基于VMD-FFT和 Inception-v4的钻井泵轴承振动信号特征提取方法研究

打开文本图片集

中图分类号：TE926 文献标志码：A doi：10.3969/j.issn.1001-3482.2026.01.002

Abstract：To address the issue of feature dispersion in mud pump bearings caused by multi-source coupled interferences （such as structural resonance andfluid impact noise）under high pressure，high load，and strong corrosionconditions，an intelligent diagnostic method that integrated variational mode decomposition（VMD）， fast Fourier transform （FFT），and the Inception-v4 network was proposed.The adaptive VMD algorithm was employed to decompose the vibration signals offive typical bearing faults in mud pumps （including raceway spalingand cracks inthe innerring of thebearing）into intrinsic modes.It effctivelyseparatedthe weak fault features from the strong background noise and used FFT to map the optimized time-frequency features into a two-dimensional energy spectrum diagram，achieving the geometric representation of the unique fault mode of the mud pump. It constructed an Inception -v4 model enhanced by transfer learning，which simultaneously extracted deep features of the spectrogram through multi-scale convolution kernels and accurately identified the early damage of the mud pump bearing.The experiment was conducted using the full operating condition data of the on-site mud pump.This method achieved an identification accuracy of97% for five types of faults， significantly outperforming traditional methods.The constructed VMD-FFT-Inception fusion framework has overcome the reliance on manual feature engineering in mud pump fault diagnosis，providing a highly generalized solution for the intelligent operation and maintenance of petroleum drilling equipment.

Key Words：drilling pump; bearing; vibration signal; feature extraction

钻井泵作为石油钻探系统的核心动力设备，其轴承长期承受高压泥浆冲击、高载荷转矩及强腐蚀介质的多重极端工况耦合作用，造成疲劳失效风险显著加剧[1-3]。（剩余9339字）