The Steel Ball's Surface Defect Detection Based On Machine Vision

In recent years,machine vision technology is the fastest developing image engineering technology. The defect detection of steel ball's surface is an important attempt for its application in industrial production and has important practical significance.At present, the majority steel ball manufacturers at domestic are still using human vision to check the steel balls. That is, a large number of workers use the visual method for the detection of steel balls under incandescent lamps, and then conduct a simple classification of them. This approach is not only with a significant inaccuracy but is susceptible to subjective factors. Different people have different testing standards. Even the same person at different times will have different testing standards. This may lead to different qualities of classified steel balls. Thus, when the products are exported, it will greatly reduce the products'prices and competitiveness with foreign products. Moreover, facing the light for a long time not only causes harm to the eyes, but also brings collimation error easily, which may result in missed or false detects of defective steel balls.Concerning the deficiencies and shortcomings of human vision, this thesis proposes an online detection of bearing steel balls based on machine vision technology. That is, to use a camera instead of eyes to obtain the steel ball's information. It also means using computers to store and process information. In this way, the accessed and stored information is all two-dimensional information. Then this information is converted into three-dimensional data by using a computer and corresponding software for programming. And finally, all the steel balls'surface defects are detected.Experimental platforms are established independently. Two perpendicular tracks are designed according to the testing requirements. The CCD camera which is installed in the tracks can collect dynamic images of the steel balls'surface. After digital processing, analysis and identification, the area of defects is calculated and the steel balls are graded. A light source system is designed independently which adopts the red LED diffused light source and the arched bowl structure in order to overcome the defects of the steel ball's reflective surface and ineffective surface image. This also avoids the interference of external light source and obtains a clear and real image. This thesis also uses the characteristics value of colorful images RGB component , analyzes the RGB characteristics value of defect points and non-defect points respectively, and has found a new method for defect extraction. The programming has realized fast accurate extraction and calculation for steel ball's surface defect. The overlapped images are eliminated according to the relationship of the steel ball's plane image and actual image. And the steel ball's three-dimensional images are also reconstructed, making the ultimately calculated defect area more consistent with the actual defect area.