On-Line non-intrusive Monitoring of Particulate Solid Materials in Gas Flowlines Using Acoustic Sensor and ML Techniques

Abstract

This paper describes initial steps towards developing a real-time quantitative particulate solids’ (sand) monitoring system for
Multiphase flowlines based on acoustic monitoring and machine learning techniques. The concentration and the velocity of the
solids were varied during experimental trials. A conventional contact microphone mounted externally to a production flowline bend was used for recording the emitted acoustic signal. Features extracted from the signal were used as input to Time Delay
Neural Network (TDNN) with solids concentration and velocity label to a training set. The TDNN achieved low values of
normalised root mean square error (NRMSE) for all the models compared to the classical neural network.