Harnessing the Power of Full-Waveform Inversion for Smarter Construction and Engineering Projects

Last Updated 1 month ago by Kenya Engineer

30 seconds summary

Full-Waveform Inversion (FWI) is an advanced seismic imaging technique that uses the entire seismic wavefield (amplitudes, phases, scattered energy) to iteratively build highly detailed subsurface models.
By more fully leveraging wave propagation physics, FWI enables much finer resolution of velocity, density, and structural heterogeneities than traditional methods. In construction and engineering, this means smarter site characterization (e.g., tunneling, foundation design, subsurface hazard assessment), reduced risk of surprises, and more efficient project planning.
Challenges include high computational cost, the need for good initial models, and handling nonuniqueness, but integrating AI, regularization methods, and high-performance computing is helping overcome these.

Full-waveform inversion (FWI) is a sophisticated geophysical imaging technique originally developed and refined in the oil & gas and exploration geophysics industry. Its goal is to build a high-fidelity model of the subsurface, especially seismic velocity and related elastic parameters, by iteratively fitting full seismic wavefields (not just first arrivals) to recorded data. The result is a detailed, quantitative 3D (or even 4D) model of subsurface properties, often with resolution and reliability beyond what classic methods can offer.

However, the value of FWI is not restricted to hydrocarbon exploration. Recent trends in civil engineering, infrastructure, geotechnical design, and construction projects are showing growing interest in using advanced seismic imaging techniques to better characterize near-surface geology, detect anomalies (faults, cavities, weak zones), and reduce subsurface risk during planning, tunneling, foundation design, and more. In such contexts, “smarter” construction means better subsurface knowledge, reduced surprises, lower contingency costs, and safer execution.

For a company like EIFGeosolutions, whose core competence lies in advanced seismic imaging, adopting or adapting FWI to engineering and construction domains represents both a technical opportunity and a strategic expansion. Below, we examine how FWI can be repurposed (or extended) for engineering purposes, the challenges, the benefits, and how EIFGeosolutions’ current infrastructure and expertise might enable that shift.

EIFGeosolutions: Technology, Capabilities & Positioning

Before diving further into FWI’s engineering potential, it’s instructive to profile EIFGeosolutions (EIF), what they do, how they operate, and what they already bring to the table.

Core Business and Scope

EIFGeosolutions positions itself as a provider of advanced seismic data processing and imaging services primarily for oil & gas exploration. On its homepage, the company states it delivers precise subsurface models tailored to exploration contexts.
In the “Technology” section, EIF lists that it specializes in FWI and RTM (Reverse Time Migration) for land, marine, and ocean-bottom-node (OBN) seismic datasets.
The company leverages DUG Insight software (from DownUnder Geosolutions) and McCloud HPC (High Performance Computing as a service) to accelerate computations. EIF is careful to highlight their use of high-performance compute infrastructure to reduce throughput times on compute-intensive tasks like high-frequency FWI, least-squares RTM, etc.
EIF also emphasizes its experience across a variety of seismic domains: land, marine, OBC, OBN, vintage data reprocessing, multiparameter FWI, shallow water de-multiple, fold-belt near-surface FWI, etc.
Client testimonials on EIF’s “Experience” page emphasize that the firm has delivered velocity models that align with selected well markers, achieving structural closures that were elusive with prior datasets.
Geographically, EIF maintains a UK registered office (London) and regional offices in Pakistan, Poland, Malaysia.
As a corporate entity, EIFGeosolutions Ltd was incorporated in December 2020 and is registered as a private limited company in the UK (company number 13053244)

Strengths & Differentiators

From their website, EIF’s differentiators and strengths include:

Access to HPC Resources — using DUG McCloud, which allows them to scale computing loads and reduce turnaround times for heavy imaging tasks.
Software Proficiency & Flexibility — strong usage of DUG Insight suite, combined with the ability to cherry-pick other tools (e.g., geometry/statics licensing from ExtremGeo, Flatiron) when needed.
Diverse Seismic Experience — working across domains (land, marine, OBN), dealing with vintage, challenging data, multiparameter FWI, near-surface imaging, etc.
Client focus & QC — consistent emphasis on workflow customization, quality control, meeting deadlines, and proper model validation with markers/wells.
Scalability & Energy Efficiency — the McCloud platform is touted not only for power but also energy efficiency, which is attractive for large-scale inversion operations.

Thus, EIF is not starting from scratch: they have both methodological and infrastructural capability. The question is: how to translate and extend that into the realm of smarter construction and engineering projects.

FWI Applied to Engineering & Construction: Potential Use Cases

While FWI is well-known in exploration geophysics, its use in engineering and construction is still niche but growing. Below are some core use cases where FWI (or modified variants) might deliver real value in civil/structural/underground engineering.

