Soil-Structure Interaction: Advanced Modelling for Deep Basements and Foundations

1. Introduction: The Critical Interface of Soil and Structure

The built environment is expanding into increasingly complex frontiers. 

As urbanization drives the demand for vertical growth in dense metropolises, the engineering challenges have shifted downward, necessitating deep basements and heavy foundations that penetrate complex subterranean stratigraphy. 

In this domain, the traditional boundaries between structural and geotechnical engineering are not merely blurred; they are effectively dissolved. 

The behavior of a superstructure—its natural frequency, its deformation under wind or seismic loads, and the distribution of internal forces—is inextricably linked to the behavior of the ground it rests upon. 

This reciprocal relationship, where the response of the soil influences the motion of the structure and the motion of the structure influences the response of the soil, is defined as Soil-Structure Interaction (SSI).1

Historically, engineering practice has often decoupled these two domains to simplify analysis. 

Structural engineers have frequently modeled foundations as fixed supports—rigid, unyielding boundaries that effectively assume the earth is an infinitely stiff solid. 

Conversely, geotechnical engineers have often modeled foundations as flexible loading patches on a soil half-space, neglecting the stiffening contributions of the superstructure’s shear walls and floor diaphragms. 

For low-rise buildings constructed on competent bedrock, this uncoupled approach is a reasonable simplification that yields conservative results. 

However, the modern engineering landscape is populated by supertall towers resting on soft alluvial clays, deep multi-level basements in seismically active zones, and infrastructure projects that thread through congested underground networks. 

In these scenarios, ignoring SSI is no longer a conservative simplification; it is a potential source of catastrophic error, significant economic inefficiency, or gross miscalculation of performance.2

The consequences of neglecting SSI are manifold. In seismic design, the flexibility of the foundation soil lengthens the fundamental period of the structure, potentially shifting it into a spectral region of higher or lower acceleration depending on the earthquake characteristics. 

The dissipation of energy through the soil—radiation damping—can significantly reduce the seismic demand on the structure, a benefit often lost in fixed-base models. 

In static design, the differential settlement of a piled raft foundation is governed not just by the soil stiffness, but by the rigidity of the raft and the superstructure above it. 

A flexible raft will conform to the soil settlement trough, while a stiff raft will bridge over weak spots, redistributing loads to the edges or to stiffer pile elements. 

Failing to capture this redistribution can lead to the under-design of piles in critical locations or the cracking of the raft itself.3

This report serves as an exhaustive, expert-level guide to the advanced modelling of SSI, specifically tailored for deep basements and foundations. 

It moves beyond the elementary “springs and dashpots” approach to explore high-fidelity continuum modelling using 3D Finite Element Analysis (FEM). 

We will scrutinize the critical limitations of traditional constitutive models like Mohr-Coulomb and advocate for the adoption of advanced formulations like the Hardening Soil model with Small-Strain Stiffness (HSsmall) and Soft Soil Creep models. 

We will dissect the mechanics of deep excavations, exploring the three-dimensional “corner effects” that 2D plane strain analyses miss. 

We will examine the design of piled rafts and barrette foundations, debating the merits of volume elements versus embedded beams. 

Finally, we will validate these theoretical concepts against real-world data from landmark projects such as the Taipei 101 tower and the Shanghai Tower, and look toward the future of Machine Learning and Digital Twins in geotechnical engineering.

2. Fundamentals of Soil-Structure Interaction (SSI)

To model complex systems accurately, one must first grasp the fundamental physics governing the interaction. 

SSI is a coupled phenomenon, but for the purpose of analysis and understanding, it is generally categorized into two primary mechanisms: Inertial Interaction and Kinematic Interaction.3 

While often analyzed separately in substructure methods, they occur simultaneously in reality, and understanding their distinct contributions is vital for selecting the appropriate modelling strategy.

2.1 Kinematic Interaction: The Filtering Effect

Kinematic interaction refers to the modification of “free-field” ground motion by the presence of stiff foundation elements. 

When seismic waves propagating vertically through the soil column encounter a rigid embedded foundation—such as a deep basement box or a cluster of stiff piles—the foundation cannot conform to the complex, short-wavelength distortions of the soil. 

Instead, the foundation averages the motion, effectively filtering out high-frequency components.

This filtering manifests in two primary ways. First, Base Slab Averaging occurs because the foundation slab is rigid relative to the soil. 

Wave incoherence over the base area means that different points on the slab would essentially want to move in different directions at the same time if they were free soil particles. 

The rigid slab forces them to move as a unit, averaging these incoherent motions and reducing the overall input acceleration, particularly at high frequencies.4 

Second, Embedment Effects play a significant role. Ground motion amplitude generally decreases with depth below the free surface. 

A deep basement interacts with soil layers at various depths, not just at the surface. 

The effective input motion driving the building is an average of the ground motions over the entire depth of embedment, which is typically less intense than the surface motion.3

Current guidelines, such as FEMA P-2091 and ASCE 41-17, provide simplified procedures to account for these kinematic effects by allowing the reduction of the response spectrum used for design. 

However, these simplified reductions are often conservative. 

Advanced 3D FEM modelling captures kinematic interaction naturally by modelling the wave propagation through the soil medium and the diffraction around the stiff foundation elements.5

 

2.2 Inertial Interaction: The Response of Mass

Inertial interaction is driven by the mass and flexibility of the superstructure. 

When the building vibrates due to an earthquake, it generates inertial forces (base shear and overturning moment) that are transmitted back into the soil through the foundation. 

