Can Control Charts Be Drawn for Only One Batch per Month per Variety? —— Five-Step Method for Selecting Three Types of Alternative Charts for Short-Cycle SPC

A quality engineer at a machining company has been staring at a control chart for two months, feeling frustrated. The machining center has to produce over a dozen different parts in rotation, with each variety being produced in a batch of two to three hundred pieces once a month, and then the setup is changed. According to the training manual, an X̄-R chart requires 4-5 consecutive pieces for each subgroup and at least 20-25 subgroups to calculate reliable control limits, which means over 100 pieces of data from the same variety, the same tooling, and the same parameters. However, in the short-cycle production environment, a variety might only have 200 pieces produced in a month, with tool changes and material batch changes in between, resulting in only a few dozen pieces of "consistent state" data. With insufficient data, the control limits can only be guessed—either too wide (missing alarms) or too narrow (false alarms).

The correct approach is not to abandon the control chart but to switch to a chart type that makes the data "comparable": by converting data from different varieties and specifications to a common baseline and plotting them on the same chart, the number of points and sensitivity can be restored.

1. Why Do Conventional X̄-R Charts Fail in Short-Cycle Environments?

The statistical foundation of control charts is to estimate the common cause variation using "within-group variation" and then use the movement of subgroup means to detect special causes. The premise for this logic to hold is that there is enough data from the same state.

An X̄-R chart requires 4-5 consecutive pieces for each subgroup and at least 20-25 subgroups, meaning over 100 pieces of data from the same variety, the same tooling, and the same parameters. In a short-cycle environment, a variety might only have one batch of 200 pieces per month, with tool changes and material batch changes in between, resulting in only a few dozen pieces of "consistent state" data. Insufficient data means that the control limits can only be guessed—either too wide (missing alarms) or too narrow (false alarms).

The correct response is not to give up on control charts but to switch to a chart type that makes the data "comparable": by converting data from different varieties and specifications to a common baseline and plotting them on the same chart, the number of points and sensitivity can be restored.

2. Three Types of Alternative Charts, Matched to Scenarios

First Type: I-MR Chart (Individuals-Moving Range Chart)

This is suitable for single-piece production, extremely slow processes, or critical characteristics where only one piece is retained each time, such as heat treatment batches, large castings, or workstations that produce only a few pieces per shift. It uses the moving range between adjacent points to estimate variation, eliminating the need to form subgroups. Accumulating 20-25 individual points is sufficient to create the chart. The trade-off is lower sensitivity compared to X̄-R charts, and the moving ranges between adjacent points are autocorrelated, so not all eight criteria can be applied for detecting out-of-control conditions. Generally, only "points out of bounds" and continuous trend rules are used.

Second Type: Standardized Z Control Chart

This is suitable for multiple varieties produced on the same equipment, using the same process and measuring instruments, but with different nominal dimensions. The method involves setting a baseline (historical mean or target value from the drawing) and a variation scale (historical standard deviation or a fixed proportion of the tolerance band) for each variety, then converting the actual measurement to Z = (actual value - baseline value) ÷ variation scale, and plotting all varieties on the same chart. This way, a dozen varieties can share the same control limits, points accumulate quickly, and comparability between varieties is established. The prerequisite is that the variation scales of these varieties are indeed similar, and the measuring instrument has sufficient resolution—this must be confirmed through MSA, typically requiring that the instrument variation does not exceed 30% of the total variation.

Third Type: DNOM Deviation Chart (Deviation from Nominal Value Chart)

This is suitable for parts of the same family with varying dimensions, such as a series of shafts with different lengths. The method involves using the nominal value from the drawing as the zero line and plotting the deviation (actual measurement - nominal value). It is simpler than the Z chart, as it does not require dividing by the standard deviation, making it easier for operators to accept. The trade-off is that it requires the tolerance bands of parts in the same family to be roughly the same; otherwise, the reasonable deviation range will differ, leading to distorted control limits.

