Beyond robots: AI automation in manufacturing is about visibility and flexibility

In late February 2020, weeks before the World Health Organization declared a global pandemic, MIT professor David Simchi-Levi, in an article cowritten with Pierre Haren, predicted the closure of manufacturing plants that would lead to toilet paper shortages and supply chain chaos globally.

Simchi-Levi, who leads the MIT Data Science Lab, made those projections with the help of his “Risk Exposure Index,” a mathematical model that predicts the impact of disruptions (like pandemics or volcanic explosions) on a supply chain.

According to Simchi-Levi, streamlining manufacturing, which is, as the pandemic demonstrated, particularly susceptible to disruption, requires a shift in how companies think about risk: Rather than measuring the likelihood that a disruption will occur, the focus should be on building resilience—which he defines as the ability to respond quickly when and where shocks do occur—and which requires much more flexibility than most manufacturing processes have today.

So, how to implement a flexible, resilient approach and ensure products are on shelves to meet consumer demand, rain or shine? With an immense amount of data and a little help from AI.

Software solutions and simulating risk

“AI is not about automation, it’s about bringing intelligence to automation,” explained Azita Martin, VP and GM of retail AI at accelerated computing company Nvidia.

Martin pointed to Omniverse, Nvidia’s simulation software, which creates a “digital twin” of a factory—an accurate digital representation of the physical environment.

Then, in what Nvidia calls the “industrial metaverse,” companies can test out different factory layouts and make optimization decisions aided by real-time updates on the physical space, she said. That’s a solution that can double throughput at pilot distribution centers, Martin added.

Meanwhile, machine health company Augury is making predictions about mechanical failures before they happen, in an attempt not to replace technicians working on those machines, but to help factories avoid disruptions caused by malfunctioning machines.

“If you’re able to tell a manufacturer that one of their key components is going to have a failure months or weeks before it actually fails, you can put it on a planning horizon,” Brian Richmond, head of solution architecture at Augury, told Retail Brew. “It gives you the ability to plan the plant, and quit letting the plant plan you.”

Augury’s sensors, which collect data about vibration and temperature to diagnose malfunctions, are currently used in factories for everything from consumer packaged goods manufacturing to lumber and paper, Richmond said.

That technology is designed to augment human work rather than replace it, but there’s no denying that over time, AI-powered automation will change the landscape of factory work.

“Long term, it’s about giving [companies] the ability to make decisions with the most amount of data…as soon as possible,” Richmond told us.

What companies do with that data will impact workers like technicians, usually in the form of more customized training and upskilling, which ultimately make workers more valuable, rather than less, he explained.

Nvidia’s Martin said upskilling opportunities aren’t just for skilled workers. She pointed to automation combined with upskilling as a way for companies to deal with high turnover rates in places like warehouses and factories.

“Upskilling really is going to [create] career opportunities,” Martin said. “There will be some transition, but we’ve never seen that people are going to be fully replaced.”

The robot factor

That’s not to say that building agile factories is all about software.

Ed Mehr, a former SpaceX engineer and the CEO and co-founder of Machina Labs, is tackling the challenge of making factories that are product-agnostic, to reduce manufacturing costs and increase flexibility.

“Tomorrow, if Samsung or Bosch wants to make a new version [of a stove]...they have to invest a lot of money in making new molds, which can cost up to a million dollars, 50 weeks of lead time, and up to $80 million for new equipment,” he explained.

Machina’s AI-powered sheet metal “roboforming” eliminates that risk by enabling manufacturers to make changes quickly, he said.

Working up: Mehr says companies implementing this type of automation at production facilities will need to hire more skilled workers, such as software and robotics engineers, but that there will be other trickle-down impacts in the form of reshoring and job stability.

Automated factories are cheaper to operate, and companies will likely bring production closer to their end consumers in the US, Mehr said.

“Because these factories by nature are more flexible…it’s less likely that if a product [becomes] obsolete, a worker will lose their job, because you can just reprogram the factories,” he concluded.

It’s that type of flexibility that Simchi-Levi says is necessary to avoid the type of supply chain chaos caused by Covid-19. It isn’t easy to achieve—data collection and uniformity, on top of visibility issues with suppliers, make end-to-end visibility difficult—but it will ultimately allow companies to predict and respond to demand, he told us.

“If you can improve forecast accuracy, you can improve everything in your supply chain.”