Yield Analysis

Yield is the most direct connection between quality and profitability. Every percentage point of yield loss translates to scrap, rework, or downgrading costs. A multi-step process with 99% yield at each of 10 steps has a rolled throughput yield of only 90.4% — nearly 1 in 10 units requires rework or is scrapped.

Yield analysis reveals hidden factories — the rework loops and sorting operations that consume resources without adding value. A process with 98% final yield might look acceptable, but if 15% of units required rework to achieve that yield, the hidden factory is consuming 15% of production capacity.

Accurate yield prediction requires knowing the process distribution relative to specifications. Overestimating yield (by assuming normality when the process is skewed) leads to production planning that underestimates rework burden and overcommits delivery capacity. Underestimating yield wastes capacity by scheduling unnecessary backup production.

EntropyStat enables more accurate yield predictions by computing the fraction conforming from the actual process distribution rather than a normal approximation. For multi-characteristic parts where each dimension has a different distribution shape, the EGDF estimates each characteristic's conformance rate independently and accurately.

Rolled throughput yield calculations are particularly sensitive to distributional accuracy because errors compound across steps. If the normality assumption overestimates yield by 0.5% at each of 10 process steps, the rolled throughput yield error is approximately 5% — which can translate to significant planning and cost errors.

The ELDF enhances yield analysis by detecting whether process variation comes from a single homogeneous population or from distinct subpopulations. If a process has 97% yield with two clusters — one at 99.5% yield and one at 90% yield — the improvement strategy is fundamentally different than if the 97% comes from a single broad distribution. EntropyStat's cluster detection provides this diagnostic information automatically.