Research On Vision-based On-line Detection Algorithm For Surface Defects Of Hot Rolled Steel Bar

The detection technology for hot rolled steel bar surface defects is one of the key technologies to increase product competence and improve production process. However, traditional non-destructive testing technology for surface defects is hard to satisfy the requirement of high speed detection for hot steel bar. In order to detect surface defects on-line, the detection technology of surface defects based on machine vision appeared. This technology has high speed detection and accuracy. Besides, it can reappear the situation of production surface quality. So many companies make a huge investment to research this technology. So far, the detection technology for hot rolled steel bar surface defects based on machine vision is well-established in developed countries, and they have had relative detection system being in use. However, the research in this field of our country just started and has large gaps with developed country. This affects the market competitiveness of our country’s hot steel bar product to some degree. So the develop and research of this technology is in urgent need.Firstly, the hardware system was researched. The whole layout scheme for steel bar surface defects detection was designed. The selection of camera number was analysized. Then the light system was designed. According to the longitudinal resolution requirement, the type and specific model of camera was selected. The focus of lens was calculated according to the lateral resolution and the specific model was selected. Afterwards, we compared the characteristics of different light sources and chose appropriate light source type. The calculation of depth of field verified that the selected hardwares were correct. Then the imaging equipment was introduced and imaging experiment was carried out. The parameters’effect to imaging was analized. Then different types of steel bar surface images were listed. Besides, the imaging result of hot rolled steel bar surface images was analyzed. We concluded three factors of affecting the imaging of steel bar surface. Lastly, the characteristics of steel bar surface images were analized qualitatively and quantitatively.To extract steel bar surface image, we proposed modified local border search algorithm. This removes useless background information existed in original images and retains steel bar information. It reduces image processing data and avoids the steel bar border being considered to defects. We analyzed the type of noise in steel bar surface images and built image degradation model. The noise in images is mainly gauss noise. The denoise effect of different filter algorithm was compared, and we obtained the most appropriate method for this project. The ideal low pass filter was used to denoise. Through the denoise effect comparision of rectangle filter and circular filter, we concluded that the rectangle filter was the best.Afterwards, the characteristic of pits in images is analyzed and we obtained that it’s more effective to use column pixels of images. We proposed a detection algorithm for pit defects based on trigonometric function and Weber contrast. Then the modified method for gray level of image, the cycle selection of sine kernel function and threshold selection were discussed. The detection result is good, but the algorithm is limited to the size of pit defects, so a pits defects detection algorithm based on lower envelope Weber contrast (LEWC) was proposed. Then the Weber’s law and its application were introduced. Besides, we led into the concept of lower envelope Weber contrast and introduced detailed detection algorithm. The simulation experiments indicated that this algorithm is very effective to detect surface pits defects of hot rolled steel bar and it is not affected by the size of defects.A hot rolled steel bar surface defects detection algorithm based on local annular contrast (LAC) was proposed. This algorithm can detect some common defects occurred on hot rolled steel bar. such as pits, scratches and overfills. Firstly, the common characteristic of these defects was analyzed. That is there is big gray level contrast between the defects and local background image. This is the foundation of the algorithm. Then the concepts of local annular background and local annular contrast were introduced, and the relationship between detection threshold and gray mean value of local annular background was obtained. This make the threshold adaptable and the detection result more accurate. Finally, the detailed algorithm process was introduced and we did experiment simulation. The real time test indicated that the algorithm can guarantee the on-line surface defects detection for hot rolled steel bar.At last, In order to test the effect of the proposed algorithm in actual hot rolling line, we programmed the defects detection algorithm and put it into the linear camera. The overall frame and interface of the software system were introduced. The main contents for the camera re-development were analized. To verify the effect of the algorithm, we did off-line test on grinder to judge whether the detection results with the algorithm in camera and experiment simulation are the same. One piece of steel bar product was put on grinder and made do cyclic ground motion to simulate the steel bar rolled on hot rolling line. The result indicated that the effect of the algorithm in camera was the same with the experiment simulation, so the algorithm is workable. The effect of different light strength to the quality of images was discussed. Then the detection system was put on hot rolling line to make on-line test. The outcome demonstrated that the surface defects detection system could detect the usual defects produced on steel bar. Besides, the reali-time was good, and it can be employed in industrial application.