{"id":55,"date":"2018-02-08T14:37:57","date_gmt":"2018-02-08T05:37:57","guid":{"rendered":"http:\/\/ctc-230148001434.kir.jp\/_assets\/?page_id=55"},"modified":"2022-07-27T07:49:21","modified_gmt":"2022-07-26T22:49:21","slug":"image_processing","status":"publish","type":"page","link":"https:\/\/www.humo.co.jp\/en\/technology\/image_processing","title":{"rendered":"Visual Inspection Technology"},"content":{"rendered":"
\r\n\t
\r\n\t\t
\r\n\t\t\t

Visual Inspection Technology<\/h3>\r\n\t\t\t

Visual inspection of mirror-finished parts (patent pending)<\/h4>\r\n\t\t\t

Even the slightest defect on mirrored or curved surfaces is not overlooked.
More certain than the human eye.<\/h5>\r\n\t\t\t

As the term “mirror-finished parts” implies, the surface of the workpiece is finished like a mirror and reflects light, making it difficult to detect defects such as dents and scratches. It was said that it was difficult to stabilize inspection accuracy.
\r\nIn order to automate this process, we have developed a method to detect defects based on the distortion of the reflected pattern by projecting the pattern onto the workpiece. Since our method can detect defects without moving the workpiece, the detection conditions are stable and even defects that are difficult to detect with the human eye can be reliably detected.
\r\nIn addition, human judgment is uneven in judging the pass\/fail of detected defects. Our system eliminates the unevenness of the human eye by using a small number of machine-learning judgments to target certain patterns.\r\n<\/p>\r\n

 <\/p>\r\n\r\n\t\t

\r\n\t\t\tInspection Details\uff08PDF)<\/a>\r\n\t\t<\/div>\r\n\r\n\r\n

 <\/p>\r\n

 <\/p>\r\n

 <\/p>\r\n\t\t\t

Visual inspection of processed parts<\/h4>\r\n\t\t\t

No defects are overlooked with reliable image analysis technology.<\/h5>\r\n\t\t\t

Defects such as holes of several tens of microns that occur on the machining marks of metal parts are checked by the human eye, and it has been difficult to automate this process. This is because it is difficult to find defects that are assimilated with machining marks, and defects are judged based on sensory criteria such as “this much” rather than NG if there is even a slight defect.
\r\nOur image processing can separate and process processing marks and automatically detect defects by using an ingenious light source and image analysis. By using machine learning to make an intermediate judgment + an algorithm that optimizes judgment criteria such as size, shape, and color to make a final judgment, we were able to make a judgment of several microns with only about 100 samples registered, and to make a judgment with reliable criteria that is not “this much.\r\n<\/p>\r\n

 <\/p>\r\n\r\n\t\t

\r\n\t\t\tInspection Details\uff08PDF)<\/a>\r\n\t\t<\/div>\r\n

 <\/p>\r\n

 <\/p>\r\n

 <\/p>\r\n\r\n\t\t\t

Visual inspection of crystal blanks<\/h4>\r\n\t\t\t

Finding defects on transparent and translucent objects.
Toward Image Detection Beyond the Human Eye<\/h5>\r\n\t\t\t

Crystal blanks, such as glass pieces ranging from several millimeters to less than one millimeter in length and width and several tens of microns to several microns in thickness, are translucent or transparent, which means that light is polarized and transmitted through them, making it difficult to detect scratches or discoloration on them. Our automatic sorting machine can detect them automatically by using an ingenious light source and algorithm.
\r\nDefects that can be detected by this automatic sorter include chips, scratches, chipping, cracks, stains (discoloration), and abnormal corner angles and areas. In order to detect multiple defects, an imaging environment (light source) appropriate for each type of defect is required. We can detect multiple defects simultaneously by separating and processing multiple images based on color information from a single image under multiple light sources. Also, although lapped crystal blanks are translucent, defects are detected using an algorithm that cancels out the effects of material texture and individual differences.
\r\nFrequency, capacitance, and leakage current measurements have common pass\/fail criteria in the form of numerical values, but image inspection has ambiguity in pass\/fail judgment, and it is said that there is nothing better than the human eye. However, since the human eye has an unevenness in detection, an automatic sorting machine has been desired. Our equipment will continue to evolve with the goal of “someday surpassing the human eye.\r\n<\/p>\r\n

 <\/p>\r\n\r\n\t\t

\r\n\t\t\tInspection Details\uff08PDF)<\/a>\r\n\t\t<\/div>\r\n\t\t<\/div>\r\n\t<\/div>\r\n<\/section>\r\n
\r\n\t
\r\n\t\t