This causes the foundation to translate and rotate, further deforming the surrounding soil. 

This deformation is not elastic; it involves hysteretic energy dissipation in the soil and the generation of new waves that radiate away from the foundation.

Two key phenomena result from inertial interaction. Period Lengthening occurs because the soil support is flexible, not fixed. 

This flexibility essentially softens the global structural system, lengthening the fundamental period of the structure compared to a fixed-base assumption.3 

For structures on soft soil, this period shift can be substantial, moving the building’s response into a different part of the design response spectrum. While often beneficial (moving away from the peak spectral acceleration), it can be detrimental if it shifts the period into a resonance band with the site soil period.

Radiation Damping (or geometric damping) is the second critical effect. As the foundation pushes against the soil, it acts like a piston, generating stress waves that propagate away from the structure to infinity. 

These waves carry kinetic energy away from the building, effectively acting as a potent energy dissipation mechanism. 

This is distinct from material damping (friction within the soil structure) and can result in total system damping ratios significantly higher than the conventional 5% used for structural materials.3

2.3 Modelling Approaches: Direct vs. Substructure

The engineering community typically employs one of two main philosophies to model these effects: the Direct Method and the Substructure Method.2

The Direct Method (Continuum Approach) involves modelling the soil, foundation, and structure together in a single, coupled Finite Element (FEM) or Finite Difference analysis (e.g., using PLAXIS 3D, FLAC3D, or ABAQUS). 

This approach is the most rigorous as it captures both kinematic and inertial effects simultaneously. 

It naturally accounts for material nonlinearity in the soil (plasticity, pore pressure generation), geometric nonlinearity (uplift, P-delta), and interface behavior (gapping, sliding).6 

It is the preferred method for complex deep basements where the geometry is irregular and simplified impedance functions cannot capture the true stiffness. 

However, it is computationally expensive and requires careful handling of boundary conditions—such as viscous or absorbing boundaries—to prevent artificial wave reflections from the model edges contaminating the results.7

The Substructure Method decouples the problem into distinct, manageable steps. 

First, a Kinematic Interaction Analysis is performed to determine the Foundation Input Motion (FIM) by analyzing the response of a massless foundation to free-field motion.4 

Second, an Impedance Function Analysis calculates the dynamic stiffness (springs) and damping (dashpots) of the foundation-soil system.8 

Finally, a Superstructure Analysis is conducted by applying the FIM to the structure supported by the calculated springs and dashpots.4 

This method is computationally efficient and is the basis for many code provisions, such as ASCE 41-17 and NIST GCR 12-917-21.2 

While excellent for regular foundations, it relies on the principle of superposition, which strictly holds only for linear systems, making it less ideal for highly nonlinear soil behavior or complex interaction geometries where the Direct Method prevails.

3. Advanced Constitutive Soil Modelling

The reliability of any SSI analysis, regardless of the method used, is entirely dependent on the quality of the constitutive model used to represent the soil. 

In the context of deep excavations and heavy foundations, the traditional linear-elastic or simple elastic-perfectly plastic models are often grossly inadequate and can lead to dangerous misconceptions about safety and performance.

3.1 The Limitations of the Mohr-Coulomb Model

The Mohr-Coulomb (MC) model is the ubiquitous workhorse of geotechnical engineering due to its simplicity and the wide availability of its parameters ($c, \phi, E, \nu$). 

However, for complex interaction problems like deep excavations, it exhibits critical deficiencies.10 

First, it assumes Linear Elasticity up to failure, implying a constant stiffness ($E$) regardless of the stress state or strain level. Real soil, however, exhibits stress-dependent stiffness (it becomes stiffer at depth due to confining pressure) and strain-dependent stiffness (it is very stiff at small strains and softens as strain increases).

Second, the MC model often predicts Unrealistic Heave in excavation bottoms. This occurs because it typically uses the same modulus for unloading as it does for loading ($E_{ur} = E_{50}$). 

In reality, soil is much stiffer when unloaded (rebound) than when loaded virginly. Using a loading modulus for the unloading condition at the base of an excavation results in excessive predicted upward movement. 

Finally, the MC model lacks Plastic Hardening. It acts as an elastic-perfectly plastic material, failing to capture the gradual accumulation of plastic strain before failure, which is crucial for accurate settlement predictions in complex loading paths.13

3.2 The Hardening Soil (HS) Model

To address the shortcomings of the linear elastic-perfectly plastic approach, the Hardening Soil (HS) model was developed. 

This is an elastoplastic model that introduces a hyperbolic stress-strain relationship and distinguishes between different stiffness moduli:

• Secant Stiffness ($E_{50}$): The stiffness at 50% of peak strength, used for primary loading.
• Oedometer Stiffness ($E_{oed}$): The stiffness utilized for one-dimensional compression calculations.
• Unloading/Reloading Stiffness ($E_{ur}$): The stiffness applied during unloading and reloading paths, which is typically set to 3-5 times higher than $E_{50}$.

This distinction is vital for deep excavations. As soil is excavated, the remaining soil below the cut enters a state of stress relief or unloading. 

By using the stiffer $E_{ur}$ for this zone, the HS model predicts much more realistic (and significantly smaller) heave compared to the MC model.14 

Furthermore, the HS model accounts for the stress-dependency of stiffness using a power law formulation:

$$E_{50} = E_{50}^{ref} \left( \frac{c \cos\phi – \sigma’_3 \sin\phi}{c \cos\phi + p^{ref} \sin\phi} \right)^m$$

Where $m$ is the power exponent (typically 0.5 for sands and 1.0 for clays). 

