A predictive model controller is used to control a drinking water chlorination process.
Predictive model controllers are commonly used on DCS platforms and as supervisory control systems. Recently, two single loop predictive model controllers with model-based feedforward compensation were implemented using a standard function block library available with the ControlLogix system from Rockwell Automation. This controller significantly reduces process variation when compared with PID or similar common controllers, providing engineers with significant insight into the process and the control approach and allowing for easy startup, debugging and maintenance.
Predictive controllers have for the past few decades been serving as a cornerstone of 'advanced' or 'supervisory' control. A predictive model controller uses a process model that is built to predict future values of controlled variables based on current controller output and feedforward signals. The controller output, or manipulated variable, is typically calculated such that the controlled variable follows a desired trajectory.
The layout of the instrumentation and the chemistry of the process require implementation of predictive model control to minimise residual chlorine content in drinking water delivered to the customers. Generally, different process gains during high and low flow conditions and varying process deadtime require controller gain scheduling and on-line adjustment of model deadtime.
In the remainder of this article, basic theoretical elements of predictive model control, implementation strategy in ControlLogix system, the chlorination control architecture, objectives and the actual controller implementation will be discussed.
Predictive model control
The fundamental idea behind predictive model control is to use information generated by a model of the controlled process to control the controlled variables as close to the target or set point or to control the entire process to a specific objective other that just a set point. The type of model that is used most frequently is a dynamic model. Such a model describes the process behaviour over a period of time and predicts values of the controlled variable based on current and past values of the manipulated variables and feedforward signals. If more than one model, or relationship, between one pair of analog values is used, then we refer to such a model and resulting controller as a multivariable model or predictive model controller, respectively.
Controlled variable variation is reduced by a multivariable predictive model controller which takes into account changes in MV and DV using the process model.
First order lag with deadtime process model is a transfer function-based model with three parameters:
* Process gain = steady state magnitude of change of CV as a result of unit change of MV.
* Process deadtime = elapsed time from MV step change to the initial reaction of CV to the change.
* Process time constant = measurement of how fast CV will approach the steady state after the initial deadtime period. The CV reaches about 63,2% of its steady state value in a period of time equal to the time constant.
The significance of the model parameters, the Process Gain = 0.6, time constant = 30 seconds and deadtime = 50 seconds is shown in Figure 2.
The internal model control, or IMC, block diagram shows how the process model is used in the control algorithm.
The complete multivariable predictive model controller is shown in Figure 5.
It is not the purpose of this article to show all the design details and justifications for the configuration and the choice of models. This configuration and model types were chosen to fit the control application.
The predictive model control configuration as shown in Figures 3 and 5 does not provide an anti-windup feature in case the manipulated variable calculated by the controller exceeds minimum or maximum limits imposed by the HLL block and ultimately the instrument, or actuator, which in this case is a vacuum driven chlorinator. To address this issue, use is made of the Model State Feedback configuration of the predictive model controller:
The diagram shown in Figure 6 represents identical control algorithm when the MV is within bounds set by the HLL function. If the MV calculated by the controller is greater than the maximum limit or less than the minimum limit the model state feedback configuration updates the internal model states and model predictions according to the value of the output actually sent to the process, ie, the maximum or minimum limit, whichever applies. When the calculated MV comes out of saturation the correct value of the manipulated variable is immediately calculated.
The feedforward controller FF is implemented using the realisable ratio of models between DV - CV model and MV - CV model. The FM portion of the feedforward controller compensates for rejection of the same disturbance twice: once by the feedforward block FF and again by the feedback controller, in case the deadtime of the feedback signal is longer than the deadtime of the feedforward signal. The final configuration for one predictive model feedback controller with predictive model feedforward signal is shown in Figure 7.
Implementing MPC Using ControlLogix function blocks
ControlLogix from Rockwell Automation is a PLC-based processor system which allows programming in ladder code and function block diagrams. Therefore, sometimes, the system is referred to as a 'hybrid' system. Unlike many other PLC systems ControlLogix allows for deterministic, ie, fixed timing, execution of ladder subroutines and the function block diagrams.
A ControlLogix program can be written as, for example, a ladder code executing sequential logic, like communication watchdog, parameter setting and calculations, mode switching, initialisation, fault handling, etc, while the function block diagram may be called by the ladder routine at some point of its execution and executed as part of the overall program. The entire program then may represent a multivariable predictive model controller, for example, with appropriate timing, bumpless mode switching, control algorithm execution and other necessary features ensuring correct and robust operation in realtime. The system maintains and updates all values, including model predictions in a tag data table which is possible to monitor in realtime or log using a data collection system.
The key function blocks used to generate the multivariable predictive model controller were: Lead/Lag (LDLG), Deadtime (DEDT), Selector (SEL), Gain multiplication (MUL), Addition (ADD), Subtraction (SUB) and High/Low limit (HLL).
Water chlorination control
The measured variables are water flow and residual chlorine. The only manipulated variable is the rate of chlorine addition into the water flow. This is the last treatment before water reaches the customer, ie, private homes and businesses. Maintaining the residual chlorine set point is of essential importance in protecting public health. The typical set point range for residual chlorine is 1,5-2,5 ppm depending on many factors, one of them being the projected chlorine loss which depends on the distance to the most remote residences supplied with water from the chlorination station and the climate. This ensures that even in the most distant homes chlorine content does not drop below 0,5 ppm and therefore ensures water safety. Residual chlorine content variation below 1 ppm and above 5 ppm causes risk to public health. Too much chlorine is unnecessary from the treatment perspective and may cause allergic reaction or eye irritation in short term and possibly cancer risks with long term exposure to high chlorine content in drinking water. Too little chlorine means risk of micro-organisms surviving, with some of them able to cause various kinds of human diseases.
The main control objective is to minimise variation of residual chlorine in drinking water. Consequently, the residual chlorine set point may be reduced to a minimum value, where enough chlorine is injected to provide safe water and overdosing is prevented. Economical savings from using less chlorine are also significant.
For more information contact Jeff Sandison, Rockwell Automation, 011 654 9700, [email protected]
| Tel: | +27 11 654 9700 |
| Email: | [email protected] |
| www: | www.rockwellautomation.co.za |
| Articles: | More information and articles about Rockwell Automation |
© Technews Publishing (Pty) Ltd | All Rights Reserved
printer friendly version