跨模态核聚球空间学习的掘进机轴承故障诊断方法

苏树智; 桂阳; 马天兵; 朱彦敏; 武康辉

doi:10.11999/JEIT260494

跨模态核聚球空间学习的掘进机轴承故障诊断方法

doi: 10.11999/JEIT260494 cstr: 32379.14.JEIT260494

1.
安徽理工大学计算机科学与工程学院淮南 232001
2.
安徽理工大学煤炭无人化开采数智技术全国重点实验室淮南 232001
3.
安徽理工大学机电工程学院淮南 232001

基金项目: 国家自然科学基金(52374155, 52574189, 52504161)，安徽省自然科学基金(2308085MF218)，安徽省高等学校自然科学研究项目(2024AH050399)

详细信息

作者简介:
苏树智：男，教授，研究方向为人工智能、模式识别、计算机视觉及医学数据分析等

桂阳：男，硕士生，研究方向为故障诊断

马天兵：男，教授，研究方向为机械电子工程、故障诊断、数字孪生等

朱彦敏：女，讲师，研究方向为机器学习、多模态模式识别、故障诊断等

武康辉：男，硕士生，研究方向为故障诊断

通讯作者:
马天兵　dfmtb@163.com

中图分类号: TP391
计量
- 文章访问数: 25
- HTML全文浏览量: 8
- PDF下载量: 6
- 被引次数: 0
出版历程
- 修回日期: 2026-06-15
- 录用日期: 2026-06-15
- 网络出版日期: 2026-06-23

Bearing Fault Diagnosis of Roadheader via Cross-modal Kernel Fusion-sphere Space Learning

1.
School of Computer Science and Engineering, Anhui University of Science & Technology, Huainan 232001, China
2.
State Key Laboratory of Digital Intelligent Technology for Unmanned Coal Mining, Anhui University of Science & Technology, Huainan 232001, China
3.
School of Mechatronics Engineering, Anhui University of Science & Technology, Huainan 232001, China

Funds: National Natural Science Foundation of China (52374155, 52574189, 52504161), Anhui Provincial Natural Science Foundation(2308085MF218), Natural Science Research Project of Anhui Educational Committee (2024AH050399)

