AGF-UKAN：一种基于语义分割模块的高速转子轴振动特征提取框架

打开文本图片集

中图分类号：TP301.6；TP751 文献标识码：A

文章编号：2096-4706（2025）18-0029-06

Abstract：Inrecentyears，visual vibration measurement hasbeen favored in industrialequipment monitoringdue toits advantagesofnon-contact，long-istanceandeasyinstalation.However，inomplexindustrialevironments，thebackground complexityandnumerous interferencesmakeitdifculttorealizeigh-precisionsegmentationbytraditionalmethods，which in turtriggers measurement erors.Tothisend，thispaperproposesasegmentationetworkbasedontheAtetionMechanismand dynamic gating fusion of AGF-UKAN. Based on theoriginal UKAN architecture，the network designs a new type of up-sampling modulelightweight mixing.Atthesametime，it introducesadynamicgating fusion module，andadaptivelyadjuststhe fusion weightoftheencoderandthedecodercharacteristicsinthedecodingstage，effctivelyimprovingthesegmentationaccuracyand robustness.Ithispaper，extensiveexpeimentalvalidatinisardoutonvarietyfpublicdtasetsasellasself-soudrotor axis datasets，andtheresultsshowthatAGF-UKANsignificantlyoutperformsthetraditionalU-Netandotherbenchmark networks intermsofDSCandmoUindexes，whichfullydemonstratesitsremarkableefectiveness incomplexindustrialenvironments.

Keywords： image segmentation; Deep Learning; visual measurement; vibration extraction

0 引言

在工业设备的运行监控与故障诊断[中，转子轴的振动测量[具有至关重要的作用。（剩余7905字）