USING ARTIFICIAL NEURAL NETWORKS TO PREDICT THE QUALITY AND PERFORMANCE OF OIL-FIELD CEMENTS

Inherent batch-to-batch variability, aging, and contamination are majo r factors contributing to variability in oil-field cement-slurry perfo rmance. Of particular concern are problems encountered when a slurry i s formulated with one cement sample and used with a batch having diffe rent properties. Such variability imposes a heavy burden on performanc e testing and is often a major factor in operational failure. We descr ibe methods that allow the identification, characterization, and predi ction of the variability of oil-field cements. Our approach involves p redicting cement compositions, particle-size distributions, and thicke ning-time curves from the diffuse reflectance infrared Fourier transfo rm spectrum of neat cement powders. Predictions make use of artificial neural networks. Slurry formulation thickening times can be predicted with uncertainties of less than +/-10 percent. Composition and partic le-size distributions can be predicted with uncertainties a little gre ater than measurement error, but general trends and differences betwee n cements can be determined reliably. Our research shows that many key cement properties are captured within the Fourier transform infrared spectra of cement powders and can be predicted from these spectra usin g suitable neural network techniques. Several case studies are given t o emphasize the use of these techniques, which provide the basis for a valuable quality control tool now finding commercial use in the oil f ield.