摘要

摘要: 针对传统掘进机轴承故障诊断方法难以有效感知跨模态多尺度故障信息的问题，本文提出了跨模态核聚球空间学习(Cross-modal Kernel Fusion-sphere Space Learning, CKFSL)的掘进机轴承故障诊断方法。该方法首先在高维核空间中采用对偶极值点锚定和极向近邻分配机制来捕获核空间中具有相似同构信息的故障样本集群以形成核聚球。随后设计了新颖的自适应二叉剖分策略，该策略根据核聚球内部故障样本的几何跨度进行自适应剖分，形成微近邻核聚球空间，在微尺度上实现了对掘进机轴承故障样本的高同构性流形聚合。为进一步刻画微近邻核聚球间的广域拓扑关联，本文提出广域拓扑同构约束，该约束在微近邻核聚球空间中量化故障样本的秩次散度与相对散度因子，从而构建微近邻核聚球间的广域动态同构图，实现微近邻核聚球广域拓扑关联的重构。最后整合跨模态故障样本的局部流形同构性与广域拓扑关联，形成了CKFSL模型的目标优化函数，实现模型对跨模态多尺度故障信息的感知。本文在理论上推导出微近邻核聚球空间投影方向解析解，进而利用空间投影来直接获得掘进机跨模态故障样本的跨模态核聚球空间同构特征，该特征不仅能够有效感知掘进机轴承跨模态多尺度故障信息，且具有良好的鉴别力。在自建的AUST掘进机数据集和帕德博恩轴承数据集上进行的实验结果表明了CKFSL方法的有效性。
- 掘进机轴承 /
- 故障诊断 /
- 空间学习 /
- 多尺度故障感知
Abstract: Objective Traditional roadheader bearing fault diagnosis methods often struggle with high-dimensional and non-linear multi-sensor data, failing to effectively perceive cross-modal, multi-scale fault information or integrate local and global structural features. To address these limitations, this paper proposes a novel Cross-modal Kernel Fusion-sphere Space Learning (CKFSL) method. By perceiving cross-modal multi-scale fault information, the proposed method extracts highly discriminative features from roadheader bearing cross-modal fault samples, improving the accuracy of roadheader bearing fault diagnosis. Methods The CKFSL method first maps roadheader bearing cross-modal fault samples into a high-dimensional kernel space via implicit transformation. It employs dual extremal point anchoring and polar neighbor allocation mechanisms to capture clusters of fault samples with similar isomorphic information, forming kernel fusion-spheres. Subsequently, an adaptive binary partitioning strategy is designed based on the geometric span of internal fault samples to tighten isomorphic boundaries, constructing micro-neighbor kernel fusion-spheres and achieving highly isomorphic manifold aggregation at a microscopic scale. A micro-neighbor kernel fusion-sphere space is further formed to re-evaluate local isomorphism (Fig.1). To characterize wide-area topological correlations, a wide-area topological isomorphism constraint is proposed, which constructs a wide-area dynamic isomorphism graph among micro-neighbor kernel fusion-spheres (Fig. 1). Finally, an objective optimization function is formulated within the space learning framework, which integrates local manifold isomorphism and wide-area topological correlations of roadheader bearing cross-modal fault samples, as illustrated in the CKFSL diagnostic flowchart (Fig. 2). The analytical solution for spatial projection is theoretically derived to obtain discriminative cross-modal kernel fusion-sphere space isomorphic features from roadheader bearing cross-modal fault samples. Results and Discussions The proposed CKFSL is first validated on the self-built AUST roadheader bearing cross-modal fault dataset, with the experimental platform shown in Fig. 3. The average recognition rates with increasing training fault samples are illustrated in Fig. 4. On the AUST dataset, CKFSL achieves a 99.49% recognition rate with only 70 training fault samples, reaching 100% as the number of training fault samples increases. Table 1 summarizes the standard deviations under different training fault sample sizes, demonstrating that CKFSL maintains the lowest standard deviation and superior robustness compared to the other seven algorithms. Furthermore, three-dimensional fault feature distributions are presented in Fig. 5, confirming that CKFSL effectively partitions highly overlapping fault samples into distinct clusters, overcoming the boundary confusion of compared algorithms. To verify the generalization capability, the proposed CKFSL method is further evaluated on the public Paderborn dataset, with the experimental setup shown in Fig. 6. As depicted in Fig. 7 and Fig. 8, CKFSL achieves 100% average recognition accuracy across four complex fault categories, significantly outperforming comparative methods that struggle to surpass an 85% recognition rate for the F4 fault category. Conclusions The proposed CKFSL effectively overcomes the inability of traditional roadheader bearing fault diagnosis methods to perceive complex multi-scale fault information. By utilizing the wide-area dynamic isomorphism graph learned within the micro-neighbor kernel fusion-sphere space, CKFSL integrates local manifold isomorphism and wide-area topological correlations of roadheader bearing cross-modal fault samples. This process enables the CKFSL to extract highly discriminative cross-modal kernel fusion-sphere space isomorphic features, thereby improving the accuracy of roadheader bearing fault diagnosis and ensuring the reliability and continuous operability of the roadheader.
- Roadheader bearing /
- Fault diagnosis /
- Space learning /
- Multi-scale fault perception.

