AbstractThe Research described in this Dissertation investigated monitoring in petroleum engineering and created new models for intelligent monitoring to alleviate risk. As part of that work, a defined set of rules was created for an intelligent monitoring system based on trend analysis of historic data. That system would predict probable future conditions of trait ‘states’ and so predict potentially catastrophic situations.
Initially, work investigated the state of the art and literature. The literature review was published, and research gaps were identified. One gap was a need for an intelligent monitoring system for the petroleum engineering industry to predict potential situations that may lead to catastrophic failures. That idea was presented to the Petroleum Institute, and they approved the direction of the work. Fault tree analysis (FTA) was conducted, and a fault tree model of a crude oil distillation unit was created to investigate potential hazards. This demonstrated that it was possible to create a simple monitoring system to identify faults. Failure mode and effect analysis (FMEA) for crude oil distillation was then investigated.
A Visual Basic program was created to complement Boolean logic calculations for the fault tree that identified faults in a component.
Multi-sensor data fusion was used and resulted in improved estimates. Pattern recognition was applied to historical data to match known situations. Using multiple sensors also improved observability.
Algorithms were produced and models tested based on lessons learned from historic catastrophic incidents. Catastrophes are normally attributed to combinations of several minor errors or failures, and seldom because of a single-point failure. The Fault Tree system predicted possible fault conditions and catastrophic failure.
A new prototype system was investigated that could make predictions from the patterns in events leading to failure. The prototype monitoring system was developed and applied to a distillation processing unit, utilising a fault tree model to forecast the current state of the unit and identify the root causes of failure. The prototype monitoring system enabled the identification of root causes but had limitations due to its inability to forecast future trait states. A new intelligent monitoring system was introduced that makes accurate predictions using sensor-driven data. The new intelligent system utilised algorithms generated by sensors and used a machine-learning algorithm to determine the most appropriate model.
|Date of Award||20 Mar 2023|
|Supervisor||David Sanders (Supervisor), Giles Tewkesbury (Supervisor) & Mohamed Hassan Sayed (Supervisor)|