Machine learning approach for risk-based inspection screening assessment

Abstract

Risk-based inspection (RBI) screening assessment is used to identify equipment that makes a significant contribution to the system's total risk of failure (RoF), so that the RBI detailed assessment can focus on analyzing higher-risk equipment. Due to its qualitative nature and high dependency on sound engineering judgment, screening assessment is vulnerable to human biases and errors, and thus subject to output variability and threatens the integrity of the assets. This paper attempts to tackle these challenges by utilizing a machine learning approach to conduct screening assessment. A case study using a dataset of RBI assessment for oil and gas production and processing units is provided, to illustrate the development of an intelligent system, based on a machine learning model for performing RBI screening assessment. The best performing model achieves accuracy and precision of 92.33% and 84.58%, respectively. A comparative analysis between the performance of the intelligent system and the conventional assessment is performed to examine the benefits of applying the machine learning approach in the RBI screening assessment. The result shows that the application of the machine learning approach potentially improves the quality of the conventional RBI screening assessment output by reducing output variability and increasing accuracy and precision.