This formulation allows the model to naturally simulate the increasing stiffness of soil with depth, a fundamental characteristic of real ground profiles.15

3.3 The Game Changer: Hardening Soil with Small-Strain Stiffness (HSsmall)

Even the standard HS model has a flaw: it assumes elasticity is constant within the “elastic” range. 

However, decades of soil dynamics research have shown that soil is incredibly stiff at very small strains ($\epsilon < 10^{-5}$) and its stiffness degrades non-linearly as strain increases. 

This small-strain stiffness is the “true” elastic stiffness of the soil skeleton before any particle rearrangement occurs.

The HSsmall (HSS) model incorporates this behavior by adding two additional parameters to the standard HS formulation:

1. $G_0$ (or $G_{max}$): The small-strain shear modulus. This is physically related to the speed at which shear waves travel through the soil and can be derived directly from Shear Wave Velocity ($V_s$) measurements using the relation $G_0 = \rho V_s^2$.15
2. $\gamma_{0.7}$: The shear strain level at which the secant shear modulus $G_s$ degrades to 70% of its initial value $G_0$.15

Why is this critical for deep basements?

In a deep excavation scenario, the soil immediately behind the retaining wall undergoes large strains and plastic failure. 

However, the soil field farther away—the “far-field”—undergoes very small strains. The standard HS or MC models would underestimate the stiffness of this far-field soil, predicting a wide, shallow settlement trough that extends far back from the excavation. 

The HSsmall model, by recognizing the high stiffness at small strains, predicts a narrower, deeper settlement trough that typically matches field inclinometer and settlement data much better. 

Empirical comparisons have demonstrated that HSsmall outperforms both MC and standard HS significantly in predicting ground surface settlements behind retaining walls.11

 

Unlike the Mohr-Coulomb model which assumes constant stiffness up to failure (a flat line on a stiffness vs. strain plot), the HS Small model follows a distinct decay curve. 

It begins at a high maximum stiffness ($G_0$) at very low strains ($10^{-6}$), then degrades non-linearly following an S-shaped curve as strain increases towards engineering levels ($10^{-3}$ to $10^{-2}$). 

This decay captures the hysteretic damping of the soil even before plastic failure occurs, providing a unified framework for both static settlement and dynamic earthquake response.

3.4 Soft Soil Creep (SSC) Model

For projects situated in soft, compressible clays—such as the London Clay, Bangkok Clay, or the soft deposits of Mexico City—consolidation and secondary compression (creep) are major factors governing long-term performance. 

The Soft Soil Creep (SSC) model is essential in these environments. Unlike the HS model, which is largely time-independent regarding the viscous properties of the soil skeleton, the SSC model explicitly accounts for the volumetric creep strain that occurs under constant effective stress.19 

This is vital for calculating the long-term settlement of high-rise buildings, where creep settlement can continue for decades after construction completion.20 

Using a non-creep model in these soils often leads to a significant underestimation of total settlements over the lifespan of the structure.

4. Modelling Deep Basements & Excavations

The construction of a deep basement is a violent event in the life of the soil. 

It involves the removal of massive overburden pressure, the installation of rigid walls, and the lowering of groundwater tables. 

Modelling this sequence requires a rigorous “Staged Construction” analysis in FEM, as the path of loading determines the final state of stress and deformation.

4.1 Top-Down vs. Bottom-Up Sequences

The construction sequence chosen by the contractor significantly influences the final displacements and must be replicated in the numerical model.

• Bottom-Up Construction: This traditional method involves excavating to the full depth, often supported by temporary steel struts or ground anchors, and then constructing the permanent basement structure upwards from the mat foundation. This exposes the soil to significant stress relief before the permanent stiff structure is in place.
• Top-Down Construction: In this method, the permanent floor slabs are cast as the excavation proceeds downwards. These slabs act as extremely stiff struts, minimizing wall deflection.21

In FEM software like Plaxis 3D, this must be modelled step-by-step: (1) Install Wall, (2) Excavate to Level 1, (3) Activate Slab 1, (4) Excavate to Level 2, etc..22 

Modelling the excavation “in one go” (often called “wished-in-place”) will grossly underestimate the wall deflections and ground settlements because it ignores the progressive softening of the soil and the accumulation of plastic strains during the intermediate stages.

4.2 The “Corner Effect” and Plane Strain Ratio (PSR)

Most excavation analyses are performed in 2D Plane Strain for efficiency. This assumes the excavation is infinitely long and the cross-section is uniform. 

However, real excavations are finite boxes. The corners of the box are significantly stiffer than the center because the wall is supported in two orthogonal directions, creating a stiff “shell” effect. This is known as the Corner Effect.24

Wall deflections near the corners are observed to be much smaller than at the center of the wall span. 

To quantify this, the Plane Strain Ratio (PSR) is used:

$$PSR = \frac{\delta_{3D, max}}{\delta_{2D, plane-strain}}$$

 

Studies in Taipei and Jakarta have shown that the PSR is low (0.2 – 0.5) near the corners and increases asymptotically to 1.0 as the distance from the corner increases. 

For a standard excavation, the “corner effect” can extend for a distance of $0.5B$ to $1.0B$ (where $B$ is the excavation width).25

 

The implication for design is profound. A 2D analysis essentially models the worst-case scenario (center of the long wall) for the entire perimeter. 

While conservative for the middle, it leads to massive over-design for the corner areas. 