In addition to these three types of charts, a practical approach is to freeze the control limits: after confirming process stability through a capability study, calculate a set of fixed control limits using historical data and freeze them for use. Subsequent batches only plot points without recalculating the limits until the process or tooling changes. This is suitable for products with small monthly batches but highly consistent processes, provided that the control limits are backed by capability study data rather than arbitrary guesses.

3. An Example

In a machining workshop, a machining center alternates between producing 12 different types of shafts, with each variety having a batch of about 300 pieces per month. The key characteristic is the outer diameter. The quality engineer first confirmed that these 12 varieties use the same equipment, process, material, and measuring instrument, with similar tolerance bands, and classified them as the same product family. Subsequently, a variation scale based on the median value from the drawing was set for each variety, and the actual outer diameter measurements were converted to Z values, which were then plotted on the same Z chart. Initially, the within-group variation of 20 consecutive pieces from each variety was used to estimate the combined standard deviation, and after accumulating 25 points, the control limits were calculated.

The chart ran for three weeks and reported two anomalies: one was a series of 7 consecutive points on one side after a tool change for a particular variety, and the other was an increase in variation due to higher hardness in a new material batch. These two issues were previously obscured in the mixed calculation chart but were now detected, leading to a genuine belief in the chart's effectiveness on the shop floor.

4. Five Steps to Implementation

Step 1: Grouping

Group the parts based on four conditions: equipment, process route, material, and measuring instrument. The tolerance bands within the same group should be similar. Unclear grouping is the primary reason for the failure of short-cycle SPC.

Step 2: Selecting the Chart Type

• For single-piece or extremely slow processes, choose the I-MR chart.
• For multiple varieties in the same family with similar variation scales, choose the Z chart.
• For parts of the same family with varying dimensions, choose the DNOM deviation chart.
• For products with highly stable processes that have passed capability studies, use frozen control limits.

Step 3: Estimating Parameters

The width of the control limits must come from the process's own variation and not be replaced by specification limits. When estimating combined limits for the same family, base it on within-group variation, and ensure there are at least 20-25 points to support the calculation. Confirm the measurement variation ratio using MSA first.

Step 4: Building the Chart and Tailoring Out-of-Control Rules

Select the appropriate out-of-control rules based on the chart type. Use continuous chain rules cautiously with I-MR charts. Clearly define the response path, responsible person, and response time for out-of-control conditions; otherwise, the chart will be ignored even if it reports issues.

Step 5: Rolling Recalculation and Version Management

Recalculate the control limits at fixed intervals (e.g., quarterly or every 25 points). When varieties are added or removed, process parameters change, or measuring instruments are replaced, freeze the old chart, rebuild it, and maintain version records in the documents.

5. Three Common Pitfalls

First: Using Control Limits as Specification Limits

The sources of these two lines are entirely different: specification limits come from customer requirements, while control limits come from the process's own variation. Plotting them on the same line means abandoning both stability assessment and conformance assessment.

Second: Forcing Data from Different Families Together

Combining control limits for parts with different tolerance bands, measuring instruments, and materials results in limits that are not applicable to any of the parts. It is better to group them more finely than to merge them for convenience.

Third: Applying the Same Out-of-Control Rules to Downgraded Charts

The sensitivity of I-MR and Z charts differs from that of X̄-R charts. Directly applying the eight criteria will lead to a surge in alarm frequency. If a few checks do not identify issues, the chart will be ignored.

Short-cycle environments need charts that understand their data, not just stricter charts. Correctly grouping, selecting the appropriate chart type, and accurately estimating parameters are essential for transforming control charts from "wall decorations" into effective "sentinels."

Having multiple varieties with small batches is not a reason to abandon control charts. Selecting the right chart type can effectively monitor short-cycle processes.

Knowledge code: 6.3.1

Version: v20260915

Author: QTank QTank is dedicated to providing systematic professional knowledge, methodologies, and practical tools to quality management practitioners, helping companies continuously improve their quality capabilities.