Machine Learning can often be used in place of complicated traditional mathematical methods in process industry applications. ML basically is automated statistics and statistics can be used for everything with the respect of that it is a slight difference on how to use it in different companies making copies impossible.

In this particular case we are talking about an advanced production environment, where we have applied machine learning for soft sensors and model based control where real sensors don’t work or break down. 

Purpose

Södra Cell Värö, one of the world’s most modern paper pulp mills, wanted to estimate future pH values in their process and use this for automated process control.

Data sources

Sensor data from production, including various flows, temperatures and pH measurements.

Video showing current PH and estimation 20 minutes in to the future together with uncertainty, added acid and base in the measured flows. 

Technologies

Gaussian processes, Recurrent neural networks, Model based control.

Outcome

Tenfifty’s ML experts have produced a solution that can predict the pH values in the process 20 minutes ahead, based on process dynamics and historical data. 

The result of the project will be used by Södra to build models of their process and improve monitoring and control, with reduced cost and environmental impact as end result.

Read more in this blogpost (in Swedish).

 

 

Author

Johannes Öhlin
Co-founder and Tech lead
Johannes Öhlin is co-founder of Tenfifty. He has a solid background within system development and architecture with a focus on artificial intelligence. As a programmer Johannes masters everything from analysis, requirements and design to deployment and he is also a very experienced technical project manager.