Artificial intelligence senses, learns, reasons, acts, and adapts to the real world without explicit programming. These abilities make it an increasingly important contributor to the development of Industry 4.0. In fact, those contributions are happening right now in manufacturing, with applications such as predictive maintenance, collaborative and context-aware robots, and automated quality testing.
The advent of artificial intelligence (AI) goes back more than a half-century; it has ebbed and flowed in terms of expectations, spiking particularly high in the 1980s, followed by “the winter of artificial intelligence” when forecasts for its rapid advance failed to be realized. Today the ebbing has resolved into a powerful torrent of promise with recent successes that have convinced both research and financial communities that it will be truly transformative. The current development may lead not only to smart machines that will take over tasks currently done by humans, but also to new applications and technologies.
This trend is not only exciting; it’s necessary. A McKinsey study notes that highly developed economies with a high GDP per capita will increasingly rely on automation based on AI to achieve growth targets because “automation fueled by AI is one of the most significant sources of productivity.”1
This finding supports the notion of some experts that AI will not only transform industries such as manufacturing, but also facilitate a paradigm shift in macroeconomics. In this view, how nations manage the opportunities emerging from technologies based on AI will be the key factor in economic growth, eclipsing both capital and labor. Data is the basis of this belief—data as a resource like oil or gas.
What Is AI?
“AI is often not represented correctly,” asserts Heiko Claussen, principal key expert at Siemens. “It’s often used as synonym to machine learning, which is not the case. Here’s a definition of AI that represents it well: AI senses, learns, reasons, acts, and adapts to the real world without explicit programming.”
That definition makes it clear where AI fits in the scope of the digital factory, because AI enables complex applications in varying environments with flexible rather than hard coded low level programming. This quality is important for its role in Industry 4.0, where different machines have to interact with and adapt to each other and the devices they are building, understanding the situation they are in and how they can contribute to the next manufacturing level.
According to Claussen, AI will enable totally new applications going forward. “Today, it is generally cost prohibitive to specifically design and produce a unique part for a particular customer. AI will help to bring this about.”
Claussen says that AI has started to enable co-working between humans and machines, which is not yet very common. “While humans generate a lot of uncertainty in the domain, if a machine is intelligent and adaptable to a changing environment, it can cope with that uncertainty, as well as collaborate with humans and use our tools.”
Trends and Applications
There are three major trends affecting the future landscape of AI applications: open source machine learning frameworks, hardware accelerators for AI, and novel machine learning approaches.
- Open Source Machine Learning Frameworks
Rather than developing proprietary machine learning toolboxes there is a surge in open source frameworks such as TensorFlow, Caffe, Chainer etc., that make it easy for startups and other innovators to prototype novel AI applications. Established algorithms, e.g., AlexNet for image classification, can simply be downloaded and applied in different context.
- Hardware Accelerators for AI
An increasing number of dedicated hardware accelerators for machine learning tasks are emerging to fulfill the demand of AI applications. General CPUs are not optimized for this specific workload and are now facing competition from GPUs, FPGAs, and dedicated ASICs. Prominent examples are Nvidia’s Volta, Intel’s Lake Crest and Google’s Tensor processing unit.
- Novel Machine Learning Approaches
Since the success of deep convolutional neural networks on image classification tasks, deep learning is dominating many machine learning applications. Deep learning refers to learning by multilayered artificial neural networks that are modeled on the human nervous system. State-of-the-art algorithms surpass human accuracy in image classification. A related approach that is attracting attention is deep reinforcement learning. This method optimizes a reward based on feedback from a system it is interacting with. This allows machines to compete and learn from each other and thus rapidly improve their performance. This method has been applied by Googles AlphaGo that managed to win against 18-time world champion Lee Se-dol.
In manufacturing, AI will be used for a host of applications, including predictive maintenance, collaborative and context-aware machines, and process optimization. Other areas where AI is likely to be prominent or is already in use include software programming, connected e-mobility (e.g., autonomous vehicles), and medical diagnostics.
“Siemens is a very strong player in AI and has been for a long time,” notes Claussen. “We have many products that utilize AI across a host of industries.”
Siemens researchers were among the first in the 1990s to harness artificial neural networks for innovative solutions such as the optimization of energy-intensive steel plants. Today, roughly 200 experts at Siemens Corporate Technology (CT) are occupied with data analysis and neural networks as the company’s AI research continues to be a high priority pursuit.
Case in Point: Gas Turbine Autonomous Control Optimizer (GT-ACO)
Siemens has made significant progress in applying neural network technology to artificial intelligence. For example, the company’s Software Environment for Neural Networks (SENN) is being continuously refined and adapted to new and evolving applications, including the optimization of gas and wind turbines.
Siemens Power Generation Services and Siemens Corporate Technology have jointly developed a system that continuously optimizes the operation and control of combustion in gas turbines. Based on AI from CT, the system, which is known as Gas Turbine Autonomous Control Optimizer (GT-ACO), has been installed at a Siemens H class gas turbine of a key customer in Asia. It has been tested over a year now in close cooperation with the customer to gain experience over all seasons, as ambient conditions affect operation. Tests are close to completion, up to now with positive results. The powerful H class gas turbine is the flagship of the Siemens fleet, based on the most advanced technology.
Improvements in overall gas turbine operation are difficult to achieve because lower emissions characteristically result in shorter service life. At low emissions high-energy combustion oscillations can suddenly occur, causing material fatigue and increased wear.
Tests on a number of gas turbine types have demonstrated that GT-ACO works. At the first test an expert set the turbine manually to minimum emission of nitrogen oxides, then artificial intelligence took over control of the combustion unit. “Two minutes after switching on, the value had dropped by 20 percent and combustion dynamics did still stay within the limits” reports Hans-Gerd Brummel, who is responsible for GT-ACO development at Power Generation Services and is a pioneer in turbine remote diagnostics and maintenance at Siemens.
The primary goal of using AI in turbines up to now is to minimize emission of nitrogen oxides without any danger of suddenly occurring high combustion oscillations. To achieve this, GT-ACO’s neural model alters the distribution of fuel in a turbine’s burners. However, these settings vary depending on factors such as location, gas composition, and local weather conditions. Consequently, GT-ACO needs a few weeks of learning on each turbine before it can autonomously make beneficial changes to the controls. GT-ACO can also be used to partially compensate for gas turbine ageing, because the technology combines the collective knowledge of gas turbine thermodynamics in the form of physical models with machine learning.
“Customers are showing great interest in our new technology,” says Brummel. “Given the high proportion of renewable energy in the supply grid now, gas turbines often have to step in to stabilize the grid frequency.” In this constantly changing operating environment, there is growing risk of increased oscillation amplitudes increasing wear. Brummel is confident, however, that “GT-ACO can help in this case by focusing the optimization effect on damping vibrations.”Have an Inquiry for Siemens about this article? Click Here >>