In pop culture, the combination of business interests and artificial intelligence is something to be feared. It brings to mind Skynet, the malevolent neural network from the Terminator movies that goes to great lengths to destroy its human makers.
The reality is different, though. We’re not living through a hostile takeover by the all-pervasive Skynet. Instead, we’re seeing individual businesses looking at the multiple AI solutions on the market with an eye to striking a better balance between operational efficiency and customer satisfaction. We take advantage of it every time we check out new products recommended by Amazon.com, which uses AI to extract as much information as possible from members’ transactions. We have fun with it when we browse Netflix, which uses AI to predict what viewers might like to watch next.
We’re also increasingly likely to encounter it at work, since businesses of all types are finding ways to use it in industrial, retail, and service operations. In this essay, we’ll focus on the impact that AI is having on the manufacturing sector. We’ll also consider a few of the ways that entrepreneurs and engineers have integrated machine learning – an application of AI that gives machines the ability to learn from experience without being explicitly programmed to do so – into the manufacturing process.
Improving the process
To date, manufacturers have been able to introduce AI into three aspects of their business: operational procedures, production, and post-production. We’ll start by taking a look at operational procedures – that is, at the ways in which machine learning can improve the course of production itself.
One company that’s taken this path is Fanuc, a Japanese manufacturer of industrial robotics and automation technology. Fanuc uses deep reinforcement learning, a type of machine learning solution developed by Preferred Networks that enables its robots to teach themselves new skills quickly and effectively, without the need for precise and complex programming.[1]
These robots improve their performance on difficult tasks such as picking up small objects, which would normally require programmers to spend many hours compiling precise and complex instructions, by filming themselves and loading the video footage into a deep learning system. This system analyzes the data to determine which approaches and which actions contribute the most to a successful outcome. In turn, this analysis helps the machines achieve a higher level of performance more quickly and more efficiently than they might have done if the company had assigned the job to human programmers.[2]
Shohei Hido, the chief research officer at Preferred Networks, explained the benefits of this approach to Technology Review in 2016. Once a Fanuc robot begins to practice a task, he said, “[after] eight hours or so, it gets to 90% accuracy or above, which is almost the same as if an expert were to program it.” This greatly shortens the learning curve, he added. “It works overnight,” he commented. “The next morning, [the robot] is tuned.”[3]
In Fanuc’s case, machine learning has allowed the company to turn out robots that help factories optimize the speed, cost, and efficiency of their operations. But it’s not just about streamlining. Machine learning could eventually enable manufacturing plants to react quickly to changing instructions, rather than relying on standardization.[4] It also has implications for worker safety, in that it can help identify the factors contributing to accidents and also develop solutions to prevent their reoccurrence.[5] It can, for example, analyze video footage to determine when workers fail to wear hard hats or follow other safety regulations.[6]
Improving the outcome
Machine learning can also lead to improvements in production outcomes – that is, in the quality and versatility of products that a plant manufactures.
For example, the German conglomerate Siemens has made AI an integral component of some of its best-performing turbines – specifically, natural gas-fired units in South Korea and the United States. These turbines have been equipped with GT-ACO (Gas Turbine Autonomous Control Optimizer), a solution developed by the company’s Learning Systems research team that uses data from multiple smart sensors make precise and rapid adjustments to fuel valves.[7] GT-ACO users can use virtual reality goggles to view an image of the turbine that incorporates real-time data from the sensors.[8]
According to Volkmar Sterzing, the leader of the research team, the ability to monitor changes in the combustion process will help buyers reduce harmful emissions and optimize fuel consumption. “To ensure that a gas turbine runs optimally, you always have to search for a balance in which several undesired effects such as combustion dynamics, loss of efficiency and emissions are kept as low as possible,” he was quoted as saying in a company statement. “If you improve one variable, you will worsen a different one. Artificial intelligence knows how to find the sweet spot.”[9]
There are significant benefits for customers in this sweet spot. As Dr. Norbert Gaus, the head of research in digitalization and automation at the company’s corporate technology unit, noted: “Even after experts had done their best to optimize the turbine’s nitrous oxide emissions, our AI system was able to reduce emissions by an additional 10-15 %.”[10]
What’s more, improved quality won’t just be something for big corporate buyers. Machine learning has already enabled Amazon and other large retailers to reduce the time needed to deliver products to all customers.[11] Eventually, it could also allow manufacturers of clothing and shoes to turn out customized items at prices that are competitive with merchandise made on standardized production lines.[12]
Augmenting the outcome
Meanwhile, AI doesn’t just affect production and products. It can also allow manufacturers to expand the relationships they have with their customers beyond the point of sale.
