My role at Siemens Corporate Technology is to envision new ways to use artificial intelligence (AI) and graph learning to help automation engineers do their jobs faster and better, and with fewer errors. Graph learning is a form of deep learning that performs analytics on network data. To get an idea of what graph learning is, think of a social network, where you have potentially billions of relational data points. The network keeps creating newer and newer connections—that is, more and more structure that is not always visible or completely understood. Graph learning seeks to identify those patterns and the trends in the data in order to learn the properties of networks through their structure.
Automation engineering is the task of integrating, via software, various sensors, actuators, and controls for automating a real-world process, such as a factory. With so much complexity in automation engineering, graph learning is becoming essential to the process. Automation engineers are supported by a suite of software tools including integrated development environments (IDE), hardware configurators, compilers, and runtimes. These tools, however, focus on the automation code itself, leaving the automation engineer unassisted in decision making. That, in turn, can lead to longer time requirements for software development and commissioning because of inherent imperfections in the decision-making process.
I’m not saying automation engineers can’t make decisions quickly or correctly. The fact of the matter is that so much data is involved in the automation process that sheer human cognitive power is not enough.
That’s why over the last couple of years I have been focusing on Cognitive Engineering. That is, the use of AI to assist automation engineers during their decision making. One example of using AI is classifying automation engineering patterns for better reusability, thus making repeatability of valuable processes much easier. Another example of using AI is to find similar automation code snippets to assist the software development process; i.e., grab useable pieces of code from all over the data set. Also, AI is used for reasoning about the hardware selection of sensors, actuators, and other devices. All of these instances involve huge amounts of data.
Giant network data sets are the reason that graph learning is core to the Cognitive Engineering concept. Automation projects are very large physically and astronomically large digitally. Imagine all the equipment, sensors, actuators, robots, and software that engineers have to think about, obtain, organize, and install to build a factory. The process of handling all that generates data that has an inherent high-dimensional structure. The data can be mined by graph learning algorithms to discover known and unknown patterns in the data. Graph learning can further be used to anticipate the automation of engineers’ decisions and provide time-saving recommendations for their next action, classify source code, and recommend better hardware configurations.
Graph learning also allows a deep look into old data that never was mined but most certainly contains important discoveries. Say a team of engineers built a factory ten years ago, and by now a lot of the collected data is sitting untouched. Maybe no one on the team has the time now to go back and look at the coding. But AI-powered graph learning could mine all of that information, and that will be a huge help for building the next factory, because the graph learning will recognize what functions were done well and be able to show the optimal way to do it now based on everything learned from the previous factory.
The AI behind Cognitive Engineering, however, is not making any decisions—it’s making suggestions. Cognitive engineering puts AI and humans working side-by-side, humans make the decisions, after graph learning has done a gigantic amount of cognitive work—more work than an army of engineers could have done over several years.
As the Industrial Internet of Things (IIoT) expands, so will the necessary graph learning, because more and more digital relationships will be built between all those things, giving rise to enormous data sets. Siemens is pushing for digitalization in cities, buildings, and manufacturing processes, and all of these benefit from the development of AI, machine learning, and graph learning. The more the IIoT expands, the more people are looking to AI, and the more its use will accelerate.
Editor’s Note: Arquimedes Canedo is a Principal Key Expert Scientist at Siemens Corporate Technology where he drives the focus area of Automation Engineering. In 2017, he was awarded the Siemens Inventor of the Year. He has more than 60 technical publications in ACM/IEEE conferences and journals and holds more than 70 patents.