Particle filter based noise removal method for acoustic emission signals

Abstract

This paper discusses the application of a statistical noise removal technique – Rao–Blackwellised particle filter (RBPF) for signal to noise ratio (SNR) enhancement of acoustic emission (AE) signals. RBPF is a recursive Bayesian method for dynamic system state estimation. Compared with other signal filtering methods, RBPF offers the advantage of broad-band signal cleansing by directly modeling the internal dynamics of the concerned physical system and statistical characteristics of the signal noise. In doing so, signal filtering can be related to the dynamic characteristics of the underlying physical system, rather than a purely mathematical operation. RBPF also outperforms a few other statistical signal filtering methods such as Kalman filter, with the ability of handling nonlinear system and non-Gaussian noise problems. Another feature of RBPF is its ability to allow real-time on-board signal processing. In this paper, moment tensor analysis was performed first to generate simulated baseline AE signals. The simulated AE signal was subsequently superimposed with noise to demonstrate the effectiveness of the RBPF method in filtering noise in the AE signals. The results show that the performance of the RBPF in SNR enhancement is very promising. Finally, RBPF is also applied to real AE data obtained from experimental tests and apparent improvement to the SNR in AE feature study is observed.