Near-Surface Characterization & High-Resolution Geophysics

For many construction projects whether tunnels, deep foundations, slope stabilization, underground parking, hydropower dams, or subways the critical zone of interest is the upper few hundred meters (or even tens of meters). Traditional geotechnical methods (boreholes, CPT, shallow seismic refraction, MASW, etc.) provide point or 1D/2D constraints, but often lack dense 3D coverage.

FWI can be adapted to “near-surface FWI” to produce high-resolution, spatially continuous maps of seismic velocity (Vp and Vs) and even density or attenuation (Q). These can be used to:

Detect weak zones, voids, or cavities (e.g. karst, subsidence features).
Identify lithological or cementation contrasts.
Map fracture zones or faulted zones that may destabilize tunnels or excavations.
Provide geomechanical inputs (stiffness, elastic moduli) for finite element or boundary element modeling.
Serve as baseline models in ground-improvement or grouting planning.

However, near-surface FWI is more challenging than deeper exploration FWI, because the signal frequencies are higher, heterogeneity is more intense, and data coverage is more variable.

Tunneling, Excavation & Underground Works

Underground projects (tunnels, shafts, mining, subway systems) are inherently risky because unknown anomalies (e.g., water-bearing fractures, weak zones, voids) can cause cost overruns or safety hazards. By performing FWI-compatible seismic surveys (e.g. active sources, multi-offset geometry) prior to excavation, the project team could generate a 3D “pre-excavation velocity/elasticity model.” This model can guide tunnel alignment, support design, and anticipate problematic zones.

In addition, repeated surveys over time (akin to 4D seismic) could detect changes (e.g. compaction, water infiltration, subsidence) post-excavation, providing feedback for monitoring.

Deep Foundations, Piling & Ground Improvement

In projects involving deep piles, micropiles, soil anchors, or ground improvement (jet grouting, compaction, etc.), detailed subsurface models are crucial to design and safety. FWI-based inversion can help refine the distribution of stiffness and density variations around piles or beneath foundations, which can help detect zones of weakness, layering, or soft inclusions that may unduly deform under load.

Dam Foundations, Embankments & Cut-and-Fill Projects

Large civil infrastructure, such as dams, embankments, or large earthworks, benefit from continuity in subsurface models. FWI-derived models may help detect zones of potential slippage, variations in material, discontinuities, or anomalous layering under the footprint. This is especially valuable in seismic zones, where detailed velocity/elastic models feed into ground motion and dynamic deformation modeling.

Geohazard & Site Risk Mitigation

In landslide-prone zones, areas susceptible to liquefaction, or where karst/void risk is high, FWI can help map anomalies that are difficult to resolve otherwise, potentially locating zones of weakness or discontinuity in the subsurface that standard geophysics might miss.

Integration with Geotechnical & Geomechanical Models

Perhaps the most powerful synergy is combining FWI-derived velocity/elastic parameter models (3D) with borehole data, CPT, pumping tests, resistivity, and other geotechnical logs. This leads to hybrid, multiscale, multiphysics site models that feed directly into structural, geomechanical, and finite-element simulations. In that sense, FWI becomes a complementary geophysical “lens” to reduce the uncertainty in engineering models.

Key Challenges & Adaptations for Engineering Use

The promise of FWI in construction is compelling—but bringing it into mainstream usage requires overcoming several challenges. Below are major technical, operational, and economic barriers, and how they might be addressed.

Challenge 1: Acquisition Design & Signal Penetration

High frequencies needed: Near-surface FWI requires high-frequency sources and receivers, which reduce penetration depth. Achieving sufficient signal-to-noise ratio at depth or through overburden is nontrivial.
Dense spatial sampling: Reliable FWI demands dense spatial sampling in both sources and receivers (broad azimuth, many offsets). In many civil sites, constraints (land, budget, access) restrict acquisition geometry.
Low-frequency content scarcity: Standard sources often lack very low-frequency (< 5 Hz) content, which helps invert for smooth, large-scale velocity variations. Without low frequencies, the inversion may suffer from cycle-skipping and converge to incorrect minima.
Heterogeneity / scattering: The near-surface is highly heterogeneous and scattering, which complicates modeling and inversion.

Mitigations / Strategies:

Use specially designed sources (vibroseis, weight drop, accelerated mass) tuned for broader bandwidth.
Multi-scale inversion: begin with lower frequency content, then progressively add higher frequencies.
Use extension strategies (wave-equation extensions, source extensions, offset extension) to mitigate cycle-skipping.
Hybrid inversion: use tomography or travel-time inversion to build a smooth starting model before FWI refinement.
Sparse or optimized acquisition design methods (e.g., optimized experimental design, OED) can reduce acquisition burden while retaining inversion quality. For example, a 2025 study in GJI proposes a method that reduces source layout by ~50%, compresses model parametrization, while retaining performance.

