The short answer
Trending
A trend is the historian made visible, and reading one well is a core skill for anyone who runs or troubleshoots a process. Trends are built from pens that belong together, such as level with pump states, pressure with flow, and residual with dose, each scaled to a fixed engineering range rather than auto-scaled, over a time range long enough to show the pattern and short enough to show the detail. Raw samples show what happened; averaged data hides the fast events; compression settings decide what was stored in the first place. Overlays of yesterday against today, cursors for reading values at a moment, and alarm and event markers on the time axis turn a curve into an explanation, and the patterns of cycling, drift, steps, and noise each point at a class of cause.
Key points
- Pens that belong together: the cause and the effect on one trend, not one value alone.
- Fixed scales in engineering units; auto-scale turns a flat line into a mountain range.
- Know what the historian stored: compression and averaging decide what the trend can show.
- The time range is chosen for the question: hours for a control problem, weeks for drift, a year for seasonality.
- Alarm and event markers on the time axis explain the curve.
- Cycling, drift, steps, and noise are different problems; the trend shape is the first diagnosis.
Building a trend that says something
- Pens together: a wet well level with the pump run states below it shows cycling and lag response in one picture; a level alone shows a sawtooth with no explanation. Pressure with flow, chlorine residual with the dose and the flow, dissolved oxygen with air flow and blower speed, tank level with the pumps that fill it.
- Fixed scales: each pen scaled to its engineering range or a sensible fixed span, so that the same trend looks the same tomorrow. Auto-scaling stretches a tenth of a psi of noise to fill the screen and makes a stable process look wild.
- Stacked or overlaid: overlay pens with related magnitudes; stack pens with different units or discrete states so they do not obscure each other.
- Time range for the question: a control loop problem in an hour or two, a diurnal pattern in a day or a week, drift in a month, seasonality in a year.
- Discrete states as step lines, analog values as lines, and bad quality shown as a gap or a marker rather than a line drawn through nothing.
- Saved trend groups for the questions that recur, named for what they answer, on the display where they are needed.
What was stored
A trend can only show what the historian kept. A tag historized on change with a deadband of one percent will not show a fluctuation smaller than that; it will show a straight line between the samples that exceeded it. A tag stored as a one-minute average will not show a five-second pressure transient; it will show a slightly lower minute. Compression that discards points along a straight line shows a straight line through a period that may have wandered within the tolerance. When a trend looks too smooth, too steppy, or shows nothing where something happened, the first question is how the tag is collected and compressed, and the answer is in the historian configuration. For diagnosis of a fast event, look at the raw samples and, if the tag is averaged, at a faster-collected tag or the controller itself.
Reading the shape
| Pattern | What it looks like | Likely causes |
|---|---|---|
| Cycling | A regular oscillation, sawtooth or sinusoid | On-off control between limits, which is normal; PID tuned too aggressively; a control valve with stiction; two controllers fighting |
| Drift | A slow steady change over hours or days with no process reason | Sensor fouling, a wet leg filling, a reference shifting, a slow leak, temperature effect |
| Step | A sudden change to a new steady value | A setpoint or configuration change, a range change, a pump or valve state change, a calibration, an instrument replaced |
| Noise | Rapid small variation with no pattern | Electrical interference, air in a line, a turbulent measurement point, a damping setting removed |
| Flat line | No variation at all | A frozen value from lost communication, a failed transmitter, a tag no longer collected, or a filter so heavy it hides everything |
| Spikes | Single samples far from the rest | Radio interference, a keyed radio, a power event, a communication glitch, a sample collected during a transition |
| Daily pattern | A curve that repeats each 24 hours | Demand, temperature, a schedule; normal, and the baseline for spotting a change |
Cursors, overlays, and markers
A cursor reads the value of every pen at one instant and is how a trend becomes a number: the level when the pump started, the pressure when the alarm came in. Two cursors give a difference and a duration: the fill rate between two levels, the run time of a pump. An overlay puts the same tag from a previous day or week on the same axes, which is the fastest way to see that today is different, and by how much. Alarm and event markers on the time axis, with the alarm journal and the operator action log, connect the curve to what people and the system did: the setpoint change that started the oscillation, the acknowledgment that came ten minutes after the alarm.
Habits
- Look at the trend before touching the process. Ten minutes on a trend saves an hour of adjusting.
- Widen the time range before concluding. A problem that started yesterday looks different from one that has been growing for a month.
- Check the pen that did not change. A pressure that fell while the flow stayed constant tells a different story from one where both fell.
- Note the sample rate and the compression on any trend used for a decision.
- Export the data for anything that will be argued about; a screenshot cannot be re-analyzed.
- Build the trend group once and save it; the next person with the same question should not have to rebuild it.
Frequently asked questions
- Why does my trend show a straight line through a period when I know the value moved?
- The historian compression or deadband discarded the samples because they stayed within the tolerance, and the trend draws a line between the samples it kept. Tighten the deadband for that tag if the detail matters, and look at the controller or a faster-collected tag for the past event.
- How long a time range should I use to diagnose a control loop?
- A few cycles of whatever the loop is doing: for a loop cycling every five minutes, an hour; for a wet well filling and draining every twenty minutes, a couple of hours. Long enough to see the pattern repeat, short enough to see the shape of each cycle. Then widen it to see when the pattern started.
- Should trends on the operator display use the historian or the live data?
- The historian for anything longer than a few minutes, so that the trend survives a display change and matches what was stored. Live buffers on the client are fine for a short real-time trend on a faceplate, and they disappear when the display closes.
- What is the best way to compare two similar sites?
- The same trend group with the tags of each site, at the same fixed scales and time range, side by side or overlaid. Differences in cycle time, run time, or pressure profile between two stations that should be alike point at a problem at one of them.
Related topics
- Trends on the HMITrend displays that answer questions: embedded sparklines on process displays, pre-configured trend groups, the axes and time spans that make a trend readable, and the difference between a trend for operating and a trend for investigating.
- Historian CompressionHow historians store years of data on a disk: exception deadbands at the collector, swinging door compression at the archive, what each discards and what it keeps, the settings that turn a trend into a staircase, lossless alternatives, and how to check that the compressed data still tells the truth.
- Historian Data CollectionDeciding what the historian stores and how often: polled versus exception collection, the deadband per tag, rates for control loops, levels, totals, and compliance values, timestamps at the source, quality codes, calculated tags, and the tag list review that keeps a historian from filling with noise and missing what matters.
- Reporting from the HistorianTurning history into the reports a utility must produce: the monthly operating report, compliance reports for the regulator, daily operator summaries, pump run and energy reports, the calculations behind daily minimums, maximums, and totals, handling bad quality and gaps, and building reports once so they run every month without an engineer.
- HistoriansWhat a process historian does that a database does not, how data gets into it, how compression and retrieval work, how to size and keep it, and the settings that quietly throw data away.
- Values Frozen On ScreenThe HMI shows numbers that are not changing while the plant runs. Whether the controller, the communication driver, the tag server, or the display stopped, and the order to check them in.
Direct contact
Have a controls question?
Reach Eric Sullivan directly about anything on this site, a controls or automation topic, or one of his personal projects.