Advanced 3D FEM analysis allows for the optimization of the support system near corners, potentially reducing strut sizes or wall thickness in these zones, leading to substantial material savings.26

4.3 Coupled Flow-Deformation Analysis (Dewatering)

Deep excavations usually require dewatering to maintain a dry working environment. Lowering the water table increases the effective stress in the soil ($\sigma’ = \sigma – u$). 

This increase in effective stress causes settlement, even without any physical soil removal. This is a coupled Hydro-Mechanical (HM) problem.

Advanced modelling must couple groundwater flow with mechanical deformation. In Plaxis, this is typically done using a “fully coupled” flow-deformation analysis, where pore pressures and deformations are calculated simultaneously at every time step.28

Key Insight: Case studies show a fascinating interaction between dewatering and excavation. Dewatering inside the pit actually reduces wall deflection. 

This is because the dry soil inside the pit has a higher effective stress and thus higher stiffness and passive resistance. 

However, dewatering also causes drawdown outside the pit, which increases the settlement of adjacent ground surfaces and structures.29 

Balancing these opposing effects—using the benefits of internal dewatering while mitigating external drawdown via cutoff walls or recharge wells—is a key aspect of sophisticated SSI design.

5. Modelling Deep Foundations: Piles and Barrettes

When the basement raft alone cannot support the load, or when settlement control is paramount, deep foundations are employed. 

Modelling these elements in 3D FEM presents unique choices regarding element types, each with its own trade-offs between accuracy and computational cost.

5.1 Modelling Elements: Volume vs. Embedded Beam vs. Plate

1. Volume Elements:

• Description: The pile is modelled as a solid cluster of 3D finite elements (continuum).31
• Pros: This is the most accurate representation. It captures the physical volume of the pile, the exact shape (whether circular or rectangular), and the complex stress changes around the pile shaft and toe. It allows for detailed analysis of local stress concentrations.
• Cons: It is computationally expensive. It requires massive mesh refinement around the pile perimeter to capture the interface gradients, significantly increasing the model size and run time.
• Best For: Large diameter bored piles, Barrettes (rectangular diaphragm wall panels used as piles) 32, or analyzing local group effects where detailed pile-soil-pile interaction is critical.

1. Embedded Beam (Pile) Elements:

• Description: The pile is modelled as a line element that does not occupy volume in the mesh but interacts with the soil mesh through special interface “interaction” functions.33
• Pros: Extremely efficient. The mesh does not need to conform to the pile geometry, allowing for the modelling of large pile groups (100+ piles) in a reasonable timeframe.34
• Cons: It does not displace soil volume (no “blocking” effect). It is generally less accurate for lateral loading of large diameter shafts or non-circular shapes unless calibrated carefully.
• Best For: Large pile groups, Piled Rafts where the global behavior is the priority over local stress concentrations.

1. Plate Elements:

• Description: 2D surface elements used to model walls or thin rafts.
• Best For: Modelling Barrettes as “plates” to capture their high bending stiffness in the major axis and their significant surface area for friction.35

Comparative Insight: For a Barrette foundation, volume elements are theoretically superior because they capture the high side friction generated by the large surface area and the “corner friction” effects. 

However, recent studies comparing modelling strategies have shown that embedded beams, when their interface perimeters are adjusted to equivalent values, can show good convergence with volume element models. 

This offers a viable path for optimizing full-building models: use volume elements for a single-pile calibration study, then switch to calibrated embedded beams for the full global model.34

5.2 Interface Properties ($R_{inter}$)

The interaction between the concrete pile and the soil is rarely perfectly bonded. There is usually a thin zone of disturbance or slippage. 

In FEM, this is modelled using an Interface Element characterized by a strength reduction factor, $R_{inter}$.37

The interface properties are derived from the adjacent soil properties:

• $\tan(\phi_{interface}) = R_{inter} \cdot \tan(\phi_{soil})$
• $c_{interface} = R_{inter} \cdot c_{soil}$

Typical values for $R_{inter}$ are critical for accurate capacity prediction:

• Steel-Soil Interfaces: Values typically range from 0.5 – 0.7, reflecting the smoother surface of steel sheet piles or casings.
• Concrete-Soil Interfaces: Values are generally higher, ranging from 0.7 – 1.0. Cast-in-place concrete is rough and interlocks well with the soil.
• Bored Piles/Barrettes: A crucial nuance exists here. While concrete is rough, the construction process using bentonite slurry can leave a “mudcake” on the excavation wall, potentially reducing skin friction. In such cases, a lower $R_{inter}$ might be warranted. Choosing an appropriate $R_{inter}$ is often the primary “knob” used to calibrate FEM models against pile load test data.37

5.3 Piled Rafts and Load Sharing

In a traditional pile design, the piles are designed to carry the entire load of the structure. In a Piled Raft Foundation, the philosophy is different: the raft (mat) bears on the soil, and the piles act primarily as “settlement reducers.”

The key metric is the Load Sharing Ratio ($\alpha_{pr}$):

$$\alpha_{pr} = \frac{\sum P_{piles}}{P_{total}}$$

FEM analysis allows engineers to optimize this ratio. 

Instead of designing piles to carry 100% of the load ($\alpha_{pr} = 1.0$), one might design them to carry 60-70%, allowing the raft to carry the remaining 30-40% via direct soil bearing. 

This can reduce the number or length of piles significantly, leading to cost savings.

Measured Data: In the case of the Shanghai Tower, measured data showed that the piles carried significantly less load than predicted by simplified methods. 

