Short answer: the basic Shewhart control chart catches sudden spikes — a point outside the limits — but is nearly blind to slow drift, which is exactly how brewing quality usually fails: gradual yeast aging, a slowly fouling heat exchanger, creeping dissolved oxygen. CUSUM and EWMA charts are built to surface that slow creep early, accumulating or smoothing small deviations into a clear signal long before any single point breaches a limit. And capability (Cp/Cpk) turns “we passed” into “how much margin do we actually have.” These are the SPC charts a brewery’s dashboard usually lacks.
This is part three of the brewing data-visualization series. The previous post’s conformance-over-time trend raised the question of slow drift; this post is the SPC that makes drift rigorous — and it builds directly on your existing QC control charts.
The control chart’s blind spot
The Shewhart control chart is rightly a brewery staple: plot a value with control limits, and a point outside them flags a problem. It’s excellent at what it does — catching a sudden shift, a one-off out-of-spec batch. But its alarm is a single point breaching a limit, and that makes it nearly blind to the failure mode brewing actually suffers most: slow drift. Yeast that ages a little each generation, a heat exchanger fouling over months, dissolved oxygen creeping up as a seal degrades — none of these trips a limit on any one batch, yet over twenty batches the process has moved meaningfully off target. The chart stays green while the beer slowly changes.
CUSUM — accumulate the small stuff
The cumulative-sum chart fixes the blind spot by adding up deviations from target. Each batch contributes its small distance from where it should be; a process truly on target wanders around zero, but a persistent small drift — even one far too small to notice batch-to-batch — accumulates into a clear, rising slope. CUSUM turns “twenty tiny nudges in the same direction” into one unmistakable line. For a brewery chasing a slow consistency problem, it often reveals the drift weeks earlier than the control chart, which is weeks of off-target beer you didn’t have to make.
EWMA — smooth to the underlying level
The exponentially weighted moving average is the gentler cousin: it averages recent batches with more weight on the latest, smoothing out single-batch noise to show the underlying level. Where CUSUM is sharp at detecting that a drift has started, EWMA is good at showing where the process currently sits once you’ve decided to track a parameter closely. Many breweries run EWMA on the values they care most about — attenuation, a key flavour compound, fill level — as a calmer companion to the noisy raw control chart.
Capability — how much margin do you actually have?
The third tool answers a question pass/fail never does: how close to the edge are we? Process capability compares your process spread to your spec width. Cp asks whether the variation even fits inside the spec; Cpk also accounts for whether you’re centred — a process can be tight but sitting near a limit, passing today on luck. A healthy Cpk means comfortable, centred margin; a low Cpk is an early warning that you’re one small drift away from failures, even if nothing has failed yet. It reframes quality from “did this batch pass” to “how robust is our process” — the same shift from snapshot to system that runs through process intelligence.
Where this breaks
The honest section. These tools assume a meaningful target and stable measurement — CUSUM drift is only real if your target is right and your lab method isn’t itself drifting; a creeping CUSUM can be a failing pH meter, not a failing process, so rule out measurement first. They reward over-reaction if misread — a CUSUM or EWMA signal means investigate, not immediately adjust; chasing every wiggle by tweaking the process (over-control) adds variation rather than removing it, a classic SPC trap. Capability needs enough data and rough normality — a Cpk from ten batches, or from a wildly non-normal distribution, is a number with false confidence. And more charts can overwhelm the floor — running CUSUM, EWMA and capability on every parameter buries the signal; pick the few parameters where slow drift genuinely hurts and chart those well, and label the unfamiliar charts so the team trusts them.
The bottom line
The Shewhart chart catches sudden failures; brewing mostly fails slowly, so a brewery’s SPC needs more. CUSUM accumulates small deviations into an early, obvious drift signal; EWMA smooths to the underlying level; and capability (Cp/Cpk) turns “we passed” into “how much margin we have.” Use them on the few parameters where slow drift actually costs you, rule out measurement drift before process drift, treat signals as cues to investigate rather than tweak, and label them for the floor. Next, drift of a different kind — flavour over weeks of ageing: shelf-life trajectories.
Frequently asked questions
What is the difference between a control chart, CUSUM and EWMA? A Shewhart control chart flags a single point outside the limits — good for sudden shifts, blind to slow drift. A CUSUM (cumulative sum) chart accumulates small deviations from target, so a persistent small drift builds into a clear slope long before any single point breaches a limit. An EWMA (exponentially weighted moving average) smooths recent points to reveal the underlying level. Both are built to catch the slow creep a basic control chart misses.
What is process capability (Cp and Cpk) in brewing? Process capability compares how tightly your process varies against how wide your spec is. Cp asks whether the spread fits inside the spec at all; Cpk also accounts for whether the process is centred on target. A high Cpk means comfortable, centred consistency; a low Cpk means you’re relying on luck to stay in spec even if no single batch has failed yet. It turns “we pass” into “how much margin do we have.”
Why does a brewery need more than a basic control chart? Because brewing quality usually fails slowly, not suddenly — gradual yeast drift, a slowly fouling heat exchanger, creeping dissolved oxygen. A Shewhart chart only alarms on a point outside the limits and can stay quiet through a months-long drift that ruins consistency. CUSUM, EWMA and capability analysis are designed to make that slow movement visible early enough to act.