Nonlinear and Inelastic Structural Analysis

Introduction and Context

Most elementary structural analysis is based on linear assumptions: material properties remain constant, displacements are small, structural geometry does not change significantly, and the response is proportional to the applied load. These assumptions are valuable because they lead to simple equilibrium equations and superposition. However, real structures frequently violate one or more of them.

Concrete cracks and crushes, steel yields and strain-hardens, soil stiffness changes with stress, cables develop geometric stiffness, and slender members may lose stability as their deformed geometry amplifies internal forces. These phenomena require nonlinear and, often, inelastic analysis.

Nonlinear analysis is therefore not merely a more complicated form of linear analysis. It changes the mathematical character of the problem. The stiffness may depend on displacement, stress, strain, loading history, or contact conditions. Equilibrium may have multiple solutions, and the response may include yielding, unloading, snap-through, bifurcation, or collapse.

In advanced structural engineering, nonlinear analysis is used for performance-based seismic design, assessment of existing buildings, collapse analysis, bridge evaluation, reinforced-concrete design, steel plastic analysis, cable structures, offshore structures, and research into structural robustness.

Learning Objectives

After studying this chapter, the reader should be able to:

1. distinguish geometric, material, and boundary-condition nonlinearity;
2. formulate nonlinear equilibrium and incremental-iterative equations;
3. explain elastic, plastic, viscoelastic, and damage-related structural response;
4. apply yield criteria, flow rules, hardening laws, and return-mapping concepts;
5. explain Newton–Raphson, modified Newton, displacement-control, and arc-length procedures;
6. interpret load–displacement curves, tangent stiffness, limit points, and instability;
7. conduct a basic nonlinear analysis and assess convergence and modelling quality;
8. evaluate professional, regulatory, ethical, and practical implications of nonlinear results.

1. Meaning of Nonlinearity and Inelasticity

A structure is linear when its response satisfies both homogeneity and additivity. If load vector P produces displacement vector u, then a linear system satisfies:

where K is a constant stiffness matrix. The principle of superposition follows: the response to P₁ + P₂ equals the sum of the responses to P₁ and P₂.

In a nonlinear system, the equilibrium relationship is more generally written as:

Here, R is the residual force vector, u is the displacement vector, q represents internal variables such as plastic strain or damage, P is a reference load vector, λ is a load factor, and F_int is the internal resisting-force vector. Equilibrium is satisfied when R = 0.

The tangent stiffness is the derivative of internal force with respect to displacement:

Unlike the linear stiffness matrix, K_t changes during the analysis and may become nonsymmetric when the formulation includes nonconservative forces, friction, or certain nonassociative constitutive laws.

Nonlinearity may be classified as follows.

Inelasticity is a broader concept concerning irreversible deformation or energy dissipation. Plastic deformation is inelastic, but inelastic response can also result from cracking, damage, friction, crushing, and viscous mechanisms.

2. Geometric Nonlinearity

2.1 Sources of geometric effects

Geometric nonlinearity arises when changes in geometry significantly affect equilibrium or strain. It is important when displacements are not negligible relative to member dimensions, when axial forces act on laterally displaced members, or when instability is possible.

Two commonly distinguished effects are:

• P–Δ effect: a global second-order effect caused by axial load acting through the lateral displacement of a member or frame;
• P–δ effect: a local second-order effect caused by curvature and local bowing within a member.

For a column carrying compressive force P and having lateral displacement Δ, the additional overturning moment is approximately:

where M₂ has units of force-length. This expression is a second-order contribution and is meaningful only when the force and displacement are defined consistently in the same configuration.

2.2 Stability and critical load

For an ideal pin-ended elastic column, Euler’s critical load is:

where E is Young’s modulus, I is the second moment of area, and L is the effective length for the stated end conditions. The equation assumes a perfectly straight, prismatic, elastic member with ideal end restraints and concentric loading. Real columns are affected by imperfections, residual stresses, material yielding, and connection flexibility; therefore, Euler load is a theoretical bifurcation value rather than a universal design resistance.

As the applied load approaches a critical condition, the tangent stiffness associated with a deformation mode may approach zero. This explains why small additional loads can produce large displacement changes.

2.3 Total and updated Lagrangian descriptions

A total Lagrangian formulation refers all quantities to the original configuration. An updated Lagrangian formulation refers increments to the current configuration. Both can represent large-displacement behaviour when consistently derived. The choice affects implementation, numerical efficiency, and treatment of rotations and stresses, but not the underlying physical requirement that equilibrium and compatibility be satisfied in an appropriate configuration.

