基于卷积神经网络的盾尾间隙测量系统研究.docx

?

?

基于卷积神经网络的盾尾间隙测量系统研究

?

?

赵栓峰耿龙龙丁志兵黄涛许倩

摘?要：针对隧道施工过程中盾构机盾尾间隙变化引起的机器磨损与注浆问题，通过分析国内外采用传统的人工计量方法测量盾尾间隙的诸多不足，提出一种基于卷积神经网络的盾构机盾尾间隙测量方法和装置。该方法通过单目相机拍摄位于管片上的激光标定点与盾尾间隙局部图像信息并传入计算机中，采用卷积神经网络的图像超分辨率重建方法对采集到的图像进行超分辨率重建，得到高分辨率图像，结合卷积神经网络与数字图像处理技术检测盾尾间隙，进行盾构机盾尾间隙的数值估计，实现盾尾间隙的非接触测量方式，使得盾构护盾间隙自动化测量能够用于盾构的全周期中，有助于实现盾构施工的精确化和自动化。通过西安地铁五号线施工现场的实际测试结果表明，程序运行稳定，盾尾间隙估计准确，最大误差为±5mm。

关键词：盾尾间隙;人工计量;卷积神经网络;激光标定;数字图像处理;数值估计

：TN98?34?????????????：A?????????????：1004?373X（2020）23?0146?04

Abstract：Inviewofthemachinewearandgroutingcausedbythevariationofshieldtailclearanceintheprocessoftunnelconstruction，aconvolutionalneuralnetworkbasedmethodanddeviceformeasuringshieldtailclearanceofshieldmachineareproposedbyanalyzingtheshortcomingsofthetraditionalmanualshieldtailclearancemeasurementmethodsathomeandabroad.Inthemethod，amonocularcameraisusedtocapturelocalimageinformationofthelasercalibrationpointsandtheshieldingtailclearance，andthenthelocalinformationistransmittedtoacomputer.Theimagesuper?resolutionreconstructionmethodbasedonconvolutionalneuralnetworkisusedtoreconstructthecollectedimageswithsuper?resolutionsoastoobtainhigh?resolutionimages.Theshieldtailclearanceisdetectedincombinationwithconvolutionalneuralnetworkanddigitalimageprocessingtechnology，andthevalueofshieldtailclearanceofshieldmachineisestimatedtoachievethenon?contactmeasurementmethodofshieldtailclearance，whichmaketheautomaticmeasurementofshieldtailclearancebeusedinthewholeperiodofshield，andishelpfultorealizetheaccuracyandautomationofshieldconstruction.AftertheactualtestontheconstructionsiteofXianMetroLine5，theresultsshowthattheprogramrunsstably，andtheestimationofshieldtailclearanceisaccurate（maximumerrorof±5mm）.

Keywords：shieldtailgap;manualmeasurement;convolutionaln