The Hot-dip Galvanized Sheet Surface Defect Online Detection System

With the rapid development of China’s economy,the demand for high-quality hot-dip galvanized sheets is increasing.However,surface defects can seriously degrade the quality of products.Therefore,accurate surface defect detection of hot-dip galvanized sheets is very important.At present,the surface defect detection of hot-dip galvanized sheet is still mainly based on manual inspection,which is not only low in efficiency,but also has high missed detection rate.Research and development of high-efficiency hot-dip galvanized sheet surface defect detecting equipment and system has become a focus of iron and steel enterprises.This paper proposes to design and implement an online detection system for surface defects of hot-dip galvanized sheets,aiming at improving production efficiency and improving product quality.Firstly,the related background,research status,purpose significance and surface defect detection principle of hot-dip galvanized sheet surface defect inspection are introduced.The common types of defects on the surface of four kinds of hot-dip galvanized sheet are listed,and according to the detection of a large-scale hot-dip galvanizing plant.Objectives and technical requirements,clarified the various indicators that the system needs to achieve,and carried out overall design planning for the system.Secondly,based on the overall design planning scheme,the hardware equipment is selected one by one according to the specific indicators that need to be achieved,including industrial cameras,lenses,lighting systems,industrial computers and other equipment.Thirdly,the implementation scheme of the system software part is designed.The image processing part mainly adopts the target detection model based on SSD(Single Shot MultiBox Detector)algorithm.Considering the diversity and complexity of the surface defect image of hot-dip galvanized sheet,the original SSD algorithm is appropriately improved.Including the production of new data sets,the use of more advanced base networks and migration learning with pre-trained models,while designing image acquisition and pre-processing,database storage and PC software,you can pass the defect information to the host computer at any time.The software is rendered.Finally,through on-the-spot investigation,on-site installation equipment,continuous debugging and improvement,respectively selected Inception v2,Inception v3 as the base network,found that the improved SSD algorithm based on Inception v3 network has achieved the expected effect through model training.8 kinds of surface defects such as raw material rolling injury and finished product rolling injury have better recognition effect.Not only the detection speed is fast,but also the detection precision is high.To a certain extent,it has been able to help a large-scale hot-dip galvanizing plant to quickly find the surface defects of hot-dip galvanized sheet.Identify and record the type and location of defects to help staff quickly assess the quality of hot-dip galvanized sheets,improve production efficiency,and verify the feasibility of theoretical analysis.