Humans have an incredible capability to see and analyze patterns. The setup of the data historian and trend chart can either hide or enable the recognition of essential patterns. The check list can help you identify the root cause of poor loop performance and a potential solution. Also included is a module to compute load and setpoint response metrics online.
If the data historian compression is too large, disturbances, limit cycles, and noise will not be visible. To assess loop performance, the compression of the process variable (PV) should be less than 1/5 of the control band where the control band is the allowable PV error around setpoint (SP). To detect limit cycles, the compression of the controller output (OUT) must be less than the control valve or variable frequency drive (VFD) deadband, resolution, and threshold sensitivity. The limit cycles in the OUT tend to be a saw tooth. The limit cycles in a flow or liquid pressure tends to be a square wave. Limit cycles in level and temperature tend to be a saw tooth. For controller gains greater than 1, patterns are more apparent in the OUT or the manipulated flow. To analyze noise, the compression of the PV must be less than 1/5 of the noise amplitude. To monitor changes in noise symptomatic of changes in the sensor or process, the unfiltered PV should be plotted. A reduction in noise may indicate a coated thermowell or electrode. An increase in noise may indicate bubbles in gas streams or droplets in vapor streams or a decrease in the degree of mixing.
If the time span is too short, load and setpoint responses, slow rolling oscillations in integrating processes, and limit cycles will not be evident and the source of a disturbance will not be detectable. The response will look like a sloping horizontal line. If the time span is too long, the trend will look like a squished jagged series of spikes around setpoint. Patterns and what started first will not be recognizable.
To detect limit cycles from backlash and stiction, the time span should be about 20 times the integral time. To detect slow rolling oscillations in integrating processes from too low of a controller gain or reset time, the time span should be about 100 times the integral time. Equation C-14a in the appendices for InTech 2012 Jan-Feb article “PID tuning rules” shows the relationship that triggers these oscillations. To determine what flow caused an upset and see the PID response, the time span should be about 20 deadtimes. To identify a temperature or composition disturbance, the time span should be about 100 deadtimes.
Since nearly all process inputs are flows, the first flow to change on a unit operation is most like the source of the upset to the unit operation. Most upsets start out as flow changes from poor valves, poor controller tuning, batch operations, manual actions, on-off actions, sequences, and trips. Changes in temperature or composition are much slower and more difficult to detect. The disturbance will be seen in the OUT or flow manipulated by a loop affected by the change.
To see the setpoint response including the settling time, the time span should be greater than 10 times the rise time. The rise time is the deadtime plus the setpoint change divided by the PV rate of change. The rate of change (ramp rate) is best measured online but can be estimated as the near-integrating process gain multiplied by the change in controller output from the setpoint change. The future PV value plus the PV ramp rate described in the July 29, 2012 Control Talk Blog the “Future PV Values are the Future” should be on the trend plot with the loop PV, SP, and OUT for setpoint response analysis.
The following checklist does not cover all of the potential uses of trend charts but will get you started with some effective trend charts
- Is the historian PV compression less than 1/5 the control band and noise amplitude?
- To see limit cycles, is the historian OUT compression less than the control valve or VFD deadband, resolution, or threshold sensitivity?
- To estimate loop deadtime is the historian sample time less than 1/5 the deadtime?
- To prevent aliasing is the historian sample time less than 1/2 the oscillation period?
- For noise analysis, is the unfiltered PV on a trend chart?
- For disturbance analysis, are all process and utility flows for unit operation besides loop PV, SP, and OUT on the same chart?
- For setpoint response analysis, is the PV ramp rate and future PV besides loop PV, SP, and OUT should on the same chart?
- Is the time span about 20 deadtimes to analyze flow disturbances?
- Is the time span about 20 times the reset time to see limit cycles?
- Is the time span about 100 deadtimes to analyze temperature disturbances?
- Is the time span about 100 times the reset time to see slow rolling oscillation in integrating processes?
- Is the integrated error and peak error for load disturbances and the rise time, overshoot, undershoot, and settling time for setpoint changes on a trend chart with the loop PV, SP, and OUT?
A DeltaV version 8.3 composite template library module I developed can be used to provide the metrics mentioned in the last checklist item by connecting the PV error and setpoint, setting the module execution time, and enabling load or setpoint metrics. The user can set the settling error to determine rise time and settling time and the setpoint threshold (minimum change in setpoint) to trigger setpoint metrics. A screen print of the module is seen in the jpeg file Loop-Performance-Metrics-Module and the module CALC block expression is in the text file Loop-Performance-Metrics-Calc