Production actuals

What are production actuals?

Production actuals are the recorded data from an actual production run: measured cycle time, booked hours, scrap, and other outcomes logged as a part was actually made, typically in a shop's ERP or MES system. They look like the ideal ground truth for improving a pre-calculation or training a predictive model: the machine made the part, and the recorded duration is real.

But real is not the same as representative. A longer observed program duration may indicate an inaccurate estimate, or it may reflect maintenance, operator intervention, shift effects, tool problems or other operational disruption. Recent research into CNC cycle-time anomaly analysis explicitly filters exactly these kinds of disturbances before computing actual program-duration distributions for further analysis.¹

Not every outlier should be discarded

That does not mean every unusual value should be deleted. A recurring tool failure may itself reveal that a process choice is not robust, and a consistently difficult setup may point to a genuine structural problem rather than noise to be discarded. The useful distinction is closer to representative variation versus explained exception versus unexplained anomaly than to a simple good-data/bad-data split.

For manufacturing AI and for post-calculation alike, production actuals need context and provenance: a measured result is evidence of what happened once, and learning what is likely to happen next requires understanding why.

Are production actuals the same as post-calculation?

Production actuals are the raw recorded data; post-calculation is the activity of comparing that data against the original estimate.

Can production actuals be used directly to train a predictive model?

Only with care: disturbances such as maintenance or operator intervention need to be distinguished from representative process variation first.

Why might two recordings of the same operation differ?

Material variation, tool wear, operator intervention, or a interruption can all affect a single recorded run without indicating an incorrect estimate.