Artificial intelligence already helps individual factories improve production, safety, efficiency and other metrics while lowering costs. Marrying AI and cloud technology can supercharge those benefits and offer manufacturers faster time to value, better visibility into supply chains and dynamic design, proponents say.
AI in the cloud could put an end to the manufacturing labor productivity slump. But where to turn for lessons? Try hedge funds, Formula One racing and cucumber-sorting operations.
AI has long been attractive in manufacturing for solving complex problems and going beyond alarms to intelligent insights, said UptimeAI CEO Jagadish Gattu. “Manufacturing operations and equipment are complex,” he said.
Moving to the cloud—software and services that run on the Internet instead of only on a company’s own network—supercharges AI in several ways. Specifically, the move:
- Enables access to new real-time data;
- Offers massive scalability;
- Makes for more computing power at more affordable prices, and
- Enables access to subject matter expert knowledge across different manufacturing sites.
“We’re living in a productivity slump,” Baber Farooq, head of product strategy for procurement solutions at SAP, said in reference to the often-cited stats from the U.S. Department of Labor and the Bureau of Labor Statistics, as well as an often-quoted Deloitte/MAPI study. “Many factors are causing that. For the last 15 years, the adoption of technology in manufacturing has not been happening at the pace that we need to reverse that [productivity slump] trend.”
Over the last 15 years, cloud processes have come to maturity and helped drive productivity for IT.
“The promise of cloud computing has been the promise of scalability—how can I scale to multiple sites, multiple customers?” he said. “AI is the continuation of that growth. AI holds the key for manufacturing to capitalize on the gains cloud computing has brought to AI.”
AI in the cloud helps manufacturers save money, adapt to change and become aware of emerging market trends, said Oliver Christie, head of Voltare Consulting. “Because we have access to AI tools, we can reduce costs, increase quality, speed up time to end result and optimize how products are built,” he said. “We’re not changing the product, just changing how it’s made. We’re able to adapt more quickly to new situations, such as changes in tariffs. Obviously, the pandemic has changed everything. Manufacturing needs to be aware of new market situations and new market trends.”
Manufacturers are already using artificial intelligence, but combining AI with the cloud provides access to a wide variety of excellent algorithms that are being constantly updated, Christie said.
“The biggest benefit in the cloud is the marketplace of algorithms, with access to a huge number of different algorithms taking different approaches to your data,” he said. “You can pick the best one out there from around the world.”
Combining AI with the cloud allows manufacturers to open a fire hose of data and glean benefits, said Joe Gerstl, director of product management for manufacturing execution systems (MES) at GE Digital.
“When you’re dealing with data on premises, you have limitations,” he said. “You have only so much space. You don’t have time to process it all. In the cloud, you can have so much data, not just big data but very rich, raw, and very thick. When I say rich, it’s raw data. It is all the data. We have customers that have 10 years of data in their manufacturing data cloud. It’s not summarized or aggregated unless you want it to be.
“Because you have so much space and it’s so cheap to store this data, you can get all your data on the cloud. The whole point of AI is to learn. It can learn patterns. The more data you can feed it, in terms of richness and thickness, the better it’s going to be at predicting things and providing the powerful analytics you need and can use,” he added. “When you apply AI to that data, you can make results more accurate because the system has more history to look at. It can learn faster, easier and smarter.”
Rewarding early adopters
Early adopters are able to be more accurate now in their predictions, Gerstl said. They can achieve improved operational equipment efficiency (OEE), better estimate when orders will be complete, and better predict and prevent problems, he said.
“They’ve had time to learn from the data, tweak the AI, and tweak their models,” he said. “They can see trends and take action faster than before. People who are further along are just better at it.”
Early adopters also are seeing benefits as non-technical workers within their companies are able to mine the data for insight, Gerstl said.
“These ‘citizen data scientists’ are able to create some very powerful analytics that help them do their jobs better and faster, make products with higher quality, and result in less equipment down time,” he said.
One way to inspire factory managers, citizen data scientists and others to become early adopters is to show them the possibilities, Christie said.
One engaging example is using a $50 computer running AI to sort cucumbers (watch a related Mediacorp video).
