新型残差网络乏燃料剪切机刀具磨损状态监测模型构建  

A Novel Residual Network Model for Tool Condition Monitoring in Spent Fuel Shearing Machines

作　　者：陈甲华 王平平[1] CHEN Jiahual;WANG Pingping(School of Economics Management and Law,University of South China,Hengyang 421001,China;Hunan Provincial Key Laboratory of Emergency Safety Technology and Equipment for Nuclear Facilities,Hengyang 421001,China)

机构地区：[1]南华大学经济管理与法学学院,衡阳421001 [2]南华大学核设施应急安全作业技术与装备湖南省重点实验室,衡阳421001

出　　处：《核电子学与探测技术》2023年第6期1148-1154,共7页Nuclear Electronics & Detection Technology

基　　金：湖南省教育厅重点项目(19A443);湖南省社科项目(14JD51)。

摘　　要：为了解决乏燃料剪切机强背景噪声工作条件、刀具磨损样本少、磨损状态识别准确率低的问题,构建了一种基于软阈值与金字塔挤压注意力改进残差网络的刀具磨损监测模型,利用高效金字塔挤压注意力ESPA模块获得多尺度特征信息的精细特征图,并将软阈值加入ESPA模块中作为非线性转换层,过滤背景噪声相关的特征。实验证明,该模型识别准确率优于单独的ESPANet模型与深度残差收缩网络DRSN模型,能够有效识别剪切机刀具磨损状态。To solve the problems of strong background noise working conditions of spent fuel shearing machines,few tool wear samples and low accuracy of wear status identification,a tool monitoring model based on soft thresholding and pyramid squeezing attention(PSA)improved residual network is developed.The efficient pyramid squeezing attention(ESPA)module is utilized to obtain fine feature maps with multi-scale feature information,and soft thresholding is added into the ESPA module as a nonlinear transformation layer to filter the features related to background noise.The experiments demonstrate that the recognition accuracy of this model is better than the ESPANet model and the deep residual shrinkage network model.

关 键 词：乏燃料剪切机 刀具磨损 状态监测 残差网络 金字塔挤压注意力 软阈值 

分 类 号：TL248[核科学技术—核燃料循环与材料]