3. Material Nonlinearity and Inelastic Constitutive Behaviour

3.1 One-dimensional elastic–plastic model

For a uniaxial material, total strain may be decomposed into elastic and plastic parts:

where ε is total strain, εₑ is recoverable elastic strain, and εₚ is irreversible plastic strain. For linear elasticity:

where σ is stress and E is Young’s modulus.

A perfectly elastic–plastic material has:

where σᵧ is the yield stress. If f < 0, the response is elastic. Plastic loading occurs when f = 0 and the loading conditions permit further plastic flow.

For a bilinear hardening model, the post-yield tangent modulus Eₜ is smaller than E. A positive Eₜ represents strain hardening; Eₜ = 0 represents perfect plasticity. In numerical models, a small positive tangent often improves stability, but it must be physically justified rather than selected only for computational convenience.

3.2 Multiaxial plasticity

For three-dimensional stress states, yielding is represented by a yield function:

where σ is the stress tensor and q denotes internal hardening variables.

For ductile metals, the von Mises yield criterion is commonly written as:

where J₂ is the second invariant of the deviatoric stress tensor and σᵧ is the current yield stress. Equivalently, using principal stresses:

Yielding occurs when σ_eq = σᵧ. This criterion is appropriate mainly for isotropic ductile metals and does not directly represent pressure-sensitive materials such as concrete, rock, or soil.

For isotropic hardening, the yield surface expands as plastic deformation accumulates. For kinematic hardening, the surface translates in stress space, which is important for cyclic loading and the Bauschinger effect. Combined hardening includes both expansion and translation.

3.3 Flow rule and hardening

A plastic potential g defines the direction of plastic flow. Under an associated flow rule:

where dεₚ is the plastic strain increment and dλ is a nonnegative plastic multiplier. For associated plasticity, g = f. Nonassociated flow uses g ≠ f, which is often more realistic for frictional geomaterials but may produce a nonsymmetric tangent matrix.

The Kuhn–Tucker conditions are:

They express, respectively, admissibility, nonnegative plastic loading, and complementarity between plastic flow and the yield surface.

3.4 Concrete and reinforced concrete

Concrete has markedly different tensile and compressive behaviour. In tension, cracking may occur at relatively small strain and stiffness may reduce sharply. In compression, nonlinear hardening may be followed by softening and crushing. Reinforcing steel generally exhibits yielding, strain hardening, and cyclic Bauschinger effects.

Important modelling choices include:

• discrete cracks versus smeared cracking;
• tension stiffening between cracks;
• compression softening;
• fracture-energy regularization;
• confinement effects;
• bond-slip between reinforcement and concrete;
• cyclic degradation and pinching.

A local softening stress–strain curve can lead to mesh-sensitive finite-element results because deformation localizes into an element whose size controls the dissipated energy. Crack-band approaches, nonlocal models, or fracture-energy-based formulations are used to reduce this dependence.

4. Incremental Formulation of Nonlinear Equilibrium

At load step n, suppose equilibrium is satisfied at displacement uⁿ and internal variables qⁿ. At the next step, the desired solution satisfies:

Because this equation is nonlinear, it is normally solved iteratively. Let uᵢ be the displacement estimate at iteration i. Linearization gives:

with the sign convention dependent on how the residual is defined. If R = λP − F_int, then:

and the updated displacement is:

The iterations continue until force and displacement convergence criteria are satisfied.

A common force criterion is:

where ||·|| is a selected norm and ηᵣ is a prescribed tolerance. A displacement criterion may be written:

The denominator must be protected against zero or very small displacement values. Convergence tolerances should be related to the required engineering accuracy, not chosen arbitrarily.

5. Newton–Raphson and Alternative Solution Procedures

5.1 Full Newton–Raphson method

The full Newton method forms and factorizes the current tangent stiffness at every iteration. It has rapid, approximately quadratic convergence near a regular solution, provided the tangent is consistent and the starting estimate is sufficiently close.

Advantages include strong local convergence and good performance for strongly nonlinear constitutive behaviour. Disadvantages include high computational cost and possible difficulty near limit points or singular stiffness states.

5.2 Modified Newton method

The modified Newton method retains the tangent stiffness from the beginning of a load step or updates it less frequently. Each iteration is cheaper, but convergence is generally slower and may become linear rather than quadratic. It can be effective when factorization dominates total computational cost and the tangent does not change drastically.

5.3 Load control

Under load control, the load factor is prescribed and displacement is solved. This is satisfactory on stable ascending branches but may fail at a limit point where displacement changes greatly for little or no load increase.