Challenge 2: Computational Load & Scalability

FWI is notoriously demanding in computation, due to the need to run full wave-equation simulations (forward and adjoint) many times over large spatial grids. For near-surface engineering FWI, the grid spacing must be fine (to resolve high frequencies), which further increases computational cost.

Mitigations / Strategies:

High-performance computing (HPC) and cloud scaling: Use distributed computing, GPU acceleration, and cloud/HPC services (like DUG McCloud which EIF already uses).
Algorithmic improvements: Use Gauss-Newton (or extended Gauss-Newton) variants, approximated Hessians, preconditioning, or quasi-Newton methods to accelerate convergence. For instance, an extended Gauss-Newton technique relaxes diagonality constraints in the Hessian to improve robustness and reduce computational bottlenecks.
Regularization / hybridization: Use robust constraints (e.g., total variation, Tikhonov, p-variation) to stabilize inversion and reduce iteration counts. A technique combining acoustic/elastic FWI with TGPV regularization helps manage sharp interfaces and smooth backgrounds simultaneously.
Sparse / source encoding & compression: Use source-encoding strategies (e.g. simultaneous multiple source encoding) to reduce the number of forward/adjoint runs.
Domain decomposition / multi-scale inversion: Start with coarse meshes or smoothed models, then progressively refine.

Challenge 3: Non-Uniqueness & Local Minima

FWI inversion is fundamentally non-linear and ill-posed. It’s prone to local minima (cycle-skipping), parameter trade-offs (e.g., between velocity and density), and sensitivity to noise or incomplete data.

Mitigations / Strategies:

Use good starting models (from tomography, well logs, or geological constraints).
Use multi-parameter inversion cautiously—e.g., invert velocity first, then density.
Use robust misfit functions (e.g. L1 norms, Huber norms, dynamic time warping) or Bayesian approaches to reduce sensitivity to outliers. A recent Bayesian-based FWI formulation explicitly handles measurement errors and spurious observations.
Use wavefield-reconstruction strategies and constraint-based formulations like alternating direction method of multipliers (ADMM) to decouple wavefield and parameter estimation (e.g. for elastic FWI) and improve stability.
Regularize cross-parameter trade-offs (e.g., fix Vp/Vs ratio or impose prior constraints) in multiparameter inversion.

Challenge 4: Integration with Engineering Workflow & Model Validation

Geotechnical and structural engineers have expectations regarding model validation, uncertainty quantification, and ground-truth calibration (boreholes, CPTs, lab tests). For FWI-derived models to be “trusted,” they must be integrated with traditional geotechnical data, and uncertainties must be clearly communicated.

Mitigations / Strategies:

Conduct joint inversion or multimodal assimilation (FWI + resistivity + gravity + borehole data) to reduce ambiguity.
Use perturbation or Monte Carlo methods to estimate model uncertainties.
Always anchor inversion results to ground-truth (boreholes, CPT, geotechnical logs) in critical zones.
Provide interpretative layers: e.g. stiffness zones, uncertainty zones, hazard flags, etc., which engineers can readily use.
Present output not just as velocity fields, but as derived engineering parameters (e.g. Young’s modulus, shear modulus, Poisson’s ratio) for direct input into structural or geomechanical models.

Strategic Roadmap: How EIFGeosolutions Can Expand into Smarter Construction

Given EIF’s starting strengths, here’s a candidate roadmap for how it might transition or augment its offerings to better capture the construction and engineering market using FWI-based techniques.

Step 1: Pilot Projects in Near-Surface FWI

Start with pilot projects in small-to-medium engineering sites (e.g. tunnel alignment, dam foundations, deep foundations) to validate the methodology.
Collaborate with civil engineering firms, contractors, or academic institutions to gather “ground truth” data (boreholes, CPT) for validation.
Tailor acquisition parameters (source, receiver spacing, bandwidth) specifically for near-surface inversion rather than exploration-scale acquisition.

Step 2: Develop a Workflow Adapted for Engineering Contexts

Create a modular workflow that goes from raw seismic, pre-processing, tomography initialization, FWI refinement, quality control, and to engineering parameter generation (e.g. elastic moduli).
Build automations and GUI tools so that non-expert users (civil engineers or site teams) can visualize and extract meaningful engineering outputs.
Implement uncertainty quantification modules by leveraging Bayesian or ensemble FWI methods.
Offer hybrid inversion (joint geophysics + geotechnical) to reduce risk.

Step 3: Infrastructure, Compute & Cloud Packaging

Leverage or extend existing compute partnerships (like DUG McCloud) to provide “on-demand FWI as a service” to engineering clients.
Create packaged offerings: e.g. “FWI for Tunnel Planning,” “FWI for Foundation Design,” “FWI for Earthworks & Embankments.”
Optimize compute pipelines for engineering-scale volumes (e.g. smaller domains, high frequencies) so that pricing and turnaround times are competitive.

