Sinkholes and other underground anomalies are not only a major hazard to Florida’s commercial and residential property owners; they are also a serious concern for those who manage the state’s roads, bridges, and other infrastructure.
Detecting and remediating sinkholes and other weaknesses before construction of a bridge or building has obvious benefits. However, traditional methods for finding and modeling these anomalies have clear limitations.
Recently, researchers at the University of Florida and Clarkson University, in partnership with the Florida Department of Transportation (FDOT) Research Center, developed a new, more nuanced non-destructive technique for sinkhole detection. Phases I and II of the project proved the viability and practicality of the method, as well as its limitations. Phase III is underway, with the goal of developing a 3D method for assessing sites for sinkholes and other karst features in a more comprehensive, timely, and accurate manner than ever before.
Phase I: A More Holistic View
The most common traditional methods for sinkhole detection at a construction site are invasive ground-penetrating tests, such as standard penetration tests (SPT) and cone penetration tests (CPT). However, because these methods have a very limited range, they uncover only small amounts of soil and rock (< 0.1 percent), soil/rock layering, and anomalies like sinkholes. Also, soil properties can change significantly over relatively short distances. Conventional testing can miss these changes, including sinkholes, even if the anomaly is just a few meters away.
“Invasive techniques work pretty well if your site is relatively consistent,” said Dr. David Horhota, FDOT State Geotechnical Materials Engineer and Project Manager for the research project. “But in Florida they rarely are. And even then, you can’t do enough boring to get a complete picture. You always have to infer the information between the borings.”
Ideally, evaluation of a site would begin with a wider ranging geophysical non-destructive test (NDT) to give technicians and engineers a more detailed, holistic view of the site and an accurate map of potential problem areas. Then, traditional invasive tests could be conducted at just those areas to obtain more detailed information.
Several NDT approaches are already in use to identify sinkholes. They range from gravity, electrical resistivity (ER), and ground penetrating radar (GPR), to electromagnetic wave methods and traditional seismic wave methods. However, these all have limitations in identifying and quantifying variability.
The goal of this research was to develop a timely, more comprehensive, higher resolution NDT to detect small sinkholes and other karst features to provide a complete underground map of a location.
This would not only be valuable for assessing the existence and limits of sinkholes. It would also allow engineers to plan borings and soundings more efficiently. This type of detailed information is needed to develop successful, cost-effective remediation programs to protect existing infrastructure long-term and address issues before new construction.
A Bigger Picture with Full Waveform Inversion
Before developing a new field test or software, the researchers knew they had to define the most accurate and useful NDT method for discovering underground anomalies. In Phase I of the project the team explored the use of Full Seismic Wave Fields and developed a technique called Full Waveform Inversion (FWI).
They chose seismic testing because it has the potential to provide a more comprehensive view of anomalies. But seismic testing also has the advantage over other geophysical methods (ground penetrating radar, electrical resistivity, etc.) because the results can be directly related to geotechnical parameters typically used by engineers, making the output most usable to practitioners.
“You get the advantages of both,” Horhota said. “You have a geophysical test that can map out a large area and the result gives you an engineering property that is directly applicable to our work.”
Traditional seismic reflection/refraction tests have significant drawbacks, however. For instance, they do not use the entire measured seismic wave field, which gives an incomplete picture and makes it difficult to accurately locate sinkholes and other soft spots.
By contrast, the FWI technique developed in Phase I uses the entire measured seismic wave field, and because later points in the wave field provide information on low-velocity zones—soft underground material—it gives a more complete picture of potential trouble spots.
“We have so much variability underground in Florida and many existing test methods can give you false readings or don’t provide the whole picture,” said Dr. Michael McVay, University of Florida researcher and Principal Investigator for the project. “The higher resolution picture provided by FWI is significant.”
Phase I testing showed that FWI was successful in identifying the location and extent of both known and unknown sinkholes at several sites in two dimensions. Analysis of simulated data sets showed that FWI accurately characterized embedded air- or water-filled voids. Results from real data sets proved that FWI did a good job of characterizing various site conditions including an embedded concrete culvert, low-velocity anomalies, open chimneys, and naturally occurring embedded voids.
“The contrast is very clear,” Horhota said. “Soft spots—including sinkholes, air- and water-filled voids, or very loose soil—appear in blue. The surrounding rock shows up in sharp red.”
To ground-truth their findings, the team did either soundings or borings over the locations where the FWI test showed anomalies.
“We showed exactly that there was next to zero-strength material in those areas,” Horhota said. “It worked.”