5.4 Displacement control

A selected displacement degree of freedom is prescribed while the load factor becomes an unknown. This method can trace portions of a response curve that load control cannot follow, particularly when displacement is monotonic.

5.5 Arc-length control

Arc-length methods constrain the combined increment in load and displacement. A simplified constraint may be represented as:

where Δs is the prescribed arc-length radius and α scales the load factor relative to displacement units. The method can pass limit points and trace snap-through or snap-back response, although it requires careful selection of the constraint and may follow an undesired equilibrium branch.

6. Return Mapping and Constitutive Integration

At each integration point in a finite-element model, a trial elastic state is first calculated using the strain increment. If the trial stress satisfies the yield condition, the step is elastic. If it violates the yield condition, a plastic correction is applied to return the stress to the admissible yield surface.

For a one-dimensional perfect elastic–plastic material:

1. calculate trial stress, σ_trial = σ_n + EΔε;
2. evaluate f_trial = |σ_trial| − σ_y;
3. if f_trial ≤ 0, accept the trial state;
4. otherwise project the stress to the yield stress with the appropriate sign and update plastic strain.

The algorithm must update stress, plastic strain, hardening variables, and the consistent tangent. A consistent algorithmic tangent improves global Newton convergence because it is the derivative of the numerically integrated constitutive response, not merely the continuum tangent.

For finite-element analysis, constitutive integration is local, whereas global equilibrium is assembled over the structure. A failure in local integration can produce apparently mysterious global nonconvergence; therefore, material-point diagnostics are essential.

7. Plastic Analysis of Frames

Plastic analysis idealizes a structural member as elastic until plastic hinges form. A plastic hinge is not a literal hinge at first yield; it is a region in which rotations increase substantially while moment remains near the plastic moment capacity.

For a rectangular reinforced-concrete or steel section, the plastic moment must be obtained from a compatible stress-resultant calculation. In general:

where A is the cross-sectional area, σ(y) is the stress distribution, and y is measured from the chosen reference axis. The stress distribution must satisfy axial-force equilibrium and material constitutive assumptions.

Plastic collapse of a frame occurs when sufficient plastic hinges form a mechanism, subject to equilibrium, compatibility, and kinematic admissibility. The collapse load predicted by ideal plastic analysis may be unsafe if local buckling, fracture, cyclic degradation, connection failure, or second-order effects are ignored.

8. Worked Example: Second-Order Amplification

Consider a pin-ended elastic column with:

• length L = 3.0 m;
• axial load P = 600 kN;
• elastic modulus E = 200 GPa;
• second moment of area I = 8.0 × 10⁻⁶ m⁴;
• first-order lateral displacement Δ₁ = 12 mm = 0.012 m.

The Euler critical load is:

Substitution gives:

A commonly used approximate amplification factor for a linearized second-order estimate is:

This approximation assumes an elastic member, a suitable first-mode representation, and no significant material yielding or geometric imperfection beyond that represented by Δ₁.

Thus:

The amplified displacement is approximately:

The first-order moment associated with the axial load and initial displacement is:

The approximate amplified moment is:

This example is not a substitute for a geometric nonlinear analysis. It demonstrates the sensitivity of response as P/Pcr increases and shows why stability effects must not be ignored in slender compression members.

9. Applied Case: Nonlinear Seismic Assessment of an Existing Frame

Background

Consider an illustrative four-storey reinforced-concrete office frame designed under older code provisions. An assessment is required because the building has irregular infill walls, uncertain reinforcement detailing, and a proposed change of occupancy.

Problem

A linear elastic analysis predicts acceptable member forces, but it cannot establish whether the frame develops a stable, ductile mechanism under severe earthquake demand. Potential brittle shear failure and soft-storey behaviour are concerns.

Analysis approach

A two-dimensional nonlinear model is developed with:

• fibre-based beam-column sections;
• nonlinear shear checks and, where necessary, explicit shear springs;
• concentrated or distributed plasticity calibrated to section response;
• nonlinear joint-panel behaviour;
• cracked concrete stiffness;
• reinforcing-steel yielding and strain hardening;
• geometric nonlinearity for P–Δ effects;
• measured or conservatively estimated material properties.

A gravity-load step is first applied and converged. Lateral load is then increased using a displacement-control procedure. The analysis records base shear, roof displacement, inter-storey drift, hinge rotations, concrete compression damage, reinforcement strain, and residual drift.

Findings

The frame initially responds almost elastically. Cracking reduces stiffness at moderate drift. Beam plastic hinges form at selected floors, but an unexpected concentration of drift develops at the first storey because infill walls terminate above it. When P–Δ effects are included, the post-peak resistance decreases. A shear-critical column reaches its limit before a stable beam-sway mechanism forms.

