand in the Case Western Reserve University (CWRU) Bearing Data Center.
One-dimensional vibration signals are widely used for gearbox fault diagnosis to perform maintenance timely and then reduce various losses. The fault diagnosis accuracy of those classifiers is determined by the features extracted from the vibration signals. These typical deep neural networks (DNNs), e.g., convolutional neural network (CNN) and convolutional autoencoder (CAE) have been applied in machinery fault diagnosis. However, the feature learning of DNNs on the nonlinear vibration signals is still a big challenge for gearbox fault diagnosis. In this article, a new CNN, multiscale fusion global sparse network (MFGSNet) is proposed for feature extraction from vibration signals and gearbox fault diagnosis. First, a novel kernel dynamic fusion method based on multiscale convolution is proposed to extract defect features of vibration signals; second, a global dense connection method is proposed to generate features from shallow and deep layers; finally, a spare feature selection layer is embedded in the deep network to perform feature selection and dimension reduction on the shallow and deep features. The experimental results show that MFGSNet has good fault feature extraction performance on gearbox vibration signals. It performs much better on gearbox fault diagnosis than these typical DNNs, e.g., residual network (ResNet) and dense connection network (DenseNet).
Early detection and diagnosis of the chiller sensor drift fault are crucial to maintain normal operation for energy saving. Due to the complex physical structure and operation conditions, sensor drift fault in the chiller system is difficult to discover. To improve the energy efficiency and operation reliability of the chiller system, this paper proposes a novel chiller sensor drift fault diagnosis method using deep recurrent canonical correlation analysis and k-nearest neighbor (KNN) classifier. A deep bidirectional long short-term memory recurrent neural network-based deep recurrent canonical correlation analysis (BLCCA) model is developed, which can automatically extract the nonlinear and temporal features from raw operation data in the chiller system. Based on the proposed BLCCA model, a residual generator is designed to generate the directional residual vector. The cumulative residual vector method is employed to improve the detectability of the sensor drift fault. An efficient KNN-based method is applied to classify the residual vector and judge the faulty sensor. Different distance measures and neighbor numbers are further analyzed to optimize the fault diagnosis performance. The proposed fault detection and diagnosis (FDD) method is validated by using a data set which has been collected from an actual chiller system. Three different state-of-the-art fault diagnosis methods are used for comparison with the proposed method. The comparisons of the experimental results demonstrate that this method achieves significant fault diagnosis performance in terms of diagnosis accuracy, recall, and F measure (F1 score).
In engineering applications, normal signals are often much easier to obtain than fault signals, which makes training dataset appear class imbalance to affect the training of deep autoencoder. Besides, the anti-interference and robustness of the deep autoencoder are extremely susceptible to the influence of the noise environment, resulting in insufficient state feature extraction. Moreover, the selection of hyperparameter has no theoretical guidance, and needs a lot of experiments and manual experience to determine the appropriate parameters. To address the aforementioned shortcomings, this paper proposes a novel deep autoencoder and hyperparametric adaptive learning for intelligent fault diagnosis of rotating machinery. Firstly, a data augmentation strategy based on overlapping segmentation, rotation and jitter, called ORJ data augmentation strategy, is proposed to increase the data capacity of training samples and reduce the influence of class imbalance on deep autoencoder training. Then, a lossless constraint term and a weight nonnegative constraint term modification loss function are proposed to construct a deep sparse autoencoder with lossless and nonnegative constraints (SAE-LNC), which improves the robustness of the model and makes the extracted state features have higher quality. Finally, the artificial bee colony algorithm (ABC) is used to adaptively learn the hyperparameters of deep SAE-LNC to improve the intelligence and accuracy of the diagnostic model. The experimental results of vibration signals collected from bearing test-bed show that, compared with other diagnostic methods, the proposed method can effectively capture high-quality state features, enhance the robustness of feature extraction process and obtain higher diagnostic accuracy.
