cracks you

cannot rule out

Problem statement

Embankment dams — massive compacted earth structures that hold back most of the reservoirs in the American West — work by maintaining a continuous low-permeability barrier of clay or silt. Any crack through that barrier becomes a preferential flow path, and concentrated seepage through a crack is the mechanism behind internal erosion, described by Reclamation as "one of the leading historical causes of embankment dam failure." The unsolved problem is detection: cracks caused by differential settlement, hydraulic fracturing, or foundation transitions "can be a few mm to 10s of centimeters wide, near the crest or 100 plus ft below the surface," and many "do not appear at the surface, are obscured by vegetation, or self-seal at the surface while remaining open internally." Reclamation states plainly that "existing methods cannot reliably confirm or rule out the presence of subsurface cracks."

Why this matters

Reclamation alone "manages more than 330 reservoirs that collectively store over 140 million acre-feet of water" across 17 western states, much of it impounded by embankment dams, and the challenge notes these are "structures that millions of people depend on for water, power, and flood protection." Because detection is unreliable, the decision problem inverts: as the guidelines put it, "the lowest risk option may be to assume they could be present and perform proactive repairs," so "dam owners may spend a significant amount of time and money repairing a structure because cracks are assumed to exist." Money that could go to dams with confirmed defects is spent on precautionary grouting and cutoff walls, while dams with real transverse cracks — the orientation that "can form continuous seepage paths from upstream to downstream" — may go unrepaired. Note also that the physics is not the gap: Reclamation says "the physics and mechanics that drive cracking of embankment dams is well understood." Engineers can predict where cracks should form (foundation transitions from compressible to incompressible material, wide foundation benches, interfaces with rigid structures, steep-walled trenches) but cannot confirm whether one is there.

What’s been tried and why it hasn’t worked

Reclamation "commonly uses tools such as electrical resistivity tomography, seismic refraction, seismic surface wave methods, GPR, and electromagnetic conductivity," and reports that "while valuable, these methods detect primarily macro-scale changes in stiffness or density and often cannot resolve narrow, discontinuous, or deep cracks" — a sensitivity cliff exactly at the feature width that matters, and worst in "the clayey and silty soils where cracks most often occur." Visual inspection fails for the same reason the problem exists: surface expression is rare "even in cases with large amounts of differential movement," and surface cracks may be hidden by vegetation or infilling. Settlement instrumentation indicates the potential for cracking, not the presence of a crack. Excavation does confirm cracks but is "considered invasive because excavating into or near a dam poses its own risks, that must typically be mitigated through reservoir and operational restrictions," and even then "these types of invasive investigations do not always yield a definitive way to rule out the possible existence of a crack." Finally, the deployment environment defeats the geophysics that might otherwise work: "buried utilities, conduits, drainage systems, monitoring instrumentation, metal reinforcement, riprap, and other infrastructure can generate signals that obscure or mimic those from subsurface cracks," so resistivity, GPR, and EM surveys "may struggle to distinguish between anomalies caused by cracks versus those caused by utilities, metal objects, or other installed features."

What would unlock progress

The prize is not for a new sensor alone but for a method that can rule out a crack — a negative result with quantified confidence, which is a different inferential product than an anomaly map. Two directions look open: physics that responds to an open void rather than to bulk stiffness/density contrast (for example, methods keyed to fluid or gas transport through the crack, or to the nonlinear/contact response of a discontinuity under excitation, rather than to average material properties); and inversion that explicitly models the known cultural clutter — because utility and instrumentation positions are documented for these dams, the interference is a known nuisance signal that could be forward-modeled and subtracted rather than treated as noise. Reclamation's own criteria reward this framing, weighting "Consideration of Interference and Site Constraints" at 15% of the score and offering a 5% bonus for "Data Fusion / AI-ML Value."

Entry points for student teams

A team can build the physical analogue that this field lacks: a lab-scale compacted clay/silt embankment section with an emplaced synthetic crack of known aperture and depth, then benchmark whichever geophysical modality they can borrow (surface waves, resistivity, GPR) against a ground truth nobody else has, quantifying the detection floor as a function of aperture, depth, and moisture. A second team could work purely on the discrimination problem — synthesize forward models of a crack versus a buried conduit or a length of rebar in the same background medium, and test whether a classifier can separate them, since being wrong about which anomaly is which is the specific failure Reclamation names. A third, lower-equipment path is decision analysis: model the cost of precautionary repair versus the expected cost of a missed crack, and derive the detection sensitivity and false-alarm rate a method would need to change a dam owner's decision — a result useful to the field even without hardware. Relevant skills: geophysics, geotechnical engineering, signal processing, inverse problems, and decision analysis.