Machines that learn

Artificial intelligence may transform manufacturing, but adoption is slow

Lindsay James
21 June 2017

5 min read

Manufacturers recognize that artificial intelligence offers an exciting future, enabling greater automation, improved predictive maintenance and a move to mass customization. While adoption so far remains slow, experts agree that the combination of human expertise and industrywide collaboration will pave the way for success.

Advances in artificial intelligence (AI) – defined by San Francisco-based computing company NVIDIA as “human intelligence exhibited by machines,” are being made at breakneck speed. Al comes into play
every time we ask Siri or Alexa a question, view a recommendation by Netflix or add a friend suggested by Facebook.

While AI helps drive many everyday consumer interactions, its power has only recently been felt among businesses.

“AI has reached a tipping point in what it can do for enterprises,” said Mark Purdy, managing director and chief economist at Accenture Research in London. “This is thanks to developments in processing power, data storage, data retrieval, sensors and algorithms. As a result, businesses are now able to optimize processes with intelligent automation systems, augment human labor and physical capital and propel new innovations.”

Business AI breakthroughs are everywhere. Computer scientists at Stanford University’s AI Laboratory in California have trained an algorithm to visually diagnose potential skin cancers. Microsoft has demonstrated a speech- recognition system that makes the same or fewer errors than professional transcriptionists. Scientists at MIT’s Computer Science and AI Laboratory in Massachusetts have mined data from more than 3 million taxi rides to develop a smarter way to move people around Manhattan. And major automakers have used deep learning, a machine learning implementation technique, to create autonomous vehicles that scan, analyze and then respond to their surroundings, aiding drivers in optimizing their decisions and actions.


The manufacturing sector, however, is lagging. In an article for media intelligence company Meltwater, Brent Dykes, director of data strategy at Utah-based software company Domo, said that “analytics maturity is a key milestone on the path to being successful with AI.” According to global consulting firm McKinsey, however, manufacturing industries to date have only captured about 20%-30% of the potential value of data and analytics – and most of that has occurred at a handful of industry-leading companies.



Forrester, a global business and technology research and advisory firm, said that much of this existing value is in preventive maintenance, a specialty of global factory automation equipment producer FANUC. The company is running a Zero Down Time (ZDT) application on its new FIELD system, which collects data from more than 6,000 robots in 26 factories and analyzes it with a machine-learning application. Any issues that could lead to a failure are highlighted, and FANUC sends parts and support to address the issue before downtime occurs.

“FANUC’s FIELD system enables companies to utilize the vast amount of data available to them,” said Steve Capon, technical manager at FANUC UK. “Manufacturing is set to become more intelligent than ever before. By using AI, the scheduling of predictive maintenance requirements to reduce downtime is a reality.”


There is huge potential in other areas too, such as improving factory automation.

FANUC is running a Zero Down Time application on its new FIELD system, which collects data from more than 6,000 robots in 26 factories and analyzes it with a form of AI known as machine learning. (Image © FANUC)

“This is an area of huge opportunity for any company, including Boeing,” Harish Rao, the company’s senior director of data analytics, wrote in Boeing’s February 2017 “Innovation Quarterly” newsletter. “Complex jobs can be automated to improve productivity, quality and safety while helping to meet delivery schedules. Data from sensors on machines can be connected with traditional data, such as design, inventory and safety records, to optimize tasks. Instead of simply identifying a task to be automated, a deep learning model can analyze all the data, determine patterns and recommend the best task for automation.”

Recognizing the potential in AI-driven factory automation, FANUC has invested US$7.3 million (900 million yen) in Japanese deep-learning specialist Preferred Networks (PFN) to create a robot that uses AI to train itself in new tasks.



“FANUC is the first company of its kind to integrate AI technology into its products in this way,” said Shohei Hido, chief research officer at Preferred Networks’ California office. “We’ve been trailing AI technology in bin-picking robots, using deep learning to estimate which would be the most successful point inside a bin to pick an object from. This process would usually require an engineer to spend around two weeks at a factory to tune the rule-based system. But by using the AI-enabled system, the robot can learn how to pick any kind of object in a few days, with over 90% accuracy. We expect this technology to be implemented by manufacturers later this year.”


AI also has the potential to facilitate the creation of more adaptive and agile enterprises.

“If manufacturing has progressed from an era of mass to lean, the appetite today is for ‘custom mass manufacturing,’” said Sreenivasa Chakravarti, head of manufacturing innovation and transformation at Tata Consultancy Services. “This implies a lot more responsiveness and quick, on-the-spot decisions. As more product features get postponed toward the latter half of operations, systems will need to respond more rapidly than ever before. This is possible only with machines that learn and adapt to these needs.”

Global sportswear brand adidas is already experimenting with AI-driven custom production, with plans to open a “SPEEDFACTORY” in Atlanta by the end of 2017. Adidas is keeping the specifics secret, saying only that the SPEEDFACTORY will use data to connect the different parts of its production process in a smart way. Robotic technology and 3D printing also will combine to create an automated, decentralized and flexible manufacturing operation. “This allows us to make products for the consumer, with the consumer, where the consumer lives in real time, unleashing unparalleled creativity and endless opportunities for customization in America,” said Eric Liedtke, the company’s group executive board member, in a press release.


For all of this potential to be realized, however, manufacturers will need to address a major skills gap.

“No matter how good an AI system is, without the human capital to apply and maintain it, manufacturers won’t benefit,” said Robert Atkinson, president of the Information Technology and Innovation Foundation (ITIF), based in Washington, DC. “McKinsey has estimated a shortfall of 140,000 to 190,000 data scientists in the US alone by 2018, as well as an even greater shortage of managers and analysts with the analytical skills needed in a big-data world.” Milind Lakkad, global head and executive vice president of manufacturing at Tata Consultancy Services, also sees a lag between the current state of the manufacturing industry and the pace at which AI is advancing.


According to McKinsey, manufacturing industries to date have only captured about 20%-30% of AI’s potential value.

“For a production facility to be effective, it’s not just the individual machine or equipment that needs to support AI,” he said. “The success of AI will require a close understanding of the dependency of the behavior of one piece of equipment on the other to plan sequenced operations.” To achieve this understanding, Lakkad said, three parallel needs must be met.

“Open and standard protocols for information interchange are required to facilitate cognitive learning,” Lakkad said. “Plus, there’s a need for sustained efforts on the part of all OEMs to bring a larger segment of their installed base ‘under management’ via the smart route. The issue of safety prognosis under such conditions also needs to be addressed. None of this can be done at a company level, but will require the entire industry to collaborate.”


While the manufacturing sector plays catch up, advances in AI will continue to evolve.

“Certainly, I think we’ll be amazed at what’s possible a few years down the road,” said Bill Franks, chief analytics officer at the Atlanta branch of the data management and services firm Teradata. “In the end, AI will only lead to more efficiency, more consistent quality and less waste or spoilage for manufacturers.”

Related resources


Register here to receive a monthly update on our newest content.