With evaluating the randomness and the nonlinear dynamic change of time sequence effectively, multiscale fuzzy distribution entropy (MFDE) is proposed to extract fault features from vibration signal. However, it is unsuitable for multivariate signals, which contain more abundant fault information. Through changing the coarse-grained and calculation process of MFDE, a novel entropy named as multivariate multiscale fuzzy distribution entropy (MMFDE) is proposed, which not only has the characteristics of MFDE, but also considers the within- and cross-channel correlations. By applying MMFDE to two kinds of simulation signals, the results show the proposed entropy is stable and reliable. A novel fault diagnosis method for rotating machinery is proposed by extracting fault features with MMFDE, selecting sensitive ones with Fisher score (FS), and identifying working state with support vector machine (SVM). Two experiments show the better diagnosis performance of the proposed rotating machinery fault diagnosis method than the related methods with the accuracy of 98.95% and 96.80% respectively.
With the rapid development of sensor and computer technology, deep learning has received extensive attention in the field of fault detection with powerful nonlinear feature extraction capabilities. However, the feature extracted by deep learning contains different fault information volume. Fault detection performance would be affected by features with less fault information. Motivated by this, the deep high-dimensional features extracted from the deep belief network are analyzed, and an index for measuring fault information volume is proposed to select the deep highly-sensitive feature (DHSF) with a large amount of fault information volume as the feature to be detected. Based on DHSFs, Euclidean distance is used for fault detection, and the moving average window function is used to reduce burst noise interference and improve detection performance. Finally, a numerical case and Tennessee Eastman process demonstrate the advantage of the proposed method in fault detection results compared with other methods.
Accurate fault diagnosis is essential to ensure the stability and reliability of machine systems. Deep learning (DL) techniques have demonstrated potential in realizing intelligent fault diagnosis. However, the obtained bearing signals in the actual environment involve various interferences and are consistently nonstationary. Moreover, it is difficult to simultaneously acquire a large amount of balanced data to ensure suitable training. Consequently, traditional DL-based methods cannot achieve a satisfactory diagnostic accuracy. To overcome this problem and enhance the diagnostic performance in complex situations, this paper proposes a method to realize effective intelligent bearing fault diagnosis. A novel index is established to evaluate the quality of the demodulation spectra, along with a novel compressed sensing deep autoencoder based on compressed sensing. Additionally, a multiscale neural network architecture incorporating a multiscale coarse-grained procedure is adopted. Experimental results and comparative analysis with existing intelligent diagnosis methods demonstrate the effectiveness and robustness of the proposed method.
Data driven-based intelligent fault diagnosis methods, as a promising approach, have been widely employed in the health management and maintenance decision of rotating machinery. However, the domain shift phenomenon caused by internal and external interference inevitably exists in practical application scenarios, which significantly deteriorates the performances of the intelligent diagnosis model. And the preparation of label information in real complex scenes is usually time-consuming and expensive. To overcome these challenges, a novel unsupervised domain adaptation framework named deep multi-scale adversarial network with attention (MSANA) is introduced for machinery fault diagnosis. It is established based on two main components, one is the shared feature generator, which is constructed by two novel multi-scale modules with attention mechanism, and the other part is a fault pattern recognition module composed of two differentiated discriminators. While the multi-scale module is used to obtain rich features through different internal perceptual scales, the attention mechanism determines the weights of different scales, which promotes the dynamic adjustment performance and adaptive ability of the model. Then, decision boundary assisted adversarial learning strategy is employed to eliminate domain distribution differences and obtain domain-invariant features. A total of ten rolling bearing-based transfer scenarios and six gearbox-based transfer scenarios are adopted to evaluate the transferability of the proposed MSANA model, and the cross-domain transfer results show that it has superior transferability and stability.
