Two phase medium identification using ultrasonic tomography technique

Abstract

The use of tomographic techniques has been widely used in pipeline and oil industry. These techniques have potential applications for flow visualization and measurement in producing wells. One of the important processes is in multiphase characterization; that serve in monitoring, measuring or controlling industrial processes. Multiphase represents the condition of more than one medium phase. The identification for two phase medium is carried out in this research. Research on industrial tomography process revolved in obtaining estimated images in cross section of a pipe or vessel containing or carrying the substances in the process. Ultrasonic tomography technique is one of the categories in process tomography. A simple tomography system can be built by mounting a number of sensors around the circumference of a horizontal pipe. In this research, sixteen pairs of 40 kHz ultrasonic sensor have been non-invasively mounted around the pipe. The characteristic of the sensor is an important factor that needs to be considered. Grease was used as the coupling material to mount these ultrasonic sensors. The output data from the sensors were processed to obtain the information of the spatial distributions of liquid and gas in an experimental column. Time of Flag (TOF) method has been chosen to extract the data from the ultrasonic signals. Analysis on the transducers’ signals has been carried out to distinguish the observation time between the longitunal (straight) propagation waves and the Lamb waves. The information obtained from the observation time is useful for further development of the images. The Linear Back Projection (LBP) algorithm has been applied to obtain concentration profiles or also called tomograms. The results obtained through LBP were filtered using Gaussian Filter and Enhancement Filter Technique. From the filtered images, further development was made by extracting features information such as mean, standard deviation, skewness, kurtosis, energy and entropy. Two approaches were applied for classification purposes using single and combination of features. Comparison between K-Nearest Neighbor (k-NN) and Linear Discriminant Analysis (LDA) classifiers have been made. From the observation, non-linear classifier (k-NN) produced a better result over linear classifier (LDA). Furthermore, it has been found that combination of features gives better performance over single feature classification.