A Study On Source Separation Of Machining Acoustic Emission Signal Based On Independent Component Analysis

In machining process monitoring systems based on acoustic emission (AE) sensor, sensing signal is the mixture of several AE sources that are related to different monitoring objectives, i.e., source from tool breakage (concerning tool condition), source from chip fracture (concerning chip condition), source from shear slip deformation (concerning work piece condition) and source of noise. The signal processing technologies in previous process monitoring systems are mostly for the purpose of extracting AE source concerning only one objective condition, and achieve this purpose by considering other sources as noises and eliminating them according to their features in frequency domain. Such strategy is defective not only in that it obstructs multi-objective process monitoring by filtering valuable information concerning other objective conditions, but also in that it proves less effective when the desired source share common frequency bands with the undesired. In this thesis, an Independent Component Analysis (ICA) technique is employed as a possible solution to the problems raised above. The main contents are as below.(1) The theoretical model, optimization criterions and typical algorithms of ICA are illustrated. The solvability conditions and uncertainty of separation results are introduced.(2) An ICA system for AE process monitoring is established. The hardware of the system consists of AE sensors, signal conditioning device and DAQ card. The software of the system is comprised of signal acquisition/playback module, spectral analysis module and ICA module. (3) Experiments are performed in which AE signals are recorded during the fracturing of YG6X and HT250 test pieces, based on which multi-channel machining AE signals are simulated by linearly combining three condition sources, namely, source from tool breakage, source from chip fracture, and source of noise. The established ICA system is employed in the attempt to separate these sources from the simulated monitoring signals, and the separation result is evaluated.(4) The signal features of both source AE signals and monitoring AE signals are analyzed, based on which a pre-processing technique is proposed to treat additive noise of sensing system, and a post-processing technique is proposed to solve the problem of source identification.The study of this thesis demonstrates that ICA is capable of simultaneously separating from mixed AE monitoring signals sources concerning different monitoring objective, even if the sources share common frequency band. This study not only brings new solution for machining process monitoring, but also extends the application range of ICA.