What is Trending?
Trending is the graphical display of process values over time, allowing operators, engineers and maintenance teams to see how a measurement develops, recognise patterns and detect deviations early. A trend is a line chart with time on the horizontal axis and one or more tags, such as temperature, pressure or flow, on the vertical axis. Trending is a core function of SCADA and HMI systems and of every historian, which makes it one of the most widely used tools in industrial automation. Where a number on a process display shows only the current state, a trend shows where that state came from and where it is heading.
🧠 What exactly does a trend show?
A trend chart consists of a few fixed elements. Setting them up deliberately gets far more information out of it:
- Pens - each line in the chart is called a pen and belongs to one tag; the term dates from the era of paper chart recorders with ink pens
- Time axis and time window - for example the last 15 minutes, 8 hours (one shift) or 7 days; zooming in and scrolling back are part of it
- Value axis and scale - a separate scale per pen, preferably matched to the normal operating range rather than the full measuring range
- Reference lines - the setpoint, alarm limits and specification limits, so the distance to a limit is immediately visible
- Data quality - good systems attach a quality status to every value, such as good, bad or uncertain in OPC; a trend should show a gap or a marked section when communication was lost, rather than a flat line suggesting a stable process
- Cursor or ruler - a vertical line that reads the values of all pens at a single point in time
The difference between a number and a trend is the difference between where and where to. A temperature of 82 °C says little on its own; a temperature that has risen from 75 to 82 °C in ten minutes and is still climbing calls for action. That is why trending directly supports the operator’s situational awareness.
🔍 What types of trends are there?
| Type | Description | Typical use |
|---|---|---|
| Live trend | Real-time chart of current values, often from a short buffer in the SCADA server | Start-up, adjustments, testing a control loop |
| Historical trend | Looking back at stored data from the historian over days to years | Fault analysis, reporting, audits |
| Embedded trend | Small trend placed directly in a process display next to its value | Monitoring in line with ISA-101 principles |
| Multi-pen trend | Several tags at once, such as pressure, temperature and flow of one unit | Seeing relationships between variables |
| Overlay | Several time periods or batches on top of each other, aligned to their start time | Comparing a batch with a reference batch |
| XY plot | Two variables plotted against each other instead of against time | Pump curve, relationship between valve position and flow |
| Control chart (SPC) | Trend with mean and control limits at ±3σ | Quality monitoring with SPC |
In modern systems live and historical trends merge: the same trend shows the last few minutes from memory and switches seamlessly to the historian when you scroll back. Overlays are especially popular in the process industry with batch processes, where a good reference batch serves as the “golden” profile.
🔧 How does data end up in a trend?
Behind every trend sits a chain of measuring, sampling, storing and retrieving. Each step determines what you eventually see.
1. Sampling. The SCADA system or historian reads the process value periodically. According to the Nyquist–Shannon sampling theorem, you must sample more than twice per period of the fastest relevant fluctuation. Sample too slowly and aliasing occurs: a fast oscillation appears in the trend as a slow wave that does not exist in reality.
2. Filtering and compression. A historian does not store every value it reads. A deadband (exception) filter first removes noise, then compression such as the swinging door algorithm discards points that lie on a straight line. On retrieval, the historian interpolates between stored points. A compression setting that is too loose makes short spikes invisible, and during fault analysis that may be exactly the information you are looking for.
3. Timestamps. Placing trends from several sources side by side requires synchronised clocks. Synchronisation via NTP is sufficient for most process trends; for finer sequence-of-events analysis, such as in substations, Precision Time Protocol is used.
4. Retrieval for display. A one-year trend with values every second contains over 31 million points, while a screen is only about 1,000 to 2,000 pixels wide. Historians solve this with plot functions: the PI System historian divides the period into intervals (typically one per pixel) and returns for each interval only the first, last, highest and lowest value, plus any exceptional status values. Every spike stays visible without millions of points crossing the network. A plain average per interval would smooth spikes away.
Indicative sampling intervals per signal type:
| Signal | Typical interval | Notes |
|---|---|---|
| Temperature in a vessel or furnace | 1–10 s | Slow process because of thermal mass |
| Pressure and flow | 0.1–1 s | Can change quickly, important for control |
| Tank level | 1–5 s | Usually slow, except in small buffer vessels |
| Energy and utility meters | 1–15 min | Often stored as a counter reading or 15-minute value |
| Vibration of rotating machines | kHz range | Belongs in a condition monitoring system; the historian only receives indicators such as RMS values |
🏭 What is trending used for in practice?
- Fault analysis - what happened in the minutes before an alarm or trip? Trends are the basis of every root cause analysis and of alarm reviews under alarm management
- Control loop monitoring - a PID controller that keeps oscillating, a process value that never reaches its setpoint or a sawtooth pattern pointing to a sticking valve is immediately visible in a trend
- Quality control - did process conditions stay within tolerances throughout the whole batch or run?
- Optimisation - seeing what a change to setpoints or recipes does to consumption, cycle time or OEE
- Predictive maintenance - a slowly rising bearing temperature or motor current announces wear weeks in advance
- Reporting and accountability - energy consumption, emissions and process data per day, month or year
Example: an industrial furnace. The tag Oven1.Temp is read every second and
stored in the historian. The process display shows an embedded trend of the last two hours, with the 850
°C setpoint and the alarm limits as horizontal lines. The operator notices that two hours ago the
temperature briefly exceeded the upper limit without an alarm following. A multi-pen trend with gas flow
and valve position shows that the gas control valve opened too far at that moment. The engineer adjusts
the controller parameters and schedules an inspection of the valve; a week later the historical trend
confirms that the spike no longer occurs.
