A BANK OF KALMAN FILTERS FOR FAILURE DETECTION USING ACOUSTIC EMISSION SIGNALS

An acoustic emission signal is represented by a series of decaying bursts caracterized by random amplitudes, varying times of ocurrences and characteristic decay times. If the decay time τ were known we would apply a Kalman filter in order to obtain the time ocurrence estimation. When the value of τ is not known a priori we propose a finite set of possible values and a Kalman filter is designed for each model. A decision test is used to select the best filter of the bank. Since we can associate the decay time to the distance between the failure and the sensor, the method allow us to localize the place where the failure is.