Trouble: Flow Controller is unstable!

Transcription

1 Page 1 of 1 By Jim Henry, Ph.D, P.E. University of Tennessee at Chattanooga 2004 Figure 1. Schematic diagram of the Filter Wash System Introduction Describing the equipment Controller Tuning Done off-line as per textbook or instructor Operational Situation Prior "normal" operation Demonstration Show that your controller works well Equipment Upgrade Situation came up A problem that Problems Real life difficulties Assignment get FOPDT "System Identification:" Step-test to References Smith and Corripio and SEM Hints How to do the analysis Instructor resources

2 Page 1 of 2 By Jim Henry, Ph.D, P.E. University of Tennessee at Chattanooga 2004 Our POTW ("Publicly Owned Treatment Works") sewerage treatment plant in Chattanooga, TN, has a large filter press to filter out the sewerage sludge solids in order to send the solids to the city landfill. The filtrate water is then processed further before returning to the Tennessee River. The filter press operates in a repeating batch mode. Between batches, the filter media and plates must be washed. The manufacturer of the press specifies that the water flow rate to the washing nozzles has to be maintained between 7 and 10 lb/min. The flow rate of the wash water is maintained by a variable speed centrifugal pump. The motor driving the pump is a variable-speed motor. The speed of the pump is under feedback control to maintain the desired flow rate to the washing nozzles. The controller is a proportionalintegral controller. A diagram of the pump, washing nozzles and control system is shown below. Also shown on the diagram are 2 other "manual" or "auxiliary" wash lines that are used by operators in the plant to wash up spills when needed. Figure 1. Schematic diagram of the Filter Wash System

3 Page 1 of 2 Back Operational Situation Plant personnel have been happy with the system (including the control system) for years. Ever since plant start-up, we ve been able to maintain the wash-water flow rate within specifications. Here are graphs of the wash water flow rate (sampled by FT-301) and Pump Speed (from FRC-301 output, expressed as %) under closed-loop feedback control for two recent periods. Flow (lb/min) at FT-301 vs time (sec) Flow (lb/min) at FT-301 vs time (sec) Pump Speed from FRC-301 Output Figure 2. Performance while 2 auxiliary washing stations are in use Pump Speed from FRC-301 Output Figure 3. Performance while 1 auxiliary washing station is in use The controller tuning parameters were set by installation contractor and have not been

4 Page 2 of 2 adjusted. The value of Controller gain is 4.0 %/lb/min; the integral (reset) time value is 1.0 second.

5 Page 1 of 2 Equipment Upgrade Situation Back to "Trouble" Unfortunately, with a recent improvement in operating procedures and equipment upgrades, we frequently have no need for either auxiliary washing station to be used. When neither auxiliary washing station is in use, our operators tell us that the wash-water flow rate is very erratic. Here are graphs of the wash-water flow rate (sampled by FT-301) and Pump Speed (from FRC-301 output) under closed-loop feedback control for a recent period with neither auxiliary washing station in use. No auxiliary washing station in use Flow (lb/min) at FT-301 vs time (sec) Pump Speed from FRC-301 Output Figure 4. Performance while no auxiliary washing station is in use Figure 4 shows that the desired flow to the wash-water nozzles goes above the desired maximum of 10 lb/min and below the desired minimum of 7 lb/min. Our maintenance chief is not happy about this.

