The Journey to Autonomous Manufacturing

10 Feb 2020 | DataProphet

Where have we come from and where are we going in industry 4.0?

In reflecting around the concepts of industry 4.0, you’re thinking about where production systems have come from, and the changes we’ve made to improve throughput and efficiency. It starts off with robotic process automation, where you take some repetitive process and you apply some mechanization to improve the throughput rate of that step.

The interesting thing about this process is that it doesn’t actually make the quality much better, it does certainly make it less variable. But if you’re going to produce 1000 defects with robotic process automation, you’re gonna make 10,000 defects in about the same amount of time. So great for throughput, but not so good for improving your quality. Then we’ve started shifting these paradigms to become more data-led in our production systems, where we draw in production information from our data systems like historians and so on – to inform the analysis that we’re doing on this process.

We come up with better control limits and, ultimately, this just makes the process automation job a little bit harder, because you have to control the process within these finite bounds. It makes the production system a little bit more expensive, but it does have the upside of reducing the number of quality defects that are experienced.

Looking at those two and thinking about where it goes next, we’ve got to think about how we make these adaptive changes as quickly as possible, to understand as much of the system from start to finish as possible. Such that we can improve the final result. That’s the trick because that last step is what really matters the most. And you’ve got to make your production system flexible, to tolerate as much variance in upstream processes as possible, without compromising on the final quality that’s produced at the end result.

So what industry 4.0 does is solves that quality problem. Given the variance of supply of materials, be it raw materials, the input of a foundry through to sort of semi-completed goods from tier-one automotive component manufacturers. You’ve got to take all of that and produce a perfect motor vehicle at the end of the line or a perfect widget – given some of these changes. So where we’re going with industry 4.0 is much more flexible, almost autonomous manufacturing systems that are correcting in as quick as time as possible, even ahead of real-time, and in places to produce the best quality at the lowest cost.

How does Autonomous Manufacturing make a production system better?

The journey to autonomous manufacturing is quite complicated – it’s not trivial about just turning it on. But in that step, we’re sitting in the guidance space where we’re able to make corrective suggestions to your production team to improve your quality and what this looks like is; go and improve process X, value Y from this value into this range.

We try to make that range as wide as possible to accommodate as much variance in your process as we observe. In that guidance, we make your production system better for two reasons:

We’re reducing your manufacturing risk because the quality result is assured given some larger variances of input material.
Improving your total system efficiency because less scrap means greater production capacity and better parts produced to the end of the process.

Where are most plants today in their journey to autonomous manufacturing – and what is the best next step?

I think most plants today are drawing data from the production systems to their engineering or production teams. Folks are following their own inquisitiveness looking through the data to try and come up with some kind of optimization improvement or system improvement. I don’t think that there is a holistic use of data from the start of a process to the end of the process to do overall system improvement. There are two reasons for this:

I think that the systems are too complex to express in terms of classical engineering descriptions of these processes. So, an engineering model that would be able to handle the material from the start of the process through to the finished goods at the end of the process is just too complex to express analytically and too complex to solve with traditional methods.
The traceability of the component through this large process is very difficult to achieve. Unless that system has incredibly rigorous sampling and tracking of component flow, folk aren’t able to naively just join that data from step A right through to step Z.

I think that solving this requires two things:

A flexible data system that allows you to express that relationship between the start of a process and the end of the process without having to enforce the rigorous traceability that would be required to do it very simply.
Look at the process with a slightly different view, how to use a quality result from each of these steps to make a final quality improvement at the end of the process. Remembering that final step result is in one that matters the most and that as you move through your manufacturing process, your manufacturing risk is in fact increasing, because you’re spending all of that time and money transforming these goods, just to have a failure at the end. And that’s a big waste.

Can an old production system ever get to Autonomous Manufacturing?

The really exciting thing about this paradigm of autonomous manufacturing is that it’s specifically designed to work with the existing process. The only thing that changes is becoming more data-led: Use production data and quality data to make the small set of process changes that result in improved quality.

The journey to autonomous manufacturing is, in fact, predicated on having an existing production system there. Although you can work in Greenfield spaces, and then good ways of doing that. But you’ve got a good production system, you’ve just got to make sure that you get enough data out of it to describe the process and then we can have a real substantial impact on your system.

What is the difference between reactive and prescriptive manufacturing?

Reactive manufacturing works where you have a quality failure at the end, and you go and make a set of reactive changes to your system. In order to correct that error. There are two things about this:

The quality defect has happened already.
You keep producing poor-quality goods until you actually solve the root cause.

A prescriptive system is very different. So what we’re saying is to make the small change now to avoid a future defect failure, a future quality failure. So it’s a small set of corrective actions in anticipation of a quality cost that is not ever realized. If you act on prescriptions, you never incur that cost of non-quality throughout the rest of your production line. Whereas reactive you wait for the quality failure, and all those costs, and then you spend extra money in your system to go and correct for those failures.

What are the best practices manufacturers need to consider in their journey to Autonomous Manufacturing?

The three best things that you can do to your production system now to prepare you for autonomous manufacturing are:

Getting your data and save it. As soon as you’re saving your production data, the opportunities in the future for you to use that data are there and wide and many – and that’s the most important thing you can do.
Improve your quality system. As long as you’re recording not only the fact that a defect occurs but also the details of the defect – and turns out that’s really important to diagnose what the root cause problem is. This drives our AI system to do that automatically for you. So yes, you’ve had a defect and this defect has a type to it, a location to where it occurs, and it has some kind of description that that gives you a little bit more information than it’s just nonconforming.
Come and chat to us about how you get your system installed and what we do in terms of our first steps to running a pilot at your plant.