PROBABILITY-DISTRIBUTION OF REPEATABILITY OF INDUSTRIAL ROBOTS

The accuracy with which an industrial robot brings the load to a position and holds it there is perhaps the most important characteristic of an industrial robot. Many researchers have consequently been interested in this field during recent years. A common method for characterizing an industrial robot's ability to return to a position is to use the terms "accuracy" and "repeatability," where accuracy characterizes the degree to which the actual measured value corresponds to a commanded value and repeatability the closeness of agreement between repeated measured values, under the same conditions, to the same commanded value (ISO definitions). The normal approximation is regularly used when calculating the repeatability. A test on this assumption for six different industrial robots is reported in this article. Two approaches for this test are used: one looks at the shape of the frequency function for the repeatability figures measured, and the second uses a chisquare test on the six data sets. The different tests show that there are small chances that the deviation of an industrial robot will follow a normal distribution. It seems to be a trend that the deviation has longer tails than the normal distribution. Simulation is used to elaborate on the consequences of the invalid assumption of normality in the definition of repeatability. The conclusion is that it is reasonable to use the normal approximation when there is no strong evidence that the deviation distribution is negatively skewed.