The raft-soil contact was more effective than anticipated, taking a larger share of the burden. This underscores the conservatism of traditional design methods that ignore the bearing capacity of the raft.41

6. Case Studies: Validation of Advanced Modelling

Theoretical models are powerful tools, but field data is the ultimate arbiter of truth. 

The following case studies demonstrate the successes and challenges of SSI modelling in complex real-world scenarios.

6.1 Taipei 101: The Corner Effect and PSR

Project: Taipei 101, formerly the world’s tallest building.

Challenge: The project required a deep excavation (22.3m deep) in the soft, compressible clays of the Taipei Basin.42

Modelling:

Initial 2D plane strain analyses overestimated the wall deflections, predicting severe damage to adjacent structures. However, engineers performed 3D FEM analyses that accounted for the “Corner Effect.” 

These 3D models predicted significantly smaller deflections near the corners of the excavation.

Outcome:

Field monitoring confirmed the 3D predictions. The Plane Strain Ratio (PSR) was observed to be well below 1.0 near the corners, rising to ~1.0 only at the center of the long sides of the excavation.42 

This validated the concept that the 3D stiffness of the “box” shape provides significant restraint.

Key Lesson: In square or rectangular excavations with low aspect ratios (Length/Width < 2-3), 2D analysis is overly conservative. 

3D analysis allows for the optimization of support systems, particularly in the corners where wall movements are naturally restricted.

6.2 Shanghai Tower: Time-Dependent Settlement

Project: Shanghai Tower (632m), located on deep, soft clay deposits in Shanghai.

Challenge: Predicting the long-term settlement of a supertall structure supported by a massive piled raft.

Data & Analysis:

Several calculation methods were used to predict the final settlement.

• Predicted Settlement (ELPLA/FEM): Estimates ranged from 101 mm to 143 mm.41
• Measured Settlement (During Construction): Measurements taken at 75% load showed settlements of approximately 60 mm.41

At first glance, the measured data appeared to show that the models over-predicted settlement. 

However, this discrepancy highlights the importance of time-dependency. The measured values were “during construction.” In soft clays, consolidation and creep continue for decades. 

The total final settlement will likely approach the predicted values as excess pore pressures dissipate and secondary compression occurs. 

Furthermore, the load sharing analysis revealed that the raft was carrying a larger portion of the load than simple designs assumed, relieving the piles and altering the settlement profile.

 

Key Lesson: Comparing “final” FEM predictions with “during construction” monitoring data requires careful adjustment for the degree of consolidation. 

It also validates the concept of the piled raft, where the raft contributes significantly to bearing capacity.

6.3 Crossrail (London): Tunnelling in Clay

Project: Crossrail deep excavations and tunnels in London Clay.

Challenge: Predicting surface settlement troughs to protect historic masonry buildings in London.

Insight:

Traditional empirical methods (Gaussian curves) often overestimate the width of the settlement trough in stiff clays. 3D FEM analyses utilizing the Small-Strain Stiffness model (HSsmall) were able to predict the “peaked” nature of the settlement trough much better than the standard Mohr-Coulomb model.44

Key Lesson: The “greenfield” settlement (no buildings) differs from the settlement with buildings. The stiffness of existing buildings modifies the ground movement—an SSI effect in itself. 

Stiff masonry buildings effectively “bridge” over the settlement trough, suffering less damage than predicted by greenfield curves.46

7. Seismic SSI Provisions & Performance-Based Design

Modern seismic codes have moved towards Performance-Based Seismic Design (PBSD), which relies heavily on accurate SSI modelling.

7.1 ASCE 7-16 and ASCE 41-17 Provisions

Standard provisions now explicitly address SSI. ASCE 7-16 Chapter 19 provides simplified analytical methods for SSI, allowing for the reduction of the base shear to account for soil flexibility and damping. 

ASCE 41-17 (Seismic Evaluation and Retrofit of Existing Buildings) goes further, mandating the use of SSI in certain performance-based evaluations.9

Key provisions include:

• Foundation Damping: Codes allow the inclusion of radiation damping in the analysis, often capped at a certain percentage (e.g., 20%) to prevent unconservative damping values.9
• Flexible Base Springs: The use of uncoupled springs is standard, but for large mats, coupled springs or continuum modelling is encouraged to capture the “rocking” mode correctly.

7.2 FEMA P-2091 Recommendations

The FEMA P-2091 guide provides practical advice for implementing these provisions. 

It highlights that while SSI generally reduces seismic demand (base shear), it increases displacement demand. 

Therefore, if the design is governed by drift limits, SSI might actually control the design. 

It also warns against the “fixed-base” fallacy for tall buildings on soft soil, where period lengthening can shift the structure into a more energetic part of the response spectrum.3

8. Emerging Technologies: The Future of SSI

As computational power grows, the field is moving beyond “one-off” FEM models toward integrated, intelligent workflows.

8.1 BIM to FEM Automation

Traditionally, building a 3D FEM model was a tedious, manual process prone to geometric errors. The new frontier is BIM-to-FEM interoperability.

• Workflow: A parametric BIM model (e.g., in Revit) containing the excavation geometry, piles, and stratigraphy is passed through a script (e.g., Dynamo or Python) or a dedicated tool (like Seequent Leapfrog) to automatically generate the PLAXIS 3D geometry.47
• Benefit: This allows for rapid iteration. If the architect moves a wall or deepens the basement, the geotechnical engineer can update the FEM model and re-run the analysis in minutes, not days. This “live link” ensures the geotechnical model is always synchronized with the structural design.49

 

8.2 Digital Twins and Real-Time Monitoring

The Geotechnical Digital Twin connects the physical excavation with its virtual FEM counterpart via real-time sensor data (inclinometers, strain gauges).

