Signal analysis of the accelerometer with advanced methods

This study aimed to identify non-stationary signals produced by accelerometer used for vibration analysis and prediction. First, the experimental data set was generated by reading the accelerometer values of all three axes (x, y, z) from the accelerometer on the inertial measurement unit. This experimental data obtained through the inertial measurement unit contains various errors. These errors can be studied in two separate groups as deterministic (systematic) and stochastic (random). In this study, the methods are operated in the time domain, are preferred. The reason for this is to avoid methodological errors (such that Fourier transform's boundary value and smearing problems) that are caused by the conversion techniques used in transformation to the frequency domain. These errors increase in each conversion and in the reverse-conversion operation. Firstly, Kalman filter is examined for two different models and speed and position information are estimated from experimental data. In the mean time, by examining convergence properties of the filter, norms of Kalman gain and state error covariance matrix are converged to certain value, that proves the minimization of the estimation error. Characterization of stochastic errors was performed by calculating the time varying variance by Alllan variance method.