InsightFab
Knowledge Base/CUSUM and EWMA Control Charts: Powerful Tools for Detecting Small Shifts
SPC8 min read

CUSUM and EWMA Control Charts: Powerful Tools for Detecting Small Shifts

Traditional Shewhart control charts quickly detect large shifts, but often miss subtle process drifts. CUSUM and EWMA are specifically designed to monitor and detect these small shifts; this article explains their application and interpretation.

Scenario

Your X-bar control chart looks perfectly normal, with no points out of control limits and no Western Electric Rules triggered. However, the customer return rate has quietly risen by 0.3%, 0.3%, 0.3% each month... Three months later, the customer calls to complain.

The process is indeed drifting; it's just that the Shewhart control chart doesn't see it.

Blind Spots of Traditional Control Charts

Shewhart control charts (X-bar, I-MR) only consider the current point each time. They are very sensitive to sudden large shifts (catching ±3σ immediately) but are slow to react to slow, continuous small shifts (e.g., a 0.5σ shift per batch)—it takes many points to trigger an alarm.

CUSUM Control Chart

CUSUM = Cumulative Sum

Method: Cumulatively sum the difference between each point and the target value. If the process continuously shifts in the same direction, the cumulative sum will keep increasing, triggering an alarm when it exceeds the decision limits.

Key Parameters:

  • K (Reference Value): Usually set to 0.5σ, to detect shifts of 1σ or more.
  • H (Decision Limit): Usually set to 5σ, to control the false alarm rate.

Advantages:

  • Very sensitive to sustained small shifts.
  • Can precisely pinpoint when the process began to change.

Disadvantages:

  • More difficult to interpret than Shewhart charts.
  • Less sensitive to sudden large shifts than Shewhart charts.

EWMA Control Chart

EWMA = Exponentially Weighted Moving Average

Method: Calculate a weighted average. Recent data has more weight, while older data has less weight, rather than only looking at the latest point each time.

Key Parameters:

  • λ (Smoothing Parameter): 0 < λ ≤ 1
- λ = 0.1: Emphasizes history, sensitive to small shifts, slow to large shifts.

- λ = 0.4: Balanced, a common setting.

- λ = 1: Equivalent to Shewhart, only considers the current point.

Advantages:

  • Easier to calculate and interpret than CUSUM.
  • Flexible λ adjustment.

Disadvantages:

  • Parameter selection affects performance, requiring adjustment based on the magnitude of the shift to be detected.

Comparison of Three Control Charts

CharacteristicShewhart (X-bar)CUSUMEWMA
Detect Large Shifts (> 2σ)FastestSlowModerate
Detect Small Shifts (0.5-1.5σ)SlowFastestFast
Calculation ComplexitySimpleModerateSimple
InterpretabilityHighLowModerate
Suitable for BeginnersYesNoYes

When to Use Which

Use Shewhart (General Cases): when the primary process risks are sudden anomalies or equipment failures.

Use CUSUM or EWMA (Advanced Cases):

  • when the process has a tendency for slow drift (e.g., tool wear, material batch differences).
  • when customers are highly sensitive to small shifts (e.g., medical, aerospace).
  • when the process is already confirmed stable, but increased monitoring sensitivity is desired.

Practical Recommendation: Run Shewhart + EWMA concurrently: one to catch major issues, the other to detect small drifts.

Golden Quote

"Shewhart is a smoke detector, CUSUM/EWMA is a carbon monoxide detector—both should be installed, because different dangers require different sensors."

Want to try it yourself?

Every tool mentioned in this article is available on InsightFab — just upload a CSV to analyze.

Go to Tools