Deep neural networks (DNNs) are popular in process monitoring for its remarkable feature extraction from data. However, the increased dimension and correlation of the process variables degrade performance of these DNNs in feature extraction of data. This paper proposes a sparsity and manifold regularized convolutional auto-encoders (SMRCAE) for fault detection of complex multivariate processes. SMRCAE can learn high-level features from the data in an unsupervised way. A sparsity-and-manifold-regularization term is integrated into the learning procedure of SMRCAE, which allows SMRCAE to perform feature selection and capture intrinsic data information. Moreover, a depthwise separable convolution (dsConv) block is used to reduce the computational cost. Two typical fault detection statistics, namely Hotelling’s T-squared (T2) and the squared prediction error (SPE), are developed on the feature space and residual space of SMRCAE, respectively. The performance of SMRCAE is evaluated on an industrial benchmark, i.e., Tennessee Eastman process (TEP) and a real process of industrial conveyor belts. The experimental results show the feasibility of SMRCAE in extracting representative features for process fault detection. The average fault detection rate of SMRCAE is 92.03% and 100% on the two cases, respectively.
Nonlinear process modeling is a primary task in intelligent manufacturing, aiming at extracting high-value features from massive process data for further process analysis like process monitoring. However, it is still a challenge to develop nonlinear process models with robust representation capability for diverse process faults. From the new perspective of the correlation between process variables, this paper develops a nonlinear process modeling algorithm to adaptively preserve the features of both global and local inter-variable structures, in order to fully exploit inter-variable features for enhancing the nonlinear representation of process operating conditions. Specifically, a unidimensional convolutional operation with a self-attention mechanism is proposed to simultaneously extract global and local inter-variable structures, wherein different attentions can be adaptively adjusted to these two structures for the final aggregation of them. Besides, cooperating with a two-dimensional dynamic data extension, the unidimensional convolutional operation can represent the overall temporal relationship between process samples. Through stacking a collection of these convolutional operations, a ResNet-style convolutional neural network then is constructed to extract high-order nonlinear features. Experiments on the Tennessee Eastman process validate the effectiveness of the proposed algorithm for two vital process monitoring problems—fault detection and fault identification.
This paper deals with the development of a novel deep learning framework to achieve highly accurate rotating machinery fault diagnosis using residual wide-kernel deep convolutional auto-encoder. Unlike most existing methods, in which the input data is processed by fast Fourier transform (FFT) and wavelet transform, this paper aims to learn important features from limited raw vibration signals. Firstly, the wide-kernel convolutional layer is introduced in the convolution auto-encoder that can ensure the model can learn effective features from the data without any signal processing. Secondly, the residual learning block is introduced in convolutional auto-encoders that can ensure the model with sufficient depth without gradient vanishing and overfitting problems. Thirdly, convolutional auto-encoders can learn constructive features without massive data. To evaluate the performance of the proposed model, Case Western Reserve University (CWRU) bearing dataset and Southeastern University (SEU) gearbox dataset are tested. The experiment results and comparisons verify the denoising and feature extraction ability of the proposed model in the case of very few training samples.
Real-world data, especially in industrial processes in which a property only fully retains normal historical data, are usually unlabeled or incorrectly labeled. Moreover, data from industrial processes are often time-dependent and associated with event sequences. These characteristics make process monitoring based on a current gauged sample unreasonable, thus requiring performance improvement. To overcome this bottleneck, this study proposes a deep unLSTM network to extract features with memory information from unlabeled data. The network comprises three parts, namely, encoder, code layer, and decoder. A normal historical training dataset is converted into a 3D input-form, and each layer outputs the 2D information, which contains all the states of an event sequence for connection to the next LSTM layer to the reconstruction layer. Additionally, a slicing operation is developed to provide accurate features rich in current memory information when constructing statistics. What’s more, except traditional process monitoring statistics, a similarity statistic based on global hidden layer features is proposed to indicate the working state of the current process more comprehensively. Compared with other deep models used on the Tennessee Eastman and real steel plate processes, the proposed deep unLSTM network, which can extract the features with memory information, is more suitable for industrial fault detection.
