On-line drill system parameter estimation and hazardous event detection

The project aims to develop statistical methods for automatic detection of hazardous events in oil and gas drilling operations. Initially, only two particular hazardous events are considered. The first is called washout and it means the appearance of a hole in the drill pipe which may compromise the safety and efficiency of the operation as well as equipment integrity. The second event is called mud loss and it means the loss of drill fluid due to a leakage in the drill system to the surrounding rock formation. As the project progresses, more complex scenarios may be considered, involving multiphase flow, influx of gas from the formation, accumulation of rock cuttings around the drill pipe, wear of the drill bit, plugged bit nozzles, the degradation of the motor, etc. or extending the initially one dimensional model to two or three dimensions for increased accuracy.

The following initial objectives of the project have been identified:

Surrogate model

A forward model for the pressure, density, and the velocity of the fluid in the drill system is available, but due to proprietary or theoretical reasons, this model may have to be regarded as a black box. The evaluation of this model involves solving a system of partial differential equations (PDE) which, even with the most efficient solver, may be computationally too expensive for many modern inference methods that require large numbers of model evaluations. One task is to study whether computationally less expensive surrogate models could be derived. They should be simple enough to enable fast evaluation but at the same time accurate enough to capture the aspects critical to the successful detection of the hazardous events. Preliminary studies suggest that suitable surrogate models could be obtained by considering an electrical circuit analogue of the drill system.

System parameter estimation

In addition to the hazardous events, there are other non-hazardous and unexpected (i.e. non-deterministic) events in the drill system that affect its behaviour, most importantly the changes of rock formation and the characteristics of the drill bit. For the successful detection of the hazardous events, it is necessary to correctly identify these non-hazardous phenomena to know accurately how the observed data would behave in the absence of washout or mud loss. Preliminary studies suggest that this could be done using some highly efficient approximate computational methods for hidden Markov models (HMM) such as the nonlinear variations of the Kalman filter. Because reasonable prior knowledge of the drill system is available, the Bayesian approach to inference is adopted.

Hazardous event detection

Provided that the parameters of the drill system have been estimated accurately enough, it should be possible to detect the hazardous events. Primarily Bayesian methods are considered. For both types of hazardous events it is expected that the event can only be detected retrospectively yet soon enough after the beginning of the event to avoid any severe consequences. In the context of HMM the estimation of events occurring before the time of the latest observations is known as smoothing. Because the detection of the hazardous events should occur as soon as possible, it is likely that linearisation based methods, such as the variations of Kalman filter algorithms are not accurate enough and instead sampling based sequential Monte Carlo methods are needed. Another algorithmic smoothing framework whose suitability for detecting the hazardous events should be studied is the so called 4D-VAR that has been applied in meteorological applications.

Multilevel inference

Fewer compromises in the computational accuracy have to be made with sampling (Monte Carlo) based inference methods, but this comes with the cost of substantially larger number of forward model evaluations. The surrogate model (see above) addresses this problem, but again it comes with the cost of compromising the accuracy due to the use of approximate models. A theoretically sound approach to compensate for this error is the use of multilevel (Monte Carlo) methods. One goal in this project is to develop a multilevel algorithm that successfully combines the results of the above-mentioned sub-problems, i.e. surrogate model, system parameter estimation and hazardous event detection in a way that allows on-line inference. This is essential if more complicated two- or three-dimensional PDE models of multiphase flow in the drilling system are to be considered.

Formal applications should be submitted via the University of Bath’s online application form https://www.bath.ac.uk/samis/urd/sits.urd/run/siw_ipp_lgn.login?process=siw_ipp_app&code1=RDUMA-FP01&code2=0012

More information on applying for a PhD at Bath may be found here http://www.bath.ac.uk/guides/apply-for-a-doctorate/

Anticipated start date: 1 October 2018