Visual Inspection Technology Research For Airbag Appearance Defects

With the rapid improvement of the automotive industry,more and more attention is paid to the vehicle safety sets.The airbag is the most important one in the passive safety apparatus.Every airbag needs to be checked carefully before equipped in the car,the sizes and the surface defects are critical projects in the inspection.The size inspection ensures that the airbag can be assembled accurately,and the surface defect inspection ensures that the airbag can work regularly.Traditionally,these kinds of inspections are completed by human vision,which is very slow and error prone.This thesis focuses on building a machine vision system to inspect the sizes and the surface defects of the airbag.The image acquisition method,the image processing algorithm,and the surface defect detection approach are studied and a complete software is built for the vision inspection system with high accuracy,fast speed and flexible.The main research works include:In order to get the images of big airbags efficiently,a multi-cameras synchronous acquirement and tiling method is proposed.The parameters of the vision system,as well as their influences on the measurement are studied.The hardware of the vision system,which includes the acquisition module,the illumination module and the computing module are designed,as well as the overall workflow of the system.A multi-cameras calibration and mosaic method is studied.By designing a series of image processing algorithms,the desired dimensions of the airbag can be measured,and based on the imported CAD drawing,the quality of the airbag can be classified automatically.A classification approach based on the Convolutional Neural Network is studied to detect the defects on the deformed and textural surface.By analyzing the features of the defects,as well as the comparison of the data augmentation,the network architecture and the network training parameters,a 14-layer Convolutional Neural Network is designed,which obtained a 97.24%classification accuracy of the defects on the airbag surface.According to the software engineering methodology,the main parts of the software process are realized,such as the requirement analysis,the overall design,the functional module design,the operation and the testing.An intelligent software for vision inspection on airbag is built,including the dimension measurement and the surface defects detection.As a result,the static repeatability of the dimension measurement is less than 0.2 millimeters,and the absolute error is less than 0.4 millimeters.The rate of missed inspection and the overkill are 0.069%and 2.048%for the defect inspection.The maximum processing time of a single-piece airbag is 2467 milliseconds where the image preprocessing time is 819 milliseconds,the size inspection time is 589 milliseconds,and the defect inspection time is 912 milliseconds,which satisfies the requirements of the automatic inline airbag inspection.