There’s a reason so many manufacturers have turned to machine vision for decades now. Simply put, vision technology is a beauty thing with its ability to consistently and properly handle the tedious and repetitive task of detecting errors. However, not all vision systems are created equal. Of course, there are significant differences between the traditional “put it down and forget it” vision systems and those with deep learning abilities.
“We are only at the beginning of this revolution with great potential for AI to help with manufacturing,” said Dr. Amy Wang, co-founder and vice president of systems at AI-developed device vision company, Cogniac. The Cogniac system combines the latest AI research, human-computer interaction tools, and large-scale data management to make computer vision easier, more accurate, and scalable, enabling information manufacturers to extract from growing image data and video streams.
Dr. Amy Wang, Cogniac co-founder and vice president of systemsThe real potential depends on business leaders understanding and accepting it, explains Wang. “It’s about embracing the opportunity for continuous improvement – something that is woven into the DNA of the best manufacturers. Fortunately, that mindset goes very well with AI and deep learning, ”she says. “If you look at using a vision system with AI as a point project that you do work and walk away that approach leaves too many opportunities on the table.”
Selecting individual applications
There is a big step in what can be automated by deep learning – things that are not possible at all before. However, putting an AI system in an existing process and expecting it to work can be frustrating. “There are usually process changes to take full advantage of the in-depth learning capabilities. AI needs to be data-driven – meaning people need to record images. You want to take care of data, you want to collect data, you want to use this data to its full advantage, ”she says.
Cogniac addresses the necessary changes by first understanding the workflow – from the parts coming into the various signals the company considers to deficiencies – to design an effective solution that is ready to learn and advance. “Subject matter experts, the people who know what they are looking for, have an important role to play in reporting images to create good models,” says Wang. “The level of commitment to managing the change process from the outset makes or breaks a project.”
The typical workflow should start by identifying a business problem, uploading images into the system, labeling a small set of data, and then enabling the system start training the module automatically. “Along the way, operators will see what the model is doing, and make adjustments as needed,” she says. “For example, if you see the defect in some way, the lighting is not right, or the camera angle is not correct, you can change your cameras. You can change the priority of your business in terms of what actions you want to pursue. ”
Case in point? A customer at Cogniac inspects 7-foot x 10-foot panels coming down the line every four seconds. It needs to identify very small defects such as very small cracks or dents on the panel. “We have 25 high-resolution cameras taking small pictures as it comes down the conveyor belt,” she says. “These images are embedded into the system and incorporated into the model. The splits could be anywhere on the panel and can be in any shape. The variability is quite large, sometimes they are very small splits. The system detected issues that the company did not even recognize at first, allowing them to change quality priorities. ”
Wang tells IndustryWeek that by looking at the incoming images, a model of the Cogniac system can predict those images. If the model is not confident due to something new (i.e. substantive or environmental change), the system will create a warning asking for help. “Helping the model through change, leading to the system learning and improving over time,” she says. “Our system has a very handy interface allowing users to create workflows for their surveys. Then we have AI-backed labels to help subject matter experts with image labeling. Once a company has its first successful application, it is almost an addict to find other opportunities. it’s easy to do more with it. ”
The best way to scale is to use experiments within the production environment so that the visual system works on images that align correctly with product expectations.