Rapid forward modeling of logging-while-drilling neutron-gamma density measurements

Abstract

Neutron-activated gamma-ray (neutron-gamma) logging-while-drilling (LWD) measurements deliver bulk density estimates without using a chemical source. The assessment of bulk density is based on neutron-induced non-capture gamma rays, corrected for neutron transport by combining particle counts acquired at two gamma-ray detectors and two fast neutron detectors. Particle counts from all four detectors are necessary to deliver one density measurement whose accuracy compares well to that of the gamma-gamma density instruments. Thereafter, borehole environmental effects are mitigated with empirical corrections based on Monte Carlo (MC) modeling. Such corrections should be avoided for standoff values greater than 0.63 cm (0.25 in) because they are no longer independent of formation properties. Neutron-gamma density measurements are also influenced by bed-boundary and layer-thickness effects. Thinly bedded formations, invasion, high-angle/horizontal (HA/HZ) wells, and enlarged boreholes can all mask true formation bulk density when implementing conventional petrophysical interpretation. Although MC methods accurately simulate 3D environmental and geometrical effects, they are computationally expensive and are thus impractical for real-time interpretation. Layer-by-layer bulk density can, however, be estimated using rapid numerical simulations coupled with inversion procedures. We have developed a rapid modeling algorithm to accurately simulate LWD neutron-gamma density measurements. Simulations are based on a theoretical, albeit realistic, LWD neutron-gamma density tool operating with a 14.1 MeV pulsed neutron source. The algorithm uses flux sensitivity functions and first-order Taylor series approximations to simulate particle counts at each detector before they are processed with a density estimation algorithm. Rigorous benchmarks against the Monte Carlo N-particle code in vertical and HA/HZ wells, across diverse solid and fluid rock compositions, thin beds, and in the presence of invasion, yield average density errors of less than 1% ([Formula: see text]) in approximately [Formula: see text] the time required of MC modeling.