Design Of Automatic Sampling System Of Molten Steel Based On Machine Vision

The typical steel production process includes raw material processing,iron making,steel making and steel rolling.In order to make each link of the production line meet the working quality standards,it is of great significance to carry out online quality inspection.The analysis of molten steel is an important quality detection link in the process of steel production.This link samples a little molten steel from the high-temperature furnace,pours the molten steel into the mold for cooling and forming,analyzes the samples after forming,and gives the quality evaluation results to further guide production.In view of the high risk factor and harsh working environment of existing manual sampling,this paper has implemented the automatic sampling system design of molten steel based on Machine Vision.The system firstly adopts mechanical arm to complete the sampling of molten steel,and then adopts structural light 3D reconstruction to complete the defect detection of sample,so as to ensure the effectiveness of sampling.The main research achievements and innovations of this paper are as follows:1.The positioning and grasping system of molten steel samples based on machine vision is designed.In order to realize the positioning and grasping function of the manipulator for the molten steel sample,the CCD is used to obtain the image of the molten steel sample,and the partition feature extraction algorithm of the sample image is designed to realize the positioning of the grasping position and grasping direction in the image coordinate system.Then through camera calibration,hand-eye calibration,and the mapping function of the end rotation axis,the transformation of the grasp position and direction in the image coordinate system to the grasp position and attitude in the coordinate system of the manipulator base is completed.Different experiments are used to analyze the accuracy of data processing in each step of positioning and grasping.The proposed sample image partition feature extraction algorithm can accurately identify the geometric center of the upper and lower regions of the sample.The relative error is less than 1% after the known distance in the checkerboard image is mapped to the distance under the manipulator base system by the internal and external parameters of the camera and the hand-eye calibration matrix.For different locations of molten steel samples,the positioning and grasping position deviation of the manipulator is no more than 2mm,which meets the positioning and grasping requirements.2.A defect detection system of molten steel samples based on fringe projection is designed to ensure the effectiveness of sampling.The system first reconstructs the 3D of the molten steel sample,and then calculates the volume of the molten steel sample based on the 3D data partition integration.By comparing the volume with the standard sample,it evaluates whether the current sampling is effective.The system is simple in structure,and adopts multi-frame projection,multi-frame acquisition and label recognition to "control" the synchronization of system projection and acquisition timing sequence,without adding additional hardware timing sequence control units.In phase acquisition,a phase calculation method of double low frequency guiding high frequency is proposed,which effectively avoids the phase unwrapping error caused by the abrupt surface deformation and the non-uniform reflectance of the sample.The experimental results of label recognition show that the algorithm has high recognition speed and accuracy,and can be used to capture the sequence of deformed fringes at different times.The qualitative and quantitative evaluation results of 3D reconstruction show that compared with the phase calculation method of low frequency guiding high frequency and double frequency heterodyne,the proposed phase calculation method has higher 3D reconstruction accuracy.Several samples of molten steel with different defect degrees are used to evaluate the effectiveness of the system,and the system can accurately judge whether the sampling is qualified.