Research On Wedge Defect Detection System Based On Fusion Of Deep Learning And Machine Vision

As a key part of the wedge-type prestressed anchor system,the quality of the wedge is an important factor that directly affects the establishment of effective prestressing force and building safety.At present,the defect detection of wedge is mainly realized by manual method,which is inefficient and difficult to meet the needs of high-speed automatic production.As a non-contact detection method,machine vision detection has the advantages of high efficiency and repeatability,which can make up for the shortcomings of manual detection.Therefore,this paper applies machine vision technology to the detection of wedge defects,measuring the size of the wedge and detecting the tooth defects of the wedge at the same time.Firstly,aiming at the shortcomings of manual inspection method,this paper designs a set of visual inspection system,which can realize the automatic inspection of wedge defects.The system includes mechanical motion unit,image acquisition unit and system software unit.The system first realizes the automatic loading and unloading of the wedge through the mechanical motion unit,then collects the image of the wedge parts through the image acquisition unit,and transmits the image to the upper computer system software in real time.The upper computer detects and identifies the wedge image by visual algorithm,and feeds the test results back to the PLC control system of the mechanical motion unit.The PLC system controls the loading and unloading of the mechanical system according to the photoelectric sensor signal and the test results,thus realizing the automatic detection of the wedge defects.The test of the production line shows that the system can detect the defect of the wedge in 0.25 seconds.Secondly,aiming at the size defect detection of wedge,this paper proposes a size detection algorithm based on corner detection and edge extraction.Firstly,the system is calibrated by camera calibration.Secondly,the measurement area is segmented by image preprocessing and ROI extraction.Then,the high-precision positioning of the measurement points of the wedge size is achieved by corner detection,edge extraction and line fitting at sub-pixel level.Finally,the size of the wedge is measured and detected according to the measurement method.Experiments show that the accuracy of the algorithm can reach 98.1%,which meets the detection requirements of the system.Finally,aiming at the tooth defect detection of wedge,an algorithm based on edge detection and decision tree was designed in the previous experiment of the system,and the accuracy rate was 95.79%.However,the accuracy of the algorithm depends on the feature parameters extracted manually,and the robustness is not strong.Therefore,in the latter part of this paper,the system is optimized.For the first time,deep learning is applied to the field of wedge defect detection,and a tooth defect recognition method based on residual network with Focal Loss is proposed.The experimental results show that the accuracy is as high as 99.61%,which meets the detection requirements of the system.Generally speaking,this paper finally realizes a wedge defect detection system based on fusion of deep learning and machine vision,which achieves the industrial-level detection speed and accuracy,meets the needs of enterprise’s automatic detection of wedge defects,and has important academic research and engineering application value.