(302v) Fuzzy Steady State Decomposition Based Multi Model Control of Nonlinear Processes Applied to pH Control

In chemical industry many processes are operated at the verge of instability or at a sensitive condition where a small perturbation leads to a phenomenal change in the process behavior, for optimal production objectives. This proves detrimental and may lead to the failure of control loop when the operating conditions slowly drifts to a neighboring region where the process behaves in an entirely different manner. A single linear controller will definitely fail in controlling such processes.

Solutions proposed in the literature for this challenging problem can be broadly classified into two, a nonlinear model based controller or a multi model controller. Developing a nonlinear model for the entire domain of operation is a tough task and hence this approach has some limitations. The later one is based on a steady state decomposition of the operating region into different linear regions. Different controllers are employed for each linear region, and an appropriate manipulative input move is determined by the current operating region. Exact decomposition of the steady state into multiple linear regions is done by ignoring the effect of ever changing process behavior (as time goes the steady state mapping of a process may change as in pH neutralization processes) and external disturbances. Another draw back is approximating the nonlinear mapping may require many linear regions. These problems can be surpassed by decomposing the operating space into overlapping linear regions. This can be done by fuzzy c-means clustering on the steady state data. This also drastically reduces the number of linear regions to approximate the nonlinear steady state mapping, whatever the degree of nonlinearity is. An integral action is introduced into the control algorithm which nullifies the steady state offset due to mismatch between the plant and local model.

In this work we applied fuzzy steady state decomposition based multi model control to a waste water neutralization process characterized by both steady state as well as dynamic nonlinearities. A new sensitivity metric is used to find the belonging of current operating region to the previously identified cluster and further control implementation.