14 March 2023 in IEEE Transactions on Instrumentation and Measurement
Intelligent fault diagnosis methods based on deep learning have attracted significant attention in recent years. However, it still faces many challenges, including complex and variable working conditions, noise interference, and insufficient valid data samples. Therefore, a novel deep transfer learning bearing fault diagnosis model is designed in this work by fusing time-frequency analysis, residual network (ResNet) and self-attention mechanism (SAM). A multiscale time-frequency feature map (MTFFM) and global statistical feature matrix (GSFM) of vibration signals are first constructed using wavelet packet transform (WPT). A deep feature extraction network combining ResNet and SAM networks is then designed to realize the fused extraction of local and global time-frequency features. Finally, we construct a joint loss function by the combination of multi-kernel maximum mean discrepancy (MK-MMD) and the domain adversarial neural network (DANN) to optimize the depth feature extraction network, which improves the cross-domain invariance and fault state discrimination of depth features. The proposed optimization method fully exploits the advantages of high-dimensional spatial distribution difference evaluation and gradient inversion adversarial strategy. Its effectiveness is demonstrated through variable working condition transfer fault diagnosis tasks, showing superior performance compared with other intelligent fault diagnosis methods.
15 November 2022 in IEEE Internet of Things Journal
In rotating machinery fault diagnosis, domain adaptation (DA) transfer learning-based framework has been attracting great attentions to tackle the problem of inconsistent feature distribution and insufficient labeled fault feature data. However, most of the existing approaches mainly focus on either the cross-domain distribution alignment or manifold subspace learning, which faces two critical limitations: 1) it is hard to overcome the feature distortions when aligning the distribution in the original feature space and 2) subspace learning is insufficient to decrease the distribution divergence. To address the above limitations, this work proposes an intelligent fault diagnosis scheme based on supervised DA with manifold embedding and key features selection. It first applies maximal overlap discrete wavelet packet transform (MODWPT) to process the vibration signals and performs the statistical feature extraction. In order to ensure that the selected key features are conductive to domain adaption, the fault discriminative ability and domain invariance of the features are investigated based on the domain differences and Laplacian score. Then, it presents a new supervised domain adaption with manifold embedding for the distribution alignment in manifold subspace by taking the class information and neighboring relationships into account. Finally, an intra classifier is learned to predict the unlabeled target domain. The proposed fault diagnosis scheme is evaluated using a set of practical data sets of motor and bearing. The extensive experimental results demonstrate that it significantly outperforms the comparative models and achieves much more effective fault diagnosis under different working conditions.
1 November 2017 in IEEE Sensors Journal
Complex wireless sensor network applications as those in Internet-of-Things or in-network processing are pushing the requirements for energy efficiency and data processing drastically. Energy-aware task allocation becomes crucial to efficiently distribute the tasks of the applications for the nodes to extend the network lifetime. In this paper, we propose a distributed optimal on-line task allocation algorithm, by considering the energy cost of communicating, computing, sensing, and sleeping activities, to optimally balance the workload distribution among the sensor nodes. Through an in-depth analysis, this paper proves that the optimal partition solution for each node consists of at most two partition cuts with the corresponding weights. This observation enables the proposed on-line algorithm to maximize the network lifetime with low execution complexity. The simulation results show that our proposed algorithm extends the network lifetime by 12.26 times compared with the strategy of no scheduling, which is 2.22 times more than previous off-line task allocation methods. Moreover, the energy spent on executing the on-line algorithm is so small that it can be neglected.
1 March 2023 in IEEE Sensors Journal
Cross-domain fault diagnosis based on transfer learning has been popularly developed to overcome inconsistent data distribution-caused degradation of diagnostic performance. However, the existing methods are typically suffering from a class imbalance of domains and lacking sufficient fault data because it is difficult to obtain the real industrial data of machinery under fault conditions. To address these problems, this work proposes a new cross-domain bearing diagnosis framework based on transferable features and manifold embedded discriminative distribution adaption. First, it applies the maximal overlap discrete wavelet packet transform to process the vibration data and extract different statistics-based features. Then, to enhance the domain adaptation performance, it designs a transferability evaluation based on the adjusted rand index and maximum mean discrepancy to quantify the fault discriminability and domain invariance of the features. After that, it proposes a novel manifold embedded discriminative joint distribution adaptation method to perform cross-domain feature discriminative joint distribution alignment in a Grassmann manifold subspace. Finally, it utilizes a random forest classifier to train the cross-domain fault diagnosis model. To verify the performances of the proposed methods, extensive experiments have been conducted on two real rolling bearing datasets. The results demonstrate that the proposed methods can achieve the desirable diagnosis results and significantly outperform comparative classical transfer learning-based models when there is the class imbalance between source and target domains.
21 February 2019 in IEEE Internet of Things Journal
The Internet of Things (IoT) has been recognized as the next technological revolution. It faces two challenges: how to achieve energy efficient communication for the battery constrained devices and how to connect a very large number of devices to the Internet with low latency, high efficiency and reliability. To address these problems, this work proposes two methods based on Kalman filter, termed as EPKF (extensions of predicable Kalman filter). They locally reduce the unnecessary transmission (access) of end devices to the network (Internet) utilizing the spatial and temporal correlations with low algorithmic overhead. Each transmitting device (TD) independently controls its transmission using the temporal correlation; and the receiving device (RD) exploits the spatial correlation among the TDs to further improve the reconstruction quality. The reconstruction problem in the RD is nonlinear. To reduce the computation complexity, an in-depth analysis of the local estimate error is conducted and the approximated linear solutions are thereupon obtained. They are fundamental methods applicable to any IoT monitored/controlled physical system that can be modeled as a linear state space representation. The pedestrian-position application is used as a case study to demonstrate the efficiency in the simulation. Remarkably, the EPKF methods using the linear combinations of the local estimates from multiple TDs reduce the transmission rate to 10%, while achieving the same reconstruction quality as using KF in the traditional manner.
16 September 2015 in IEEE Transactions on Wireless Communications
Energy efficiency is a primary concern for wireless sensor networks (WSNs). One of its most energy-intensive processes is the radio communication. This work uses a predictor combined with a Kalman filter (KF) to reduce the communication energy cost for cluster-based WSNs. The technique, called PKF, is suitable for typical WSN applications with adjustable data quality and tens of picojoule computation cost. However, it is challenging to precisely quantify its underlying process from a mathematical point of view. Through an in-depth mathematical analysis, we formulate the tradeoff between energy efficiency and reconstruction quality of PKF. One of our prominent results for that is the explicit expression for the covariance of the doubly truncated multivariate normal distribution; it improves the previous methods and has generality. The validity and accuracy of the analysis are verified with both artificial and real signals. The simulation results, using real temperature values, demonstrate the efficiency of PKF: without additional data degradation, it reduces the communication cost by more than 88%. Compared to previous works based on KF, PKF requires less computational effort while improving the reconstruction quality; compared with the techniques without KF, the advantages of PKF are even more significant. It reduces the transmission rate of them by at least 29%. Besides, it can be integrated into network level techniques to further extend the whole network lifetime.