Crystal clear

Can I trust the measurement? That’s a question that operating personnel at water treatment plants constantly ask themselves. “The people in the control room must keep a constant eye on all the graphs tracking things like turbidity, pH and conductivity,” says Lars Bondzio, business development manager at the Endress+Hauser product center for liquid analysis. “From experience they can decide whether measurements for a given application are within the usual range.” Outliers or other anomalies could point to problems with the process – or to deposit formation, sensor drift or aging.

“The sheer number of measurements they have to assess can be overwhelming,” says Julia Mildner, head of the service innovation group. Then there is the shortage of skilled and experienced workers, which means that shifts are often understaffed. This is where Endress+Hauser aims to help out with an intelligent assistance system. It uses machine learning algorithms to model the plant operators’ knowledge and automatically check the plausibility of each measurement. To do this, the AI also pulls in historical measurements, diagnostic and verification information from the sensors, and environmental data. “For example, the algorithms can ascertain whether high turbidity levels are the result of heavy rain, meaning everything’s okay, or whether action is required,” Mildner explains.