A trend chart is only a few hundred pixels wide, but the data behind it may contain millions of points. Something has to reduce that data to fit, and the reduction method decides whether the spikes survive the trip to the screen. Min-max decimation is a rendering technique built specifically to keep them: for each column of pixels it keeps the highest and lowest values that fall there, so a transient that hides for a fraction of a second is still drawn. It is the answer to why a plain averaged trend can smooth an event out of existence while a min-max trend shows it.
Min-Max Decimation in one line: Min-max decimation is a plotting technique that reduces dense time-series data to screen resolution by keeping, for each horizontal pixel column, the minimum and maximum values that fall within it. Because it always retains the extremes of each column, spikes and dropouts survive the downsampling instead of being averaged away as they would be with a plain mean.
A trend plot maps time across a fixed number of horizontal pixels, and each pixel column can only show so much. When the time range holds far more data points than there are pixel columns - which is the normal case for any meaningful span of a fast tag - many raw points must map to a single column. The renderer cannot draw them all, so it has to choose a representative for each column. That choice is the whole game, because it decides what the viewer sees and, just as importantly, what they never see.
The tempting default is to average all the points in a column into one value and plot that. For a slow, smooth signal this is fine, but for a spiky one it is quietly destructive. A column that spans, say, a full minute of data might contain hundreds of calm readings and one sharp two-second spike. Averaging that column produces a value barely above the calm level, so the spike is drawn as though it never happened. The chart looks clean and is wrong.
This is not a rare edge case; it is the central hazard of visual downsampling. The events an operator most needs to notice on a trend - a pressure surge, a momentary dropout, a valve slamming - are exactly the brief, extreme excursions that an averaging reduction erases. A trend that hides its transients gives false confidence, showing a placid history over a process that was actually spiking. The whole point of a spike-preserving decimation method is to prevent that false confidence.
Min-max decimation solves the problem by refusing to collapse each column to a single value. For every pixel column it finds two values among the raw points that fall there: the minimum and the maximum. It plots both, typically as a thin vertical line or filled band spanning from the low to the high of that column. The result is not a single wandering line but an envelope that traces the full range the signal covered within each slice of time, so the top of every spike and the bottom of every dropout is drawn.
This guarantees spike survival by construction. If a two-second surge occurred anywhere inside a column's time span, its peak becomes that column's maximum and is drawn, no matter how brief it was against the surrounding calm. The envelope may be thin where the signal was steady and tall where it swung, and that variation in thickness is itself informative - it shows at a glance where the process was quiet and where it was volatile, information a single averaged line cannot convey.
The contrast with averaging is stark and worth stating plainly. An averaged downsample answers where was the signal roughly, and hides how far it strayed; min-max decimation answers how far did the signal range, and never hides an extreme. The tradeoff is that min-max draws a band rather than a crisp line, which is slightly busier visually, but for any application where a missed transient matters that is a price well worth paying. It keeps two carefully chosen points per column instead of one misleading average.
For a cloud SCADA platform, min-max decimation is what makes long trends both fast and honest. When a user in Merobix pulls up a week of a fast pressure tag, sending every raw sample to the browser would be slow and pointless - there are far more samples than pixels. Reducing the data with min-max decimation, whether in the query layer or the renderer, ships a compact representation that still contains every excursion, so the trend loads quickly and yet a two-second surge from Tuesday night is still visible on the screen.
This directly serves how operators actually use trends, which is to catch exceptions. Someone scanning a month of history for anomalies is looking for the spikes and dropouts, and a rendering method that averaged those away would defeat the entire purpose of the review. Min-max decimation ensures the anomalies are the very things preserved, so a fleeting over-pressure or a momentary loss of signal shows up in the trend rather than being smoothed into a reassuring but false flat line.
It is worth distinguishing this plotting technique from storage-side reduction. Min-max decimation is about what to draw for a given range on a screen of finite width, and it is applied when the trend is rendered; it does not change what the historian keeps. That separation means the full-resolution data remains available for numeric queries even as the visual is decimated for display. The plotting method and the retention policy answer different questions, and min-max decimation is squarely the visual one - making sure the picture of the data tells the truth about its extremes.
When more data points map to a pixel column than can be drawn, averaging collapses them into one value near their center. A brief spike is a small contribution to a column full of calm readings, so the averaged value barely rises and the spike is drawn as if it never happened. The chart looks clean but omits exactly the transient event a viewer most needed to see.
Instead of one value per pixel column, it keeps two: the minimum and the maximum of the raw points that fall in that column, drawn as a vertical band from low to high. Because the peak of any spike within a column becomes that column's maximum, it is always drawn no matter how brief it was. The result is an envelope that shows the full range the signal covered.
It is a display method. It decides what to draw for a given time range on a screen of finite pixel width, and it is applied when the trend is rendered rather than when data is stored. The historian still keeps its data at full resolution for numeric queries; min-max decimation only governs how that data is reduced to fit the plot without hiding its extremes.
Merobix reads your field devices into a cloud SCADA - the real thing behind these terms, live in days from any browser.