Advanced regulatory control seeks to incorporate knowledge of process dynamics, disturbances, constraints, and objectives to increase process efficiency and capacity. The PID power and flexibility enables an incredible spectrum of creative opportunities to achieve these goals. Here we look at techniques to preemptively deal with upsets and setpoint changes, protect against abnormal conditions, improve process linearity, compensate for dead time, eliminate limit cycles, reduce valve deadband, improve valve resolution and threshold sensitivity, extend valve rangeability, and optimize process operation. We also consider when to move up to model predictive control.
Perspective
The most common advanced control technique is feedforward control that is designed to preemptively correct for a disturbance. In industrial applications, feedforward control is far from perfect so feedback control is essential. Most of the problems with feedforward control stem from incorrect or missing dynamic compensation. Separate open loop tests should be conducted where the disturbance variable and manipulated variables are stepped. If a time constant plus dead time model of both responses is adequate, the dynamic compensation simplifies to a feedforward dead time and lead-lag block.
The feedforward dead time is the disturbance variable dead time minus the manipulated variable dead time. If this value is negative, dynamic compensation of dead time is not used and the feedforward gain needs to be reduced because the feedforward correction will be late and some feedback action will already have occurred.
The feedforward lag time is set equal to the disturbance variable time constant. The lead time is then set equal to the manipulated variable time constant. The ratio of manipulated variable to disturbance variable time constant is the lead-lag factor. If this factor is greater than one, there is a kick in the feedforward to compensate for the excessive slowness in the manipulated variable. If open loop tests cannot be done or higher order dynamics (e.g. secondary lag) alter the initial response or an integrating or runaway response occur, then an addition of lead or lag time to make the feedforward response faster or slower may be the simplest solution. This chapter will show how the closed loop oscillation for a disturbance can be used to decide what term to adjust and in what direction.
If the feedforward arrives too soon, an inverse response is created that is particularly disruptive. In this case, the feedforward can do more harm than good. Since timing cannot be perfect, a feedforward signal that is a late is better than one that is early especially considering PID controllers are typically tuned with a gain that is far below the maximum. Thus, the better solution is to make sure the feedforward signal can handle major disturbances with the largest dead time and slowest time constant by being liberal with the use of dead time and lag time in the manipulated variable path. Knowledge of how time constants and dead times change with production rate can be used to schedule the setting in the feedforward dynamic compensation.
Setpoint feedforward does not require dynamic compensation. Setpoint feedforward only offers a significant advantage if there is less than 50% of the final change needed as a step change in PID output for a setpoint change. This scenario can be caused by a structure of proportional on process variable (I on error and PD on PV), the product of the beta setpoint weight factor and PID gain less than 0.5 for a two degrees of freedom (2DOF) structure, or the product of the setpoint lead to lag ratio and PID gain less than 0.5 for a structure of proportional on error (e.g. PI on error and D on PV). These values are offered as a guideline only. Specific application conditions need to be considered.
The other major considerations to be addressed are disturbance measurement noise, error, and rangeability. Noise can be amplified by a feedforward lead time greater than the lag time. A threshold sensitivity limit may be a better solution than adding a filter in terms of screening out insignificant changes. If the rangeability of a flow measurement is insufficient, a flow calculated from valve position should be used for low flows.
Output tracking and the remote output mode enable the PID output to be scheduled to deal with abnormal situations. This function is called an open loop backup because the feedback control is temporarily suspended to guarantee a predetermined correction. Control is returned to the PID feedback control when a protective state is assured. The PID output can be set at a limit to achieve the fastest possible setpoint response without upsetting other loops by setpoint rate limiting of the manipulated flow. The PID output is then set to a final resting value (captured from last batch or startup) and control is returned to the PID when a process variable value one dead time into the future is projected to reach the setpoint.
Intelligent integral action can stop limit cycles (equal amplitude oscillations) from deadband and resolution or threshold sensitivity limits, and stop variable amplitude oscillations from wireless devices, at-line and off-line analyzers, slow secondary loops, or slow final control elements, and enable directional move suppression for optimization. Move suppression is set with analog output (AO) block or lower loop setpoint up and down rate limits to provide a gradual and less disruptive approach to an optimum and a fast getaway for abnormal conditions. Move suppression is also useful to prevent unnecessary crossing of the split range point that is location of severe discontinuities in the transition from one manipulated variable to another. The elimination of unnecessary crossings not only reduces process variability but also improves process efficiency from eliminating cross neutralization of reagents or alternately heating and cooling.
The principal methods of optimization in advance regulatory control involve the use of valve position control and override control. Often these work together to providing a sequential optimization addressing the most pressing constraint.
In valve position control (VPC), process control loop valve(s) are pushed to a maximum or minimum throttle position to reduce energy use by lowering a compressor discharge pressure or variable pump speed, reduce raw material costs by reducing expensive reagent flow, or maximize production rate by increasing feed flow. The main problems with valve position control is the slow response from integral only control and the lack of tuning rules to prevent interaction and the response to limit cycles in valve position. The limit cycle period is large and only apparent when there are no disturbances and a long trend chart time frame is used. Intelligent integral action, directional move suppression, and feedforward control can make VPC more effective and easier to implement. The November 2011 Control feature article "Don't Over Look PID in APC" provides more details and a summary of opportunities.
In override control, the lowest or highest output of VPC representing valve constraints and/or multiple PID controllers representing process variable constraints and targets, such as high level or high pressure or a desired flow, is selected either as a setpoint for a PID feed or to directly manipulate a final control element (e.g. control valve or variable speed drive). The principle difficulty with override control is the proper setup of external reset feedback to prevent an unselected controller from winding up or walking off, tuning to deal with constraint dynamics, and operator understanding of the order of constraints in the sequential optimization.
Valve position control is also used to simultaneously stroke a small control valve in parallel with a large control valve. The small valve provides better resolution and threshold sensitivity and the large valve provides capacity. The VPC problem of being too slow is addressed by the use of a compound feedforward signal. The first part of the feedforward signal is typically a flow feedforward. The second part is an innovative use of split ranging. The output of a splitter block provides a feedforward signal that increases with process PID output when the splitter output signal to the small valve has reached a high output limit. When the splitter output signal to the small valve is less than the high output limit, the VPC feedforward signal is zero. The split range point should be set to equalize the open loop gains for each valve throttled. For example, if the large valve has 4 times more capacity than the small valve, the split range point would be 80%.
Model predictive control (MPC) of continuous processes offers an inherent dynamic compensation of feedforward signals and constraints, a simultaneous optimization of constraints, incorporation of economics (e.g. feed and energy costs), and a potential analysis of the contribution of each constraint to the MPC output. Advanced regulatory control is used for batch operations and as a low cost and quick solution for simple optimization problems since only a configuration change is needed.