“With very little technology and open-source software, you can set up something that was unheard of, or prohibitively expensive, 20 years ago and put people into the mindset of how you can train AI,” he said.
Christie recommends his clients consider buying or building such a machine, or at least watching the video to get line management and workers thinking about the possibilities.
Switch to AI uneven industry to industry
The telecom and retail industries are ahead in adopting AI in the cloud, Gattu said. In manufacturing, automotive, energy and food and beverage are standouts, he added.
The equipment life cycle in a particular sector is one factor in how quickly AI in the cloud is adopted, he said.
The switch to artificial intelligence—cloud or otherwise—is slower in sectors that keep equipment a long time, as well as in highly regulated sectors, such as energy, he added.
“Another reason for delays in scaling in domain-specific industries, such as process manufacturing, is that AI solutions are data-science-centric and less domain/application oriented,” Gattu said. “A plant engineer should not have to learn about neural networks to improve the operations, just as a driver using self-driving car does not have to know about deep learning. That’s why our plant-monitoring solution uses a purpose-built AI engine to solve the needs of plant engineers and manufacturers.”
Seeing benefits no matter firm’s size
While many early adopters have been large manufacturers, small and mid-size companies are also gaining benefits, Gerstl said.
GE Digital is starting to create what it calls starter kits with out-of-the-box AI to sell to smaller companies, he said.
One small company in the UK was able to use GE Digital’s tool to solve a quality issue that had baffled executives and shop-floor workers at the company for a long time, he said.
Another company using GE Digital’s tool had a consumer complaint with one of its products, Gerstl said. Without the analytics provided by the tool, solving the problem would have taken six months. “We did it in two weeks—and it took that long because it was the first time,” he said. “We had all the data accessible and in a proper format and we had the tools that allowed us to get to the bottom of it.”
Improving efficiency and productivity
AI in the cloud can help manufacturers improve poor-performing plants by learning what better-performing plants are doing correctly, Gattu said.
“What we generally see is that one plant in the United States has 15 days of downtime out of every year and another plant in Algeria in the same enterprise has 20 days of downtime,” he said. “The difference between 15 and 20 days can be millions of dollars.”
For example, when a piece of equipment is leaking, three possible actions might be possible to repair the leak, Gattu said: With one, the leak could be fixed in five days. With another, seven days. And with the other, only two days.
“Our AI solution is learning which of these actions is solving these problems faster,” he said. “When the issue comes up for the fourth time, the manufacturer knows he should go with the third recommendation because that’s able to solve the problem in two days. You can take that knowledge and present it to the person operating the plant in Algeria—you can transfer learning from one use case and make it available to other operations.”
Achieving greater efficiency is a key benefit, Gerstl said. For example, some food and beverage manufacturers achieve efficiencies in the 90s.
“They have to be efficient to remain competitive,” he said. “With this data, we can predict relationships and trends. In one case, we discovered a direct relationship between performance speed and quality: After a certain point, the faster you went, the worse your quality got.”
By combining the cloud with new or existing AI, manufacturers can compare results and individual variables leading to those results—machine to machine, factory to factory and potentially among other manufacturers in the same sector—to achieve better productivity and efficiency, Christie and others said.
Google bought AI firm DeepMind in 2014 and was able to deploy AI in the cloud to reduce cooling costs by 40 percent in its large data centers, he said.
“If you have many machines doing the same job, you can see what’s working for one machine and optimize it machine to machine and factory to factory,” he said. “AI in the cloud is the fastest way of optimizing across the board. If you’re connecting all those machines to the cloud, it’s easier to ask the questions and get input. Once you connect machines to the cloud, you can connect to every machine globally. If you wanted to sell information to other manufacturers doing something similar, that’s valuable data. If there’s no direct competition, sharing data would help you both.”
Making scaling easier
AI in the cloud offers flexible scaling, Gattu said.
“AWS, Google and others specialize in how to keep these systems running,” he said. “They have automatic scaling: If you have more work, they automatically scale to more machines. If you don’t have work, they automatically scale down and you save on costs.”
Companies that only have AI on premises are then “stuck with the hardware and can’t scale easily,” Gattu said. “Cloud gives you a lot more flexibility and access to more powerful servers. You don’t have to buy all the high-end compute power and then get stuck with those servers or have to upgrade every two years. With cloud, updating is easy because your vendor is getting new machines. You can scale enormously—to petabytes—in the cloud. On premises, you have to struggle to get that kind of scale without losing performance.”
