Model based estimation of uncertainty for shock calibration of accelerometers.

In contrast to strain gauges, for instance, accelerometers actually belong to the group of reusable sensors; yet, under extreme conditions, especially in tests with shock load, often single sensors show irreversible changes in performance, are damaged or totally destroyed. Thus it is recommended to recalibrate all accelerometers after each test. An efficient, time-saving method and equipment for such recalibrations was developed by the Federal Institute for Materials Research an Testing (BAM). This apparatus comprises a Hopkinson-bar and a pendulum striker. A precision laser-doppler-vibrometer serves as reference. Its uncertainty is negligible compared with other uncertainties especially those related to parameters of spectral evaluation (sampling rate, shape of excitation pulse, width of window) and to the quality of the sensor signal (noise, drift and offset). In order to estimate the uncertainty of calibration results the real sensor is numerically simulated by a damped spring-mass-system. The response of this virtual sensor is superimposed by artificial noise, drift and offset of selectable quantity. The result is compared with the analytically calculated frequency response of the sensor model. This simple method enables a realistic estimation of uncertainty with regard to amplitude and phase as a function of frequency for given test conditions and settings.