• Mechanism: The framework consists of five dimensions: Physical Space, Virtual Space, Data, Services, and Connections. If the real-time monitoring data deviates from the FEM prediction (e.g., wall deflection is 20% higher than predicted), the Digital Twin can use Inverse Analysis to automatically adjust the soil parameters (e.g., reduce stiffness $E$) until the model matches reality.50
• Value: This provides a dynamic “early warning” system that is far more sophisticated than simple threshold alarms. It predicts future performance based on current calibrated reality, allowing for proactive risk management.51

8.3 Machine Learning (ML) in Settlement Prediction

Running a complex 3D FEM model can take hours or even days. 

Machine Learning models (like Random Forest, XGBoost, or ANN) can predict settlements in milliseconds.

• Hybrid Approach: Engineers are now using FEM to generate thousands of synthetic “training data” scenarios. An ML model is then trained on this data. This “surrogate model” can be used by designers to instantly explore the sensitivity of settlements to different parameters (e.g., “What if the wall is 10% stiffer?”) before committing to a final, rigorous FEM run. Studies have shown that Random Forest and XGBoost models often outperform traditional ANNs in terms of training speed and accuracy for these tabular datasets.52

9. Conclusion: The New Standard of Care

The era of designing deep basements using simple earth pressure coefficients ($K_a$, $K_p$) and independent spring foundations is drawing to a close. 

The complexity of modern urban infrastructure, combined with the stringent requirements for limiting damage to adjacent assets, demands a holistic approach to Soil-Structure Interaction.

Key Takeaways:

1. Embrace Continuum Modelling: For deep, complex excavations, 3D FEM is no longer a luxury; it is a necessity to capture corner effects, dewatering impacts, and construction sequences accurately.
2. Abandon Simple Elasticity: The Mohr-Coulomb model is insufficient for deformation analysis in deep excavations. Advanced models like HSsmall that capture stiffness degradation and stress-dependency are the new standard of care.
3. Validate with Data: Models are only as good as their calibration. Case studies like Taipei 101 and Shanghai Tower prove that while we can predict trends well, precise magnitude prediction requires rigorous calibration against field data and an understanding of time-dependent effects.
4. Integrate and Automate: The future lies in the seamless flow of data from BIM to FEM to Digital Twin, allowing for designs that are not just safe, but optimized, adaptive, and intelligent.

By mastering these advanced modelling techniques, engineers can confidently descend deeper into the earth, securing the foundations of the vertical cities of tomorrow. 

The integration of structural and geotechnical analysis is no longer an academic ideal—it is the practical requirement for the safe and efficient delivery of the world’s most ambitious infrastructure projects.

