Removable Tensor Strainmeter and Vector Tiltmeter System for Use With Forward and Inverse Methods for Characterizing Deformation During CO2 Injection

Abstract:

Injecting fluids into a well deforms the enveloping rocks in a complex pattern that increases in magnitude and expands outward with time. While this evolving strain field creates space needed to store these fluids, it can also signal problems. Fault slip occurs when stresses caused by injection reach a critical value, and maintaining stresses below a critical stress state is important for limiting the risk of faulting and subsequent leakage. Since it is impossible to measure stresses directly, the approach is to measure displacement or strain, and then calculate stress change. The geodetic research community has developed borehole strainmeters capable of measuring the horizontal strain tensor with high resolution (>1 nanostrain), but these require permanent installation and are too expensive to be abandoned after short term studies.

A far less expensive, removable instrument capable of measuring four components of strain and two components of tilt has been developed. Each sensing component employs non-contact eddy current transducers capable of measuring nanometer displacements. While not as precise as permanent borehole instruments, this new removable system should be able to resolve ground deformations associated with 0.5 to 1 microstrain per day rates expected at a proposed CO2 injection site. This system should also be well-suited for aquifer monitoring as well as for some geophysical signals.

Finite element techniques are used to simulate a field injection test within the Bartlesville sandstone reservoir at the Avant field CO2 storage analog site, Oklahoma. These models suggest that measuring strain change at shallow depths, on the scale of 100s of ft, can be used to monitor the proposed water injection during a water flooding operation at a depth of approximately 1700 ft. A set of stochastic optimization algorithms are then used to iteratively generate a sequence of parameter estimates, and a high performance cluster computer efficiently evaluates this computationally expensive forward model for each set of parameters. The set of converged parameter estimates is then used to find the expected value of each parameter, as well as the uncertainty of each expected value as required by the measurement limitations (noise, model error, spatial/temporal constraints) we impose on the dataset.