Condition Monitoring and Management from Acoustic Emissions

This paper analyses acoustic emission energy signals acquired under mixed load conditions with one induced fault. With Mean field independent components analysis is applied to an observation matrix build from successive acoustic emission energy revolution signals. The paper presents novel results that provide remarkable automatic grouping of the observed signals equivalent to the grouping obtained by human experts. It is assumed that the observed signals are a non-negative mixture of the hidden (non-observable) non-negative acoustic energy source signals. The mean field independent component analysis incorporates those constraints and the estimates of the hidden signals are meaningful compared to the known conditions and changes in the experiment. Most important is the estimate of the load independent wear profile due to the induced fault. The strength of this signature increases as the load progress and disappear as the induced fault is removed – this result has not been achieved with classical independent components analysis or principal components analysis.