Industrial AI Update

In manufacturing today there can be a gap between the promise of AI that we hear about in the news cycle and getting real ROI from technologies involving artificial intelligence. Today on the podcast, we’re talking to an expert in Industrial AI. Bryan DeBois is the director of industrial AI at RoviSys, where he helps manufacturers realize the value from incorporating AI into their industrial processes. Bryan DeBois, welcome to the podcast.

Bryan DeBois: Well, thanks for having me.

Leah Archibald: So, what are we talking about when we talk about AI in the manufacturing industry?

Bryan DeBois: We don’t realize it, but it’s actually been around the industrial space for quite some time. If you look back, there’s some specific applications of AI that we’ve been using in the industrial space for at least 10 years, in some cases as far back as 15 years.

2 Traditional Uses of AI in Industry:

1. Anomaly Detection – In anomaly detection, an AI learns what normal looks like. It can tell you when a process is going abnormal, although it may not be able to tell you why.
2. Predictive Maintenance – In predictive maintenance, AI uses past data to predict failures that have occurred in the past. This technology is good within this limited scope, but it does not handle scenarios where things fail in new ways.

This is where we’ve been in the industrial space. We’ve been in this space of what’s called supervised learning.

Leah Archibald: And supervised learning means there is a human involved in the process who’s supervising something?

Bryan DeBois: Totally. In the case of quality, you’re training the AU system to predict the final quality of this batch. It’s gonna be based on all of these existing batches that I know. And one of the important assumptions that’s built into that model is that somebody knows what to do with that prediction. So if I can predict for you what the final quality of this batch is gonna be, does somebody on the line – an operator or a supervisor – do they know how to fix that outcome? If AI can predict that this batch is going to be of subpar quality, does somebody know how to bring the batch back up to spec? That human intervention is built into any of these predictive supervised learning approaches.

Now, the final type of AI that’s available today that is the most cutting edge is what’s called autonomous AI. This is where we leverage what’s called deep reinforcement learning, which is different than supervised learning. Deep reinforcement learning is actually able to do what people want AI to do, which is to make a decision, not just make a prediction about what could happen, but be able to say: in order to fix that, you need to do X, Y and Z.

• Supervised Learning – Needs a human to interpret the results of an AI model. Ex: Anomaly Detection, or Predictive Maintenance
• Deep Reinforcement Learning – Uses AI models to advise humans on the best action. Ex: Autonomous AI, aPriori’s aP Design

Deep reinforcement learning as a technology was a big leap forward, and it’s been around for a while now. But as you know, in the manufacturing industry, for good reason, we’re risk averse. We tend to get things typically 5-7 years after the rest of the other industries, like finance. We tend to get cutting edge those technologies later. So it’s just coming into the industrial world right now. But we can do autonomous AI today, and it’s like magic because it’s making decisions that a human operator would make in real time.

One of the things that I typically talk to customers about is we are solving the unsolvable problems with autonomous AI. We’re able to tackle problems that in the past you had to throw humans at. Or you just lived with the fact that we’ll never solve this problem. In the past it was not something that could be solved. And we’re actually tackling those problems and we’re able to solve them with autonomous AI.

Leah Archibald: So I’m thinking of someone coming to this podcast thinking: how can I take advantage of AI technology? And what I’m hearing from you is start with the high level problems that you’re facing. So when I talk to manufacturing execs they say: look, we need to cut cost. Cost is such a big variable and it rolls right into our bottom line. We’ve got to get the cost down. And then they’re also thinking: I’ve got to get this to market sooner. And I’ve got to make sure that this is manufacturable way before prototyping and testing. So it sounds to me that this smart technology is really hitting both of those.

Bryan DeBois: It is. Once you have the problem, now I can work backwards. And you mentioned about the concern about cost, and the concern about speed to market. There is an effort right now to move towards onshoring and bringing a lot of manufacturing back into the States. And we will not be able to compete with countries that can literally just throw bodies at these problems unless we adopt smart AI, autonomous AI, and these types of technologies. Those are going to be the multipliers that we need to be able to compete with the low-price labor that you can get in so many of these other countries.

Leah Archibald: I hear you saying: start with the business problem, then map the technology to it, and then come back to the business problem, because that’s really how you can see if you’ve gotten the ROI out of your AI implementation. You need to know if your AI solution is actually decreasing cost or reducing manufacturability errors, speeding my time to market, or reducing my carbon, if that’s your KPI.

Bryan DeBois: I would say that’s right. And it’s interesting that you bring up sustainability in particular, because I think that a lot of these companies have very aggressive sustainability goals that they’ve set for themselves. And some of these processes have not changed that much in 100 years. To me, the only way that you’re going to hit those types of goals is by adopting industrial AI. I don’t know any other way that you’re going to be able to do it.

Leah Archibald: If all I was doing was following the news, I would think that I’m already behind in adopting AI in my business. How much of a race do we need to be in to adopt in order to adopt and compete based on AI?

Bryan DeBois: I think that’s a really fair question. The reality of it is that there are some customers who have really gone all in on AI, and it’s really impressive the results that they’re seeing. But they are the exception. The vast majority of customers are really still pretty early on in their AI journey.

Leah Archibald: So for most of the industrial companies listening to this podcast, you’re not late to the party. Lots of room to expand. Lots of room to get ROI if you jump in right now.

Bryan DeBois: And I’ll make one last point here, which I think is important. Everyone talks about how we’re in the 4th industrial revolution. I truly believe that AI will be the outcome of the 4th industrial revolution. I think that that’s what we’ve been working towards. I’ve got a long arc of a career, and we’ve been doing these data projects for the last 20-plus years, and we’ve been collecting all this data. And the question I always get is: we have tons of data, we just don’t know what to do with it. AI is the thing that you do with it. Now we have the reason why you were gathering all that data for 20-plus years. Now we can leverage that data and make some really cool things happen.