How In-Sight 2800 Makes Vision Automation Simple
The In-Sight 2800 combines the best of machine learning and traditional rules-based vision in a fully integrated vision system. Leveraging pre-trained algorithms, In-Sight 2800 can quickly and easily be deployed in any factory environment to automate error detection.
The system is intended for use by line and automation engineers to solve challenging factory automation problems, without requiring knowledge of deep learning or machine vision. An engineer can turn on the In-Sight 2800 and have it recognize and classify defects within minutes. And, unlike some other vision systems, it can do so with an unlimited number of classes, thus solving even more advanced categorization and sorting tasks.
In-Sight 2800 allows manufacturers of any size to streamline integration, meet exact application requirements, and achieve higher product quality by offering:
- High ease of use
- Multi-class functionality
- Multi region of interest capability
1. High ease of use accommodates all skill levels and expedites deployment
Machine vision and deep learning have the reputation of being both extremely capable and correspondingly difficult to deploy effectively. However, recent advancements in factory automation technology, like the In-Sight 2800, have led to the creation of a set of tools that are now easier to use than ever.
In-Sight 2800 is designed to be straightforward to set up, with no advanced programming needed. Training the system to solve a problem is very much like training an attentive new employee on the line. The engineer shows examples with distinctions that need to be made, and embedded edge learning is able to quickly make the same distinctions.
Edge learning is a subset of deep learning in which processing takes place directly on-device using a set of pre-trained algorithms. The technology is simple to setup, requiring less time and fewer images for training, compared to more traditional deep learning-based solutions. With traditional solutions, automating many classification applications cans take days or weeks, and require hundreds of images and hours of analysis from experienced vision and deep learning experts. By contrast, deploying the In-Sight 2800’s edge learning tools takes minutes, a handful of training images, and the attention of an engineer who understands the problem they need to solve, but who doesn’t necessarily have specific vision or deep learning knowledge.
2. Multi-class functionality addresses wide range of tasks
One key competency of edge learning is its ability to quickly and reliably separate parts into categories, after being trained on labeled images of those parts in the designated categories. A common application for this function is to classify acceptable and unacceptable parts as OK/NG.
Users train edge learning classification tools by providing images of both acceptable and unacceptable parts. There is no need to mark or define what makes a part unacceptable. Instead, the tool itself weights which variations in a part are significant for making a determination, while ignoring variations that do not affect the classification. The edge learning tools, embedded within the In-Sight 2800, can also handle classifications that are much more complex that a binary OK/NG decision.
With increasing mass customization, manufacturers often stock many variations of every part. For example, wheels for luxury automobiles can come in dozens of SKUs with slight differences in pattern, color, and finish. It can take a human inspector a full minute to distinguish some of them. The time required to complete this activity is justified by the potential outcome of not doing so – installing a wheel other than the one ordered, which can result in a dissatisfied customer and lost future business.
After being trained with just a few examples of each luxury wheel design, edge learning can reliably choose the right wheel, or confirm the specified style is being installed on the vehicle. The ability to define multiple classifications provides the ability to solve a higher number of factory automation problems.
3. Multiple regions of interest (ROI) functionality focuses on essential features
To refine an inspection application, a line engineer can use their knowledge of what the significant variable areas are on the part to define specific focus areas, called region of interest (ROI). Through an intuitive interface, configuring applications on the In-Sight 2800 is simple using familiar click-and-drag tools. One drag defines a box, another moves it. The box can be locked to invariant features of the part.
ROI definition is a standard part of machine vision, but its use often requires some expertise. In-Sight 2800 makes it easy for someone without specific vision tool experience to apply. And the power of its vision tools means that any number of such ROIs can be defined, and each of these ROIs can identify any number of classes.
This makes it straightforward to perform assembly verification for complex assemblies with many different configurations and variable parts, such as printed circuit boards (PCBs). Such problems previously took an immense amount of work to decide which features should be checked to confirm that the right part had been installed, and then program the vision system to examine those features. In-Sight’s edge learning tools make these determinations autonomously, meaning the engineer can focus on higher value-add activities, like optimizing their operations.
Easy-to-use machine learning with unparalleled flexibility
Developed from years of experience in factory automation, Cognex vision tools are specifically focused on line operations requirements. The advantage of this well-honed technology becomes clearer as errors become subtler and harder to detect.
For example, a rotary capper can misthread, damage, or leave a gap when capping a bottle. Many systems can easily detect large, visible errors. The difference emerges when the gap is almost imperceptible. Other systems may pass such an error, causing potential leaks or contamination. In-Sight 2800, with its combination of edge learning and focused machine vision tools, will classify those almost invisible flaws as unacceptable.
Easy to use, trainable with a handful of images, capable of multi-class and multi-ROI operations, the In-Sight 2800 vision system is transforming factory automation.