Data-driven Fault Diagnosis For Offshore Managed Pressure Drilling Process

The petroleum exploration in offshore is of high-risk and high-investment.Due to the harsh operating environment of offshore drilling,the varied geological conditions below the seabed and the large and complex drilling control system,a timely and accurate fault diagnosis method is significant and urgent for reducing the risk of Managed Pressure Drilling(MPD).The drilling process and typical faults about MPD system that is widely used in offshore drilling are simulated based on a commercial high-fidelity simulator.To deal with the frequent switching and uncertainty of set-point in MPD system,slow feature analysis is introduced to detect the drilling faults.With the aim to prompt the accuracy of fault diagnosis,a method based on RIMER(Belief Rule-base Inference Methodology using the Evidential Reasoning Approach)is constructed to fuse all results come from divergent fault diagnosis method.The main work of this dissertation is summarized as follows:Firstly,in order to address the problem that the existing MPD hydraulic model is too simple to accurately simulate the real drilling process,the MPD simulation platform is built based on high-fidelity drilling software Drillbench.Three scenarios are designed to simulate some common phenomenon in actual MPD drilling process.Scenario one is steady operating point changes caused by the increase of well depth.Scenario two is a Kick in the wellbore.Scenario three is a performance deteriorating of the controller.The drilling data are collected and used for data-based fault diagnosis.Secondly,the characteristics of MPD control system are frequent switching and uncertainty of set-point.Those are big challenges in fault detection for MPD system.To cope with this problem,an algorithm initially utilized in process industries,slow feature analysis(SFA),is introduced to detect abnormal drilling incidents.Comprehensive evaluating those four indices provided by SFA,the driller could successfully differentiate controllable process changes,e.g.due to set-point changes,from truly abnormal events that should be considered faults.Furthermore,the evaluation of controller performance is provided for drilling operator.The simulation studies show that the SFA-based method could correctly recognize real drilling incidents,minimize false alarm and significantly reduce the non-productive time caused by misjudgments.Finally,the MPD system in offshore drilling is a large safety-critical system.To prompt the accuracy of fault diagnosis,an algorithm combining results of various methods is proposed.For the issue of existing fusion method,for instance,they could not deal with various uncertainties and update the prior knowledge about the fusion algorithm.An intelligent decision fusion approach comes up based on RIMER to fuse the results come from various fault diagnosis methods.In order to calculate the belief degree in belief rule-base and set individual matching degree at evidential reasoning process,the methodologies are successfully conceived based on Bayesian formula and confusion matrix.The simulation studies show that the RIMER-based decision fusion approach could improve the accuracy of fault diagnosis for MPD system.