Tire industry now a pacesetter in AI arena

For many years, discussions with tire manufacturers worldwide were mainly about machines, hardware, production and specific projects, according to Dan Paul, global industry manager at Rockwell Automation.

“But, when Covid hit, everything changed: the focus switched to digital transformation, sustainability and skills aspects,” Paul added in an interview, accompanied by Jordan Reynolds, principal and global director of data science, EMEA, at Rockwell’s Kalypso unit.

The sudden change, Paul said,  was largely due to the “drastically constrained labour element” with employees unable to go into factories and then the departure of many experienced people from the workforce.

“This increased appetite for investment in automation, within which the main lever left to pull was ‘digital algorithmic’ capability, so that demand for things like industrial AI went up dramatically,” said Reynolds.

So much so, added Paul, that the tire making industry is now probably outpacing the overall industrial market in terms of this accelerated adoption of AI and machine-learning technologies.

This shift is partly because the control & automation challenges within tire manufacture are really significant, said Reynolds. In tire manufacture, he explained, “you’re dealing with elastomeric materials, that are very hard to mix, to extrude and to assemble into a tire.

“These processes are quite difficult to control because the basic material behind all of it – the rubber polymer, or elastomer – is very difficult to deal with in an automated fashion.”

And, as any disruption to the automation capabilities can be “catastrophic” to tire production, manufacturers started looking at new methodologies to enhance control of their processes.

The net result is that Rockwell/Kalypso is now working on AI initiatives with most major tire makers, according to the US-based automation group’s officials. 

Current projects cover almost every stage of production, around parameters such as asset-availability, consistency of production, quality levels and scheduling accuracy, said Reynolds.

Moreover, the work involves building AI capabilities that optimise tire-assembly processes: determining, for example optimal lengths of material, pressure and/or rotational drum speed in order to get perfect alignment during splicing.

The Rockwell experts went on cite a tire-splicing use-case, which delivered 150 additional tires per machine per day in one manufacturing plant.

Before starting to apply AI, “every client had an accuracy level of between around 70% and 85%,” reported Reynolds. So, two out of every 10 tires would be out of alignment due to the “stretchy nature” of the elastomeric materials.

With AI, he said, companies can take around 18 months or so of historical data that describes, for example, what a tire-splice profile should look like.

And, with data on environmental parameters, tire-type, pressure settings, measurements, etc, they can build a predictive model with machine-learning to predict the profile for the next tire.

“So, if the profile is out of alignment, the splice is too heavy, too light or uneven, it can determine the optimal adjustments to make in order to bring it back into tolerance in real-time,” said Reynolds.

AI, he continued, can also factor in how long the material waits on the conveyor where it can contract, or its position within a roll of materials where the elastic forces change from outer layers to nearer the core.

“Then, you can use AI to look at the moving average pressure-sensitivity over time,” Reynolds noting that “no human could take in all of these variables and write PLC logic. It’s just too multi-variable and non-linear and the relationships are very unclear.”

Mixing challenge

Meanwhile, he continued, in mixing the main challenge is to get the right Mooney viscosity: “Controlling how much power goes into the mixing process, over a set amount of time and then how that power fluctuates over time. It can be a roller coaster ride.”

With mixing, said Reynolds, it’s good to have operators who are experienced in adjusting settings and control points, such as time, heat, pressure and RPM to control the power-curve.

“We essentially apply the same methodology, using historical data from previous mixing jobs to ensure the Mooney viscosity measurements come out, okay. And we learn how the power-curve associates to viscosity, according to different compounds.

“With that knowledge, we can then prescribe the right power-curve or the right set-points to get more consistent viscosity. That changes the entire quality-cycle from mixing and extrusion through to tire-building, curing, etc.”

Reynolds went on to explain how extrusion is “really difficult” as it requires controlling heat, pressure and process speed while trying to get very consistent thickness and width to within say a 1% tolerance.

Here, some tire companies are using a machine-learning modelling approach to predict what the dimensions are going to be, based on certain set points and then optimise the process with that model.

New learning

“You have to think of this as a new way of doing control,” said Reynolds. “With this new transition that we’re going through right now, most software applications have shifted away from programming and more towards learning.

“So, for example, with facial recognition on Facebook, all software applications utilise machine-learning in cases where it’s too difficult to program something but you can learn out of historical data. And the same thing is now happening in industrial automation.”

According to the Rockwell managers, certain major tire makers are already having “great success” with AI, “ultimately, producing more tires. They are scaling it up dramatically, helped by having the expertise they have brought on board to set it up.”

Asked about the barriers to wider adoption of AI and machine-learning in the tire industry, Reynolds said the main ones right now are to do with awareness, understanding, and acceptance.

Much of this requires engineering leadership, corporate leadership with understanding of these topics, said Reynolds, adding that the average tire manufacturer doesn’t have many data scientists and machine-learning engineers at their disposal.

But, he continued, some of the largest players “are employing entire data science teams with up to 200 people, as they’re taking this so seriously. Other tire companies are in different stages of maturity on that.” Adoption is realising that AI is an imperative and then acquiring the talent to set it all up said the Rockwell experts.

“This is about a journey and buying into the results that AI promises to achieve: greater levels of automation, more tires on a daily basis. It’s really about having confidence in what this is delivering in many different areas of tire manufacture” Reynolds concluded.