Qiuyang Zhou has a total of 73 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • An enhanced MCKD based on built-in information for gearbox fault diagnosis under speed-varying condition

    by Yifan Sun, Yihao Zhang, Qiuyang Zhou, Guojin Li, Yonghao Miao

    As built-in information, encoder data have demonstrated distinct advantages over vibration signals in the field of mechanical fault diagnosis. However, mechanical equipment in actual industrial settings operates under speed-varying and nonstationary conditions. This poses tremendous challenges to the direct application of conventional feature extraction methods on encoder information for equipment fault diagnosis under speed-varying conditions, especially without the assistance of additional external speed sensors. Motivated by this issue, this article proposes an enhanced maximum correlated kurtosis (CK) deconvolution (EMCKD) method for gearbox fault diagnosis using built-in encoder information under speed-varying conditions. First, the instantaneous angular acceleration signal (after second-order differencing) is resampled based on the encoder position sequence as a reference to eliminate the influence of speed fluctuations. This approach overcomes the limitation of traditional monitoring techniques that rely on extra speed sensors to provide rotational speed information. Second, time-domain synchronous averaging is proposed as the cycle estimation method for maximum CK deconvolution. This method effectively improves the accuracy of period estimation during each iteration, thereby enhancing the efficiency and diagnostic precision of EMCKD when processing complex encoder information. Finally, the proposed method is validated using both simulated encoder data and actual fault data from planetary gearboxes. Comprehensive quantitative analyses demonstrate that EMCKD exhibits superior parameter stability and robust anti-noise capability even under extremely low signal-to-noise ratios. Furthermore, two distinct experimental cases, external gear ring and planetary gear faults, are conducted, and the results indicate that the proposed method outperforms traditional methods by a significant margin.

  • Open Access

    Few-Shot Fault Diagnosis of Rotating Machinery Using Complex Convolution and Disentangled Representation Learning

    by Qiuyang Zhou, Xiaoyu Xian, Zhengyu Chen, Lei Yan, Yuming Fan, Kexin Yin

    Few-shot fault diagnosis is a challenging task in rotating machinery health monitoring because only limited labeled fault samples are available in practical industrial scenarios. Under such conditions, deep learning models are prone to overfitting and may fail to extract stable fault-sensitive features from vibration signals. Moreover, the weak fault-related components are usually coupled with operating-condition variations, background vibration, and environmental noise, which further degrades the discriminability and generalization ability of diagnostic models. To address these problems, this paper proposes a complex-valued disentangled representation learning network for few-shot fault diagnosis of rotating machinery. First, a direction-pair complex augmentation strategy is developed for triaxial vibration measurements. Two directional vibration components are selected and organized as the real and imaginary branches of a complex-valued input, which increases sample diversity under few-shot conditions. Then, a lightweight complex-valued convolution block is designed to model the coupled dynamic characteristics between different vibration directions and extract fault-sensitive representations. Furthermore, a dual-branch disentangled representation structure is developed to decompose the learned features into fault-sensitive representations and condition-related interference representations. To enhance the separability of fault embeddings under limited samples, a cosine-based disentangled representation loss is introduced, which improves intra-class compactness and inter-class discrimination while suppressing irrelevant interference information. Finally, a few-shot diagnosis strategy is constructed to identify fault categories with only a small number of labeled samples. Experimental results demonstrate that the proposed method consistently outperforms representative methods in terms of diagnostic accuracy, feature separability, and robustness, especially under extremely limited labeled samples.

  • Open Access

    Simulation-Driven Bearing Fault Diagnosis Under Fault-Free Conditions with Hierarchical Convolutional Attention Networks