Deep learning has gained a great achievement in the intelligent fault diagnosis of rotating machineries. However, the labeled data is scarce in actual engineering and the marginal distribution of data is discrepant under different conditions. Transfer learning provides a feasible way to overcome these difficulties. Considering the effect of noise on the transfer fault diagnosis, this work puts forward a new deep transfer learning network based on convolutional auto-encoder(CAE-DTLN) to implement the mechanical fault diagnosis in target domain without labeled data. In the proposed framework, CAE is used as the feature extractor as it has the ability of noise removal. Moreover, both CORrelation ALignment (CORAL) loss and domain classification loss are integrated to enhance the effect of domain confusion. The proposed model is applied to the fault transfer diagnosis of planetary gearboxes under different working loads and noise levels, and it is compared with other typical fault transfer diagnosis models. The experimental results show that CAE-DTLN has higher diagnosis accuracy and stronger generalization ability. The average diagnostic accuracy of CAE-DTLN is over 99%. Moreover, the proposed transfer learning model has better anti-noise performance.
Vibration signals have been widely used for machine health monitoring and fault diagnosis. However, due to the complex working conditions, vibration signals collected from gearbox are generally nonlinear and non-stationary, which may contain multiple time scales and much noise. Considering these physical characteristics of vibration signals, in this paper, a novel deep neural network (DNN), attentive kernel residual network (AKRNet), is proposed for multi-scale feature learning from vibration signals. Firstly, multiple branches with different kernel widths are used to extract multi-scale features from vibration signals. Secondly, an attentive kernel selection is proposed to fuse the multiple branches features, where dynamic selection is developed to adaptively highlight the informative feature maps, while suppress the useless feature maps. Thirdly, an attentive residual block is developed to improve the feature learning performance, which not only can alleviate gradient vanishing, but also further enhances the impulsive features in feature maps. Finally, the effectiveness of AKRNet for feature learning of vibration signals is verified on two gearbox test rigs. The experimental results show that AKRNet has good capacity of feature learning on vibration signals. It performs better on gearbox fault diagnosis than other typical DNNs, e.g., stacked auto-encoders (SAE), one-dimensional CNN (1-D CNN) and residual network (ResNet).
In practical engineering, data imbalance is an urgent problem to be solved for rolling bearing fault diagnosis. This paper proposes a unified framework incorporating predictive generative denoising autoencoder (PGDAE) and deep Coral network (DCN). The proposed framework mainly comprises two parts i.e. generative model and diagnosis model. The generative model PGDAE is used to generate extra fault data, and it is constructed by gated recurrent unit. By this way, unbalanced dataset can reach equilibrium. The diagnosis model DCN is used for fault recognition, which is constructed by a deep convolutional neural network with correlation alignment. Correlation alignment is used to adapts the class-specific features learned from real data and simulated data. Finally, experimental bearing data are used to evaluate the performance. The results show that the generative model can effectively generate workable fault data and the diagnosis model can accurately recognize the fault modes.
Numerous intelligent methods have been developed to approach the challenges of fault diagnosis. However, due to the different distributions of training samples and test samples, and the lack of information on test samples, most of these methods cannot directly handle the unsupervised cross-domain fault diagnosis issues. In this paper, a joint distribution adaptation network with adversarial learning is developed to effectively tackle the mentioned fault diagnosis issues. Firstly, deep convolutional neural network (CNN) is constructed to extract the features of training samples and test samples. Secondly, since the joint maximum mean discrepancy (JMMD) cannot precisely measure the joint distribution discrepancy between different domains, an improved joint maximum mean discrepancy (IJMMD) is proposed to accurately match the feature distributions. Finally, adversarial domain adaptation is also developed to help the constructed CNN to extract the domain-invariant features. Therefore, the proposed method can achieve precisely distribution matching, and extract the category-discriminative and domain-invariant features between the source and target domains. Substantial transfer fault diagnosis cases based on three rolling bearing datasets fully demonstrate the effectiveness and generalization ability of the proposed method.
Due to limited conditions of production sites, only the small fault dataset (target dataset) of the rolling bearing can be collected, which leads to the failure construction of the effective deep learning network. Aiming at the above problems, the sufficient fault dataset (source dataset) of other type of rolling bearing is introduced as the auxiliary, and thus a new transfer learning network based on convolutional neural network (CNN) is proposed. The new transfer learning network is with a new structure, and it is trained by a new training strategy, and then it is optimized by a new optimal fusion method of dropout layer 4 and L2 regularization. The measured fault signals of the rolling bearings are tested and verified, and results demonstrate that the proposed transfer learning network has low computation cost, high accuracy and strong diagnosis ability. Furthermore, it performs much better than the traditional transfer learning networks.
