Research On Defect Detection Technology Of Ship Hull Based On Machine Vision

In recent years,with the continuous improvement of hardware level,the machine vision industry has developed rapidly.At present,the cutting-edge technology of machine vision has become an indispensable part of many fields such as scientific research and education.As one of the typical applications of this technology,defect detection has attracted more and more attention of scientific research institutions.As the representative of the traditional industry,the construction level of the ship also symbolizes the national industrial level to a certain extent.In the process of ship construction,it involves the process of ship steel processing,frame processing,sectional construction,hull welding and so on.During this period,due to manual operation errors,machine failures,wear and tear,environment and other reasons,the ship hull inevitably appears cracks,holes,erosion and other defects,if these defects are not found in time,it will bring great potential safety hazards to the ship.At present,defect detection mainly relies on manual matching instruments,but this method will not only consume a lot of time and energy,but also affect the efficiency of ship construction and ship economic benefits.In this paper,machine vision technology is used to detect common defects of ships.This paper designs a defect detection scheme based on point cloud matching algorithm and a defect detection scheme based on deep learning.For the defect detection scheme based on point cloud matching,this paper first describes the basic knowledge of camera model,coordinate transformation,camera distortion,stereo matching,camera calibration and so on.Then,based on the ZED binocular camera and the open source ORB-SLAM project,the real-time three-dimensional reconstruction of ZED camera is realized.At the same time,it is found that although the accuracy of the reconstructed point cloud is slightly insufficient,the overall effect is good when verified by the laboratory ship model.In order to unify the model data into the standard point cloud format PCD of PCL(Point Cloud Library)for easy detection,this paper takes the square plane of 20 cm x 20 cm as an example to illustrate the storage methods of STL,OBJ,PLY and PCD.At the same time,the conversion algorithms between OBJ,STL,PLY format and PCD format are compiled and improved by using the three-dimensional visual library VTK(visualization toolkit).In order to get better matching results,this paper divides the matching into two steps.Firstly,SAC-IA(Sample Consensus Initial Aligment)algorithm is used to roughly match point clouds.Rough matching can provide better initialization results for next fine matching.Then,ICP(Iterative Cloest Point)algorithm is used to achieve fine matching of point clouds.Then we use the fast approximate nearest neighbor search algorithm based on Kdtree data structure to find the defect location in the fine matching results.In order to validate the effectiveness of the algorithm,three kinds of defects,deformation,hole and size error,are created on the 3D model of the actual ship compartment.The experimental results show that the algorithm can accurately locate the defects.For the defect detection scheme based on deep learning,the classical Faster RCNN network is adopted in this paper.Because the laboratory lacks the necessary ship model and can't go to the shipyard to take real data,three kinds of defects,including erosion,hole and scratch,have been made on 100 pieces of 300 mm x 300 mm marine steel plates with thickness of 1.5 mm.Each of them has about 200,and 500 image data have been taken.Then the data set is made according to Pascal VOC format.Finally,the training results show that the model can accurately identify the above three defects,which has a certain guiding significance.