Development Of Scratch Detection System Of Alumina Ceramic Substrates Surface Based On Deep Learning

The alumina ceramic substrate is an important functional device for explosive foil in military industry,and its working reliability is of great importance.Due to the poor toughness caused by the combination of ceramic materials,and the subsequent process may involve a series of complex processing processes,such as mechanical impact,acid-base corrosion,temperature impact.The alumina ceramic substrate will face problems such as high strength,high brittleness,poor uniformity,which may affect its working performance.In order to improve the reliability of military devices with alumina ceramic substrates as the key component,it is necessary to test the processing quality of the formed alumina ceramic substrates.It is difficult to evaluate the forming quality of alumina ceramic substrates by single process quality inspection due to the complex processing flow of alumina ceramic sheet.Therefore,the comprehensive processing quality of the complex process can be evaluated by detecting the surface defects(mainly surface scratches)of the formed alumina ceramic substrates.At present,scanning electron microscope(SEM)or optical microscope(OM)is usually used to detect the scratch on the surface of alumina ceramic substrates,and then the scratch is identified and detected by manual reading.Because the size of alumina ceramic substrates is usually 10-20 mm,and the scratch defect is usually 5μm.It is difficult to ensure the high consistency of the manual reading process of a single alumina ceramic substrate.The main work of this thesis is developed a scratch detection system based on deep learning to realize the automation,high efficiency and intelligence of scratch detection.(1)An image acquisition system for scratch detection of alumina ceramic substrates is designed and developed.According to the size of alumina ceramic substrates and the requirement of surface scratch detection,the image acquisition system is designed and developed,including optical system and mechanical movement part.Through the construction of the module,the surface images of alumina ceramic substrates which meet the requirements of detection accuracy can be collected in real time under the moving state,so that it can be further input into the subsequent PC software for image processing of scratch detection.(2)The scratch detection method of alumina ceramic substrates based on deep learning is studied.The two-stage algorithm Faster R-CNN and the single-stage algorithms SSD and SSD＿Mobilenet in the field of target detection were studied from convolutional neural networks.For the alumina ceramic substrates under study,a training dataset was produced by prefabricated scratch and data augmentation methods.The Faster R-CNN,SSD＿VGG16 and SSD＿Mobilenet scratch detection models were built using the same dataset,and the detection speed,mean average accuracy and detection accuracy of the three were compared.The experimental results show that the single-stage detection algorithm SSD＿VGG16 and SSD＿Mobilenet outperform the two-stage algorithm Faster R-CNN in terms of detection speed and accuracy,and SSD＿VGG16 is slightly higher than SSD＿Mobilenet in terms of accuracy and precision,but with a slight decrease in detection speed.(3)An upper computer software system for scratch detection on alumina ceramic substrates was developed.Through the graphical development platform of LabVIEW,the training,detection of deep learning network and the control of sports platform are integrated.Through the design of four key modules: algorithm setting module,function operation module,real-time acquisition module and detection and observation module,it is convenient for the operator to observe the scratch detection results in real time and train the data and model after replacement.It has a good man-machine interface and good scalability and flexibility.The results show that the system has good scratch detection ability and can meet the detection requirements of industry.