Smart machines are changing the way we manufacture.
Machine learning, a type of artificial intelligence that gives machines the ability to learn without human help, creates new connections between manufacturing design and technologies like augmented reality. But it also changes the actual practice of manufacturing.
Today, machine-learning technology is relatively easy to acquire, but getting started with real-world applications in manufacturing can be very difficult. To help firms navigate this new terrain, I am excited to appear as a guest on Digital Engineering’s Live Roundtable Discussion “Breathing Life into Digital Twins,” on Tuesday, March 14. I’ll be sharing some simple ideas for how to get started with machine learning in manufacturing.
The Digital Twin
The digital twin is one real, practical example of machine learning in action. Inexpensive, high-quality sensors make it possible now to closely monitor the manufacturing process. Using the data from these sensors, organizations can build digital simulations of the manufacturing process — the digital twin.
Simulation allows companies to safely and inexpensively explore cutting-edge ideas like the use of new materials, new product designs and new manufacturing processes. For any new scenario, twin simulations can predict cost, performance and the most likely flaws. What a company learns in simulation can be used to improve manufacturing in the real world.
Video 1: The idea behind the manufacturing digital twin
Technologies like the digital twin are not just hype. They are here, today.
CSC’s Big Data and Analytics team has built a digital twin that simulates the manufacture of hybrid cars. GE has released an industrial Internet platform (called Predix) that makes it easier to build, deploy and manage a digital twin.
It isn’t commonplace, yet, for machine learning to be embedded within manufacturing design software. Efforts in that arena are still mostly in the research and development stage. But leading-edge adopters of machine learning in the industry are using the technology to narrow down options for a design and to predict the performance of a design.
Suppose, for example, you wanted to manufacture a hybrid car that was efficient enough to drive in the city but also cheap to own. CSC’s Big Data and Analytics team built a digital twin that runs day and night looking for innovative hybrid-car options. The algorithm, which scales using our Industrial Machine Learning approach, alerts us when it finds a manufacturing design option that is likely to fit.
Video 2: CSC’s Big Data and Analytics team built a digital twin that simulates the manufacture of hybrid cars.
Combining machine learning with manufacturing design is not just hype.
Using machine learning to test possible designs, GE’s Industrial Solutions team has built a shoe-box sized circuit breaker capable of dissipating lightning-sized electrical discharge. Airbus helicopter engineers are using natural language processing algorithms to “read” design specifications and then make repair recommendations. Schaeffler is using machine-learning algorithms to analyze data from sensors embedded in a machine’s bearing components, and then make suggestions on how to improve the machine’s design. ScanDisk is using machine learning to validate hypotheses about the design and assembly of its semiconductor chips.
How to Get Started
Making sense of this technology can be difficult. When getting stared, it’s helpful for manufacturers to follow two practical rules of thumb:
First, develop a real data strategy.
Second, acquire and implement technologies using small, agile experiments.
Avoid adopting a technology or starting a project simply because you’ve heard someone else is doing the same. This will only doom you to an endless series of cookie-cutter applications. Instead, create a map of your organization and use that map to figure out the best initiatives for you. (Here’s a link to a video describing the process. Here’s another link to workshops that CSC offers if you find you need help building your map and defining your data strategy.)
Avoid biting off a machine-learning transformation all at once. Instead, run small experiments that make it easy for you to recover from mistakes. Create a hypothesis about what you think might work (here’s where having a real data strategy helps). Test those hypotheses using small experiments and learn as you go.
Figure 3: Adopt new machine-learning technology in small, agile experiments.
To learn more about this space and its exciting possibilities, attend the Digital Engineering roundtable on March 14. I’ll be joined by Chad Jackson of Lifecycle Insights and Thomas Leurent. The session will be hosted by Kenneth Wong.
Also, please don’t hesitate to post your comments and questions below. Let me know how things are going with your efforts to get started with machine learning in manufacturing and please share any insights you have that could help others do the same.
Jerry Overton is head of advanced analytics research in CSC’s ResearchNetwork and founder of CSC’s FutureTense competency, which includes the Predictive Modeling Research Group, Advanced Analytics Lab and Predictive Modeling School. Connect with him on Twitter.