Search

The intershaft bearing is located between the high and low-pressure rotors of the aero-engine, where the working environment is harsh, the load variation range is large, and the lubrication and heat dissipation are poor. The fault of the intershaft bearing is sudden and will cause the engine to hold the shaft and break the shaft, which is more harmful than the ordinary bearings. In recent years, bearing intelligent fault diagnosis methods based on deep learning have been widely applied. However, most of the existing methods are mainly proposed based on second-generation neural networks. Spiking neural network (SNN), also known as the third-generation neural network, mimics the dynamics of the biological brain and is more powerful for processing time-series information. As we know, vibration data is a typical time-series data, SNN would have stronger feature extraction potential for it. In this paper, we propose an improved spiking neural network (ISNN) for intershaft bearing fault diagnosis. Specifically, we propose an encoding method to encode raw data into spike sequences and demonstrate that the encoding method is accurate and efficient. Then we derive the gradient relation in the ISNN and mathematically prove the replacement of invalid gradients in it. Furthermore, we compensate for the loss of information in forward propagation and simplify the process of backpropagation by constructing suitable spiking neurons. Finally, we test the ISNN on the fault dataset of intershaft bearings, and the results show that the proposed ISNN outperforms previous SNNs and typical second-generation neural networks.

Various working conditions and bearing structures make remaining useful life (RUL) prediction more challenging. This paper presents a new intermediate domain support vector machine (SVM)-based transfer model for rolling bearing RUL prediction. This transfer model aims to solve the problem when the source degradation indexes are too poor to be applied by introducing the new intermediate domain. At first, high-quality feature degradation indexes are selected using the joint evaluation index and the principal component analysis (PCA) algorithm. Then, a maximum correlated kurtosis de-convolution (MCKD) algorithm is carried out to obtain the demarcation point between healthy and degradation stages. After selecting high quality domain, the objective function of intermediate domain SVM is designed, based on classical domain independent SVM, to optimize source to intermediate and intermediate to target transfer processes simultaneously. Finally, experiments using both ball and conical bearing datasets indicate that the proposed method has higher RUL prediction performance than existing models, which proves the advantage of multi-optimization transfer learning.

Deep learning (DL) theory has made great progress in the field of intelligent fault diagnosis, and the development of domain adaptation has greatly promoted fault diagnosis under polytropic working conditions (PWC). Extensive studies have been conducted to solve the problem of fault diagnosis under PWC. However, the existing fault diagnosis methods based on domain adaptation have the following shortcomings. First, multi-source information fusion is rarely considered. Second, the utilization of inherent labels is also insufficient in classification problems. To deal with the above problem, a novel multi-source dense adaptation adversarial network (MDAAN) is proposed, which leverages multi-sensor vibration information and classification label information. Specifically, the frequency spectrum of multi-sensor data are firstly employed to make full use of fault information. Afterwards, the dense convolution and fusion convolution blocks are used for deep feature extraction and fusion. Finally, a joint loss function is reconstructed under the framework of unsupervised learning, which considers the distribution differences of the features and the label information simultaneously. The experimental results from various working conditions, including still distant working conditions, all demonstrate that the proposed method can achieve state-of-the-art performances, which has shown great promise as an intelligent fault diagnosis method.

Accurately and intelligently identifying faults of the planetary gearbox is essential in the safe and reliable operation and maintenance of the mechanical drive system. Recently, fault diagnosis of planetary gearbox has acquired tremendous progress, especially with the rising popularity of deep learning (DL). However, most methods are standard supervised learning (SL) where the input is directly mapped to a fault type, and with strong feedback. Also, their learning ways are static and unlike human's learning that gradually acquires knowledge by interaction with environment. To a certain extent, these deficiencies reduce the generalization and intelligence level of DL-based fault diagnosis methods. Besides, due to harsh working conditions, signals acquired often have strong noise and nonlinear features, leading to relatively low accuracy if raw signals are used as the input directly. Thus, this work proposes a new fault diagnosis method based on time-frequency representation (TFR) and deep reinforcement learning (DRL). We first define fault diagnosis as a sequential decision-making problem in the classification Markov decision process (CMDP). Next, the vibration signals are converted to uniform-sized TF maps by synchro-extracting transform (SET) to enhance the robustness of feature representation. Finally, a diagnosis agent is built and trained in the framework of DRL to learn the optimal classification policy automatically. Experimental results show that this method not only achieves better generalization and stability with an overall accuracy of over 99.5% in single-speed load cases but also outperforms others in multi-work conditions. Index TermsClassification policy, deep learning, deep reinforcement learning, fault diagnosis, planetary gearbox.