Works cited

1. Soil-structure interaction – Wikipedia, accessed December 25, 2025, https://en.wikipedia.org/wiki/Soil-structure_interaction
2. Soil-Structure Interaction for Building Structures – National Earthquake Hazards Reduction Program, accessed December 25, 2025, https://www.nehrp.gov/pdf/nistgcr12-917-21.pdf
3. FEMA P-2091, A Practical Guide to Soil-Structure Interaction, accessed December 25, 2025, https://www.fema.gov/sites/default/files/documents/fema-p-2091-soil-structure-interaction.pdf
4. Assessment of Soil-Structure Interaction Modeling Strategies for Response History Analysis of Buildings, accessed December 25, 2025, https://www.iitk.ac.in/nicee/wcee/article/WCEE2012_5552.pdf
5. FEMA P-2091, Webinar on A Practical Guide to Soil-Structure Interaction – YouTube, accessed December 25, 2025, https://www.youtube.com/watch?v=Z_6rRtVgcME
6. Assessment of Soil-Structure Interaction Effects on Seismic Behavior of Isolator and Mass Damper Equipped Buildings – IIETA, accessed December 25, 2025, https://www.iieta.org/download/file/fid/120276
7. Comparative analysis of direct and substructure methods – ResearchGate, accessed December 25, 2025, https://www.researchgate.net/figure/Comparative-analysis-of-direct-and-substructure-methods_tbl1_388441282
8. Dynamic analysis of soil-structure interaction effects on NPP building using simplified and solid FE model of layered subsoil – Extrica, accessed December 25, 2025, https://www.extrica.com/article/20423
9. ASCE/SEI 41-17, accessed December 25, 2025, https://amplify.asce.org/content/standard/9780784440032
10. (PDF) Comparative Analysis of Hardening Soil Models and Field Measurements for Deep Excavation: A Numerical Study – ResearchGate, accessed December 25, 2025, https://www.researchgate.net/publication/374950059_Comparative_Analysis_of_Hardening_Soil_Models_and_Field_Measurements_for_Deep_Excavation_A_Numerical_Study
11. Enhancing Deep Excavation Optimization: Selection of an Appropriate Constitutive Model, accessed December 25, 2025, https://www.mdpi.com/2673-4109/5/3/41
12. Comparative Assessment of Mohr-Coulomb and Hardening Soil Models for Deep Excavation Design in Stratified Soils: Insights from Numerical and Field Data | International Journal of Computational and Experimental Science and Engineering, accessed December 25, 2025, https://www.ijcesen.com/index.php/ijcesen/article/view/3967
13. Failure Criteria Interpretation Based on Mohr-Coulomb Friction – ResearchGate, accessed December 25, 2025, https://www.researchgate.net/publication/248878143_Failure_Criteria_Interpretation_Based_on_Mohr-Coulomb_Friction
14. Eur adopted by Plaxis for analysis using Hardening soil and Hardening soil small strain models – Communities, accessed December 25, 2025, https://bentleysystems.service-now.com/community?id=community_question&sys_id=ac15f9f647dd4a909091861f536d4315
15. 16- Hardening Soil Model with Small Strain Stiffness – PLAXIS – Rocscience, accessed December 25, 2025, https://static.rocscience.cloud/assets/verification-and-theory/RS3/16-Hardening-Soil-Model-with-Small-Strain-Stiffness-PLAXIS.pdf
16. Adaptation and validation of the small strain hardening soil model for sensitive clays in eastern Canada – PhareSST, accessed December 25, 2025, https://pharesst.irsst.qc.ca/cgi/viewcontent.cgi?article=1397&context=actes
17. Deep Excavation Analysis Supported by Anchored Diaphragm Walls: A comparison of Constitutive Models – Semantic Scholar, accessed December 25, 2025, https://pdfs.semanticscholar.org/6713/def0a1cd2bd9a90a53017bab9deaf2711857.pdf
18. Evaluation of Soil Constitutive Models for Analysis of Deep Excavation under Undrained Condition – Civil & Environmental Engineering, accessed December 25, 2025, https://my.civil.utah.edu/~bartlett/PLCC/General%20Info/EVALUATION%20OF%20CLAY%20CONSTITUTIVE%20UNDRAINED%20MODELS%20.pdf
19. Soft-Soil-Creep Model* — FLAC3D 7.0 documentation, accessed December 25, 2025, https://docs.itascacg.com/flac3d700/common/models/softsoilcreep/doc/modelsoftsoilcreep.html
20. FLAC 810 Leaflet – AWS, accessed December 25, 2025, https://itasca-downloads.s3.amazonaws.com/documents/leaflets/ICG19-LFT-FLAC-810-A4-01.pdf
21. Evaluation of the wall deflections of a deep excavation in Central Jakarta using three-dimensional modeling, accessed December 25, 2025, http://homepage.ntu.edu.tw/~khyang/word/Paper/Hsiung%20et%20al.%20(2018),%20TUST.pdf
22. PLAXIS 3D 2025.1 Tutorial Manual – Seequent, accessed December 25, 2025, https://files.seequent.com/PLAXIS/Manuals/PLAXIS_3D/English/PLAXIS_3D_1_Tutorial%20Manual.pdf
23. PLAXIS 2D 2025.1 Tutorial Manual – Seequent, accessed December 25, 2025, https://files.seequent.com/PLAXIS/Manuals/PLAXIS_2D/English/PLAXIS_2D_1_Tutorial%20Manual.pdf
24. Analysis of the corner effect on excavation behaviors – Canadian Science Publishing, accessed December 25, 2025, https://cdnsciencepub.com/doi/10.1139/t98-013
25. CORNER EFFECTS IN DEEP EXCAVATIONS-ESTABLISHMENT OF A FORECAST MODEL FOR TAIPEI BASIN T2 ZONE – Journal of Marine Science and Technology–Taiwan, accessed December 25, 2025, https://jmstt.ntou.edu.tw/cgi/viewcontent.cgi?article=1859&context=journal
26. Three-dimensional numerical analysis of corner effect of an excavation supported by ground anchors – Sci-Hub, accessed December 25, 2025, https://2024.sci-hub.st/7703/177af1e9c1d95736e2c61dc2b41450e0/ahmadi2019.pdf
27. Influence of corners in excavations on damage assessment – White Rose Research Online, accessed December 25, 2025, https://eprints.whiterose.ac.uk/id/eprint/133352/15/jgere.18.00017.pdf
28. Coupling Deformation Induced by Dewatering and Deep Excavation of Subway Station in Water‐Rich Soft Soils | Request PDF – ResearchGate, accessed December 25, 2025, https://www.researchgate.net/publication/398291416_Coupling_Deformation_Induced_by_Dewatering_and_Deep_Excavation_of_Subway_Station_in_Water-Rich_Soft_Soils
29. Research on the Deformation Law of Foundation Excavation and Support Based on Fluid–Solid Coupling Theory – NIH, accessed December 25, 2025, https://pmc.ncbi.nlm.nih.gov/articles/PMC10818629/