Engineering response

Possible interventions include column jacketing, addition of walls or braces, joint strengthening, improvement of confinement, and modification of infill-wall layout. The preferred intervention is selected not only from peak strength but also from deformation capacity, constructability, cost, disruption, and robustness.

Learning points

1. A nonlinear model exposes failure modes that linear force results may conceal.
2. Modelling assumptions about joints, shear, infill, and cyclic degradation can govern conclusions.
3. A converged numerical solution is not automatically a physically reliable prediction.
4. Performance should be judged using appropriate acceptance criteria, not a single maximum load.
5. Independent review and sensitivity analysis are essential for high-consequence decisions.

10. Verification, Validation, and Interpretation

Nonlinear analysis requires three distinct forms of confidence:

• verification: the equations and computer implementation have been solved correctly;
• validation: the model represents the physical system adequately for the intended use;
• engineering judgement: the results are interpreted within uncertainty and code requirements.

Good practice includes:

• checking units and sign conventions;
• testing a simple member against a hand solution;
• comparing mesh refinements;
• examining energy balance;
• checking equilibrium residuals;
• monitoring local constitutive variables;
• testing sensitivity to material properties and damping;
• documenting load paths and convergence controls;
• distinguishing numerical failure from physical failure.

Energy quantities can be useful. External work is approximately:

and internal work is:

For a properly implemented quasistatic analysis, external and internal work should agree within numerical tolerance, subject to boundary and numerical treatment.

11. Limitations, Governance, and Professional Responsibility

Nonlinear results can appear highly precise while being uncertain. Major sources of uncertainty include material variability, construction defects, deterioration, loading history, boundary conditions, damping, fracture properties, and unmodelled failure modes.

Engineers should not report a nonlinear capacity without stating the model scope, assumptions, calibration evidence, convergence criteria, acceptance limits, and governing failure mechanism. Numerical sophistication does not compensate for incorrect structural idealization.

Relevant standards vary by jurisdiction and application. They may specify constitutive models, ductility assumptions, acceptance criteria, load combinations, cyclic protocols, and requirements for independent checking. The governing building, bridge, seismic, and material standards must therefore be consulted for a project-specific assessment.

Artificial intelligence and automated model generation may assist with parameter studies, anomaly detection, mesh-quality checks, and surrogate modelling. However, AI-generated constitutive parameters or failure predictions must be independently verified. Traceability, reproducibility, data provenance, and human responsibility remain essential.

12. Contemporary Developments and Best Practice

Current research and professional practice increasingly address:

• finite-element models with fracture-energy regularization;
• multiscale and physics-informed constitutive models;
• probabilistic nonlinear analysis and fragility assessment;
• digital twins updated using structural monitoring;
• hybrid simulation combining physical and numerical substructures;
• machine-learning surrogates for repeated nonlinear analyses;
• robust assessment under deterioration and uncertain hazards;
• resilience-based design considering repair time and functionality.

Best practice is to use the simplest model capable of answering the engineering question, increase complexity only when it changes the decision, and perform sensitivity studies on assumptions that materially influence the outcome.

Chapter Summary

Nonlinear structural analysis is required when material properties, geometry, boundary conditions, or loading history cause the response to depart from proportional elastic behaviour. Material nonlinearity includes yielding, cracking, crushing, plastic flow, damage, and cyclic degradation. Geometric nonlinearity includes P–Δ effects, large displacement, instability, and changes in equilibrium configuration.

The central nonlinear equilibrium equation is expressed through a residual force vector, and solution normally requires incremental-iterative procedures. Newton–Raphson, modified Newton, displacement control, and arc-length methods each have appropriate applications. Constitutive integration commonly uses trial states and return mapping, while consistent tangents improve global convergence.

Nonlinear results must be verified computationally, validated physically, and interpreted through engineering judgement. Convergence is necessary but not sufficient: the model must also represent the relevant failure mechanisms. Professional practice requires transparent assumptions, sensitivity analysis, compliance with applicable standards, and independent review for important decisions.

Key Terms

Nonlinearity: Departure from proportional load–response behaviour.

Inelasticity: Irreversible deformation or energy dissipation.

Tangent stiffness: Derivative of resisting force with respect to displacement.

Geometric nonlinearity: Nonlinearity caused by changes in structural geometry.

Material nonlinearity: Nonlinearity caused by constitutive behaviour.

P–Δ effect: Global second-order effect from axial load acting through lateral displacement.

Plasticity: Irreversible deformation governed by a yield condition and flow rule.

