CEE_20_02

Page 10

ARTIFICIAL INTELLIGENCE

AI at the edge has a small footprint At a recent Festo press event, Tanja Maaß, managing director at Resolto Informatik, spoke about the benefits of bringing artificial intelligence directly to the machine. Suzanne Gill reports.

W

hile artificial Intelligence (AI) is starting to play a much bigger role in industry, Maaß was quick to point out that the requirements of the machine builder (OEM) and end-user will be very different. She said: “For the end-user, AI is probably most often associated with predictive maintenance applications. But it could have many more uses – helping speed up production lines or to reduce waste, for example. “End users could also use AI technology to help as part of an intelligent energy optimisation solution and for peak load forecasting. Early anomaly detection is another good application for AI. Using an algorithm in place of a programme will result in a system that is able to learn through experience to identify anomalies in a process.” In the machine building sector AI can be utilised to create service offerings. Maaß explained: “As margins on hardware offerings reduce, having the ability to offer a service would give OEMs a valuable new repeating income stream and the use of AI will help

detect anomalies to lower the risk of breaking a contract through machine failure.” Taking this thinking to the next stage, Maaß pointed out that if it was possible to offer a 100% guarantee that a machine will always be available it could make the product-as-a-service model more attractive to end-users. Achieving 100% guaranteed machine availability requires real-time data and Maaß pointed out that any data analysed in the cloud will only ever answer questions retrospectively. “Even if you are monitoring data continuously in the cloud it will still always have a slight latency to it,” she said. “Historical data can only ever answer questions about an event that has already happened. “Live data, monitored close to the field, however, allows us to predict what will happen. These real-time machine learning solutions are able to learn healthy behaviour patterns and recognise emerging unhealthy behaviour patterns. With an AI algorithm it should be possible to not only ask for recommendations to prevent potential events, but to ask it to act automatically to avoid an event.”

SCRAITEC analytics work continuously and in real-time.

10

February 2020

www.controlengeurope.com

The SCRAITEC platform from Resolto Informatik is able to collect data, analyse it, optimise it and then present the solution in some way. Since the company was acquired by Festo several years ago, it has focused on applications that use SCRAITEC algorithms on Festo hardware, such as the CPX controller and the IoT gateway. In one such application, Resolto was approached to provide a predictive maintenance solution to allow an automotive company to predict wear and cycle time deterioration on pneumatic clamping units. This application quickly turned into an optimisation solution. “When we applied the data to SCRAITEC it identified an unexpected unhealthy behaviour pattern in one part which was found to be losing the company the equivalent of one car every day. SCRAITEC on a Festo CPX E CEC controller – with no cloud connectivity – has learned the normal operative state, regardless of the clamping unit type, and is now able to offer an early indication of anomalies which will lead to clamping unit failure. In another application, for a household electrical appliance company, Resolto was approached to find a solution to poor welds of washing machine drums. “We needed to look at data from the SAP system in addition to the plant-level data,” said Maaß. “This identified a problem linked to where the steel came from. The solution sees SCRAITEC cross-reference the origin of steel in a drum and automatically make the necessary heat and timing adjustments to the welding process for each drum.” In conclusion, Maaß pointed out that it often comes as a surprise to many that AI need not be a resource intensive application. She said: “Using a trained algorithm will have a very small footprint and requires minimal hardware for interpretation. If you have a device running close to a machine and it is well trained and has local feedback, there is probably no need to ever connect it to a central system.” ! Control Engineering Europe


Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.