    by Qiuyang Zhou, Xiaoyu Xian, Lei Yan, Yuming Fan, Kexin Yin

    Reliable and intelligent fault diagnosis of rotating machinery is crucial for the safety and stability of industrial systems. Nevertheless, the acquisition of labeled fault data is often difficult in practical applications because of the high cost of maintenance, the rarity of fault events, and the inherent safety risks associated with fault induction experiments. As a result, most real-world datasets consist mainly of healthy operating samples, which makes bearing fault diagnosis under fault-free training conditions particularly challenging. The objective of this study was to develop a simulation-driven diagnostic framework capable of identifying real bearing faults without using real fault samples during model training. To achieve this objective, pseudo-fault data were generated by superimposing periodic impulse–resonance responses, governed by theoretical bearing fault characteristic frequencies, onto healthy vibration signals. The synthesized dataset was further analyzed using wavelet packet decomposition and envelope spectrum analysis to extract discriminative time–frequency features. These features were then fed into the proposed Hierarchical Convolutional Attention Network (HCANet), which captured hierarchical multi-scale representations while emphasizing fault-related components. Furthermore, a Central Clustering Loss was employed to encourage intra-class compactness and enhance inter-class separability, thereby improving the generalization capability of the diagnostic model. Experimental validation on two bearing datasets showed that the proposed method achieved high diagnostic accuracy when tested on real fault samples, despite being trained exclusively on healthy signals and synthesized pseudo-fault samples. These results demonstrated the effectiveness of the proposed simulation-driven strategy and highlighted its potential as a practical solution for bearing fault diagnosis in zero-real-fault-data scenarios.

  • A Novel Angle-Domain Deep Deconvolution Method for Bearing Fault Diagnosis Under Speed-Varying Condition

    by Yonghao Miao, Sen Hu, Xun Wang, Yun Kong, Qiuyang Zhou

    The extraction of fault features under strong noise condition (SNC) has been successfully addressed by the deep network-based maximum correlated kurtosis deconvolution (MCKD-DeNet) method. However, under speed-varying condition (SVC), this MCKD-DeNet method exhibits significant performance limitations due to inherent deficiencies in its objective function. To overcome the diagnostic challenge under SNC and SVC, this paper presents a novel angle-domain deep deconvolution (ADD) method. Initially, the method establishes a multi-layer neural network architecture, wherein feature learning is seamlessly integrated into the deconvolution framework to enhance the deep extraction capability of fault characteristics. Subsequently, the angle-domain index average kurtosis, which can effectively measure the fault characteristics under SVC, acts as the guiding criterion for optimizing the network. Furthermore, through the implementation of adaptive weight updating and feature learning strategies, fault features are progressively extracted and strengthened. Ultimately, the effectiveness of ADD is rigorously verified via simulations and experimental analysis, which consistently demonstrate its enhanced ability to accurately extract fault features under both SNC and SVC conditions, outperforming conventional techniques in terms of robustness and diagnostic accuracy.


  • AKgram: A Robust Adaptive Framework for Bearing Fault Diagnosis Under Speed-Varying Condition

    by Chuntao Zhang, Sen Hu, Qiuyang Zhou, Yonghao Miao

    Rolling bearing fault diagnosis under speed-varying condition (SVC) presents a significant challenge due to the non-stationary nature of vibration signals. The narrowband filtering demodulation (NFD) methods, while effective under constant speeds, faces severe limitations under such non-stationary operations. Specifically, their traditional feature indicators fail to characterize the non-stationary fault impulses, leading to significant performance degradation. Furthermore, the inherent fixed frequency band segmentation strategy of these methods can also impair their diagnostic accuracy. To address these limitations, this paper proposes AKgram, a novel adaptive diagnostic method designed specifically for the SVC. First, the fluctuation trend spectrum (FTS) is derived from the signal spectrum to realize adaptive frequency band segmentation, constructing a multi-level tower-shaped boundary distribution map while preserving full spectral component integrity and overcoming fixed segmentation limitations. Then, average kurtosis (AK) is introduced as a robust indicator to quantify fault-induced impulses and extract non-stationary fault features under SVC. Finally, AK values of segmented sub-bands are visualized in the tower-shaped map to form the AKgram, with the sub-band of maximum AK identified as the optimal demodulation band for direct fault feature extraction. The effectiveness and superiority of AKgram are validated via simulation and experimental data from a variable-speed bearing test bench, outperforming the classical Kurtogram method in optimal band identification and fault feature extraction.


