A new statistical model for the calibration of force sensors
Charles P. Reeve · 1988
The National Bureau of Standards has been calibrating force sensors for many years.The objective in these calibrations is to determine the functional relationship between the applied load and the sensor response.In a typical calibration several runs are made in which identical sequences of known loads are applied to the sensor.The sensor is rotated in the loading machine between runs.The previous method of analysis incorporated a quadratic polynomial model which was fit to the pooled data.The new method presented here fits separate polynomials to the mean data and between-run differences.The 'best' degrees of these polynomials are automatically determined by algorithms which incorporate statistical tests.As a result, error contributions from several sources are quantized.Methods for computing confidence intervals for the 'true' sensor response and for a new observation are given, and methods of inverse prediction (measurement of an unknown load) based on these intervals are illustrated.