We sat down with the CEO of deevio, Donato Montanari, to discuss his previous experiences in the computer vision field, why deep learning is such a good fit for industrial automation, and how deevio can revolutionize quality control.
As of autumn 2018, wattx has founded its newest venture - deevio. In this article, we sat down with the co-founder and CEO of deevio, Donato Montanari, to discuss his previous experiences in the computer vision field, why deep learning is such a good fit for industrial automation, and how deevio can revolutionize industrial quality control.
I am an Italian engineer and have spent most of my career working in computer vision. Starting with hardware, then firmware, and software, before finally moving to management – pretty much anything that I’ve done has been related to imaging or light, one way or another.
My first job was in the US, where I worked as an IC Designer and designed image sensors as well as memories chips that go into a camera. That’s how I got started.
I’ve spent about 10 years working as an engineer in the US, and was starting to feel like it was time to move back to Europe. The company I was working for at the time, Cypress Semiconductor, offered me an opportunity to run one of their business units in Europe. It was a move, not only geographically, but also professionally, from a more technical to a business and management role.
That depends on which market you focus on. You basically have two markets: consumer and professional. I would say that anything which is related to consumer is very heavily US- and Asia-dominated.
In the professional space, meaning anything from industrial to agriculture to aerospace or astronomy, I would say, the opposite is the case. Europe is actually very strong – and Germany is definitely leading in computer vision for industrial automation. However, European companies are not doing a very good job in advertising this fact. So, when we think about computer vision, we all think about Google or Facebook. And it’s true for the consumer. But it’s not true for the professional space.
As children, when we learn what objects are, we don’t think of it in terms of rule-based programming, right? You don’t tell your 4-year old when he or she asks what a smartphone is, that it is a 12cm high and 5cm wide object with rounded corners that is 6mm thick. Instead, you point at a smartphone. Then, the child tries to classify different objects like remote controls or iPads as smartphones and you have to tell them “no, no, no” and point to an actual smartphone. The more smartphones it sees, the easier it can “distinguish” between what is a smartphone and what is not a smartphone. This is, in very simple terms, what you also do with deep learning.
It all started with one simple question: How do you install a machine vision system in a factory if you want to do quality control? Usually, the traditional way is that you write some code. You have to explain to a camera the feature set of the object you are trying to detect. This is very time-consuming and requires people that are very skilled in the subject matter. And you don’t find those easily. It is very difficult to find good machine vision or computer vision experts.
Another problem is that these programmed systems can be very inflexible. If some features change in the object you want to detect, you have to code all over again. So, we - me and the company that I used to work for - were thinking that there must be a better way of doing this. And that led us to artificial intelligence in general, and more specifically to deep learning.
Today, for every product you have to program strict rules. And if you lack rules for a new product or changes are made to an existing product, you have to change the code.
The big advantage of deep learning is that it allows you to introduce flexibility to the line. And the manufacturing trends of the last years all point to personalization.
There is this great quote from Henry Ford: “Any customer can have a car painted any color that he wants, so long as it is black.” Those days are absolutely gone. People want to have their own personalized items. And that’s what they want to buy. Which means that ever-smaller batches are manufactured that change even more often. Typically, our customers change their production lines several times a day and in this situation, having a rule-based approach to quality control is not particularly beneficial. There is just no flexibility there.
After having the initial idea, the first question was how can we do this? The problem was finding people that could take this idea about using deep learning for quality control and check whether it was feasible. And this is where wattx enters the picture.
What wattx could offer was a team of professionals with the right skill set to work on complex projects like this one, while covering all possible angles that you need to examine, with every discipline from data scientists to front-end engineers to venture developers on staff. So, we approached them early in 2018 and started to and are still working very closely with them to develop what today is known as deevio.
The problem we are trying to solve is quality control. But, and this is important, there are two kinds of quality control: it can be in-line and off-line. If you want to produce an object, you have to buy a bunch of machines, each of which produces a component of your object and you stack them one after the other to make a line: the production line for your object. All quality control that is made on the production line we call “in-line”. And every machine has its own cameras and its own controls and little parts that it produces, so these systems can be fairly complex. But then, at the end of the line, when the object is complete, there is this one person who grabs a finished object and looks at it. This is called the final control and is in the vast majority of cases done by a human worker.
And this is a very stressful job and very difficult because human eyes get tired. So the level of quality fluctuates during one shift. It also depends on the person; There are very experienced people that have done this job for many years, and there are new people on the job. It’s not homogenous. Further, human inspection traditionally doesn’t leave any track record. If I say that this phone is good and this one is bad - once I have made that decision, there is no evidence substantiating my decision.
What deevio does is to put, next to this person, our system – which is basically a camera and a computer – and instead of this person having to really pay a lot of attention inspecting a specific object, he or she puts that object under our camera, and our camera says “okay, I think this is where you have to look because there may be an issue here”. In essence, we want our technology to augment human quality inspectors.
