|Ph.D Student||Berman Arie|
|Subject||Improving Linear and N.L. Estimation Methods by Employing|
Validation Gate Tests During the Prediction
Process in Robust Tracking/Sensing Time-
Varying parameter Systems
|Department||Department of Mechanical Engineering||Supervisor||Professor Emeritus Yehoshua Dayan|
In many autonomous systems such as guided vehicles the obtained measurements are accompanied by noise and possible errors, which may lead to wrong decisions of the control and guidance system. In this work an improved method of Kalman filter with validation gates is applied for filtering unexpected high measurement noise due to clutter, glittering, or shaking of the measuring system, featuring minimum computational effort and minimum time. Optimal variable validation gates around the predicted values of the signal states - the Autonomous Guided Vehicle or its target position - reduce the unexpected noise effect. Values of measurements out of these gates are not considered and the integration of the prediction model for the tracked signal continues until a new measurement is received within the gate. The dimensions of the validation gates are determined by filter predictions error and normal noise variances of the measurement. The Kalman filter gain is related to the prediction error variance, to the probability of the unexpected high measurement noise and to its approximated covariances. From the results of the numerical examples, it is concluded that the presented algorithm is superior to other known alternative simple techniques. Convergence analysis for the filtering algorithm of a linear target tracker, using variable validation gates, is presented.
Another application of the new method is presented: robust tracking of time varying signals when there are abrupt, sudden random changes in the system parameters, the sensor gains or the transducers. These changes may occur due to hazardous environmental conditions (vibrations, temperature or pressure cycling) or special jumping inputs. Robustness of the filter is achieved by introducing a parallel cooperative controller and utilizing a new nonlinear gain-tuning algorithm for adjusting the jump parameters. By the aid of this algorithm, the filter remains stable even if the varying parameters, having unknown statistics, are outside of the original linear stability region of the nominal values of these parameters. To limit the noise of the output, the gain-tuning process is applied only to the controller whose output is deviating more than the other from the predicted value of the tracked signal and only if the differences between the two parallel outputs of the filter are over a specified threshold. The predicted value of the filter is calculated by the new method - a Kalman filter with validation gates for the two parallel outputs.