In the mechatronic system, rolling bearing is a frequently used mechanical part, and its failure may result in serious accident and major economic loss. Therefore, the remaining useful life (RUL) prediction of rolling bearing is greatly indispensable. To accurately predict the RUL of the rolling bearing, a new kind of gated recurrent unit neural network with dual attention gates, namely, gated dual attention unit (GDAU), is proposed. With the acquired life-cycle vibration data of a rolling bearing, a series of root mean squares at different time instants are calculated as the health indicator (HI) vector. Next, the to-be HI sequence is predicted by GDAU according to the existing HI vector, and then the RUL of the rolling bearing is estimated. The experimental results show that the proposed GDAU can effectively predict the RULs of rolling bearings, and it has higher prediction accuracy and convergence speed than the conventional prediction methods.

The localized faults of rolling bearings can be diagnosed by the extraction of the impulsive feature. However, the approximately periodic impulses may be submerged in strong interferences generated by other components and the background noise. To address this issue, this paper explores a new impulsive feature extraction method based on the sparse representation. According to the vibration model of an impulse generated by the bearing fault, a novel impulsive wavelet is constructed, which satisfies the admissibility condition. As a result, this family of model-based impulsive wavelets can form a Parseval frame. With the model-based impulsive wavelet basis and Fourier basis, a convex optimization problem is formulated to extract the repetitive impulses. Based on the splitting idea, an iterative thresholding shrinkage algorithm is proposed to solve this problem, and it has a fast convergence rate. Via the simulated signal and real vibration signals with bearing fault information, the performance of the proposed approach for repetitive impulsive feature extraction is validated and compared with the noted spectral kurtosis method, the optimized spectral kurtosis method based on simulated annealing, and the resonance-based signal decomposition method. The results demonstrate its advantage and superiority in weak repetitive transient feature extraction.

Efficient and accurate planetary gearbox fault diagnosis is the key to enhance the reliability and security of wind tur-bines. Therefore, an intelligent and integrated approach based on deep belief networks (DBNs), improved logistic Sigmoid (Isigmoid) units and impulsive features is pro-posed in this paper. Vanishing gradient problem is an in-herent drawback of conventional Sigmoid units, which usu-ally occurs in the back-propagation process, resulting in that the learning considerably slows down and the classi-fication rate is reduced. To solve this problem, Isigmoid units are designed to combine the merits of unsaturation from Leaky Rectified Linear (LReL) units. The results of handwritten digit recognition experiments show the supe-riority of Isigmoid over Sigmoid on convergence speed and classification accuracy. Since impulses contain much useful fault information, especially for early failures, an integrated approach using the optimized Morlet wavelet transform, kurtosis index and soft-thresholding is applied to extract impulsive components from original signals to improve the diagnosis accuracy. Then the features ex-tracted from original signals and impulsive signals are employed to train and test the DBNs with Isigmoid, Sigmoid and LReL units for comparison. Finally, the results of plan-etary gearbox fault diagnosis show that Isigmoid has high-er comprehensive performance than conventional sigmoid and LReL.

In the mechanical transmission system, the gear is one of the most widely used transmission components. The failure of the gear will cause serious accidents and huge economic loss. Therefore, the remaining life prediction of the gear is of great importance. In order to accurately predict the remaining life of the gear, a new type of long-short-term memory neural network with macro-micro attention is proposed. First, some typical time-domain and frequency-domain characteristics of vibration signals are calculated respectively, such as the maximum value, the absolute mean value, the standard deviation, the kurtosis and so on. Then, the principal component of these characteristics is extracted by the isometric mapping method. The importance of fusional characteristic information is filtered via a proposed macro-micro attention mechanism, so that the input weight of neural network data and recursive data can reach multi-level real-time amplification. With the new long short-term memory neural network, the health characteristics of gear vibration signals can be predicted based on the known fusion features. The experimental results show that this method can predict the remaining life of gears and bearings well, and it has higher prediction accuracy than the conventional prediction methods.