Machine Learning in manufacturing!

Major companies such as GE, Siemens, Intel, Funac, Kuka, Bosch, NVIDIA and Microsoft are all making substantial investments in machine learning approaches to enhance all aspects of development. The technology is used to minimize labor costs, decrease product errors , reduce unplanned down times, increase transition times, and increase the speed of production.

According to TrendForce, so-called ‘smart manufacturing’ (approximately industrial IoT and AI) is expected to increase noticeably in 3 to 5 years.

This article will concentrate on the use of cutting-edge AI by three of the leading companies in the manufacturing world to make interesting changes to factories and robotics. It will concentrate on two key themes:

*The numerous ways in which machine learning is currently used in production are

*What outcomes the innovations are delivering for the businesses highlighted (case studies, etc.)

Most big companies that produce machine learning software for manufacturing often use the same software in their own manufacturing, based on what our research indicates. For all of those developments, this makes them the inventor, the test case and the first customers. This is a pattern that we’ve also seen in other trends in industrial business intelligence.

For smaller manufacturers, this same in-house AI production approach may not be feasible, but it seems to be both feasible and (in many cases) favored for giants such as GE and Siemens to negotiate with outside suppliers. In any case, the examples below will prove to be useful representative examples of AI in the manufacturing sector.

SIEMENS

For decades, the German conglomerate Siemens has been using neural networks to track its steel plants and increase production. The company maintains that this realistic experience has given it a leg in designing AI for manufacturing and industrial applications. In addition , the company claims to have invested about $10 billion in US tech companies (through acquisitions) over the last decade.

In March 2016, Siemens released Mindsphere (in beta), the key competitor of GE’s Predix offering. Mindsphere — defined by Siemens as a smart cloud for industry — enables machine manufacturers to track machine fleets for service purposes around the world. IBM’s Watson Analytics also merged into the tools provided by their service at the end of 2016.

Like GE, Siemens aims to track, document and evaluate everything from design to execution in manufacturing to identify challenges and solutions that people might not even know about. The German Government referred to this general dynamic of “Industry 4.0.”

AI’s success storey Siemens regularly highlights how the emissions of particular gas turbines have improved more than any person could have. “While experts have done their best to optimize the emission of nitrous oxides from the turbine,” says Dr. Norbert Gaus, Head of Digitalization and Automation Research at Siemens Corporate Technology, “our AI system has been able to reduce emissions by an additional 10 to 15 per cent.”

Siemens’ new gas turbines have over 500 continuous temperature, strain, stress and other variable sensors. All of this information is fed to their neural network-based AI. Siemens argues that their device is learning how to constantly change fuel valves to establish optimum combustion conditions on the basis of real environmental conditions and the current state of the equipment. More combustion results in less undesirable by-products.

One of the many ways that Siemens sees that their technologies can ultimately be used is with a product called Click2Make, a production-as-a-service technologies. Through companies with a thorough understanding of all available resources and highly adaptable robots, the goal is ultimately to make mass customization goods possible.

How it will work is that a company determines that they want to create a particular limit run item, such as a special coffee table. The company would apply its design and the system would automatically begin a bidding process between the facilities that have the equipment and the time to process the order. It will allow suppliers to draw up production plans automatically and sell them to potential customers in real time. The target is a fast turn from concept to execution.

GE

General Electric is the 31st largest corporation in the world in terms of sales and one of the largest and most diverse manufacturers on the planet, producing anything from massive industrial machinery to home appliances. It has more than 500 factories all over the world and has just begun to turn them into smart facilities.

In 2015, GE released its Brilliant Manufacturing Suite for Consumers, which had been field testing in its own factories. The system takes a systematic approach to monitoring and processing all in the production process to identify potential problems before they occur and to detect inefficiencies. The first “Brilliant Factory” was built that year in Pune , India, with an investment of $200 million. GE claims to have increased the performance of the equipment at this facility by 18%.

With that data, the Predix deep learning capabilities can spot potential problems and possible solutions. GE spent around $1 billion developing the system, and by 2020 GE expects Predix to process one million terabytes of data per day.

GE now has seven Brilliant Factories, powered by their Predix system, that serve as test cases. It claims positive improvements at each. For example, according to GE their system result in, their wind generator factory in Vietnam increasing productivity by 5 percent and its jet engine factory in Muskegon had a 25 percent better on-time delivery rate. They claim it has also cut unplanned downtime by 10–20 percent by equipping machines with smart sensors to detect wear.

FANUC

Although GE and Siemens are heavily focused on applying AI to build a holistic manufacturing process, other companies specialising in industrial robotics are concentrating on making robots smarter.

Fanuc, a Japanese company that is a pioneer in industrial robotics, has recently made a strong push for greater communication and AI use within their equipment. In 2015, Fanuc purchased a 6 per cent stake in the AI Start-up Chosen Network for $7.3 million to incorporate deep learning into its robots.

A partnership with Cisco and Rockwell Automation to build and deploy FIELD (FANUC Intelligent Edge Connection and Drive) was announced in early 2016. It is defined as an industrial internet network for the production of goods. Only a few months later, Fanuc joined forces with NVIDIA to use their AI chips for their “Factory of the Future.”

Fanuc uses deep reinforcement learning to help some of its industrial robots train themselves. They perform the same task over and over again, improving each time until they have achieved adequate accuracy. By partnering with NVIDIA, the aim is to allow multiple robots to learn together. The idea is that what could take eight hours for a robot to learn, eight robots will learn in one hour. Fast learning means less downtime and the opportunity to manage more varied goods in the same factory.

While humans had to initially program any particular action taken by an industrial robot, we gradually created robots that could learn for themselves. More and more robots will be able to transfer their skills and learn together in the future. Robot systems with relatively repetitive tasks (fast food robots being a good candidate) are low-hanging fruits for this kind of transfer learning.

CONCLUSION

Automation, automation and complex analytics have been used by the manufacturing industry for many years. For decades, whole organizations and research fields have been looking at manufacturing data to find ways to minimize waste and increase performance. Manufacturing is now a fairly streamlined and technically advanced industry.

As a result — unlike some industries (such as taxi services) where the introduction of more advanced AI is likely to cause major disruption — the near-term application of new AI technology in the manufacturing sector is more likely to look like evolution than revolution.

Greater industrial networking, more widely distributed sensors, more efficient analytics, and improved robotics are all capable of constraining noticeable but modest changes in performance or versatility.

These newer implementations of machine learning have resulted in relatively small reductions in equipment failures, increased on-time delivery, minor improvements in equipment and quicker training times in the dynamic world of industrial robotics. These changes can seem small, but when added together and distributed over such a wide sector, the overall potential savings is substantial. According to the UN, the global value added by manufacturing (net production of manufacturing after subtraction of intermediate inputs) amounted to $11.6 trillion in 2015. That’s why businesses are spending billions on improving AI resources to squeeze a few extra percentage points out of various factories.

In the longer term, the complete digital integration and the advanced automation of the entire design and development process might open up some fascinating possibilities. Customization is uncommon and costly, whereas high-volume mass-produced goods are the dominant model of manufacturing, as the expense of redesigning a production line for new products is always excessive.

Consumers have, for the most part, been able to trade because mass-produced goods are so much cheaper. If the technology that makes manufacturing more versatile is widely introduced, allowing customization to become cheap enough, it could lead to a real shift in many markets. Instead of most shoes coming in a dozen sizes, they could be made in an infinite number of sizes — each order custom-fitted, designed, and delivered within hours of the order being placed.