How-to
Verifying HMI Historical Trend Retention Before Committing
Short answer
An 8-hour trend sampled every 5 minutes spans 96 intervals — 97 stored points if both endpoints are kept — so the panel needs a historical buffer of at least 97 samples per tag that repopulates the trend after screen changes. Verify the buffer depth, the interval setting, and restoration behavior on the bench before you build the screens.
On this page
- How many samples does an 8-hour trend at a 5-minute interval need?
- Does the trend keep its history when the operator changes screens?
- Where do the historical samples live on the panel?
- What sample interval and buffer depth should you configure?
- How do you verify retention before committing the project?
- Does the history survive a power cycle?
- What if the panel only offers a short real-time trend?
- Questions asked
An 8-hour trend sampled every 5 minutes needs 96 sample intervals — 97 stored points if the buffer keeps both endpoints — and a panel that restores those samples after the operator leaves the screen and comes back. Both halves are testable on the bench in under a day, and both fail silently in production: a trend object that only draws what it samples while its screen is open will show a fraction of the window you promised, with no error raised. Do the arithmetic first, then prove the retention behavior before any screens are built.
How many samples does an 8-hour trend at a 5-minute interval need? #
It needs 96 sample intervals, which is 97 stored points when the buffer keeps both window endpoints — so a panel whose documented buffer depth per tag is below 97 samples cannot display the full window, however the screen is drawn.
Carry the units through:
- Window: 8 h x 60 min/h = 480 min
- Intervals: 480 min / 5 min = 96 intervals
- Stored points: 96 + 1 = 97 points
One detail bites here: specifications quote buffer depth sometimes in intervals, sometimes in stored samples, and a 96-sample limit leaves the oldest endpoint off an 8-hour display. Get the number in stored points per tag and compare it against 97.
| Window | Interval | Intervals | Stored points |
|---|---|---|---|
| 8 h | 5 min | 96 | 97 |
| 8 h | 1 min | 480 | 481 |
| 24 h | 5 min | 288 | 289 |
Decision: at least 97 stored points per tag meets the requirement, anything less fails it, and anything more is margin you will want the day someone asks for 24 hours.
Does the trend keep its history when the operator changes screens? #
Only a historical trend fed by a sampling buffer does; a real-time or loop faceplate trend redraws from the moment the screen opens, so after a screen change an 8-hour window shows only the minutes since you returned.
The short loop trend — typically 15 or 30 minutes on hybrid controller faceplates — holds no cache, restarts every time the screen is opened, and offers no scrolling back. The historical trend mode retains samples in panel memory and repopulates the configured trend screen after navigation; that is the behavior an 8-hour window requires.
On platforms that separate the functions, the trend object creates no storage by itself: a data-sampling (logging) object writes the samples in the background, and the historical trend object reads that buffer. If the documentation describes only the short loop trend, the requirement fails on that mode, and the history has to come from the sampling object or from outside the panel.
Where do the historical samples live on the panel? #
In panel RAM unless the platform logs to internal flash or a memory card, and that choice decides what survives a reboot and what the operator can replay on the panel itself.
Three destinations, three behaviors:
- RAM sampling buffer — repopulates the trend after screen changes, clears at power-down.
- Internal flash or memory card — survives reboots; check the documented capacity and what happens when the log fills and rotates.
- Removable storage — archive files; on some platforms these open only in PC software, not in the panel's viewer, so confirm on-panel replay before relying on them.
Byte count is rarely the constraint. Worked size for the requirement: 97 points x 8 bytes per record (a 4-byte floating-point value plus a 4-byte timestamp) = 776 bytes per tag, and 12 analog tags = 12 x 776 = 9,312 bytes, under 10 kB. The panel has that memory many times over; what it may lack is a documented depth of 97 samples per tag. Specify on the sample count, not on memory size.
What sample interval and buffer depth should you configure? #
Configure the 5-minute interval the requirement states, a buffer depth of 289 stored points per tag so the same buffer also covers 24 hours, and an 8-hour display window — depth beyond the window costs almost nothing.
The interval comes from the phenomenon, not from the display. Resolving a cycle of a given period takes at least two samples per cycle, so a 5-minute interval resolves at fastest a 10-minute cycle; to see a cycle's shape, sample five to ten times faster than that. A steady temperature loop is comfortable at 5 minutes; pressure or flow oscillations of a few minutes are not.
For the depth: 24 h x 60 min/h / 5 min = 288 intervals, 289 stored points. Set the buffer there if the platform allows it; if the documented cap forces a choice, 97 stored points is the minimum that meets the 8-hour requirement. Keep the interval fixed and time-based so the samples land on the clock.