High precision and fast fault diagnosis is an important guarantee for the safe and reliable operation of machinery. In recent years, due to the strong recognition ability, data-driven fault diagnosis technology based on deep learning has attracted enormous attention. The fault diagnosis module proposed by many scholars have achieved excellent recognition results, but some of them are too complex to deploy in practice, due to the high costs. In this paper, an efficient feature extraction method based on the Convolutional Neural Networks (CNN) was proposed, and the high-precision fault diagnosis task was completed used a lightweight network. Firstly, a One-Dimensional CNN (1D-CNN) is used to extract the multi-channel features for the input original signals, to generate the feature maps which are highly consistent with the Two-Dimensional CNN (2D-CNN). Secondly, a strategy of lightweight model is proposed based on the principle of dynamic convolution and separable convolution. Thirdly, a Spatial Attention Mechanism (SAM) is used to adjust the weights of the output feature maps. Finally, the model completes the diagnosis task through the learned features. The performance of the model was verified under different operating conditions and noise environments. The experimental results demonstrate that the proposed method has excellent anti-noise ability and domain adaptability.
Rolling element bearings are critical components in industrial rotating machines. Faults and failures of bearings can cause degradation of machine performance or even a catastrophe. Therefore, it is significant to perform bearing fault diagnosis accurately and effectively. Deep Learning based approaches are promising for bearing diagnosis. They can extract fault information efficiently and conduct accurate diagnosis. However, the structure of deep learning networks is often determined by trial and error, which is time consuming and lacks theoretical guidance. Besides, the traditional deep learning approaches have low diagnosis accuracy and learning efficiency. To address these problems, this paper proposes a rolling element bearing fault diagnosis approach based on principal component analysis and adaptive deep belief network with Parametric Rectified Linear Unit activation layers. In the proposed approach, particle swarm optimization is integrated to obtain an optimal DBN structure with high accuracy and convergence rate. Experiments on tapered roller bearings and comparison studies with state-of-the-art methods are conducted to demonstrate the effectiveness and accuracy of the proposed approach.
Intelligent fault diagnosis based on deep neural networks and big data has been an attractive field and shows great prospects for applications. However, applications in practice face following problems. (1) Unexpected and unseen faults of machinery in real environment may be encountered. (2) Large collections of healthy condition samples and few fault condition samples result in the imbalanced distribution of machinery health conditions. This paper proposes a novel deep metric learning model, where machinery condition is classified by retrieving similarities. The trained deep metric learning model can learn and recognize new faults quickly and easily to address the first problem. As core of deep metric learning, a novel loss function called normalized softmax loss with adaptive angle margin (NSL-AAM) is developed for second problem. NSL-AAM can supervise neural networks learning imbalanced data without altering the original data distribution. Experiments for balanced and imbalanced fault diagnosis are conducted to verify the ability of the proposed model for fault diagnosis. The results demonstrate that the proposed model can not only extract more distinctive features automatically, but also balance the representation of both the majority and minority classes. Furthermore, the results of experiments for diagnosing new faults are reported, which proves the capability of the trained model for open-set classification.
For the application of deep learning in the field of fault diagnosis, its recognition accuracy is limited by the size and quality of the training samples, such as small size samples, low signal-to-noise ratio and different working conditions. In order to solve above problems, one novel method for fault classification is proposed based on a Bidirectional Long Short-Term Memory (Bi-LSTM) and a Capsule Network with convolutional neural network (BLC-CNN). The Bi-LSTM is utilized to achieve the feature denoising and fusion, which is extracted by CNN. The fault diagnosis with insufficient training samples is carried out by the capsule network. The influence of sample size on the method is discussed emphatically. The effectiveness and superiority of the proposed method are validated through analyzing the data of bearings and gears under different working conditions with different noise. The results indicate that the proposed method has good performance and immunity to noise.
