Feature learning method of acoustic emission signal during single-grit scratching on BK7 in brittle regime

Abstract

Acoustic emission (AE) signal is released from a material as it undergoes deformation and fracture processes. AE signal provides enough information for monitoring of the interaction between tool and material. AE RMS, AE count rates, and frequency characteristics are always employed in AE-based monitoring techniques. Monitoring sensitivity is not satisfied because information implies in AE signal is lost or weaken to some extent for the present monitoring parameters. When a hard-brittle material is removed in brittle regime, burst-type AE (b-AE) associated with crack and chip is predominant compared with low-amplitude continuous-type AE. It will be more objective and physically meaningful if b-AE event is taken as a unit to be monitored. This paper is dedicated to the method of monitoring b-AE event in AE signal. Ignoring background noise, an AE signal can be looked as a convolution of b-AE events and corresponding coefficients. b-AE events correspond to brittle transient behaviors and include information of fracture mechanism. Coefficients represent intensity and location of crack and chip during removal processes. The two elements of the convolution contain different aspects of signal feature. Considering randomness of AE signals, the convolution form will be realized by self-learning from training data set of AE signals. Dictionary-learning algorithm for decaying oscillation modes of b-AE and sparse representation procedure of an AE signal are detailed in this paper based on the convex optimization theory. AE signals in brittle regime are acquired from tests of single diamond grit scratching on BK7. Simulation and experimental results verify the correctness of the feature learning method. According to the method proposed in this paper, b-AE can be objectively monitored without the interference of varying machining parameters.