By David Tillett
A customer’s frustrations about the complexity and expense of deploying traditional machine vision systems led the founders of an Israeli company to take a new tack when developing their product.
Ofer Nir, VP products and marketing for Inspekto, said that over the long history of vision systems’ use in manufacturing, advanced systems capable of dealing with the most difficult challenges are also the most complex and expensive. These systems are only practical where there is great production stability — a single product churned out in very large batches over long periods of time.
Nir said current machine vision systems fall into three general categories: vision sensors, which are relatively easy to deploy but can only answer basic quality inspection needs, such as presence/absence of a part; smart cameras, which can handle more complex inspection tasks, but are themselves more complex; and full vision systems, which are tailor-made solutions designed by a machine vision integrator who painstakingly selects the best combination of cameras, lighting systems and software for a single product in its particular production environment — a process that must start from the beginning for each new product or application.
Nir said his company’s updated Inspekto S70 Gen. 2 system, which launched this spring, has a three-layered AI software that mimics the way human vision works. Its capabilities overlap much of what can be done by the three traditional types of vision systems, but it stands in a category of its own.
“So, we developed a very special technology,” he said. “We call it the AMV-AI — autonomous machine vision AI. It’s not a regular machine vision AI, it’s not an inspection AI — it’s a completely different approach and an architecture for entire end-to-end machine vision based on multiple different AI engines running in tandem.”
The technology offers two major benefits, Nir said. First, it allows the use of a single electro-optic system for a wide range of applications, and second, it dramatically simplifies the deployment process for the end user.
The Inspekto S70’s strength is in the breadth of tasks that it can handle — but it is not meant to replace the most complex vision systems, Nir said.
“We’re not saying that we are solving ... the use cases nobody else can solve,” he said. “Our core statement is not ‘Use Inspekto only when everything else fails.’ Our core statement is that you have one product here … that as-is fits [a] very wide range of different quality inspection applications, but it delivers them with the simplicity and immediacy that is unprecedented.”
The AMV-AI has three layers of AI engines: The Acquisition AI layer adjusts the hardware settings to optimize the image that it captures; the Detection AI analyzes images to detect the part; and the Inspection AI checks the part for defects.
According to the company’s website, the Inspekto S70 Gen. 2 can be set up in about 45 minutes by users without any machine vision or AI expertise and needs to be shown an average of 20 to 30 samples of good parts — and no samples of defective parts — to begin inspections.
Nir said that with most vision systems, “...the end user — i.e. the manufacturers — needs to provide hundreds of samples of OK parts, hundreds of samples of not-OK parts, make sure that those samples are spreading the exact definition of what is a defect, what is not a defect, provide the machine vision integrator a very clear definition about what they are looking for.” He said this can be difficult for manufacturers to provide, because in many cases defects are rare, or defects might not follow a repeatable pattern.
“And our technology enables us to use a very, very small set of samples — OK parts only,” Nir said. “Like a human being, when we see a scratch, we understand it’s a defect. So is our technology very much mimicking the way human brains work.”
Nir said that understanding the challenges at their most basic level helped them design the AI. “So, understanding that, we broke down the process … basically, in any machine vision solution there was the three parts. The first part is getting the image, the second is understanding what you see … and the third one is making a decision about what you see — in this case, is it OK or not OK, defective, not defective, etc.”
The Inspekto S70 Gen. 2 features a new array of LEDs that can project light from multiple directions at varying intensities and for different intervals. Nir said the Acquisition AI layer controls the lighting and camera, deciding how many images it will need and with what combination of lighting schemes, then merges them to create a single, high-dynamic-range image, even on highly reflective materials. It can also compensate for vibrations in the work environment and inspect up to three parts per second.
Nir said, “…even more than that, it adjusted dynamically in real time to changing environment[al] condition, for example, day/night, suddenly there is vibration, less vibration, etc. So, this is the first element of getting the image, then understanding what we see. Usually today it is utilized by pattern-matching technology that is analyzing and doing … very intensive image processing per pixel and looking for patterns and deciding what they’re seeing. And we developed a completely different approach based on a dedicated AI engine; nobody else in the market is doing anything similar to that.”
Nir said the capabilities of the Inspekto S70 Gen. 2 offer unique advantages for his company’s potential customers.
“Benefit No. 1 is, this enables us to come to our customer and say, ‘Hey, let me prove that I [can] resolve your use case, before you make a decision. Because I have a product ready and the deployment is so simple, I come over, install it within a day and then we see if it resolves your need. Then, Mr. Customer, you make a decision.’ Just this conceptually small element is a dramatic change of the decision process of the end user.”
Nir said the second advantage is the system’s adaptability. “Today, you need this quality inspection, then you use the Inspekto S70 for this. Tomorrow, suddenly there is a different need. Any other alternative would require you to go into, from scratch, a new machine vision project. Here, what I’m telling you: ‘Hey, you have Inspekto S70 monitoring Part [SKU] A, let’s try to monitor part SKU B — most likely it will succeed.’ ”
Nir said these capabilities make machine vision accessible to new users.
“No more weeks and sometimes months of design development and deployment — suddenly it’s less than a day. No more unknown, unpredictable cost.”
Looking to the future, Nir said the trend of increasing automation will only continue. Manufacturers are already using automation to cover the shortage in skilled machine operators. And he said operators are often performing quality inspections as a side task. There is also mounting pressure to offer production flexibility and high-mixture, low-volume output.
“What we see suddenly is a spill of demand coming from, I would say, not just the big manufacturers but from the SMBs [small and medium-sized businesses], the smaller guys, the injection molding plants … and then machining, etc.”
“This trend also mandates a flexible machine vision solution, because if you do not eliminate the need for human inspection, you will not be able to automate end-to-end the production process,” Nir said.
David Tillett, associate editor
Contact:
Inspekto, Ramat Gan, Israel, 972-73-3678187, www.inspekto.com
