A triaxial accelerometer calibration technique that evades the problems of the conventional calibration method of aligning with gravity is proposed in this paper. It is based on the principle that the vector sum of acceleration from three sensing axes should be equal to the gravity vector. The method requires the accelerometer to be oriented and stationary in 6 different ways to solve for the 3 scale factors and 3 offsets. The Newton-Raphson method was employed to solve the non-linear equations in order to obtain the scale factors and offsets. The iterative process was fast, with an average of 5 iterations required to solve the system of equations. The accuracy of the derived scale factors and offsets were determined by using them to calculate the gravity vector magnitude using the triaxial accelerometer to measure gravity. The triaxial accelerometer was used to measure gravity 264 times to determine the accuracy of the 44 acceptable sets of scale factors and offsets derived from the calibrations (gravity was assumed to equal 9.8000 ms-2 during the calibration). It was found that the best calibration calculated the gravity vector magnitude to 9.8156 ± 0.4294 ms-2. This equates to a maximum of 4.5% error in terms of a constant acceleration measurement. Because of the principle behind this method, it has the disadvantage that noise/error in only one axis will cause an inaccurate determination of all the scale factors and offsets.
Measurement characterisation of error has been performed from synthetic acceleration data. A variety of kinematics computation techniques were investigated and then applied. Data acquisition specifications, including analogue to digital conversion (ADC) resolution (6, 8, 10, 12 and 14 bit) and sampling rate (100, 150, 200, 250, 300, 375, 500 and 600 Hz), have been varied to investigate the effect on the accuracy of kinematics data with respect to short (20m) displacements. The magnitude of the errors in acceleration, velocity and position are reported for the simulated data. Also, error reduction techniques, including over-sampling and oversampling/multiple point averaging/reduced data transmission, were implemented to examine their effectiveness. The results of this investigation show that MEMS accelerometers are subjected to significant amount of errors, and require accurate calibration and characterisation of errors. Error reduction techniques are also necessary to ensure accurate computation of kinematics information.