HTML全文

图 1 微近邻核聚球空间和广域动态同构图的构建

下载: 全尺寸图片幻灯片

图 2 基于CKFSL的掘进机轴承故障诊断的流程图

下载: 全尺寸图片幻灯片

图 3 AUST掘进机实验平台

下载: 全尺寸图片幻灯片

图 4 在AUST数据集上训练样本递增时不同方法的平均识别率

下载: 全尺寸图片幻灯片

图 5 基于不同方法的三维故障特征分布图

下载: 全尺寸图片幻灯片

图 6 Paderborn轴承实验平台

下载: 全尺寸图片幻灯片

图 7 在PU数据集上训练样本为240时不同方法的平均识别率

下载: 全尺寸图片幻灯片

图 8 在PU数据集上不同方法随样本递增的平均识别率

下载: 全尺寸图片幻灯片

表 1 在AUST数据集上不同训练样本下各方法的标准差

方法\训练样本数	不同训练样本对应的标准差
	100	140	180	220	260
KPCA	1.44	1.11	2.27	3.69	6.51
LLE	2.79	2.08	1.99	2.62	4.59
PCA	2.97	1.19	1.84	1.66	2.02
LPP	3.95	0.96	1.01	1.63	1.32
OCCA	12.88	13.99	11.86	12.05	17.09
LMDDA	10.01	11.18	8.85	5.24	12.68
BPCA	1.53	1.16	1.09	2.08	3.43
CKFSL	0.71	0.94	0.49	0.49	0.00

下载: 导出CSV

参考文献(22)