The more complete FWI picture presented a challenge, however. Namely, FWI was computationally expensive as it required solving elastic wave equations thousands of times. The algorithm developed to process the FWI code took two hours to analyze data from one 120-foot test line on a standard computer. It takes technicians about 30 minutes to test an average line in the field. To be efficient, the FWI data-processing time would have to be reduced to 30 minutes. This became a primary goal of Phase II research.
Phase II: Real-Time Data, Real-Time Decisions
Building on the results of Phase I, the objectives for Phase II were to develop a standalone software for 2D FWI analysis that could process data in 30 minutes or less, and to conduct a sensitivity analysis of measurements along the test array to discover any other limitations to the 2D method.
For FWI to be a practical tool, the faster software program needed a simple Graphical User Interface (GUI) for field use on a laptop computer so technicians could use that information to decide the locations of the next test lines. This ensured that as much information of layering, voids/anomalies, etc., was recovered in the field on one site visit, reducing unnecessary field testing and data processing efforts.
The software also had to automatically run on raw field data so technicians could use it without significant training.
To achieve the 30-minute threshold, the development team implemented advanced boundary conditions and solution convergence methods (e.g. gradient techniques) developed by Dr. Khiem Tran at Clarkson University, as well as variable grid dimensioning, temporal windowing, and parallel computing.
These methods overcame the limitations. At the end of Phase II, the software processed the needed data within 30 minutes. The software development team, led by Dr. Scott Wasman at the University of Florida, had also created a user-friendly GUI to easily import data and display results.
While the time threshold was achieved, the sensitivity analysis showed that the 2D FWI method required seismic data to be acquired directly over the top of the void. The team called this the “3D Effect.” Generally, voids that were near, but not directly under, the test line were distorted or may not have been identified. This was especially true if the test line was more than one diameter-width of the void away. For instance, if a void was two feet in diameter and the array line was at least two feet away, there was likely to be distortion.
This led to considerable effort for data collection and analysis, as multiple test lines were usually needed to locate unknown voids. It also spurred Phase III of the project.
“Since we discovered limitations in detection when we tested along one line of arrays, we asked ourselves, ‘Can we run the arrays in a grid pattern and map out the whole area at one time?’” Horhota said.
Phase III: Completing the Picture
Phase III is underway; the goal is to collect data in a grid pattern without setting out multiple arrays of receivers and then repeat the process along parallel lines. This should capture all the data in one pass and allow effective testing over a large volume of ground. It will also provide a third dimension to the site map, improving on the current 2D method.
That granular data can then be used to plan out an invasive boring program, improving the chances of detecting sinkholes and identifying their extent.
If successful, improvements proposed for this phase should increase the efficiency (and thereby decrease the testing time in the field) of the FWI method.
The benefit could be more efficient boring programs that result in cost savings (reduced numbers of borings) for sites with consistent geotechnical features. However, it might result in higher costs for highly variable sites which require additional borings. Nevertheless, even for sites that accrue additional costs for site investigation during design, identifying inconsistencies earlier should greatly minimize the chances of cost and time over-runs during construction due to potential unforeseen changes in site conditions.
Another concern with the added resolution is computing time. Adding a third dimension to the analysis adds that much more information to be processed, while the 30-minute threshold still has to be maintained.
“There is more information to process, but we expect computing power to keep increasing,” McVay said. “We’re also considering just analyzing certain frequencies to maintain efficiency. There are several options we are testing in this phase of the research.”
“Think of it as a sliding scale,” Horhota said. “You can adjust resolution up or down depending on your information and time needs. If you have a little lower resolution but a more complete picture in the same 30 minutes, it is still valuable.”
Even if the computing time increases due to the 3D analysis, the data is collected for a relatively large area, which minimizes returns to the site for retesting. In addition, the existing software can be used to analyze the results from individual lines of receivers in the field in real-time (less than 30 minutes) to optimize the placement of the receivers.
“Site analysis can potentially happen in three phases,” Horhota said. “First, the 3D technique gives you a large picture so you can hone in on trouble spots, then do quicker, targeted 2D analysis based on the 3D map. Lastly, you go back and ground-truth it with a typical boring program. That way you’re not shooting in the dark.”
Plans for the Future
If successful, the software developed under this project will be transferred to the FDOT State Materials Office for immediate use with its seismic equipment. McVay and Horhota, however, also see potential for broader use of these techniques.
“The equations driving this software are open source, published and accessible to anyone,” Horhota said. “We expect people to apply them in other platforms. There are many, many possible applications.”
For instance, Tran has already used the technology to locate abandoned mines beneath Ohio highways and plans to extend the method to detection of incipient slope failures.
“We want something that is useful out there in practice,” McVay said. “That’s what this research has always been about.”
Phase 3: BDV31-977-82 “Sinkhole Detection with 3-D Full Elastic Seismic Waveform Tomography”
Research in Progress listing