Predictive Modelling for Asset Availability using Artificial Intelligence

Reliability, Availability and Maintainability (RAM) studies have been performed on equipment, systems, and oil and gas production fields to predict availability targets using reliability block diagrams and equipment runtime statistical analysis. However, no integration of RAM analysis involving multiple fields can be found; multi-field data need to be updated manually, and there is no live data updating feature available, resulting in data inaccuracy and longer duration to complete RAM analysis. In this study-anchoring on the theme of integration and automation-the authors aim to improve completion time, increase the visibility of availability data for the production field and improve flexibility in data update. The main objective is to allow for accurate prediction of field availability for the following month and to expedite the correct intervention actions identification to meet the required target. The predictive model was developed utilizing Microsoft Azure Machine Learning and R Programming by utilizing availability data of field with Mean Absolute Error less than one percent. As part of machine learning improvement, it is recommended for the model to be expanded to include other fields with integration of more live data.

Keywords: Asset availability, Predictive model, Machine learning, Artificial intelligence.

Download PDF