Risk-based Steering of New Process Technology Development

For over 25 years we have had good experience with black box models, in our case special hybrids of artificial neural networks (ANN) and CG algorithms. The approach provides us with a very effective method to model the actual behavior of a process. Unrealistic assumptions and the search for an explanation why the theory is not confirmed in practice in this particular case are omitted. And above all, it is possible to investigate connections for which there is no usable theoretical approach yet.

Here are a few examples:

- When it came to the basis weight of recycled paper, our model was the fastest way to create a soft sensor.

- In multi-stage polymerization with extrusion, we predicted material properties such as notched impact strength with very high accuracy and were able to optimize the formulation from the first polymerization stage to the additives on the extruder in order to find the most cost-effective way to achieve the desired impact strength. Without the data-driven analysis, it would have been necessary to first understand the relationship between molecule structure and notched bar impact strength and then the relationship between formulation, process parameters and molecule structure.

- For a similar process, there was a viscosity soft sensor for the polymer based on a fairly good theoretical approach. After several years, however, he did not deliver correct values and no expert could be found who knew the model well enough and could adapt the calculation. The soft sensor was then replaced by a purely empirical model, which is constantly updated automatically with current data.

- During a combustion process, we were able to produce good NOx models even though the fuel consisted of a mixture of substances that was difficult to describe and unknown amounts of air flowed into the combustion chamber. We have made it possible to improve the efficiency and durability of the boiler through optimal air supply.

The producing Industry creates new valuable Products from raw materials, beside it generates a lot of Data. These Data combined with Process Knowledge and Laboratory Data give us the possibility to improve our process efficient to new limits.

The Grey-Box Approach provides us with a very effective and efficient method to improve a process or a process step.