Most current rolling bearing fault diagnostic approaches presume that sufficient labeled data are available for training. However, labeled fault data collection in realistic engineering is time-consuming and expensive, whereas the majority of data available are unlabeled, restricting diagnostic performance. To alleviate the dependence of feature self-extraction model on rolling bearing fault data and improve the feature capture ability, a bearing fault diagnosis method based on feature-enhanced generative adversarial networks with an auxiliary classifier (AC-FEGAN) is proposed in this paper. Firstly, the non-saturating loss with gradient penalty term is introduced to circumvent the gradient vanishing and improve learning stability. Then, employing identity mapping of residual networks introduced into AC-FEGAN, representative faulty features can be automatically extracted, overcoming the restrictions of shallow convolutional neural networks in feature extraction. Simultaneously, to improve feature learning ability, the self-attention module embedded in the residual network accomplishes refined feature extraction by incorporating appropriate weight matrices into the faulty feature maps. Finally, the auxiliary classifier is employed as a pre-training model for the fault diagnosis model, with generated and unlabeled samples employed to fine-tune the auxiliary classifier to identify faulty bearings. The experimental results show that the proposed method can significantly improve the quality of generated samples while eliminating the reliance of existing diagnosis methods on data volume to achieve more accurate and effective fault diagnosis.

As a well-known homologous information fusion technology, the full-spectrum is widely used in orbit analysis of rotating machinery. However, due to the average effect of the Fourier transform, the instantaneous features of the axis orbit can not be obtained in full-spectrum. This limits the application of the full-spectrum in non-stationary vibration analysis. The time– frequency representation (TFR) of axis orbit has been proved to be an effective tool for analyzing non-stationary vibration signals. In this paper, a new approach of instantaneous features extraction of harmonic axis orbits is proposed. Furthermore, based on these features, an approach to construct the purified orbit TFR (POTFR) is proposed. First, the estimated speed signal is extracted from the TFR of a single-channel signal by data fitting. Guided by the estimated speed signal, the harmonic components of interest are separated by the Vold-Kalman filter. Furthermore, the parameters of each harmonic instantaneous orbit are calculated from the time domain by using the separated harmonic signals. Based on these parameters, high-resolution POTFRs of all of the harmonic orbits are constructed. Finally, the POTFR of the original signal is obtained by combining the POTFRs of all of the harmonic components. The proposed approach is illustrated by numerical simulation and is further validated using lab experimental signals of oil whip and bistable behavior. The start-up signal from a hydroelectric generating set further demonstrates the superiority of the method.

Disassembly sequence planning (DSP) is an important part of equipment maintenance in a hydropower station. In this paper, the generation of an excellent disassembly sequence (DS) for equipment is studied. Firstly, according to the characteristics of hydropower equipment, a combination node type is added to the directed graph analysis model, and the distance factor of components in space is added to the evaluation function of DS. Secondly, a DSP strategy including the grouping and minimization of the node’s scope is adopted to reduce computational complexity. Thirdly, a novel team-based genetic algorithm (TBGA) combining teams, fast feasible solution generator (FFSG), precedence preservative crossover (PPX) mechanism, multi-point heuristic mutation (MHM) mechanism, and forward-and-backward optimization operator (FBOO) is designed for DSP. The proposed TBGA maintains global search capabilities through teams and enhances local search capabilities through individuals. In the evolutionary process, teams, MHM, and FBOO have good complementarity to improve the comprehensive performance of the algorithm. Finally, four experiments are conducted and the performance of TBGA is tested based on the comparison of a well-known genetic algorithm, simplified teaching-learning-based optimization, and simplified swarm optimization algorithm. The results show that the proposed method can get better search results in limited iterations and require only about 25% time of other algorithms.

Rolling bearings act as key parts in many items of mechanical equipment and any abnormality will affect the normal operation of the entire apparatus. To diagnose the faults of rolling bearings effectively, a novel fault identification method is proposed by merging variational mode decomposition (VMD), average refined composite multiscale dispersion entropy (ARCMDE) and support vector machine (SVM) optimized by multistrategy enhanced swarm optimization in this paper. Firstly, the vibration signals are decomposed into different series of intrinsic mode functions (IMFs) based on VMD with the center frequency observation method. Subsequently, the proposed ARCMDE, fusing the superiorities of DE and average refined composite multiscale procedure, is employed to enhance the ability of the multiscale fault-feature extraction from the IMFs. Afterwards, grey wolf optimization (GWO), enhanced by multistrategy including levy flight, cosine factor and polynomial mutation strategies (LCPGWO), is proposed to optimize the penalty factor C and kernel parameter g of SVM. Then, the optimized SVM model is trained to identify the fault type of samples based on features extracted by ARCMDE. Finally, the application experiment and contrastive analysis verify the effectiveness of the proposed VMD-ARCMDE-LCPGWO-SVM method.