Neural Net Identification of Flow Regime using Band Spectra of Flow Generated Sound

Abstract

Multiphase production log interpretation requires that the flow regime along hole in the wellbore is known. Flow regime is the cased-hole analogue of Iithology. Knowledge of the flow regime will help to interpret tool signals, will help to evaluate the flow rate on a per phase basis, and will reduce post processing load by at Ieast a factor of 10. Flow regime can be classified correctly by a neural net in up to 87% of all cases using 1/3 octave band spectra of flow generated sound. A neural net trained on commercially available tool data ('noise cuts') appears to be too sensitive to the wellbore inclination. Hence, application of automated neural net interpretation of noise logs requires a new generation of noise logging tools.