Tire Lap Defect Detection Based On 3D Vision

The function of the tire is to bear the weight of the car,transfer the traction force of the car engine and brake,it is the key component of the car.The load limit,buffering capacity and wear resistance of automobile tire are of great significance to automobile safety.Therefore,quality monitoring in tire production line is very important.The tire forming process refers to the process of automatic bonding of raw tire rubber materials through the molding machine,including the lap of multi-layer belt layer and tread.In recent years,the vision-based detection method has been widely used in defect detection of tire production line,which has improved the automation and intelligence of defect detection of production line.However,due to the demand of surface height measurement in tire lap process,the defect detection method based on two-dimensional vision has been unable to meet the demand.In this paper,a defect detection system based on 3D vision is applied to the lap joint defect in tire production line,including the hardware system of data acquisition,image processing algorithm and visual software interface based on MFC.Specifically,it includes:(1)A gradient based edge detection operator combined with morphology method was applied to realize four defect detection in the lap process of tire forming belt layer based on the characteristics of 3D data.To determine the width of the exception and two kinds of defects of serpentine migration,implements the bimodal threshold segmentation of lap joint,and based on the first-order and second-order derivatives of overlapping material left and right side of the edge detection,and then based on the detection of edge contour points sampling coordinates,compared to standard materials center coordinates determine width error and deviation size.In order to detect the defects of insufficient and excessive lap joints,edge detection based on Canny operator and sub-pixel was implemented to obtain the lap part and lap tail of the lap material,and the lap quantity was calculated by comparing the pulse number per lap of the forming machine.Finally,the measurement results based on vision and micrometer are compared in actual production environment,and the error meets the measurement requirements.(2)In order to solve the problems of too many manual parameters,complex parameter tuning and low flexibility of traditional edge detection,an improved semantic segmentation model Unet model based on deep edge monitoring is proposed.Based on the semantic segmentation model Unet,this model analyzes the disadvantage of the traditional pixel level loss to the edge contour insensitive,and proposes an inverse spatial displacement transformation network ISDTN based on the homography matrix distance measure function.This module has brought significant improvement in both ODS and IOS edge detection evaluation indexes.And in the reasoning stage will not bring additional parameters and calculation,is a plug and play efficient and convenient module.A novel loss function is proposed by combining the module and pixel-level loss function.In order to search for the joint loss optimal hyperparameters,a large number of experiments are carried out,and the performance of the original Unet and the improved Unet in five indexes is compared,and the results are improved.Finally,the effect of edge detection and semantic segmentation is shown,and the improvement effect is obvious.(3)A software system based on MFC is developed based on the defect detection algorithm,which can display the collected data and defect detection results in real time.In order to facilitate the analysis,the system added material width and lap width coordinate display chart,as well as the intermediate result analysis module,the system is easy to use,beautiful interface.