30. Hydro-mechanical coupled analyses on wall deformations caused by deep excavations and dewatering in a confined aquifer – ProQuest, accessed December 25, 2025, https://search.proquest.com/openview/17e5b514db60518d5600d9a7433771fa/1?pq-origsite=gscholar&cbl=54451
31. FEM MODELLING OF PILED RAFT FOUNDATIONS IN TWO AND THREE DIMENSIONS – Lund University Publications, accessed December 25, 2025, https://lup.lub.lu.se/student-papers/record/5204468/file/5204470.pdf
32. Three-dimensional Finite Element Analyses of Barrette Piles under Compression and Uplift Loads with Field Data Assessments – ResearchGate, accessed December 25, 2025, https://www.researchgate.net/publication/340527834_Three-dimensional_Finite_Element_Analyses_of_Barrette_Piles_under_Compression_and_Uplift_Loads_with_Field_Data_Assessments
33. PLAXIS Forum – Embedded pile (line element)and volume pile (volume element)’s interfaces, accessed December 25, 2025, https://bentleysystems.service-now.com/community?id=community_question&sys_id=e4a3ae3e1b15ce90f3fc5287624bcb80
34. PLAXIS – Sign In, accessed December 25, 2025, https://bentleysystems.service-now.com/sys_attachment.do?sys_id=d071536487b36a905bea87380cbb350f
35. Numerical and Experimental Evaluation of Axial Load Transfer in Deep Foundations Within Stratified Cohesive Soils – MDPI, accessed December 25, 2025, https://www.mdpi.com/2075-5309/15/15/2723
36. Vertical Behaviour Assessment On An In-Group Barrette Pile – Part I Pile As Volume Elements | PDF | Young’s Modulus | Pascal (Unit) – Scribd, accessed December 25, 2025, https://www.scribd.com/document/933259956/Vertical-Behaviour-Assessment-on-an-in-Group-Barrette-Pile-Part-I-Pile-as-Volume-Elements
37. PLAXIS – Frequently Asked Questions – Interface elements – Communities – Sign In, accessed December 25, 2025, https://bentleysystems.service-now.com/community?id=kb_article_view&sysparm_article=KB0047965
38. Interface Friction Parameters for the Mathematical Modeling of Shell Structures with Infill – eCommons, accessed December 25, 2025, https://ecommons.udayton.edu/context/cee_fac_pub/article/1063/viewcontent/Interface_Friction_Parameters_for_the_Mathematical_Modeling_of_Shell_Structures.pdf
39. Determining the interfaces parameters for geotechnical modelling – E3S Web of Conferences, accessed December 25, 2025, https://www.e3s-conferences.org/articles/e3sconf/pdf/2019/23/e3sconf_form2018_04042.pdf
40. Evaluation of Nonlinear Load Sharing Ratio of Pile and Raft in Piled Raft Foundation in Cohesionless Soil – Indian Geotechnical Society, accessed December 25, 2025, https://www.igs.org.in/storage/proceedings-uploads/TH-03-028-281223033603.pdf
41. Analysis of Piled Raft of Shanghai Tower in Shanghai by the Program ELPLA – GEOTEC Software, accessed December 25, 2025, https://geotecsoftware.com/jdownloads/Practical%20Examples/Analysis%20of%20Piled%20Rafts/8-Shanghai.pdf
42. THREE DIMENSIONAL ANALYSES OF DEEP EXCAVATION IN TAIPEI 101 CONSTRUCTION PROJECT, accessed December 25, 2025, http://yo-1.ct.ntust.edu.tw/jge/files/articlefiles/v2i120070703733207183.pdf
43. Three dimensional analyses of deep excavation in taipei 101 construction project, accessed December 25, 2025, https://www.researchgate.net/publication/295607185_Three_dimensional_analyses_of_deep_excavation_in_taipei_101_construction_project
44. Monitoring settlement in London Clay – Tunnels, accessed December 25, 2025, https://www.tunnelsandtunnelling.com/analysis/monitoring-settlement-in-london-clay/
45. (PDF) Predicting tunnelling-induced ground movements and interpreting field measurements using numerical analysis: Crossrail case study at Hyde Park – ResearchGate, accessed December 25, 2025, https://www.researchgate.net/publication/316946223_Predicting_tunnelling-induced_ground_movements_and_interpreting_field_measurements_using_numerical_analysis_Crossrail_case_study_at_Hyde_Park
46. Multi-Temporal InSAR Structural Damage Assessment: The London Crossrail Case Study, accessed December 25, 2025, https://www.mdpi.com/2072-4292/10/2/287
47. BIM-to-FEM: Development of a Software Tool to Increase the Operational Efficiency of Dam Construction Projects – ResearchGate, accessed December 25, 2025, https://www.researchgate.net/publication/365597478_BIM-to-FEM_Development_of_a_Software_Tool_to_Increase_the_Operational_Efficiency_of_Dam_Construction_Projects
48. Automating stability analysis using BIM-FEA-Integration – ISSMGE, accessed December 25, 2025, https://www.issmge.org/uploads/publications/51/126/386_A_automating_stability_analysis_using_bimfeaintegrat.pdf
49. Automating stability analysis using BIM-FEA-Integration | 37 | Geotech – Taylor & Francis eBooks, accessed December 25, 2025, https://www.taylorfrancis.com/chapters/oa-edit/10.1201/9781003431749-37/automating-stability-analysis-using-bim-fea-integration-beck-j%C3%BCrgens-henke
50. Intelligent Monitoring System for Deep Foundation Pit Based on Digital Twin – MDPI, accessed December 25, 2025, https://www.mdpi.com/2075-5309/15/3/366
51. An Integrated Intelligent Approach for Monitoring and Management of a Deep Foundation Pit in a Subway Station – PMC – NIH, accessed December 25, 2025, https://pmc.ncbi.nlm.nih.gov/articles/PMC9697277/
52. Machine Learning–Finite Element Mesh Optimization-Based Modeling and Prediction of Excavation-Induced Shield Tunnel Ground Settlement | International Journal of Computational Methods – World Scientific Publishing, accessed December 25, 2025, https://www.worldscientific.com/doi/full/10.1142/S021987622450066X
53. Hybrid Modeling for Ground Displacement Prediction Around Excavations: FEA and Machine Learning – ResearchGate, accessed December 25, 2025, https://www.researchgate.net/publication/392894039_Hybrid_Modeling_for_Ground_Displacement_Prediction_Around_Excavations_FEA_and_Machine_Learning

Prediction of Excavation-Induced Displacement Using Interpretable and SSA-Enhanced XGBoost Model – MDPI, accessed December 25, 2025, https://www.mdpi.com/2075-5309/15/23/4372