Comment

1 August 2006

Smart bias for supercalendar paper machine mix tank

The mix tank for a paper machine is the last point in the process where you can stabilize the consistency of the stock and the relative proportions of the various constituents of the stock. In the case of the supercalendar machine at the Port Hawkesbury site in Nova Scotia, Canada, there were three fresh feeds to the mix tank: thermal mechanical pulp, kraft, and kaolin clay. There were two types of recycled feed: recovered fiber and broke. The operator defined the proportions of all the constituents except the recovered fiber, which you must process as you make it. 

The startup configuration included a proportional-integral (PI) gap controller for the mix tank. The idea with the gap controller was to have a low gain around the set-point and higher gain away from the set-point. The time constant for the tank was approximately five to six minutes, depending on the production rate. So there was not much time to react to any disturbance. The controller worked well for periods of time but would then begin a standing wave behavior, which could last for hours or even days. This cycling impacted the performance of the paper machine since cycling of the feed meant cycling of the consistency control, which meant cycling of the basis weight of the paper we produced. 

To solve the cycling problem, we introduced a proportional-only (P-only) level controller. We installed a high-level cutoff on the recovered fiber flow to avoid overflows at the mix tank. Because we installed a high shutoff for each inlet flow to the mix tank, it was possible to open up the proportional band to more than 50% (i.e.: reduce the gain of the controller) and get very good response. Widening the proportional band introduced an off-set unacceptable to the operations group, particularly for such an important tank.

To eliminate most of the off-set, we introduced a smart bias. Using over two years worth of data from past machine performance, we developed a correlation of mix tank controller output versus feed flow to the paper machine and implemented it as the smart bias generator. We had to heavily filter the feed signal before using it as the smart bias. The results were good; the mix tank controlled with little off-set. Because there were no more instances of sustained cycle, the level controller actually controlled to set-point much better with the P-only controller than it did with the PI controller.