There are also other benefits for companies that adopt our technology. The camera, like any other camera, takes a picture. That picture is saved. Now, the big advantage is that you have a picture for every product shipped to your customer. And therefore, when the customer says “hey, you shipped me something that wasn’t correct”, you can go back and check exactly that order with its specific pictures, and find out where the fault happened.
Since we don’t have to do all this manual programming anymore, our solution is actually much cheaper than the traditional approach, as we don’t have to spend the same amount of time to reach an acceptable result. Also, the traditional approach requires several players to be involved in providing a solution for you. There is a guy selling you the components, then there is a guy who’s running and installing this solution, and then there is a guy writing the software. So basically, three different organizations have to be involved. With us, we are the only ones that need to be interacting with the customer.
It is very hard to say because we are very new. The proof is always in the pudding. You have to have a couple of years running the product in a production environment before you really can talk about accuracy. What we can say though is that we are competing against humans, and the human eye does get tired. Regardless of how good or experienced you are, if you look at similar objects for 8 hours a day, I guarantee you, your performance during the seventh hour is not the same as it was in the first. So the biggest benefit is not the accuracy but the consistency. Because whatever accuracy we get, which preliminary seems to be very high, we will be consistent for eight hours. Across shifts. Across the work week. Across the seasons, and so on and so forth.
You are correct. There are a lot of competitors - which is a good thing because if there is no competition there is no market. What makes us different is that most of those competitors focus on in-line inspection. So, they want to put a system and a camera in every single machine, whereas we focus on off-line inspection. We want to put our system next to the person who is responsible for inspections.
We still think that this is a pretty big market. Studies from BCG suggest that out of three people working in manufacturing, two of those make products while one checks the goods that the other two have produced. Which is not an optimal number, right? So, one of our advantages is that if we help this person doing their job better and faster, they can spend more time helping out at the assembly line. Which translates into a real improvement of the efficiency of the organization.
Right now, we are about 10. The majority of us have a technical background, and are mostly data scientists.
Besides not speaking German and handling the cold? (laughs) One challenge that we are trying to mitigate is that, except for me, our team is relatively inexperienced in the market we are working in. But I see that more as an opportunity than as a challenge. Because, the big advantage of people who are not necessarily experienced in a particular field is that when they get to the field, they look at it differently. Every time you look at things differently, you have the opportunity to innovate and contribute something new. That’s something - if managed properly - that becomes an asset.
One of the biggest advantages of having worked in machine vision for a long time is that you get to know a lot of the people who make the decisions. And this has proven very important. Especially if you put that together with the network that wattx has.
We are really trying to change how the industry works. And I think that every time you are setting yourself up for a big change - that’s always very exciting.
The other interesting thing, that interests me personally, is - usually, if you want to be successful with AI, you have to be very vertical. You know, the guys that have to specialize in PCB inspections, or the guys that specialize in the inspection of little vials, and so on and so forth. Since we work at the end of the line, with the people in this manufacturing environment, what we do is actually very transversal to the industries. We get to work with people who make all sorts of different things, from chemicals to plastics to automotive. So, every day there is a different challenge that we are trying to solve. There’s a lot of variance and a lot of constant learning going on, day in, day out.
I think it’s a really good opportunity. What you see is that, if you look from a corporate point of view, when you got to the peak of your career - which usually happens at an age of 45 to 55 - a lot of work has become routine. Some like that security, knowing what they will do and what to expect. It depends on your personality, right?
But for some, you’ll eventually start feeling that something is missing. The challenge, that passion, the ability to get your hands dirty.
I think that there are a lot of very skilled people who feel that. And many of them probably walk around with a great idea that they have been thinking about for years, that they would love to work on if they could only find a way. However, and here lies the problem, these people often don’t have the possibilities to work on their passion projects. They can try to launch something internally but most organizations are set up to improve on existing products and services, not build new ones. Especially if we are talking about something technical that is extremely difficult to bootstrap. On top of that, you have to figure out how to put a team together.
And that’s really where working with wattx is becoming critical because everything is already there. The team is waiting for you. All you have to do is come in and bring an idea with you, and then work with the team to make it happen. And that’s fantastic. I think that you’ll find a lot of executives who would be very happy with this environment.
What’s happening in the industrial automation world is that there is a lot of insecurity. Companies being bought, companies being sold, the Chinese market growing faster than ever, so you do have people who are in a position where they can say “okay, I can take a break from my secure executive job. I am really not struggling too much financially. Why not do this?” And one of the reasons why they don’t do this is because they don’t realize that there is a wattx.
Email works great: donato(at)deevio.ai - it’s the best way to get in touch with me. I’ll be happy to answer any questions, AND we are hiring! (laughs) We are in Berlin and we are funded.