Yield function: Function defining the boundary between elastic and plastic states.

Plastic multiplier: Scalar controlling the magnitude of plastic strain increment.

Hardening: Increase or translation of the yield surface during plastic loading.

Kinematic hardening: Translation of the yield surface in stress space.

Isotropic hardening: Expansion of the yield surface.

Plastic hinge: Region capable of sustaining large rotation near a plastic moment.

Limit point: Point at which the load–displacement path reaches a local maximum or minimum.

Snap-through: Sudden change to another equilibrium configuration.

Return mapping: Constitutive algorithm that projects a trial state onto the admissible yield surface.

Residual: Difference between externally applied and internally resisted forces.

Arc-length method: Iterative procedure that controls a combined load–displacement increment.

Verification: Demonstration that equations or software are implemented correctly.

Validation: Demonstration that a model adequately represents physical behaviour.

Review Questions

Short Answer Questions

1. Distinguish material nonlinearity from geometric nonlinearity.
2. What is the physical meaning of the tangent stiffness matrix?
3. State the Kuhn–Tucker conditions for plastic loading.
4. Why can load control fail near a limit point?
5. What is the purpose of return mapping in constitutive integration?

Long Answer Questions

1. Explain the formulation and solution of nonlinear equilibrium using the Newton–Raphson method.
2. Discuss the assumptions and limitations of Euler’s critical-load equation.
3. Compare isotropic hardening, kinematic hardening, and combined hardening under cyclic loading.
4. Explain the principal modelling issues in nonlinear analysis of reinforced-concrete frames.
5. Discuss verification, validation, convergence, and engineering judgement in nonlinear finite-element analysis.

Analytical / Application Questions

1. A steel column has an axial load equal to 0.70 of its elastic critical load. Explain qualitatively how second-order amplification affects displacement and moment, and identify the assumptions required for a simplified amplification estimate.
2. Design a nonlinear analysis strategy for an existing irregular reinforced-concrete building subject to earthquake loading. Specify material models, geometric effects, solution control, response quantities, and acceptance checks.
3. A nonlinear analysis converges numerically but predicts a highly mesh-dependent crack pattern. Diagnose the likely causes and propose modelling and sensitivity measures to improve the reliability of the conclusion.

Multiple Choice Questions with Answers

Question 1

Which condition most directly represents nonlinear equilibrium?

A. K is constant and P = Ku

B. R(u, q, λ) = 0 with internal forces depending on state

C. All loads are mutually independent

D. Stress is always proportional to total strain

Correct Answer: B. R(u, q, λ) = 0 with internal forces depending on state

Question 2

The P–Δ moment for axial load P and lateral displacement Δ is approximately:

A. P/Δ

B. Δ/P

C. PΔ

D. P + Δ

Correct Answer: C. PΔ

Question 3

Euler’s critical load is proportional to which quantity for a pin-ended elastic column?

A. L²/(EI)

B. EI/L²

C. EI/L

D. E/L²I

Correct Answer: B. EI/L²

Question 4

In perfect plasticity, the yield stress after yielding is assumed to:

A. Remain constant

B. Increase linearly without limit

C. Become zero

D. Depend only on elastic strain

Correct Answer: A. Remain constant

Question 5

The von Mises criterion is primarily associated with:

A. Ductile metals

B. Unreinforced masonry only

C. Ideal fluids

D. Timber creep only

Correct Answer: A. Ductile metals

Question 6

Under an associated flow rule:

A. The plastic potential equals the yield function

B. Plastic strain is always zero

C. The tangent stiffness is always zero

D. The material must be elastic

Correct Answer: A. The plastic potential equals the yield function

Question 7

Which method is particularly useful for tracing equilibrium paths through limit points?

A. Simple load control only

B. Arc-length control

C. Elastic superposition

D. Allowable-stress scaling

Correct Answer: B. Arc-length control

Question 8

For the worked column example, the approximate Euler critical load was:

A. 175 kN

B. 600 kN

C. 1,754 kN

D. 17,540 kN

Correct Answer: C. 1,754 kN

Question 9

A converged nonlinear solution is not necessarily reliable because:

A. Convergence proves all failure modes were modelled

B. The physical model and assumptions may still be inappropriate

C. Residual forces must always be large

D. Material properties are irrelevant after convergence

Correct Answer: B. The physical model and assumptions may still be inappropriate

Question 10

Return mapping is used to:

A. Convert force units

B. Project an inadmissible trial stress onto the yield surface

C. Remove geometric stiffness

D. Replace equilibrium with superposition

Correct Answer: B. Project an inadmissible trial stress onto the yield surface