📊 How do you recognise patterns and deviations in a trend?
Experienced operators read trends like a language. Common patterns:
- Drift - a slow, persistent shift, often caused by fouling, wear or a sensor going out of calibration
- Oscillation - a regular wave, usually from a controller tuned too aggressively or interaction between control loops
- Step - a sudden jump, for example from a setpoint change, a switching action or a failed measurement
- Flat line - a value that does not move at all may indicate a frozen measurement or a communication failure
- Noise - fast, small fluctuations that hide the real trend
For statistical assessment, the quality world uses control charts, developed by Walter Shewhart at Bell Labs in the 1920s. The Western Electric rules from the 1956 Statistical Quality Control Handbook and the eight Nelson rules from 1984 make that assessment objective. Examples of such signals:
| Rule (Western Electric/Nelson) | Signal | What it may mean |
|---|---|---|
| One point beyond 3σ | Outlier | Disturbance or measurement error |
| Two of three points beyond 2σ on the same side | Shift starting | Process is deviating structurally |
| Eight points in a row on the same side of the mean | Level shift | New raw material, different setpoint, sensor drift |
| Six points in a row rising or falling | Trend | Wear, fouling, heating up |
Today, anomaly detection and machine learning complement this: models learn the normal behaviour of hundreds of tags and flag combinations a human would miss in individual trends.
🧭 How do you set up good trends?
- Define fixed trend groups per process area. Agree per unit which tags are shown together, so every operator uses the same trends.
- Scale to the operating range. A temperature moving between 60 and 70 °C is a flat line on a 0–400 °C scale.
- Show setpoint and limits. Only then can you see how much room is left before an alarm or trip.
- Limit the number of pens. More than six to eight lines in one chart makes a trend unreadable in practice; use several stacked trends sharing one time axis instead.
- Embed trends in process displays. The High Performance HMI approach, on which ISA-101 (2015) builds, recommends showing key values with a small trend, so the direction is visible at a glance.
- Check the historian settings. Have the sampling interval, deadband and compression documented per signal type; otherwise you will later miss what you need.
- Choose the right tool. Classic trends live in WinCC, Ignition and DCS packages; for broader analysis, Grafana on a time series database such as InfluxDB or TimescaleDB is also used, often fed through a Unified Namespace.
🔐 What about security and data integrity of trends?
Trends are only useful if they are correct. Stuxnet showed what happens when they are not: the malware replayed previously recorded, normal values to the operators, so they saw a calm process while the centrifuges were being damaged. A trend that looks “too good” therefore deserves attention too.
- Read access without write rights - office users should preferably view trends via a historian replica in the DMZ, possibly fed through a data diode, rather than directly on the control network
- Integrity of stored data - in the pharmaceutical industry, trend data used for batch release counts as an electronic record; GMP, EU GMP Annex 11 and 21 CFR Part 11 then require audit trails, access control and protection against alteration, see data integrity
- Monitoring the trends themselves - a tag that stops updating or a sudden flat line may be a fault, but may also be a signal for monitoring and security investigation
❓ Frequently asked questions
What is the difference between a live trend and a historical trend?
A live trend shows current measured values in real time, usually from a short buffer in the SCADA or HMI system. A historical trend retrieves stored data from a historian and can look back months or years. In modern systems a trend moves seamlessly from live to historical data when you scroll back.
How often should a process value be sampled for trending?
The sampling interval for trending depends on how fast the process changes: often 1 to 10 seconds for temperatures and 0.1 to 1 second for pressure and flow. You must sample more than twice per period of the fastest relevant fluctuation to avoid aliasing. Vibration analysis needs kilohertz sampling and belongs in a separate condition monitoring system.
Why does a spike sometimes look smaller in a trend than it really was?
A spike in a trend can look smaller because of a slow sampling interval, a compression setting in the historian that is too loose, or averaging when a long period is displayed. Good historians therefore return the minimum and maximum value per pixel interval rather than an average. During fault analysis, always check at what resolution the data was stored.
How many pens should a trend have?
A trend stays readable in practice with roughly four to eight pens, each with a clearly distinct colour and its own scale. With more tags, several stacked trends sharing one time axis work better. That keeps relationships visible without lines tangling.
Is trending the same as a dashboard?
Trending is not the same as a dashboard: a trend shows how process values develop over time, whereas a dashboard mainly shows current indicators and summaries such as a KPI. Dashboards often include trends as one element. For operational control and fault analysis, the process trend is usually the more useful tool.
Can trending help with predictive maintenance?
Yes, trending is a simple first step towards predictive maintenance, because slowly rising values such as bearing temperature, motor current or vibration level often reveal wear weeks in advance. By placing a limit line on the trend, the maintenance team sees when intervention is needed. Advanced predictive maintenance adds statistics and machine learning on top.
📌 In summary
Trending is the visual analysis of process data over time and shows not only where a process stands, but also where it is heading. With well-chosen sampling, careful compression, synchronised clocks and clear scales, the trend becomes an indispensable tool for monitoring, fault analysis, optimisation and maintenance in industrial environments.