Implementation of an Optimization Module into a Simulation Software Package, 2001 Engineering / Finishing & Converting Conference Proceedings

L. Laperriere, L. Wasik--For the past few years, the pulp and paper industry has faced the difficult challenge of doing more with less, and commercial simulation has become a valuable tool for this purpose. Doing iomorel. often means optimizing the process, which in turn means finding combinations of process parameter-values that yield optimum measures of process performance, with the least sensitivity to process disturbances. In the context of most current commercial simulators, this view of optimization presents two important problems to be solved. First, the category of process variables often used as qualitative metrics in process performance evaluation and optimization are usually absent from commercial simulator models, i.e. paper strength or pulp color do not lend themselves to mass and energy balances. Second, optimization is often performed manually by iyplayinglo with the simulator using trial and error variables combinations, i.e. most simulators do not make use of mathematically robust optimization procedures that search for the optimal combination of process variables automatically and systematically. In a previous article we have presented a potential solution to the first problem sited herein by developing and implementing a neural network-based module that can perform independently of the classic heat and material balance of process variables upon which the model is based. In this paper, we tackle the second problem of optimizing some important process quality metrics appearing in new models by systematically searching the ihspacelg of process parameter values that yield their maximum or minimum. A simulated annealing process, which demonstrates the well-known downhill simplex method was used for this purpose and is described in this paper. An example simulation that uses a newly developed optimization module is presented and discussed.