Research On Defect Detection Of Weld Surface Based On Incremental Principal Component Analysis

In modern manufacturing,robotic welding technology is widely used in many scenarios,which can greatly improve the level of automation in manufacturing industry.Because of the constraints of the production environment and production process during the welding process,it is inevitable to produce a variety of weld defects,such as cracks,pores,splashes,etc.,these defects have a very important impact on the service life and safety of the product equipment.Therefore,it is necessary to detect the defects on the weld surface.The traditional manual detection methods some problems such as misjudgment,omission and time-consuming and laborious.The technology of online detection based on machine learning can improve detection efficiency and accuracy,and promote the development of automation level.Therefore,this paper adopts the method based on machine learning to conduct online detection research on weld surface defects.The main research contents are as follows:(1)In view of the actual production environment,the image collected contains a lot of noise,impurities and too large image resolution.The image is denoised by image preprocessing technology,and the weld area of interest is extracted from the workpiece background.Finally,the image is divided into a series of local block images containing the effective weld area.These local block images contain the main characteristic information of weld surface defects,which is of great value to the subsequent research of defect detection.(2)Batch processing algorithms such as 2DPCA,RC2 DPCA,BDPCA and Angle-2DPCA cannot meet the real-time requirements of online detection.The I2 DPCA algorithm proposed on the basis of the incremental algorithm CCIPCA has problems such as loss of image feature information and unsatisfactory dimensionality reduction effects.In view of these limitations,this paper proposes a feature extraction method based on the IBDPCA and IBlockPCA algorithms on the basis of the I2 DPCA algorithm.These two algorithms can perform feature extraction on high-dimensional mass data in an incremental iterative manner.The IBDPCA algorithm simultaneously extracts the feature information of the row and column direction of the image matrix and fuses the feature information of the row and column direction;IBlockPCA algorithm takes into account the correlation between global features,and extracts more in-depth information and local information by dividing the image into multiple sub-image blocks.These two algorithms can solve the problems such as the loss of feature information and the unsatisfactory effect of dimensionality reduction in the existing algorithms.Moreover they can obtain feature matrix with small dimension through feature extraction,which reduces the calculation cost for the subsequent classification and recognition research.(3)Through the self-built weld image data set,comparative experiments on reconstruction error,convergence,classification rate,complexity and other performance are carried out.The experimental results show that the two incremental algorithms proposed have good convergence.Compared with the existing algorithms,the proposed algorithm has a higher rate of weld defect identification under the condition of satisfying the time and memory required by the system.(4)Through the online testing experimental system for weld surface defects built,the welding platform video is online tested for weld defects.This experimental system can test the weld defects in an automatic way,and output and save the test results such as weld defect type,location and labeled image.Experimental results show that the proposed algorithm can meet the real-time and classification accuracy requirements of weld surface image online detection.