As a modeling and control specialist in the engineering technology department of a large chemical company, I was called in to solve difficult and persistent problems that had plagued the production units. I focused mostly on important individual loops in key unit operations, such as gas pressure loops for furnaces and headers, temperature loops for reactors and kilns, surge loops for compressors, level loops for boiler drums and surge tanks, pH loops for all types of equipment and streams, dissolved oxygen loops for bioreactors, polymer pressure loops for extruders, and thickness control loops for sheets.
Except for headers, kilns, plug flow reactors, gas reactors, and sheets, problems with interacting loops were minimal. I could focus on individual loop performance and on transferring as much variability as possible from a key process variable (PV) to its manipulated variables. Even for unit operations with interacting loops, a simple half decoupler with the feedforward of one loop output to another loop output would usually suffice. In distillation, my associate Tolliver did not need to do much in the way of decoupling because there was usually only one temperature loop. Two point composition control, that is, control of top and bottom composition by temperature, was rarely needed.
I generally did not have to worry about coordination or the movement of a manipulated variable affecting another loop. I could go for the gusto. For optimization, I simply used valve position controllers (VPC) to maximize feed by pushing coolant, vent, and feed valves to a maximum throttle position (Tip #97). I also used VPC to maximize the use of waste fuels and reagents by pushing an expensive fuel or reagent valve with good process dynamics to a minimum throttle position.
In petroleum refining, the gas furnaces and catalytic reactors have interactions and don’t have much back mixing reducing the advantage of the PID. There is a greater need for plant-wide optimization. Because of incredibly large plant capacities, a fraction of a percent improvement in efficiency or production rate translates to huge increase in profitability in an industry where margins are squeezed. Consequently, MPC, linear programs (LP), and even real-time optimization (RTO) are used. I know of one refinery where all control systems were moved from PID to MPC because of the in-house MPC expertise.
I have used MPC with success but have returned to PID because existing and new PID features can be used in more innovative ways than expected to meet process objectives. I have to acknowledge I am the product of Shinskey’s books (download Appendix B). I like to tackle the process on a unit operation basis.
Concept: Loop coordination objectives are used to ensure nearly identical SP responses and to decrease loop interactions. A primary optimization objective is to achieve a more efficient or productive operating point without upsetting the process. The effective use of key PID features can achieve these objectives.
Details: There are three major loop coordination objectives plus an optimization objective. The first coordination objective is to ensure the simultaneous response of flow inputs to a unit operation (e.g., ratio flow control and blending). The second is to all secondary loops to have a uniform desired secondary PV response to a setpoint change in cascade control or model predictive control (MPC). This secondary PV response consistency is particularly important for parallel unit operations, so that differences in the primary controller tuning and MPC model are indicative of discrepancies in identical unit operations. The third coordination objective is to slow down the transfer of variability from the PV to the PID output to reduce interactions. Coordination has traditionally been achieved by increasing reset time and decreasing gain. However, this tuning sacrifices disturbance rejection and rise time (Tip #71). Optimization has been traditionally achieved by integral-only valve position controllers that are difficult to tune. Today, there are key PID features (e.g., external-reset feedback) to provide an easier and more effective means of coordination and optimization. Use external-reset feedback with directional setpoint rate limits or a setpoint filter to provide the desired speed of response with disturbance tuning. For optimization loops, use a slow setpoint rate limit to provide a gradual approach to the optimum and a fast setpoint rate limit to provide fast correction for disturbances and abnormal operation. For surge loops, use directional rate limits for the analog output setpoint to provide a fast opening and slow closing surge valve. For valve position controllers that maximize coolant temperature setpoints or reactor feed setpoints to push coolant valves to the largest safe throttle position, use slow increasing rate limits and fast decreasing rate limits on the process controller setpoint being maximized (Tip #97).
Watch-Outs: Setpoint rate limits on analog outputs (AO) will confuse Maintenance and Operations when they are manually stroking a valve. AO setpoint rate limits should be turned off when in manual. Controllers without the positive feedback implementation of integral action may not have external-reset feedback.
Exceptions: Optimization must take a back seat to the correction of large, fast disturbances. To prevent equipment damage and an environmental violation, an open loop back up may be needed (Tip #91).
Insight: The response of control loops or final control elements can be coordinated to work together to meet process objectives by the use of setpoint filters and velocity limits and external-reset feedback.
Rule of Thumb: Set a setpoint filter time that the slowest coordinated loop can achieve, slow down the AO or PID setpoint rate limit in both directions to minimize interaction, and set the AO or PID setpoint rate limit slow in the optimization direction and fast in the protection direction.