A cooling load that adds up the nameplate heat of every machine assumes the whole process runs at the same time.
Few processes do. A diversity factor corrects for it: the largest load the plant actually sees at one moment, divided by the sum of each item's own peak. It is a number below 1 (electrical engineers call this ratio the coincidence factor), and it pulls in the opposite direction to a margin.
The two get mixed up because both change the design load. Margins added at each successive step compound: four 10% margins give 1.1⁴ ≈ 1.46, about 46% on the original figure. Diversity removes load that will never appear at once. A design with generous margins and no diversity carries both errors in the same direction.
The factor should come from the process schedule. Which machines run on which shift, which rooms are cleaned while others produce, how a batch moves from granulation to drying to compression. If the schedule shows the fluid bed dryer and the coating pan never run in the same hour, the chiller does not need to carry both peaks.
Where it is applied matters as much as its value. A room's own AHU has to meet that room's peak whenever its machine runs, so there is no diversity there. It belongs on shared plant: the chiller, the main switchboard, a central utility.
Write the schedule and the factor into the design basis. When production adds a third shift, anyone can see which assumption changed.