The electromechanical impedance (EMI)-based damage identification method is a non-destructive testing approach in the field of structural health monitoring. The frequency response function (FRF) of EMI can effectively reveal the health conditions of a structure. Typically, the health condition is identified by comparing the FRF of a structure to that of a baseline. However, baselines may exhibit unpredictable shifts in real applications. In this study, a new EMI-based health identification method is proposed without reference to baselines or handcrafted features. An adjustable inductive shunt circuit that can enrich the EMI dataset is connected to a piezoelectric transducer. Pre-set damage, including bolt looseness and mass variations, are selected to demonstrate damage identification. The FRFs are extracted using a phase-sensitive detection algorithm. The damage identification model is realized using a one-dimensional convolutional neural network. Experimental results show that the proposed method can identify the location of bolt loosening and mass variation with an overall accuracy of 99.24%. The proposed method can be applied for identifying the health conditions of a structure with strong nonlinearity.

Deep learning technology provides a promising approach for rotary machine fault diagnosis (RMFD), where vibration signals are commonly utilized as input of a deep network model to reveal the internal state of machinery. However, most existing methods fail to mine association relationships within signals. Unlike deep neural networks, transformer networks are capable of capturing association relationships through the global self-attention mechanism to enhance feature representations from vibration signals. Despite this, transformer networks cannot explicitly establish the causal association between signal patterns and fault types, resulting in poor interpretability. To tackle these problems, an interpretable deep learning model named the variational attention-based transformer network (VATN) is proposed for RMFD. VATN is improved from transformer encoder to mine the association relationships within signals. To embed the prior knowledge of the fault type, which can be recognized based on several key features of vibration signals, a sparse constraint is designed for attention weights. Variational inference is employed to force attention weights to samples from Dirichlet distributions, and Laplace approximation is applied to realize reparameterization. Finally, two experimental studies conducted on bevel gear and bearing datasets demonstrate the effectiveness of VATN to other comparison methods, and the heat map of attention weights illustrates the causal association between fault types and signal patterns.

Most current data-driven prognosis approaches suffer from their uncontrollable and unexplainable properties. To address this issue, this article proposes a physics-constraint variational neural network (PCVNN) for wear state assessment of the external gear pump. First, a response model of the pressure pulsation of the gear pump is constructed via a spectral method, and a compound neural network is utilized to extract features from the pressure pulsation signal. Then, the response model is formulated into an objective function to softly constrain the learning process of the neural network, forcing the learned features to have explicit physics meaning. Meanwhile, to characterize the system uncertainty, the variational inference is utilized to extend a Kullback-Leibler ( KL) divergence into the objective function. Finally, the wear state is evaluated based on the distance of learned physics features. Experimental results on an external gear pump validate the merits of the proposed method in explainable representation learning and system uncertainty estimation. It also offers a controllable and explainable perspective to understand the dynamic behavior of the system.

