ARTICLES

Download PDF

1   2    3    4

Multivariate Control

Some processes have multiple variables and consequently loops that must be managed to prevent them from conflicting with each other or alternatively requiring too much operator attention to be able to manage while keeping watch on the remainder of the operation. Multivariate controllers are developed based on a combination of engineering theory and knowledge of the process operations/thermodynamics but also through the use of models. Models are developed by collecting a large data set of operating data under a wide variety of steady state operating conditions in which a controlled disturbance is introduced to the process and the results monitored. Using this information, applications and controls engineers are able to develop algorithms or models to force the same desired response from the unit for any condition defined within the set of test conditions. One constraint of many model-based controls is that they must be used within the ‘boundary conditions’ under which they were developed, as in many cases they are not able to extrapolate beyond the ‘realm of knowledge’ used to create the model. This is especially true for models such as neural networks, expert systems and regression-based algorithms. Multivariate models are typically executed in the control computers in realtime, updating every minute or less to be able to respond on the same timescale as the process itself.

A typical use for multivariate control is on a distillation tower, where the change in feed affects the operation of the process, but also the amount of reflux, heat addition (reboiler and condensers), tray conditions (flooding, fouling), as well as temperature and pressure profiles needing to be managed. Multivariate control is also used for operations that interact with each other to prevent them from ‘competing’ against each other.

Optimisation

Being able to control one or more closely linked unit operations automatically makes it possible to minimise operating expenses to the average plant operating conditions. However, external factors frequently change. For example, the price of feed stock, fuel or product changes on a regular basis (on a different timescale to that of the process). This timescale is typically not the same realtime as a process that is a function of the process dynamics. However, many businesses want to be able to respond to the realtime environment of the business world. This is the realm of process optimisation.

Optimisation routines are often run in dedicated computers separate from the regulatory and multivariable processors that obtain their inputs from external sources such as the number of orders for plant product, inventory levels or other business inputs. These programs then provide set points or objectives to multiple unit operations or multiple processing units so that the overall facility is operating under the ideal conditions for the current business environment. The outputs from the optimisation routines are sent to the multivariate and regulatory controllers that then make the necessary modifications to the process.

1   2    3    4