The annual award ceremony of 2019 Shenzhen robots organized by Shenzhen Robotics Association was held in Shenzhen Civil Center grandly on May 29. The AOI Automatic Detection Project of Lenovo PC Server Label from Shenhui Vision (Shenzhen) Technology Co., Ltd. (Shenhui Vision) was awarded with the technological innovation award of industrial robot.
In this program, Shenhui Vision used 3D machine vision technology innovatively to detect the 3D position information of PC server. 99.5% label detection accuracy rate could be realized through character recognition + template matching. It deserved the technological innovation award in Shenzhen – a city where robot giants gathered.
Introduction to AOI Automatic Detection Project of Lenovo PC Server Label
AOI Automatic Detection Project of Lenovo PC Server Label is the label automatic detection program designed and implemented by Lenovo Shenzhen Bonded Area factory, Lenovo Group’s factory with the largest capacity of PC server worldwide. It is able to produce a total of over 700,000 PC servers per year. 4 categories of labels need detecting for each set of server. Previously, Lenovo server label was detected manually and it cost about 2-3 min to detect every category label of one server. The test quality could be hardly controlled, the nonconforming products could not be traced and the test results cannot be digitalized.
According to the traditional 2D machine vision, objects are positioned using 2D camera + light source lighting scheme or by extracting the object characteristics or through template matching for object for positioning. However, the servers produced by Lenovo’s Shenzhen Bonded Area factory are in a wide range of varieties which can be generally divided into 1U (height: 4.4 cm), 2U (height: 4.4*2=8.8cm) and 4 U (height: 4.4*4=17.6cm). Each kind of server is 420-740cm long and 370-484cm wide and with different shell colors (black and white) and surface textures. For the object positioning featured by various materials, textures and dimensions, the traditional 2D machine vision + light source plan are poor in terms of compatibility and could hardly realize compatibility.
On the basis of the label detection demands of Lenovo’s Shenzhen Bonded Area factory, Shenhui Vision designed a set of 2D + 3D machine vision solution which can be used for detecting the server labels, recognizing and detecting contents and detecting printing quality automatically.
We use Shenhui Vision HD1,000 high-speed and high-precision binocular structure light 3D camera to position PC server. By taking advantage of binocular triangle principle, 3D camera could realize 3D imaging, collect the information about object height and obtain the 3D information with higher stability and precision by using the code information of structured light. The above is not affected by such factors as the external light, object material, surface texture, color and height. Under a shooting distance of 1.2 m, the range of vision could cover the largest dimension of PC server so as to position PC server perfectly and precisely.
In consideration of large varieties of server labels that need to be compatible, the position of different label textures, characters in different sizes and characters on the server differs. The traditional OCR recognition method could hardly recognize so many types and the light source cannot be changed (erected on the mechanical arm), making it a failure to satisfy the label types of different textures. With the innovative method of character recognition + template matching, the same labels could be found among massive of labels for classification. The fixed label contents are provided with template matching mode. OCR recognition is still used for the massive of uncertain label contents. In this way, the label detection accuracy can be enhanced greatly from the original 80% to 99.5%.
After Lenovo Group’s Shenzhen Bonded Area factory used Shenhui Vision’s 2D + 3D machine vision label detection solution, the label detection time of a single server has been shortened to less than 30 s from the previous 2-3 min and the detection rate of defective label exceeded 99.5%. More importantly, the detection result can be digitalized through the solution so as to generate 3.5 PB detection data per year. The above lays a solid foundation on Lenovo Group’s comprehensive realization of intelligent manufacturing in terms of data.