Deep learning (DL) is progressively popular as a viable alternative to traditional signal processing (SP) based methods for fault diagnosis. However, the lack of explainability makes DL-based fault diagnosis methods difficult to be trusted and understood by industrial users. In addition, the extraction of weak fault features from signals with heavy noise is imperative in industrial applications. To address these limitations, inspired by the Filterbank-Feature-Decision methodology, we propose a new Signal Processing Informed Neural Network (SPINN) framework by embedding SP knowledge into the DL model. As one of the practical implementations for SPINN, a denoising fault-aware wavelet network (DFAWNet) is developed, which consists of fused wavelet convolution (FWConv), dynamic hard thresholding (DHT), index-based soft filtering (ISF), and a classifier. Taking advantage of wavelet transform, FWConv extracts multiscale features while learning wavelet scales and selecting important wavelet bases automatically; DHT dynamically eliminates noise-related components via point-wise hard thresholding; inspired by index-based filtering, ISF optimizes and selects optimal filters for diagnostic feature extraction. It's worth noting that SPINN may be readily applied to different deep learning networks by simply adding filterbank and feature modules in front. Experiments results demonstrate a significant diagnostic performance improvement over other explainable or denoising deep learning networks. The corresponding code is available at https://github.com/albertszg/DFAWnet.

Conventional piezoelectric cantilever-based vibration energy harvesters have narrow bandwidth. In this article, we develop a dual-beam piezoelectric energy harvester featuring an annular potential energy function that can harvest vibration energy over a wide spectrum under small amplitude excitations. The proposed harvester contains two conventional piezoelectric cantilevers placed orthogonal to each other which are coupled by repulsive magnetic force. We demonstrate analytically and numerically that a new annular potential energy function can be configurated with proper configuration. In the new annular stable state, the harvester can detour around the potential barrier rather than jump over it, yielding large-amplitude voltage outputs throughout a wide spectrum. Case studies were carried out and it is proved that the proposed annular stable harvester has a bandwidth of 3.9 Hz and a voltage output performance 3.01 times better than that of a conventional bistable one under excitations of 3 m/s2. The nonlinear dynamics of the proposed harvester are analyzed in detail.

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. First, most GCNs are constructed on unweighted graphs, considering importance of neighbors as the same, which is not in line with reality. Second, the receptive field of GCNs is fixed, which limits the effectiveness of GCNs for feature representation. To address these issues, a multireceptive 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.

Accurate estimation of remaining useful life (RUL) of industrial equipment can enable advanced maintenance schedules, increase equipment availability and reduce operational costs. However, existing deep learning methods for RUL prediction are not completely successful due to the following two reasons. First, relying on a single objective function to estimate the RUL will limit the learned representations and thus affect the prediction accuracy. Second, while longer sequences are more informative for modelling the sensor dynamics of equipment, existing methods are less effective to deal with very long sequences, as they mainly focus on the latest information. To address these two problems, we develop a novel attention-based sequence to sequence with auxiliary task (ATS2S) model. In particular, our model jointly optimizes both reconstruction loss to empower our model with predictive capabilities (by predicting next input sequence given current input sequence) and RUL prediction loss to minimize the difference between the predicted RUL and actual RUL. Furthermore, to better handle longer sequences, we employ the attention mechanism to focus on all the important input information during the training process. Finally, we propose a new dual-latent feature representation to integrate the encoder features and decoder hidden states, to capture rich semantic information in data. We conduct extensive experiments on four real datasets to evaluate the efficacy of the proposed method. Experimental results show that our proposed method can achieve superior performance over 13 state-of-the-art methods consistently.

This paper presents a new scheme for the control of unknown block-oriented nonlinear systems using spline adaptive filter (SAF). The principle of adaptive inverse control (AIC) is utilized as the control structure in this scheme. First, the mathematical model of the unknown controlled plant was established through SAF, which consists of a linear FIR filter and a nonlinear spline interpolation function. The controller is obtained by adaptively establishing the inverse model of linear FIR filter and nonlinear spline function, respectively. In this process, a shift form DCT-LMS with variable learning rate (VL-DCTS-LMS) algorithm is proposed in order to adapt the inverse of FIR filter, and a modified Newton–Raphson method is used for directly calculating the inverse of spline interpolation function. Considering the steady-state error resulted by the plant modeling inaccuracy, filtered error feedback is introduced into the plant input in order to eliminate the control error. The effectiveness of the proposed control scheme is verified by several numerical examples, including some additional discussions and a comparison with other control methods.

