A method to determine the uncertainties of solid particle erosion measurements utilizing machine learning techniques

Abstract

The ability to predict solid particle erosion is a vital task for the integrity management of pipelines carrying solid particles in many industries. Usually, the accuracy of predicted values is assessed by comparing them with the measured erosion of some cases where data is available. Thus, defining the uncertainties of the experimental data of erosion is crucial to evaluate the reliability of the predictions of solid particle erosion and consequently, the decisions made based on these predictions. The databases gathered by many investigators that include erosion data of various test cases have been considered for this uncertainty investigation. These cases include erosion rate data collected using the ultrasonic thickness measurement technique for gas-sand flows in 0.0508, 0.0762, and 0.1016-m elbow geometries with various particle sizes (75,150,300,550 μm) and air velocities vary in the range between 11 up to 202 m per seconds. This study aims to develop a framework for defining the uncertainties of erosion experiments. For this purpose, first, the uncertainties of repeated experiments were estimated and then analytical analysis was used to generalize these results to cases where repeated results were unavailable. Based on this idea, a correlation between erosion and the most effective parameters was developed. Then, machine learning algorithms were used to determine a correlation between the erosion rate and chosen parameters including material hardness, pipe size, particle size, and flow velocity. The framework of the current model/method is based on the combined results of analytical analysis and machine learning procedure, and the estimated uncertainties are presented by lower and upper bounds curves of erosion data.