Characterization and utilization of noisy displacement signals from simple shear device using linear and kernel regression methods

Small strain testing of dynamic soil properties such as threshold strains, γtv, are often conducted using specialized devices such as dual-specimen simple-shear, as devices configured for large strain testing produce noisy signals in the small strain range. Working with a simple shear device originally developed for large-strain testing, we extend its low-strain capabilities by characterizing noisy signals and utilizing several statistical methods to extract meaningful responses in the small strain range. We utilize linear regression of a transformed variable to estimate the cyclic shear strain from a noisy signal and the confidence interval on its amplitude. We utilize Kernel regression with the Nadaraya–Watson estimator and a Gaussian kernel to evaluate vertical strain response. A practical utilization of these techniques is illustrated by evaluating threshold shear strains for volume change with a procedure that takes into account uncertainties in the measured shear and vertical strains.