The discrepancies between measurement data and an analytical nominal m
odel for a large gap magnetic suspension testbed is accounted for by a
n uncertainty model. The results show that uncertainty bounds correspo
nding to a combination of additive and diagonal input multiplicative u
ncertainty can be obtained directly by calculating the smallest norm o
f the difference between the measured and nominal model response. Use
of the identified uncertainty model allowed a strong correlation betwe
en design predictions and experimental results. In addition, robust co
ntrollers based on the experimentally derived uncertainty model show s
ignificant stability and performance improvements over controllers des
igned with assumed ad hoc uncertainty levels.