Tommy Brennan - Vision Cognitive Technology

“The first thing to know is that machine vision is subtly yet significantly different from computer vision. Computer vision is about technology that can understand and interpret visual information,” he said. “Machine vision is all of that and something more. It’s about the tight integration of hardware and software to allow for context in specific tasks. For example, it can allow you to use AI to do inspections on a production line.”

What do you understand when you hear the term ‘vision cognitive technology’? Unless you’re deep in the weeds of high tech manufacturing, the odds are probably not much. But for Tommy Brennan, machine vision is the focus of each of his days and something he is arguably an industry leading expert in.

When it comes to producing medical devices, where tolerance for error is really low, usually a person has to visually inspect each product as it comes off the line. That person checks for cosmetic defects and anything that basically doesn’t ‘look’ right.

“That’s something that’s quite hard to get an AI to do but it’s what we do on a daily basis here. We help companies install cameras on their production lines to take the human element out and automate something that previously couldn’t be done by computer,” said Brennan.

Brennan is lead machine vision specialist at DMI, a position he came to following a 13 year period in industrial manufacturing. He brings that experience to the traditionally hard problem of getting computers to understand the world around them, something that human beings often do effortlessly.

“Computers can easily do things like dimensional inspections where they measure the size of something on a line perfectly. But humans can’t do that just by looking, they need tools such as rulers and tape measures etc. But conversely, we can tell if something is ‘off’ at a glance easily, and that’s something that’s a lot harder to teach an AI to do. That’s machine vision in a nutshell,” he said.

Brennan is from Sligo and grew up in his father’s furniture making business, where he saw specialist machinery being used to create objects more efficiently. He credits this experience as a child with lighting the spark of interest within him in engineering, and in particular with the mechanics of uniting computing with moving parts.

“I was always a bit of a nerd, building PCs and getting stuck into video games. But as time went on, I became more interested in building computers than playing games with them. I was 100 per cent the cliché of the curious kid who loved to take things apart to see how they worked, and that’s essentially what I still do,” he said.

“I’ve always had an interest in it, and I spent my teenage years helping to draw up kitchens using AutoCAD, learning about precision and accuracy. In college, that turned into a drive to learn more about automation and controls. It was all about programming and electrical design.”

While machine vision has been around in some form or another since the 1980s, according to Brennan, what’s changed in recent times to breathe new life into it is that the cost of the technology has come down, and there are a lot more ‘low code’ solutions available, making it less complex to work with.

“But what’s probably had the biggest effect is the AI side of things and the use of deep neural networks. They have revolutionised visual inspections. It used to be very difficult to come up with an algorithm that could reproduce what a person could do and that’s now relatively simple,” said Brennan.

“This is part of a general tipping point with relation to AI that society has reached. These ideas have been around for a while now but many improvements in the technology have seemingly come together at the same time to make AI suddenly much more competent.”

This aspect of technology growth - cheaper, faster, smaller and greater adoption – will have an impact on manufacturing that Brennan thinks will take a generation to fully play out.

“We’re going to see AI mature and develop, and we’re going to see it applied more cleverly to more kinds of software. The net result will be that people will be able to produce better results using less effort than they currently are able to,” he said.

“There’s also an important conversation we should be having as a society. Yes, some of these developments will make some positions obsolete but they will also create an entire new ecosystem of higher value jobs, usually with the potential for greater job satisfaction.”

He points out that while automating more of a production facility does tend to mean some roles become redundant, it also creates the need for more people who can maintain these machines, programme them, refine them and instruct them. In the end, more jobs will likely be created than lost.

“Some serious leaps in ability have occurred in recent years but it’s also important to know that we’re at the start of something here, not the end. Rolling these abilities out takes expert judgement but yes, a lot more is possible and should be explored than even a few years ago,” he said.

A crucial part of Brennan’s work with DMI is in exploring the essential appropriateness of what the company suggests as solutions to its customer’s problems. It’s not, he’s keen to point out, a vendor of any one technology and its staff aren’t salespeople.

“We’re interested in quality, not quantity. Our goal is to work with each company as an individual entity and work with what its needs are. Just putting a vision system at the end of a production line isn’t necessarily going to do that,” he said.

“If a line completes 20 complex steps and each step adds €100 to the part, then a vision system that assesses the result at the end of all 20 steps doesn’t make sense. If a mistake occurs at step two but it’s not caught until step 20, then that’s hugely wasteful. So, it’s our job to be able to know where the value is being added in the process.”