Industry 4.0: Chips against data bottlenecks

IN Blue Yonder General — 02 February, 2015

The Internet of Things, with concepts such as the connected car, unlocks previously unimagined possibilities for both manufacturers and consumers. But there is one problem: How will gigantic data quantities from sensors be transported over mobile radio networks and analyzed in real time? The solution: a chip that decides which data is relevant using the most innovative algorithms, before the data ever reaches a backend computer.

 

Fpga_xilinx_spartan © Xilinx

 

Today, there are hundreds of sensors in a typical vehicle recording everything they possibly can. But only a tiny part of the captured data is relevant for each individual analysis or forecast. The earlier that data is sorted, the faster the data can be evaluated and the more efficient the big data application that uses it will be. That is because data transfer through today’s mobile radio network is costly and can prove to be a bottleneck for big data analytics.

200 million decisions per second

A solution for this hurdle has been developed at the Karlsruhe Institute of Technology (KIT) in Germany. Using the solution, data volumes can be dramatically reduced before they get to the IT systems that analyze it. This approach uses the NeuroBayes algorithm, upon which all Blue Yonder Predictive Applications are based.

Taking this example, the elementary particle detector Belle II, which is currently being built in Japan, will create so much sensor data that it will be practically impossible to select it all and move it into a huge data warehouses.  For this reason, intelligent decisions need to be made directly as the data is generated — at the sensors themselves — and only the data that is considered relevant by the algorithm is selected. To do this, a microchip was equipped with the NeuroBayes algorithm on it. This intelligent chip finds and classifies all relevant information at incredible speed: an individual chip makes 200 million decisions per second!

From a purely physical point of view, the NeuroBayes algorithm takes up hardly any space on the chip — occupying a mere 4% of the chip’s surface. For this reason, the researchers have developed a chip with 128 NeuroBayes units, which can make 32 billion decisions per second and be integrated into a normal computer through a standard interface.

An additional advantage: the chip can be reprogrammed and can adapt its decisions to new findings, including information gained from sensors in other machines, such as cars. This holds enormous potential for the Internet of Things.

Pure science is again leading the way in big data, as in the case of the NeuroBayes algorithm itself, which originally was developed for an experiment at CERN, the European Organization for Nuclear Research. And without the Internet of Things, humanity would be missing out on one of its greatest opportunities to improve the lives of billions of people around the globe and solve some of the most pressing challenges we face in the modern world.

Prof. Dr. Michael Feindt Prof. Dr. Michael Feindt

is the mind behind Blue Yonder. In the course of his many years of scientific research activity at CERN, he developed the NeuroBayes algorithm. Michael Feindt is a professor at the Karlsruhe Institute of Technology (KIT), Germany, and a lecturer at the Data Science Academy.