Visual Inspection

When looking for a solution that replaces manual tending or loading, the only way to stay ahead of the game is robot vision. Robot applications with vision systems are needed to keep the operators safe in hazardous environments and improve their ergonomics. It can also improve cycle time, lower costs, and assure consistent quality when it comes to pick-and-place positioning or precise part assembly. Our experience, combining vision positioning equipment and several brands of robots (Mitsubishi, Fanuc, Yaskawa, Kuka, ABB, etc.), allows us to deliver a solution for any application need.

Quality control of the catalytic converters and DPF filters provides verifications on multiple levels. First and the most important is to check if every single cell and the entire product is within the necessary tolerances that provide 100% efficiency. From a process point of view, it is also a good checkpoint for controlling the cost management, as any deviations can be discovered early in the process and therefore limit the risk of producing imperfect products.
We have developed an in-line machine for complete control of blocked and partially blocked cells, or optical leaks on substrates with different geometry.

The surface of each product should be free of scratches, grease, or stains. While regular algorithms can detect flaws on monochrome backgrounds, they fail with patterned or wood-finish surfaces. To accurately identify surface irregularities on such backgrounds, we use advanced deep learning with neural networks, leveraging the latest software and hardware tools. The stability and sensitivity depend on the size of the base of input images. Usually, we train our system with a wide range of images with defined irregularities, and at least the same number of images representing good quality products.

With our experiences in 3D scanning equipment (structured light pattern projection, 2D laser profiler, etc.), area scan cameras, telecentric lens, and different types of LED lighting (ring, dome, bar, etc.), by combining it with measurements from a variety of additional sensors, we are able to transfer the time-consuming geometry inspection to in-line 100% quality control and calibration with a 3D measuring machine. The results show if a certain 3D point is a potential candidate for deformation, not only determining bad quality parts, but also identifying all chips or gauges (with data such as depth, surface area, volume, location of deformation, etc.).

Sometimes products cannot be sorted based on geometric properties or colour differentiation, like sorting different artificial teeth. In those special cases, we are using a different AI approach – deep learning – where the learning phase starts after system implementation. Usually, the goal is to differentiate between various part types. It is best for the system to scan as many pieces of each type as possible, with multiple repetitions. The greater the number of input samples, the higher the reliability achieved. It is important to note that reliability also greatly depends on the visual similarity of the parts. In general, machine sorting is several times faster than manual.

End products are normally checked to make sure they are assembled properly, incorporating all needed pieces whether these are in their correct positions. Our part assembly quality control is mostly based on area scan cameras and pattern recognition. For most tasks, Keyence and Cognex tools are used and for more complex challenges, we develop our own algorithms and software.

For traceability and genealogy purposes, we implement various human readable, linear, and 2D code laser and inkjet printing, labelling, and scanning applications. If the process requires, we also perform a quality analysis of the applied codes, OCR, or other markings.