Procex Pty Ltd: Neural Networks

Neural networks which "learn" your process can predict product quality for those properties where there are no on-line sensors. The neural network models can then be used to optimize your process. This new technology has matured and packages are now available for industrial processes.

Being able to continuously predict quality delivers the following benefits:

less off-spec product
early warning of deviations
greater output
faster changeover times
manufacture closer to quality limits
reduce costs
reduced testing

Neural networks can recognize and match process conditions to QA lab results. The relationships that are identified are updated over time as more process information and QA results are gathered. Thus, processes with variability of raw materials or other process characteristics can learn and adapt.

The accuracy of the predictions is such that the loop can be closed to give real-time process control of the predicted quality parameters.

In fact, any calculated variable can also be used to allow optimization of costs, for example.

Procex has successfully created neural network models to predict:

Strength properties of linerboard
Curl in white papers
Bleach plant pulp brightness
Pulp mill Kappa Number
TRS emissions from a lime kiln
TRS emissions from a recovery boiler
Final turbidity in a water treatment plant

The Models can be used to:

Investigate relationships between inputs and outputs
Optimize the quality and/or cost of production.

If you are interested in pursuing this application of modern technology, then Contact Procex!