Wright Williams & Kelly
Factory Explorer® v2.2
Summary: Capacity analysis predictions for start rates and line yields are biased when using a non-constant distribution for product lotsize. This error is fixed in Factory Explorer® v2.3.
Details: When a non-constant distribution (e.g. uniform, triangular, etc.) is specified for a product's lotsize, Factory Explorer®'s capacity analysis prediction of start rate will be biased low, and line yield will be biased on the high side. This error is most noticeable when a model does not include scrap, as the resulting line yield prediction can exceed 100%. The error is caused by the fact that ExpectedValue(X) is generally less than ExpectedValue(ceiling(X)), where X is a non-negative random variable. When using a distribution for product lotsize, Factory Explorer® automatically rounds any non-integral lotsize up to the next larger integer number of units. This rounding was taken into account when computing product flows and throughput rates, but not when computing predicted start rates, and hence start rates were biased on the low side. Factory Explorer® computes line yield by dividing throughput rate by start rate, and hence the resulting line yield was biased on the high side. The magnitude of the error depends on the average lot size, but for 25 unit lots it is on the order of 5%.
Workaround: Use a constant distribution for product lot size. For capacity analysis, which deals in expected values, this restriction should not be a problem. If random lot sizes are desired for simulation (to see the effect of this variability, say), use a release list file, and in that file generate the number of units contained in each release lot randomly.