This article presents a novel multidimensional feature extraction and selection method for electrocardiogram (ECG) arrhythmias classification. First, a novel multi-interval symmetrized dot pattern method (MSDP) is proposed to extract image shape features, which are combined with time-frequency features, morphological features and RR interval features to form a new feature set. Then a novel average weight based ReliefF method (WReliefF) is proposed to integrate with genetic algorithm (GA) and support vector machine (SVM) and form a feature selection method named as WReliefF-GA-SVM. The WReliefF-GA-SVM is a hybrid type feature selection method, and it can be divided into Filter stage and Wrapper stage to select features. In Filter stage, the algorithm uses WReliefF to sort the features according to the average weight. In Wrapper stage, GA and SVM are combined, and the classification performance of SVM is used as the fitness function to search the best feature subsets. Finally, the five types of ECG beats from the MIT-BIH arrhythmia database are classified by a one against all (OAA) SVM model. The results show that the proposed method achieves average sensitivity, specificity and accuracy of 99.42%, 99.83%, and 99.74%, respectively.

Rotating shaft–disk–blade (RSDB) system is one of the most important parts of turbomachinery, such as aero-engine, gas turbine and power plant. The coupling vibration of RSDB system with blade crack is vital for the blade health monitoring and crack detection of rotating blade. This study aims at addressing the dynamic modeling and steady-state coupling vibration mechanism of RSDB system with blade crack. First and foremost, on the basis of the stress state at crack section, an improved analytical breathing crack model (modified stress-based breathing crack model, MSBCM) for rotating blade is proposed. The validity of the proposed breathing crack model is verified by comparing the results obtained by MSBCM, finite element contact crack model and conventional analytical crack models. The comparative results suggest that MSBCM is of high fidelity and behaves best among the analytical crack models. Subsequently, a comprehensive dynamic model of the coupling vibration for RSDB system with blade crack is formulated on the basis of continuum beam theory and Lagrange equation. The shaft bending, shaft torsion, blade bending and blade radial deformation are comprehensively considered in this model. The validity of the proposed dynamic model is verified through comparison with finite element simulation and experimentation results. By introducing the proposed MSBCM, the dynamic coupling vibration model of the RSDB system with blade crack is formulated. At last, the steady-state coupling vibration mechanism of two typical structures for RSDB system is comprehensively investigated. It is suggested that the shaft torsional vibration is much more sensitive to blade crack than the shaft bending vibration be, which indicates that the vibration features of shaft torsional vibration may offer indicators for the presence of blade crack.

Prognostics and Health Management (PHM), including monitoring, diagnosis, prognosis, and health management, occupies an increasingly important position in reducing costly breakdowns and avoiding catastrophic accidents in modern industry. With the development of artificial intelligence (AI), especially deep learning (DL) approaches, the application of AI-enabled methods to monitor, diagnose and predict potential equipment malfunctions has gone through tremendous progress with verified success in both academia and industry. However, there is still a gap to cover monitoring, diagnosis, and prognosis based on AI-enabled methods, simultaneously, and the importance of an open source community, including open source datasets and codes, has not been fully emphasized. To fill this gap, this paper provides a systematic overview of the current development, common technologies, open source datasets, codes, and challenges of AI-enabled PHM methods from three aspects of monitoring, diagnosis, and prognosis.

As an integrated application of modern information technologies and artificial intelligence, Prognostic and Health Management (PHM) is important for machine health monitoring. Prediction of tool wear is one of the symbolic applications of PHM technology in modern manufacturing systems and industry. In this paper, a multi-scale Convolutional Gated Recurrent Unit network (MCGRU) is proposed to address raw sensory data for tool wear prediction. At the bottom of MCGRU, six parallel and independent branches with different kernel sizes are designed to form a multi-scale convolutional neural network, which augments the adaptability to features of different time scales. These features of different scales extracted from raw data are then fed into a Deep Gated Recurrent Unit network to capture long-term dependencies and learn significant representations. At the top of the MCGRU, a fully connected layer and a regression layer are built for cutting tool wear prediction. Two case studies are performed to verify the capability and effectiveness of the proposed MCGRU network and results show that MCGRU outperforms several state-of-the-art baseline models.