Deep learning, especially transfer learning, has made a great deal of extraordinary achievements in intelligent fault diagnosis. In practical situations, data shift problem is inevitable due to complicated and changeable working conditions. Ignoring this problem may result in considerable degradation of diagnostic accuracy. Thus, domain distance is measured in only one metric space, and the result of domain alignment may not be ideal. This paper proposes a novel transfer learning method named hybrid distance-guided adversarial network (HDAN) to deal with this problem. Specifically, HDAN contains two parts: a feature extractor composed of a convolutional neural network and a shared classifier. Wasserstein distance and multi-kernel maximum mean discrepancy are applied in the proposed method to measure the domain distance in different metric spaces for improving the result of domain alignment. The domain distance of the last two hidden layers is minimized to improve the efficiency of domain-invariant feature extraction. Two experiments are implemented to confirm the superiority of the proposed method. The results of two experiments demonstrate that the proposed HDAN can achieve better feature cluster performance of the same class in different domains than the methods selected for comparison.
The long-term safe operation of rotating machinery is closely related to the stability of rolling bearings. This paper proposes a rolling bearing fault diagnosis method based on refined composite multiscale fuzzy entropy (RCMFE), topology learning and out-of-sample embedding (TLOE), and the marine predators algorithm based-support vector machine (MPA-SVM). First, the RCMFE algorithm is used to extract the features of the original rolling bearing fault signal and to construct the original high-dimensional fault feature set. Then, TLOE is used to reduce the dimensionality of the high-dimensional fault feature set. The low-dimensional sensitive fault features are extracted to construct a low-dimensional fault feature subset. Finally, fault-type discrimination is performed using the MPA-SVM. The Case Western Reserve University dataset and data from fault diagnosis experiments performed on 1210 self-aligning ball bearings were used to verify the proposed method. The results demonstrate the effectiveness of the fault diagnosis method, which can diagnose bearing faults with up to 100% accuracy.
The most existing deep neural networks (DNN)-based methods for fault diagnosis only focus on prediction accuracy without considering the limitation of labeled sample size. In practical applications of DNN-based methods, it is time-consuming and costly to collect massive labeled samples. In this paper a task named few-shot fault diagnosis is defined as training model given small labeled samples in source domain and testing given small samples in target domain. We develop a novel intelligent fault diagnosis model for few-shot fault diagnosis which is using similarities of sample pairs to classify samples, rather than end-to-end classification. The proposed model contains modules of feature learning and metric learning. The module of feature learning has twin neural networks aiming to extract features from the sample pair. The module of metric learning is to predict similarity of the sample pair. The similarities of sample pairs combined the test sample with each labelled sample are utilized to complete the classification task. Label smoothing is utilized to further improve performance of classification. The performance of the proposed model is verified by two fault diagnosis cases which are bearing fault diagnosis cross different working conditions and cross bearing locations. The comparison studies with other models demonstrate the superiority of the proposed model.
Monitoring a system is often not an easy task and the best approach to address it would be to develop a monitoring system that uses data, expert knowledge, and mathematical models. Combining these three sources of information on the system is often unpractical of various reasons such as in complex systems. In this paper, a hybrid method for diagnosing single and multiple simultaneous faults, while considering unknown operating conditions, is proposed. The proposed method is integrated in a Bayesian network classifier with innovative decision rules combining statistical decisions and fault signature matrix. A heating water process is utilized for illustrative and evaluation purposes, in which several scenarios of operating conditions are simulated. The results, in terms of classification rates, show superior performance for the proposed method over a purely data-driven method. It is worth mentioning that the proposed method shows higher capabilities to isolating multiple simultaneous faults that the data-driven method fails to accomplish.