How do you verify retention before committing the project? #
Bench-test one tag end to end: configure the sampling object and the historical trend, accumulate more points than the window needs, navigate away and back, and confirm the oldest visible sample sits a full window behind the newest — before any production screens exist.
- From the panel's documentation, write down four facts per tag: supported sample intervals, buffer depth in stored points, storage destination (RAM, flash, card), and whether the trend restores after a screen change. Any line the documentation does not state is unverified — treat it as a failed test until the bench proves otherwise.
- Configure one analog tag at a fast interval, 10 seconds, so 97 points accumulate in 97 x 10 s = 970 s, about 16 minutes. That proves depth and restoration without waiting 8 hours.
- Navigate to another screen and return. The trend must repopulate to the oldest buffered sample. A view that repositions to the current time while the history stays scrollable passes; an empty trend fails.
- Power-cycle the panel and repeat the previous step. Record whether the buffer survives; if it clears and the requirement includes power interruptions, move the logging destination to flash or card.
- Run one overnight test at the real 300-second interval: after 8 hours, the oldest sample must sit about 8 hours behind the newest.
- Shake out tag wiring early the way testing HMI simulation against a live PLC describes, but run the retention steps on the panel and controller that will ship.
If the touch panel replaces an older operator interface, add those four documented facts to the checklist the way planning an HMI terminal replacement from an older operator panel structures a swap — trend behavior is exactly the capability that changes between panel generations.
One limit to keep straight: a display trend is an operator aid, not a safety record, and safety-relevant events belong in the controller's alarm and fault log.
Does the history survive a power cycle? #
Only if the samples are logged to non-volatile storage — a RAM buffer clears at power-down, so an 8-hour window restarts empty after every outage unless the platform writes to flash or a memory card.
Check three things in the documentation: whether logging to flash or card is a documented option rather than an assumption, whether the buffer is battery-backed RAM, and how the log rotates when storage fills. Write load is no concern at this interval — 5 minutes is 12 records per hour per tag — but it becomes one if intervals drop to seconds across many tags, because flash endurance is finite and the documentation should state the rated writes.
If the history must survive outages and the panel offers no non-volatile logging, hold the samples in the PLC instead; retentive controller storage is the same discipline as keeping HMI-entered timer presets in PLC memory.
What if the panel only offers a short real-time trend? #
Then the 8-hour history has to come from somewhere else: a logging option on the panel, a ring buffer in the PLC that the trend reads, or a PC historian — pick the one your documentation actually confirms, and design the screen around it.
The PLC ring buffer is the cheapest fallback. Size it the same way: 97 samples x 4 bytes for a 32-bit real = 388 bytes per tag before timestamps, and 12 tags = 12 x 388 = 4,656 bytes of controller data. The array must be retained, and any windowed statistic computed from it — mean, minimum, maximum — is valid only once the array holds the full window, the same rule that makes a mean depend on an array that retains the samples.
The screen still owes the operator a readable window: a labeled time axis, a visible span, and a scroll control they can find — the discipline of designing HMI screens for clear operator decisions. The storage decision determines whether the history exists; the screen design determines whether anyone can use it.
Questions asked
How many points does a 24-hour trend at a 1-minute interval need?
Twenty-four hours holds 1,440 one-minute intervals, so the buffer needs 1,441 stored points if both endpoints are kept. At 8 bytes per record, a 4-byte value plus a 4-byte timestamp, that is 1,441 x 8 = 11,528 bytes, about 11.3 kB per tag. The memory is trivial; the documented samples-per-tag limit is what decides whether 24 hours at 1 minute is possible.
Should I sample on change of value instead of a fixed interval?
Not for a fixed 8-hour window. Change-of-value sampling stores nothing while the signal is steady, so the time axis grows gaps and the window no longer represents 8 hours of history. Use a fixed, time-based interval, 300 seconds here, and keep change-of-value or forced sampling for event capture beside it.
Can I log to a USB memory stick and replay it on the panel?
Sometimes. Several platforms write archive files that only the vendor's PC software can open, so the panel itself cannot replay them. Confirm in the documentation that the panel's own trend viewer reads the logged files before you rely on removable media; otherwise the card is an archive for a PC, not the operator's 8-hour window.
What sample interval do I need to see a 2-minute process cycle?
At least two samples per cycle, so 1 minute is the slowest interval that shows a 2-minute cycle at all; use 20 to 30 seconds to see its shape. Then size the buffer from the window: 8 hours at 30 seconds is 960 intervals, 961 stored points.