微特电机 ›› 2026, Vol. 54 ›› Issue (9): 78-83.

• 生产技术 • 上一篇    

基于马尔科夫转移场和 Swin Informer 的风电齿轮箱故障诊断

杨　 永1,2,楼昱昉1,2,周月红1,2,刘国锋1,2,袁一帆1,2,杨　 帅1,2   

  1. 1. 运达能源科技集团股份有限公司,杭州 310000;2. 全省海上风电技术重点实验室,杭州 310000
  • 出版日期:2026-09-28 发布日期:2026-09-28
  • 通讯作者: 刘国锋(1991—),通信作者,男,研究方向为风电项目物流运输执行管理。
  • 作者简介:杨永(1974—),男,主要从事风电项目物流运输管理、进出质保、大部件更换等工作。

Fault Diagnosis of Wind Turbine Gearboxes Based on Markov Transition Field and Swin Informer Fusion Model

YANG Yong1,2,LOU Yufang1,2,ZHOU Yuehong1,2,LIU Guofeng1,2,YUAN Yifan1,2,YANG Shuai1,2   

  1. 1. Windey Energy Technology Group Co.,Ltd.,Hangzhou 310000,China;2. Zhejiang Key Laboratory of Offshore Wind Power Technology,Hangzhou 310000,China
  • Online:2026-09-28 Published:2026-09-28

PDF (PC)

8

摘要：

针对风电齿轮箱在变工况、强噪声环境下信息表达与特征挖掘难题,提出融合马尔科夫转移场、移位窗口( shifted window,Swin) 变换器和稀疏自注意力时序网络的多模态故障诊断方法,称为Swin Informer融合模型。通过马尔可夫转移场将风电齿轮箱一维故障信号编码为二维时频图像,以捕捉时间序列的动态转移特性;利用视觉神经网络的层级化窗口注意力机制,对生成的时频图像进行深层次的空间特征提取,从而强化对图像中故障模式的表征能力。 原始的齿轮箱一维故障信号被输入至变换器模型中,以充分挖掘其中的长时序依赖性。通过特征融合策略将图像中的空间特征与序列中的时域特征进行有效整合,并输入分类器实现高精度故障识别。实验结果显示,相较于传统单模态方法,本文所提Swin Informer融合模型在分类准确率与系统鲁棒性方面均表现更优,展现了良好的工程应用前景。

关键词: 风电齿轮箱, 马尔科夫转移场, 移位窗口变换器, 稀疏自注意力时序网络, 融合模型, 故障诊断

Abstract:

Addressing the challenges of information expression and feature mining in wind turbine gearboxes under varying operating conditions and strong noise environments, a multimodal fault diagnosis method is proposed that integrates Markov transition fields, shifted window(Swin) transformer, and informer, referred to as the Swin Informer fusion model.The one-dimensional fault signal of the wind turbine gearbox is encoded into a two-dimensional time-frequency image through Markov transition fields, capturing the dynamic transition characteristics of the time series. The hierarchical window attention mechanism of the shifted window transformer is utilized to perform deep spatial feature extraction on the generated timefrequency image, thereby enhancing the representation ability of fault patterns in the image. The original one-dimensional fault signal of the gearbox is input into the sparse self-attention temporal network model to fully mine the long-term temporal dependencies. The spatial features in the image and the time-domain features in the sequence are effectively integrated through a feature fusion strategy and input into a classifier to achieve high-precision fault identification. Experimental results show that, compared to traditional single-modality methods, the proposed Swin Informer fusion model performs better in terms of classification accuracy and system robustness, demonstrating promising prospects for engineering applications.

Key words: wind turbine gearbox, Markov transition field, shifted window, informer, Swin Informer, fault diagnosis

中图分类号: 

联系我们

地址：上海市徐汇区虹漕路30号
邮编：200233
电话：021-64367300-248、64821783
E-mail:wtdj@vip.163.com
版权所有 © 2020 《微特电机》编辑部 沪ICP备17053313号-1  公安备案号：31010402006356