Optical sensors are increasingly being employed for quality assurance
tasks. Such tasks require metrologically safe tools, which means that
the measuring uncertainty of the employed sensors has to be quantified
, compared and minimized. The measuring uncertainties of optical trian
gulation sensors have been found to increase up to a factor of ten, if
typical industrially used surfaces, such as steel, are probed under u
nfortunate angles. This paper proposes three techniques for reducing e
rrors incurred in measurements of triangulation sensors when the surfa
ce of the specimen is not perpendicular to the incident light beam, an
d when the characteristics of the specimen surface vary. The relative
merits of the three techniques are discussed, enabling choices to be m
ade in respect of sensor cost, size of error reduction and speed of op
eration. The third and most robust technique involves several detector
s viewing from different directions in combination with the use of pri
ncipal component analysis. Principal component analysis is a statistic
al method to obtain a robust average of all signals of the several det
ectors.