When your 8-cavity mold produces housings with different wall thicknesses from each cavity, a simple variance study is not enough. Here is how crossed COV designs reveal the full picture.
Advanced Technique · 2025-06-01 · 11 min · OpEx Excellence Team
The Multi-Cavity Challenge
Injection molding is the backbone of appliance housing production. A single mold might have 4, 8, or even 16 cavities producing identical parts simultaneously. But 'identical' is the wrong word. Each cavity has its own flow path, its own cooling channel geometry, and its own steel temperature profile. The result is systematic variation between cavities that gets mixed with random variation within each cavity. Traditional quality charts that average across all cavities can mask significant cavity-to-cavity differences. A housing might have an average wall thickness of 2.50mm with a standard deviation of 0.08mm, but if cavity 3 consistently produces parts at 2.42mm while cavity 7 produces at 2.58mm, the average tells a misleading story.
Crossed vs. Nested COV Designs
A crossed COV design is used when every combination of factor levels exists in the data. For a 4-cavity mold sampled across 3 shifts with 5 parts per combination, you have 4 x 3 x 5 = 60 measurements with every cavity appearing in every shift. This lets you estimate cavity effects, shift effects, and the cavity-by-shift interaction. A nested design is used when some factor levels are unique to certain groups. If each shift uses a different set of raw material batches, those batches are nested within shifts and cannot be compared directly across shifts.
Always use a crossed design when possible. It provides more information per data point because you can estimate interactions. Reserve nested designs for situations where crossing is physically impossible.
Variance Decomposition: A Real Example
An appliance housing manufacturer ran a crossed COV on a critical wall thickness dimension across an 8-cavity mold, 3 shifts, and 2 material lots. The variance decomposition revealed: Cavity-to-cavity variation accounted for 58% of total variance, within-cavity (part-to-part) variation was 22%, shift-to-shift variation was 12%, material lot variation was 5%, and measurement system contribution was 3%. The dominant source, cavity-to-cavity, pointed directly to mold maintenance. Inspection of the cavities revealed that cooling channels in cavities 2, 5, and 7 had partial blockages from scale buildup, creating hot spots that changed the flow and packing behavior.
Using COV Results to Drive Improvement
The beauty of COV is that it tells you where to invest your improvement effort. With 58% of variation coming from cavity differences, the highest-leverage action is to equalize cavities (cleaning cooling channels, checking vent depth, verifying gate dimensions). Investing in tighter material specifications (5% of variation) or additional operator training (part of the 12% shift effect) would have minimal impact by comparison. After cleaning the cooling channels and equalizing gate dimensions, a follow-up COV showed cavity-to-cavity variation dropped to 18% of total variance. Overall variation reduced by 54% without changing any material or design specifications.
- Before: cavity-to-cavity = 58% of total variation, frequent out-of-spec parts
- After cooling channel maintenance: cavity-to-cavity = 18% of variation, 54% reduction in overall variation
- Investment: 16 hours of mold maintenance
- Scrap reduction: 64%
Sanders Charts for Visual Analysis
Standard control charts struggle with multi-cavity data because they either pool all cavities (hiding systematic differences) or require a separate chart for each cavity (creating information overload). Sanders charts solve this by displaying variance components visually, showing the contribution of each factor and their interactions in a single view. Combined with layered control charts that can filter by cavity, shift, or lot, you get a complete picture of where variation lives and whether it changes over time.
Frequently Asked Questions
What is a crossed COV design?
A crossed design means every combination of factor levels is represented in the data. For example, if you study 4 cavities across 3 shifts, every cavity appears in every shift. This allows you to estimate not just the individual effects of cavity and shift, but also their interaction.
How many samples do I need for a COV study?
A minimum of 2-3 replicates per factor level combination is needed. For an 8-cavity mold studied across 3 shifts, you would want at least 2-3 parts per cavity per shift, giving 48-72 total measurements. More replicates improve the precision of variance estimates.
Can COV results guide DOE factor selection?
Absolutely. COV identifies which sources of variation are largest. Those sources often point directly to factors worth investigating in a subsequent DOE. If cavity-to-cavity variation is dominant, the DOE should include factors that differ between cavities (cooling, venting, gate size).
Run your own COV analysis
Load multi-cavity data into the COV tool and see the variance decomposition, Sanders charts, and layered control charts. Use the built-in examples or upload your own data.
Get started →Tags: COV, injection molding, multi-cavity, appliance housing, variance decomposition, crossed design