Manufacturers can take advantage of a variety of tools on hosted platforms, such as Azure, and start small. Once they get their applications running, they can then scale without worrying about buying the hardware or managing the software, Gerstl said.
“The hardware is very costly for on premises,” he said. “It cost $100,000 for one of our customers to set up the hardware they use on their site. The procurement process, especially at large corporations, is a real pain. It can take six weeks to get the hardware set up. In that six weeks, the company could already be set up [for AI in the cloud] on Azure or another platform.”
Codifying experts’ knowledge
Another appeal is to codify the domain knowledge of factory-level subject matter experts, many of whom are approaching retirement, Gattu said.
With AI, manufacturers can bank that expert knowledge before these experts retire and add it to their smart factory tools. “The next generation expects to get this kind of knowledge through tools,” he said.
The best AI-in-the-cloud products have the ability to capture domain knowledge and are continuously learning, Gattu said.
“You need the subject matter expert who knows what it means when pressure changes in a pump,” he said. “In our solution, we bridge the gap between the AI, the domain knowledge, and the self-learning workflows. If you really want to get the ability to learn, to explain, to understand what is really going on, you need to have a feedback loop. In our plant-monitoring system, the AI continually learns from what the user is doing, from data coming in, from maintenance actions being taken.
“Once you have that learning and you have an application that can learn on its own, the growth of that knowledge is exponential,” he added. “Today, you might be two steps ahead. Tomorrow you might be 10 steps ahead because it’s a machine that’s learning. If there is a set of knowledge you want to build in five years, you could do the same in a year with AI. It can really increase your rate of efficiency and continuously improve the organization’s operations.”
Learning from hedge funds, Formula One
In addition to sorting cucumbers, manufacturers can learn lessons from some other unlikely sources. For example, hedge funds continue to sponsor competitions to build the best algorithm based on a set of data, Christie said.
Or, consider Formula One racing.
The weather impacts both racing and manufacturing production, he said. Slight changes in temperature, wind and rain impact how a car performs; and those same changes can impact factory production in real time. Additionally, weather changes globally, such as a hurricane 1,000 miles away at a critical point in a supply chain, can also impact production.
“A car going the smallest fraction faster makes a difference,” Christie said. “They have huge amounts of data being used in real time to make real-time decisions as to what to do next. It’s a very good industry to look at: how to make fast decisions and how to change when new data is available. It becomes a mirror of what a factory should do.”
For example, fiberglass work is very sensitive to changes in temperature and humidity, he said.
An AI system could learn from outcomes when temperatures in a factory are hotter or cooler, humid or less humid, and see what enables the best outcome.
“Let the AI set the temperature and humidity,” Christie said. “Your staff, with their huge amount of knowledge, normally has a good idea. But to correct for something as large as a factory can be difficult. Manufacturers can put some sensors in and set up simple questions: ‘What’s ideal for your manufacturing performance.’”
But algorithms are not foolproof or the end-all solution. “One thing that needs to be realized is when you build an algorithm, you’re building off data and off a human’s decision as to what’s important and what isn’t,” he said. “We need to keep in mind that algorithms are not foolproof.”
Improving creative design
While humans will remain the drivers of creative design for a long time, AI with Cloud can automate and improve the process, Farooq said.
With AI in the cloud, product designers could see immediately the impact of choosing a particular component or other supply looking at time to source, overall availability, cost, tariffs, natural disaster and multiple other factors.
Envision a product designer inputting different components and raw materials into a design program and seeing the impact in real time—in terms of time needed to receive supplies, accessibility, price and the different factories where the part could be made.
“Data that exists around different supplies available could be fed into a system, and the system could make recommendations in real time affecting not only the design itself but also the type of supplies that might be used in the process,” Farooq said. “Based on that, the system can tell me what kind of part can be available at what time and from where so I can design a particular part correctly.”
No need even to run a simulation because the system updates the end results based on changing parameters, he said.
“As the design work is happening, they are being given this information proactively by the system,” Farooq said. “They don’t have to pause their work, run a simulation and come back.”