6 Page 1 of 2 Assignment Back to "Trouble" Determine what the controller tuning parameters should be in order to get better performance (no oscillation, flow rate within the specifications). This can be done based on step-response testing of the system. Fortunately, the filter press is in stand-by status today. This means that you can conduct some step-response experiments to determine the system parameters necessary to design the controller for the performance you desire. You can conduct step-response experiments by going to this web site: Below is what that page looks like. Flow-step.htm Figure 5. Web page for running experiments

7 Page 2 of 2 Fill in the form with your information in items #1 - #7, then click on the "RUN EXPERIMENT" button at #8. The results of the experiment can be analyzed to determine the system gain, K, the system s first order time constant,t, and the system s dead time, t o. [Smith, pages , Seborg, pages ]

8 Page 1 of 2 Hints Back to "Trouble" To do the analysis, examine the step response in detail where the step happens. See the figure 5 below. Figure 5 is a copy of what the results of the experiment will look like. Figure 5. Sample experimental results To expand the area of interest, you can access the data by click on the "Data as Text" icon as shown at the right. Select all the data on that page and "Paste Special" with

9 Page 2 of 2 "Unicode Text" into an Excel spreadsheet. Plotting the data will yield a graph such as in Figure 6, below. Figure 6. Detail of step response (Step input of 20% at 40 seconds) Having the system s parameters for appropriate operation configuration, use controller tuning formulas to determine the controller tuning parameters that you want to recommend. [Smith, pages , Seborg, pages ]

10 Page 1 of 2 Demonstration of Effective solution Back to "Trouble" Once you have improved tuning parameters, try them out on the system. You can conduct experiments by going to this web site: Flow-PI-control.htm At that site, you can conduct experiments in which the set-point changes or experiments in which the Manual Wash stations are turned on or off. Below is what that page looks like.

11 Page 2 of 2 #1 and #2 are your personal information #3 is the desired length of the experiment. You are limited to less than 160 seconds. #4 and #5 are the values of the tuning parameters that you have chosen. #6 is what youchoose for the set point. #7 and #8 specify the change in set point, if you choose. #7 is for the amount of the change, #8 is for the time of the change. #9 is for setting the options on the turning on or off the Auxiallary Wash Stations. When you have completed the specifications for the experiment, click "RUN EXPERIMENT" at #10

12 Page 1 of 1 Possible Problems Back to "Trouble" Instead of the "nice" response as seen in Figure 5, the experiment may give results like those shown to the right, here. These can not be analyzed. Results like this are due to the start-up effects in the system. While the system is in "stand-by" status, all the pipes drain back to the pump. So the graph shown at the right is a result of surging of water where air has been. To overcome this effect and thus get a graph that you can analyze, you can run TWO experiments consecutively. The first experiment is a "throw away," just used to fill the pipes so the system behaves normally. That is, in fact, what was done to get the results in Figure 5.

13 Page 1 of 1 References Back to "Trouble" References: Smith and Corripio, Principles and Practice of Automatic Process Control, 2 nd edition. Seborg, Edgar and Mellichamp, Process Dynamics and Control, 2 nd edition.

14 Page 1 of 2 Back to "Trouble" : The cause of the instability in the system is that the system gain, K, changes as the "disturbance" changes which are caused by the parallel users (auxiliary washing stations). With both auxiliary washing stations not in use, the system gain is highest, leading to instability in the closed-loop system. The solution is to reduce the controller gain, K c, or to increase the integral (reset) time,t i. As a practical matter, a plant engineer who was familiar with controller tuning might try reducing the controller gain by a factor of 2 or increasing the integral (reset) time by a factor of 2. Either of these probably would give good results. To run the step experiments, it is wise to choose values of the system "Input (%)" to be in the expected operating range. This can be estimated from Figures 2 and 3. In Figure 2, the Input is about 50%. In Figure 3, it is about 30%. So an initial value of Input = 30% and a step height of 25% would be a good choice. A data file with successful step experiment for both auxiliary washing stations turned off is at The graph of that data is at right. The system gain is about (19-11)/20 = 0.4 lb/min/%. The dead time is about 0.6 sec.; the first order time constant is about 0.6 sec., also, according to Smith and Corripio s "Fit 1." Using the Ziegler-Nichols quarter-decay tuning formula, the controller gain, K c, = 2.2 %/lb/min and the integral (reset) time,t I = 2 sec. Trying that in the controller, the following are the results:

15 Page 2 of 2