Vision Deep Learning Solution
Automated inspections are an important part of quality control in almost every production. Conventional image processing software require high programming effort for complex product inspections. The Deep Learning Solution can implement inspections with low effort, which we have tested in our lab on mass production terminals. Watch the case live
The challenge in using a vision system is that the programming effort of the quality control does increase exponentially to the errors. Complex applications are therefore costly to program, difficult to parameterize, and for new products, the entire application must be partially redeveloped. In this case, we have investigated in our laboratory on mass production terminals, which are complex and have a variety of different errors.
With the Deep Learning Solution, complex quality controls can be programmed with little effort. This means that the technician does not have to develop his own inspection algorithm. Only the correct object-camera distance and the light ratio have to be set.
After manually teaching in approx. 100 good and 20 bad images, the system independently calculates an inspection algorithm. If the algorithm is not correct, further classified images can be added and the algorithm recalculated. Thus, even a new product can be integrated into production without much effort.
The minimal setup effort and the small number of images needed to achieve a verification algorithm surprised even our programmers. The new technology has proven itself and we look forward to the next Deep Learning applications.
- newest camera technology from Cognex Cooperation
- Fast adaptability to new products
- Inspection algorithm with feed of low number of images