Development of Lab, Field, and In-Situ Image Collection Systems and Analysis Methods for Soil Characterization

Abstract

Image-based soil characterization methods are proven to be rapid, accurate, clean, lower-cost, and (semi-)automated procedures for determining the particle size distribution (PSD) of coarse-grained materials. A PSD is used to classify a soil as well as provide an initial estimate for various properties including its compressibility, permeability, and unit weight. Soil PSDs are utilized extensively throughout geotechnical, environmental, and construction engineering; earth sciences; and related industries. The traditional method for determining a soil’s PSD is by sieving, a time-, resource-, and energy-intensive procedure. Therefore, there has been an increased demand for determining soil PSDs using image-based methods as an alternative to sieving.

One such image-based method was developed in 2014 and is called “SedImaging.” Short for “sediment imaging,” SedImaging captures an image of a soil specimen that has sedimented through a water column. The photographed soil assembly is analyzed by a sieve-calibrated mathematical wavelet method. The results are a PSD with excellent agreement to sieving without any of the procedural disadvantages of sieving. This dissertation details two SedImaging hardware systems, “FieldSed” (2017) and “Sed360” (2020) which were developed by the author to expand upon the advantages of the original 2014 SedImaging method. These newer systems have different applications; the FieldSed is a field-portable device that was used in a large-scale soil characterization project for the Kalamazoo River, and the Sed360 is a largely automated test that has expanded the range of testable soils by SedImaging by over a factor of 2.5. The image analysis method used with the original SedImaging system has also undergone a major transformation that is described within this dissertation. When used with the Sed360, this new autoadaptive image analysis method can correctly size soil particles across the entire sand size range (as defined by the Unified Soil Classification System), as well as accurately generate soil PSDs for a range of gradations, including gap-graded specimens. The SedImaging hardware and image analysis advancements detailed in this dissertation transform this sieving alternative into a nearly fully automated, low cost, rapid, and robust soil characterization technique.

Image-based soil characterization methods can also be combined with existing geotechnical testing systems to enhance the original system’s capabilities. An example of this is with the popular cone penetrometer (CPT). A CPT was fitted with cameras that were used to photograph passing soil layers during CPT advance. Known as the “VisCPT” (short for the “vision cone penetrometer”), this newer system refines CPT soil layer delineation and is capable of even detecting thin (several centimeter thick) soil layers that are missed by CPT results. To do so, textural indices are determined for VisCPT soil images. These textural indices have been correlated to soil particle size. The latest research involving the VisCPT is presented in this dissertation. This includes replacing the multiple cameras required in the earlier VisCPT generations with a single high-resolution camera and recalibrating the textural indices analysis method. This dissertation also details exploratory research on combining the textural indices and the mathematical wavelet method (previously used mainly with SedImaging) to expand the application of both the VisCPT and SedImaging. The results of this research will be used with the latest VisCPT system to detect thin soil layers in scheduled calibration chamber testing and in earthquake prone regions of the world.