Applications of deep transfer learning to intelligent fault diagnosis of machines commonly assume symmetry among domains: 1) the samples from target machines are balanced across all health states, and 2) the diagnostic knowledge required by target machines is consistent with source machines. In reality, however, such assumptions cannot be justified as machines operate normally in most of the time with only occasional faults. As a result, the collected monitoring data from target machines contain massive healthy samples but a small number of faulty samples, and some health states experienced in source machines may never happen in target machines. Therefore, if sufficient labeled data are available with diverse health states from the source machines, only partial diagnostic knowledge can be transferred to a target machine in presence of domain asymmetry. In order to selectively transfer diagnostic knowledge across asymmetric domains, we propose an adversarial network architecture named deep partial transfer learning network (DPTL-Net). The DPTL-Net uses a domain discriminator to automatically learn domain-asymmetry factors, by which the source machine samples are weighted to block irrelevant knowledge in the maximum mean discrepancy based distribution adaptation. The performance of the DPTL-Net is demonstrated in two case studies, where the diagnostic knowledge is transferred across different working conditions of a planet gearbox and across different yet related bearings. The results show that the DPTL-Net achieves better diagnostic performance than other transfer learning methods due to its transfer capability in presence of domain asymmetry.
An inconsistent distribution between training and testing data caused by complicated and changeable machine working conditions hinders wide applications of traditional deep learning for machine fault diagnosis. In a target domain, in which labeled samples are not available (testing data), transfer learning can adopt a relevant source domain (training data) to identify the similarity between the two domains and subsequently mitigate negative effects of a domain shift. Previous studies on transfer learning mainly focused on decreasing the marginal distribution distance of two different domains or narrowing the conditional distribution distance even though marginal and conditional distributions provide different contributions to transfer tasks. The relative importance of the two distributions is difficult to dynamically and quantitatively assess. To align the two distributions (joint distribution) of two different domains, in this paper, we propose a dynamic joint distribution alignment network (DJDAN) to evaluate the relative importance of marginal and conditional distributions dynamically and quantitatively. Furthermore, compared with common metrics that use pseudo labels to calculate the conditional distribution distance, the proposed DJDAN uses soft pseudo labels to more accurately measure the conditional distribution discrepancy between different domains. Extensive experiments reveal the superiority and generalization of the proposed DJDAN for bearing fault diagnosis under different working conditions.
Deep learning-based bearing fault diagnosis has been systematically studied in recent years. However, the success of most of these methods relies heavily on massive labeled data, which is not always available in real production environments. Training a robust bearing fault diagnosis model with limited data and working well under complex working conditions remains a challenge. In this paper, a novel meta-learning fault diagnosis method (MLFD) based on model-agnostic meta-learning is proposed to address this issue. The raw signals of different working conditions are first converted to time-frequency images and then randomly sampled to form tasks for MLFD according to the protocol of meta-learning. The MLFD model acquires prior knowledge by optimizing initialization parameters based on multiple fault classification tasks of known working conditions during the meta-training process, and achieves fast and accurate few-shot bearing fault diagnosis under unseen working conditions by leveraging the learned knowledge. To comprehensively evaluate the performance of our method, a series of experiments were conducted to simulate different industrial scenarios based on the Case Western Reserve University Bearing Fault Benchmark, and the results demonstrate the superiority of MLFD in solving the few-shot fault classification problem under complex working conditions.
Recently, the deep transfer learning approaches have been widely developed for mechanical fault diagnosis issue, which could identify the health state of unlabeled data in the target domain with the help of knowledge learned from labeled data in the source domain. The tremendous success of these methods is generally based on the assumption that the label spaces across different domains are identical. However, the partial transfer scenario is more common for industrial applications, where the label spaces are not identical. This partial transfer scenario arises a more difficult problem that it is hard to know where to transfer since the shared label spaces are unavailable. To tackle this challenging problem, a double-layer attention based adversarial network (DA-GAN) is proposed in this paper. The proposed method sheds a new angle to deal with the question where to transfer by constructing two attention matrices for domains and samples. These attention matrices could guide the model to know which parts of data should be concentrated or ignored before conducting domain adaptation. Experimental results on both transfer in the identical machine (TIM) and transfer on different machines (TDM) suggest that the DA-GAN model shows great superiority on mechanical partial transfer problem.