Step 4: Develop Domain-Specific Interpretation Layers & Tools

Build interpretation modules that translate velocity/elastic parameter outputs into direct engineering metrics (stiffness zones, shear wave velocity profiles, flagged anomalous zones).
Build web-based or desktop visualization tools so that engineers can view 3D slices, plan cross-sections, extract depth profiles, generate input files for FEM/BEM codes.

Step 5: Marketing, Client Education & Positioning

Publish case studies, white papers, webinars, and application notes showing how FWI-derived models improved design or reduced risk in pilot projects.
Partner with civil engineering firms or infrastructure developers to embed FWI in their standard site characterization workflows.
Emphasize the “smarter construction” narrative—risk reduction, better subsurface knowledge, reduced surprises, and savings on contingency.

Step 6: Continuous Improvement & R&D

Track advances in algorithmic FWI (e.g. physics-informed neural networks, extended Gauss-Newton, adaptive regularization) and adopt them to reduce cost and improve robustness. For instance, research on PINNs has begun to show promise in wave propagation / inversion.
Investigate joint inversion with other geophysical methods (EM, resistivity, gravity) to broaden applicability on difficult sites.
Explore 4D/monitoring FWI for time-lapse subsurface change detection in infrastructure projects.

Illustrative Use Case: Tunneling through Heterogeneous Terrain

To make this more concrete, consider a hypothetical urban subway tunnel project in a city with heterogeneous subsurface (layers of fill, clay, sand, interspersed rock, minor faults or discontinuities). The contractor is concerned about encountering zones of weakness, abrupt stiffness changes, or voids, which can cause settlement, over-excavation, water ingress, or unexpected support needs.

A conventional approach might use boreholes every few hundred meters, CPT tests, geophysics like MASW or refraction tomography, and then design conservatively. But significant unknowns remain between points.

With a properly designed seismic acquisition (multi-offset, dense, broad-band sources), an FWI workflow tailored to near-surface conditions could:

Build a baseline velocity/elastic model covering the tunnel corridor region.
Flag anomalous zones (e.g., zones of reduced stiffness or velocity inversions suggestive of voids, karstification, or weak sediments).
Provide cross-sections and depth profiles of shear-wave velocity (Vs) for use in ground response and settlement modeling.
Help optimize alignment or support design by avoiding or anticipating problematic zones.
Optionally, after partial excavation, perform repeat surveys (4D FWI) to detect compaction, residual deformation, or water infiltration.

The end result: reduced geotechnical uncertainty, more confident design, fewer surprises in execution.

Risks, Limitations & Mitigations in Practice

While the upside is strong, it’s fair to acknowledge risks and limitations in deploying FWI for engineering. It’s not a panacea. Some key caveats:

Cost & Time: For small sites, the additional cost and computational burden may not justify marginal gains over conventional geophysics. The business case must be evaluated on a per-project basis.
Data Quality Dependencies: The method is sensitive to noise, aliasing, missing offsets, or bandwidth limitations. Poor acquisition severely degrades inversion results.
Model Bias / Non-Uniqueness: In absence of adequate constraints, FWI may produce ambiguous or geologically unreasonable artifacts.
Domain Shift: Many FWI algorithms have been tuned for deeper, lower-frequency exploration settings; adapting them to shallow, high-frequency, highly scattering domains is nontrivial.
Interpretation Gap: Engineering teams may resist adopting “black box” geophysics unless the output is clearly interpretable and validated.
Uncertainty Communication: Without robust uncertainty quantification, clients may misinterpret the inversion model as “truth” rather than a probabilistic best-fit model.

To mitigate these, a firm like EIF would need to:

Scope projects wisely (start with medium-to-high risk or high-value sites).
Invest in QC, validation, and ground-truth calibration (boreholes, CPT).
Use conservative interpretive boundaries (e.g. buffer zones around flagged anomalies).
Ensure transparent uncertainty analysis and risk communication.
Evolve their algorithmic toolbox (e.g. Bayesian FWI, regularization strategies, hybrid inversions) to improve robustness.

In Conclusion

Full-waveform inversion (FWI) holds impressive potential for “smarter” construction and engineering projects — the ability to produce high-resolution, quantitative subsurface models can shrink geotechnical uncertainty and improve design confidence. However, realizing that potential requires careful adaptation, integration, and validation in the engineering domain.

EIFGeosolutions is well placed to play a leadership role in this transition. Their existing core strengths HPC access (via DUG McCloud), software familiarity (DUG Insight), diverse seismic experience, and willingness to customize workflows provide a strong foundation. The strategic pathway involves piloting near-surface FWI projects, developing engineering-friendly workflows, embedding interpretive tools, building trust via validation, and scaling through packaged services.