  • Knowledge-Guided Multinode and Multihidden-Layer Network Architecture for High-Speed Train Axle Box Bearing Fault Feature Extraction

    by Lei Yan, Zheng Cao, Yanlong Pan, Cai Yi, Chunjun Chen, Qiuyang Zhou

    Fault feature extraction plays a vital role in monitoring abnormal conditions of high-speed trains and ensuring operational safety. However, most existing deconvolution-based methods construct objective functions in the time domain, making them highly sensitive to random impulses, while their iterative initialization is usually empirical and unstable under low signal-to-noise ratio (SNR) conditions. To overcome these challenges, this paper proposes a knowledge-guided multi-node and multi-hidden-layer neural architecture for axle box bearing fault feature extraction. A 1/3 binary tree structure covering the resonance frequency band is introduced to initialize the network, thereby enhancing the likelihood of achieving the global optimum. Furthermore, a hybrid loss function that integrates the envelope spectrum L2/L1 norm and time-domain mean kurtosis is developed to jointly characterize the fault-related information from both spectral and temporal domains. The network parameters are optimized through a layer-wise gradient descent strategy, and the weights corresponding to the maximum loss value in each node are selected for final feature extraction. Experimental validation on simulated, public, and real-world datasets under both constant and variable speed conditions demonstrates that the proposed method can effectively extract discriminative fault features, exhibiting strong robustness and adaptability under complex operating environments.

  • Open Access

    A Pre-Activated Residual Parallel Convolutional Block-Based BiGRU Model for Remaining Useful Life Prediction

    by Yifan Sun, Qiuyang Zhou, Yu Xia

    The accurate prediction of the Remaining Useful Life (RUL) of key mechanical equipment in modern industry is crucial for reducing production risks and optimizing maintenance decisions. However, existing Convolutional Neural Network (CNN)-based models lack an inherent memory mechanism, and single convolutional kernel-based CNN models fail to capture multi-scale temporal features effectively. Moreover, some existing methods fail to account for the stability of the model training process, which tends to result in prolonged training time and an elevated risk of overfitting. To overcome these problems, a pre-activated residual parallel convolutional block-based BiGRU model (PRPC-BiGRU) is proposed in this study. First, the residual parallel convolutional block (RPCB) is constructed to simultaneously extract multi-scale temporal features. Subsequently, the pre-activated convolutional structure, which applies normalization and activation function prior to convolution operations, is utilized to improve gradient propagation and training stability. Finally, experimental results using the aero-engine benchmark datasets to verify the effectiveness and superior prediction performance of the proposed PRPC-BiGRU model.

  • Bearing Fault Diagnosis Across Working Conditions via Temporal–Spectral Attention and Feature Adaptation

    by Menghan Jin, Qiuyang Zhou, Lei Yan, Wei Dong

    Reliable bearing fault diagnosis is crucial for ensuring the operational safety and efficiency of rotating machinery. However, real-world applications often involve significant variations in working conditions, which lead to domain shifts that severely degrade the generalization ability of conventional diagnostic models. Moreover, vibration signals exhibit complex temporal and spectral characteristics, and effectively extracting and integrating these features under varying conditions remains a challenging task. To address these issues, this paper proposes a novel diagnostic framework, termed TSANet (Temporal-Spectral Attention Network), for bearing fault diagnosis across varying working conditions.In the proposed TSANet, the raw vibration signals are directly used as temporal inputs, while their spectral counterparts are obtained via frequency transformation. These two representations are processed by dual-stream convolutional neural networks (CNNs) to extract modality-specific features. A tailored self-attention mechanism is introduced, where spectral features serve as keys and temporal features as queries and values, enabling selective fusion of informative patterns across time and frequency domains. To enhance robustness under unseen working conditions, TSANet incorporates domain adaptation modules that not only align the final fused latent representations but also perform separate domain alignment on the high-dimensional temporal and spectral features extracted by the dual CNN streams. This adaptation strategy enables the model to learn condition-invariant representations and improves transferability across varying working scenarios. Extensive experiments are conducted on benchmark bearing datasets under various working conditions. The results demonstrate that the proposed TSANet achieves superior diagnostic accuracy, transferability, and robustness compared to several state-of-the-art approaches. This work provides a new perspective on multimodal feature fusion and condition-invariant representation learning for intelligent fault diagnosis in complex industrial environments.