Integration of Optimization to the Design of Pulp and Paper Production Processes - Simulation and Modeling Methodologies, Technologies and Application - page 239

Simplified, on the operational level we optimize the variances of the quality
attributes of the paper and the broke dosage and the probability of under/overflows.
On the design level, we optimize the design according to the sizes of the tanks and
expectation values of the system performance.
Model Building. At the same time that the optimizer negotiated with the designer
about the problem formulation, he had to discuss with the model builder if a suitable
model for the problem can be built. In this discussion there were two main themes: is
the physical phenomenon of the problem known or is there enough data to model the
problem stochastically and can the model be simple enough that it can be calculated
fast enough in the optimization loop.
Optimization and Result Interpretation. In this case example, the tasks of problem
formulation, model building and optimization were performed simultaneously and
were highly iterative. The main focus of the case example was in optimization. The
results of the optimization are described in [30-31].
After the optimization, the results were presented to the designer as two-
dimensional Pareto optimal sets. In Fig.4, a Pareto optimal set in respect to the two
most important parameters is presented. The designer then made the decisions e.g.
between a decent investment cost and an acceptable probability of break.
Data, Knowledge and Models. The designer in this case had a wide experience in
process design, paper making, modeling and optimization. The optimizer was mathe-
matically oriented, but had only minor experience on paper making or process design.
The modeler was familiar with process modeling and optimization.
The largest data flow in the process was from designer to optimizer. The designer
had to communicate the customer requirements, the original design about the struc-
ture and operation and the freedoms and limitations for optimization in the design.
The main models for this communication were a process flow sheet and steady-state
model of the process. Making of these models was mainly a task for the designer. The
designer was able to formulate most of the limitations and requirements in numerical
form, e.g. the probability of the break may not be greater than Pmax. Due to the na-
ture of a first time project, the data transfer between the optimizer and modeler was
also huge.
Modeller was responsible for building three models: dynamic model, predictive
model and a validation model. The two first mentioned were used in optimization
while the validation model build with different simulation software was used only for
one selected design.
Practically, the problem formulation and optimization required simultaneous model
development, because there wasn't previous knowledge about feasible models.
The results of the optimization were delivered as a document containing simulation
graphs and Pareto optimal sets (one example in Figure 4) of optimization results.
Information Tools. This case example was carried out as a research project, and
therefore the engineering tools used didn't match the ones used in industry. MATLAB
was used both for the optimization and simulation for optimization. APROS process
simulator was used in validating the results of optimization.