Using Machine Learning Tools to Drive Manufacturing Innovation

Machine learning tools can prove to be a valuable asset for manufacturers to drive operational efficiencies and improvements. Manufacturing companies can use these tools and the data they provide to analyze and improve processes while testing new ideas, designs and production strategies.

Piotr Niedzwiedz, co-founder and CTO at deepsense.io, said big data and machine learning can revolutionize the future of production. He said companies already are collecting and using data to improve quality at every stage from design inception to the last step of production. “We can use machine learning and predictive analytics tools to mine data for these patterns and detect failures sooner,” Niedzwiedz said.

Louis Columbus, director of Global Cloud Product Management at Ingram Cloud, said the core technologies of machine learning align well with the complex problems facing manufacturers. Sensors collect data for processing, then are used to produce algorithms which are designed to continually learn and seek optimized outcomes in everything from supply chain efficiencies to built-to-order products. “These algorithms iterate in milliseconds, enabling manufacturers to seek optimized outcomes in minutes versus months,” Columbus said.

He said there are a number of ways machine learning can revolutionize manufacturing. First, it can increase production capacity while lowering material consumption rates. Machine learning tools also deliver more relevant data so finance operations and supply chain teams can better manage factory and demand-side constraints. He said machine learning can bring “an entirely new level of insight and intelligence into these teams” to optimize their decision making. These tools also can help manufacturers better manage their equipment by performing better maintenance, preventing breakdowns and extending the life of machinery.

“Manufacturers are turning to more complex, customized products to use more of their production capacity, and machine learning help to optimize the best possible selection of machines, trained staffs and suppliers,” said Columbus.

Jerry Overton, head of advanced analytics research at the CSC Research Network, said that while machine-learning technology is easy to acquire, organizations can be challenged with starting real-world applications. Overton said one simple way to start is with a “digital twin,” where inexpensive, high-quality sensors make it possible to closely monitor the manufacturing process. Organizations can then use data to build digital simulations of the manufacturing process (the twin) to allow them to explore cutting edge ideas, designs and processes. In one example, Overton points to GE’s Industrial Solutions team which has built a shoe-box sized circuit breaker that can dissipate lightning-sized electrical discharge and help improve the design of semiconductor chips. Other “twins” can be constructed and run through models to help find optimal product designs.

Overton said manufacturers should start by developing a “real data strategy” and then acquire and implement technologies using small, agile experiments. He said manufactures should create hypothesis about what they think might work then slowly test them and build upon the knowledge base.

“Avoid biting off a machine-learning transformation all at once. Instead, run small experiments that make it easy for you to recover from mistakes,” said Overton.