One company that has successfully incorporated machine learning into its catalog is Cummins Power Generation, an Indiana-based manufacturer of power-generating equipment, including generators and prime and stand-by systems. The company teamed up with Microsoft and Avtex several years ago to develop a remote monitoring system that collects data from Cummins products around the world. This system, known as the Power Command Cloud, “connects to millions of Cummins generators around the world, providing greater visibility into how equipment is performing and enabling refueling and performance maintenance at the exact time to maximize uptime,” Microsoft reported in 2016.[13]
This machine learning solution helps Cummins’ customers by monitoring multiple components, alerting users to trouble, and working to minimize the length and frequency of outages. But it goes a step further: It also notifies authorized service technicians of problems (both potential and actual) and of service requirements when they arise. In effect, it allows Cummins equipment to initiate its own service calls, thereby streamlining the repair process and reducing the time needed to restore normal functioning.[14]
In the long run, machine learning is likely to have positive consequences for manufacturers that incorporate it into their products. It has the potential to create new streams of revenue by giving buyers an easy way to access expert technicians whenever their devices need service or repairs.[15] And on a more general level, it lays the foundation for extending the relationship between the producer and the consumer beyond the moment at which goods or services are sold. It gives manufacturers a reason to offer their products on terms that will allow them to continue collecting information from sensor devices and analyzing it after receipt.
Conclusion
Machine learning appears to have the potential to change relationships between producers and consumers in positive ways. It can help producers by giving them ways to streamline their operations, and it can help consumers by making better products available at reasonable prices. Additionally, it can provide both parties with reasons to continue working together even after they sign purchase agreements and execute deliveries.
Furthermore, it can pave the way for additional improvements. Using machine learning to keep up the relationship between producers and consumers gives the former the ability to feed data – real operational data, collected on factory floors and not in testing facilities – back into their AI systems. These systems can then use the data to gain a better understanding of production processes and generate new solutions.
About the Author
Gregory Miller is a writer with DO Supply who covers Robotics, Artificial Intelligence and Automation. When not writing, he enjoys hiking, rock climbing and opining about the virtues of coffee.
[1] https://www.technologyreview.com/s/601045/this-factory-robot-learns-a-new-job-overnight/
[2] https://www.technologyreview.com/s/601045/this-factory-robot-learns-a-new-job-overnight/, https://www.fanuc.com/product/robot.html
[3] https://www.technologyreview.com/s/601045/this-factory-robot-learns-a-new-job-overnight/
[4] https://emerj.com/ai-sector-overviews/machine-learning-in-manufacturing/
[5] https://www.ibm.com/blogs/internet-of-things/iot-worker-insights-worker-safety-and-ai/
[6] http://ai.business/2017/03/29/the-role-of-machine-learning-in-construction-safety/
[8] http://www.ulliwaltinger.de/deep-learning-meets-virtual-reality-how-to-step-inside-a-gas-turbine/
[10] https://emerj.com/ai-sector-overviews/machine-learning-in-manufacturing/
[14] https://www.industrialiotseries.com/2018/07/25/predictive-maintenance/ , https://blogs.microsoft.com/iot/2016/11/29/cummins-power-generation-keeps-the-lights-on-with-azure-iot-suite/
[15] https://www.sas.com/en_us/insights/articles/big-data/3-internet-of-things-examples.html
Sign up for the free insideAI News newsletter.
Nice post about machine learning has significant potential for manufacturing. I think your post very helpful for more people. Thanks for sharing the information.