[1]	LI Changpeng, MA Tianbing, RUI Shi, et al. A fault identification method for cutting head of the roadheader based on parameter optimization VMD and RCMFDE[J]. Signal, Image and Video Processing, 2025, 19(4): 319. doi: 10.1007/s11760-025-03884-4.
[2]	WANG Xin, JIANG Hongkai, MU Mingzhe, et al. A trackable multi-domain collaborative generative adversarial network for rotating machinery fault diagnosis[J]. Mechanical Systems and Signal Processing, 2025, 224: 111950. doi: 10.1016/j.ymssp.2024.111950.
[3]	伍章俊, 许仁礼, 方刚, 等. 一种面向旋转机械多传感器故障诊断的模态融合深度聚类方法[J]. 电子与信息学报, 2025, 47(1): 244–259. doi: 10.11999/JEIT240648. WU Zhangjun, XU Renli, FANG Gang, et al. A modal fusion deep clustering method for multi-sensor fault diagnosis of rotating machinery[J]. Journal of Electronics & Information Technology, 2025, 47(1): 244–259. doi: 10.11999/JEIT240648.
[4]	WU Yifan, ZHAO Dandan, LI Chuan, et al. MCSANet: Cross-modal semantic alignment in multi-attribute learning for zero-shot bearing fault diagnosis[J]. IEEE Transactions on Industrial Informatics, 2026, 22(3): 2599–2609. doi: 10.1109/tii.2025.3641795.
[5]	CHEN Jianbing, SUN Tingting, and LYU Mengze. DR-PDEE for engineered high-dimensional nonlinear stochastic systems: A physically-driven equation providing theoretical basis for data-driven approaches[J]. Nonlinear Dynamics, 2025, 113(10): 10947–10968. doi: 10.1007/s11071-024-10664-1.
[6]	JI Xiaodong, YANG Yang, QU Yuanyuan, et al. Health diagnosis of roadheader based on reference manifold learning and improved K‐means[J]. Shock and Vibration, 2021, 2021(1): 6311795. doi: 10.1155/2021/6311795.
[7]	QIAN Quan, WU Fei, WANG Yi, et al. Maximum subspace transferability discriminant analysis: A new cross-domain similarity measure for wind-turbine fault transfer diagnosis[J]. Computers in Industry, 2025, 164: 104194. doi: 10.1016/j.compind.2024.104194.
[8]	王大虎, 刘畅, 王健, 等. 一种高精度并行主偏度分析算法及其在遥感图像中的应用[J]. 电子与信息学报, 2023, 45(10): 3492–3501. doi: 10.11999/JEIT220960. WANG Dahu, LIU Chang, WANG Jian, et al. A high precision parallel principal skewness analysis algorithm and its application to remote sensing images[J]. Journal of Electronics & Information Technology, 2023, 45(10): 3492–3501. doi: 10.11999/JEIT220960.
[9]	WANG S G W, PATILEA V, and KLUTCHNIKOFF N. Adaptive functional principal components analysis[J]. Journal of the Royal Statistical Society Series B: Statistical Methodology, 2025, 87(3): 603–631. doi: 10.1093/jrsssb/qkae106.
[10]	QIE Yifan, SCHLEICH B, and ANWER N. Generative adversarial networks and hessian locally linear embedding for geometric variations management in manufacturing[J]. Journal of Intelligent Manufacturing, 2025, 36(2): 1033–1062. doi: 10.1007/s10845-023-02284-0.
[11]	KERBOUCHE A, ABIDI H, LOUIFI A, et al. Enhancing fault detection reliability in PV systems through a novel approach based on interval-valued kernel PCA[J]. Electric Power Systems Research, 2026, 256: 112842. doi: 10.1016/j.epsr.2026.112842.
[12]	LI Tao, HAN Yongming, DUAN Xiaoyan, et al. Fault detection for multimode process based on local neighborhood-density standardization and ensemble serial global-local preserving projections processes[J]. Reliability Engineering & System Safety, 2025, 261: 111119. doi: 10.1016/j.ress.2025.111119.
[13]	ZOU Xinxin, XU Hao, and LIU Xinling. Manifold learning based on locally linear embedding for symmetric positive definite matrix[J]. Pattern Recognition, 2025, 172: 112691. doi: 10.1016/j.patcog.2025.112691.
[14]	ZHAO Xiaowei, WU Dongming, NIE Feiping, et al. Nonlinear locality-preserving projections with dynamic graph learning[J]. IEEE Transactions on Neural Networks and Learning Systems, 2025, 36(4): 7225–7236. doi: 10.1109/tnnls.2024.3408835.
[15]	JI Xiaodong, AN Rui, JIANG Hai, et al. Research on fault recognition of roadheader based on multi-sensor and multi-layer local projection[J]. Applied Sciences, 2025, 15(5): 2663. doi: 10.3390/app15052663.
[16]	LI Xinrui, MIN Hongqi, ZENG Yong, et al. Sparse MIMO for ISAC: New opportunities and challenges[J]. IEEE Wireless Communications, 2025, 32(4): 170–178. doi: 10.1109/mwc.001.2400201.
[17]	RAN Ruisheng, WANG Ting, ZHANG Wenfeng, et al. Autoencoder-based discriminant locality-preserving projections for fault diagnosis[J]. IEEE Transactions on Instrumentation and Measurement, 2025, 74: 3526113. doi: 10.1109/TIM.2025.3551906.
[18]	孙俊, 杨俊龙, 杨青青, 等. 联合局部线性嵌入与深度强化学习的RIS-MISO下行和速率优化[J]. 电子与信息学报, 2025, 47(7): 2117–2126. doi: 10.11999/JEIT241083. SUN Jun, YANG Junlong, YANG Qingqing, et al. Joint local linear embedding and deep reinforcement learning for RIS-MISO downlink sum-rate optimization[J]. Journal of Electronics & Information Technology, 2025, 47(7): 2117–2126. doi: 10.11999/JEIT241083.
[19]	GUAN Yang, MENG Zong, SUN Dengyun, et al. 2MNet: Multi-sensor and multi-scale model toward accurate fault diagnosis of rolling bearing[J]. Reliability Engineering & System Safety, 2021, 216: 108017. doi: 10.1016/j.ress.2021.108017.
[20]	LI Sheng, FENG Ke, XU Yadong, et al. Cross-modal zero-sample diagnosis framework utilizing non-contact sensing data fusion[J]. Information Fusion, 2024, 110: 102453. doi: 10.1016/j.inffus.2024.102453.
[21]	ZHOU Hongdi, HUANG Tao, ZHONG Fei, et al. Bearing fault diagnosis based on local manifold discriminant domain adaptation[J]. IEEE Sensors Journal, 2024, 24(7): 10504–10514. doi: 10.1109/JSEN.2024.3357809.
[22]	CHEN Lin, GE Li, JIANG Xue, et al. Boosting RPCA by prior subspace[J]. IEEE Transactions on Signal Processing, 2025, 73: 2170–2186. doi: 10.1109/TSP.2025.3569861.