Modal Analysis / Frequency Analysis — Overview

Natural frequencies, mode shapes, eigenvalue extraction methods, resonance avoidance strategies, and solver comparisons across Abaqus, ANSYS, and Nastran.

Articles in This Section

1. What Is Modal Analysis?

Modal analysis — sometimes called free vibration analysis or normal modes analysis — is the process of determining the natural frequencies (eigenfrequencies) and corresponding mode shapes (eigenvectors) of a structure. Every real structure has a set of these characteristic vibration patterns, much like a guitar string has overtones. Each mode describes the shape the structure tends to deform into when vibrating at that particular frequency.

The result of a modal analysis is typically a list of mode pairs: \( (\omega_i, \{\phi_i\}) \), where \( \omega_i \) is the i-th natural angular frequency (rad/s) and \( \{\phi_i\} \) is the corresponding mode shape vector. From \( \omega_i \) the natural frequency in Hz is simply:

\[ f_i = \frac{\omega_i}{2\pi} \]

These results are the foundation for nearly all subsequent dynamic analyses — harmonic response, transient dynamics, random vibration, and response spectrum methods all use the modal basis to reduce computational cost dramatically.

2. Why It Matters: Resonance and Vibration Qualification

The single most important reason engineers run modal analysis is resonance avoidance. When a structure is excited at or near one of its natural frequencies, the response amplitude can grow to destructive levels if damping is insufficient. Classic real-world failures include bridge deck oscillations (Tacoma Narrows), turbine blade resonance, and PCB (printed circuit board) fatigue in electronics.

In defense and aerospace applications the qualification standard MIL-STD-810 (and its NATO equivalent AECTP 400) requires that equipment demonstrate either (a) no resonance within the specified vibration bandwidth, or (b) sufficient structural integrity even when resonances are excited. Modal analysis drives this qualification process by identifying all problematic natural frequencies before physical testing begins, drastically reducing prototype iteration cycles.

3. Governing Equation of Motion

For a discretised FEM model under free vibration, the equation of motion is:

\[ [M]\{\ddot{u}\} + [C]\{\dot{u}\} + [K]\{u\} = \{0\} \]

where \([M]\) is the global mass matrix, \([C]\) is the damping matrix, \([K]\) is the global stiffness matrix, and \(\{u\}\) is the nodal displacement vector. The dots denote time derivatives.

For the undamped free vibration case (\([C] = 0\)), assuming a harmonic solution \(\{u\} = \{\phi\} e^{i\omega t}\), this reduces to the classical eigenvalue problem:

\[ \left([K] - \omega^2 [M]\right)\{\phi\} = \{0\} \]

Non-trivial solutions exist only when \(\det([K] - \omega^2[M]) = 0\), which yields the characteristic polynomial whose roots are the squared natural frequencies \(\omega_i^2\). In practice, direct polynomial expansion is never used for large FEM models — iterative eigensolvers (see Section 5) are employed instead.

4. Types of Modal Analysis

5. Mass Matrix Formulation: Consistent vs. Lumped

The accuracy of natural frequencies depends heavily on how mass is distributed within each element — this is controlled by the mass matrix formulation.

The consistent mass matrix uses the same shape functions \([N]\) as for the stiffness matrix:

\[ [M]^e = \int_V \rho\, [N]^T [N]\, dV \]

This gives a full (non-diagonal) element matrix and is more accurate for flexural modes but increases storage and solve cost. The lumped mass matrix distributes total element mass equally to each node, yielding a diagonal matrix which is cheaper to invert and better for explicit time integration, but can underestimate natural frequencies for bending-dominated problems.

Most commercial codes offer a mixed or HRZ lumped (Hinton-Rock-Zienkiewicz) approach, which preserves the diagonal structure while improving accuracy over simple lumping.

6. Eigenvalue Extraction Methods

For large FEM models (millions of DOFs), extracting even a few hundred eigenvalues requires highly efficient iterative algorithms. The dominant methods are:

Lanczos Method

The Lanczos algorithm projects the problem onto a much smaller Krylov subspace, finding eigenvalues at the extremes of the spectrum first. It is the default in Abaqus (*FREQUENCY, EIGENSOLVER=LANCZOS) and delivers excellent convergence for the lowest modes. The block variant processes multiple vectors simultaneously, which is more numerically stable and is used in ANSYS (Block Lanczos).

Subspace Iteration

An older but robust method that iterates a subspace of trial vectors simultaneously. It is unconditionally stable and handles near-repeated eigenvalues better than classical Lanczos, but is generally slower for large problems.

Power Dynamics / PCG Lanczos (ANSYS)

ANSYS offers a PCG Lanczos variant that uses an iterative (preconditioned conjugate gradient) solver for the linear systems within Lanczos steps, making it memory-efficient for very large models exceeding 10 million DOFs.

ARPACK (Abaqus AMS / open-source)

For extremely large problems, Abaqus provides the Automated Multi-level Sub-structuring (AMS) solver, which uses ARPACK internally and is faster than Lanczos when requesting many modes (>200).

7. Software Comparison

8. Common Engineering Applications

Automotive NVH

NVH engineers perform full-vehicle modal analysis (body-in-white, trimmed body) to identify structural modes that fall within the 20–200 Hz range where road and engine excitation is strongest. Targets include minimizing cabin noise at idle (engine second-order harmonic at ~40 Hz for a 4-cylinder at 1200 rpm) and booming sounds in the 80–120 Hz band.

Aerospace Structures

Flight qualification requires demonstrating that structural modes do not couple with control surface flutter or aeroelastic excitation. Satellite launch certification (ECSS-E-ST-32C) mandates that primary modes exceed the launch vehicle's maximum dynamic pressure excitation band, typically up to 100 Hz.

Rotating Machinery — Campbell Diagram

For turbines, compressors, and fans, natural frequencies must not coincide with engine order (EO) excitation lines across the operating speed range. The Campbell diagram plots natural frequency vs. rotational speed alongside EO lines; crossing points indicate potential resonance and must be avoided or verified to have sufficient margin. Pre-stressed modal analysis (with centrifugal stiffening) is essential here.

\[ f_\text{excitation} = n \cdot \frac{\text{RPM}}{60}, \quad n = 1, 2, 3, \ldots \]

Electronics and PCBs

Consumer electronics must pass JEDEC JESD22-B103 and IEC 60068-2-64 random vibration tests. Modal analysis identifies PCB modes in the 50–2000 Hz range; high-stress zones at mode antinodes indicate solder joint fatigue risk. Damping pads are positioned at mode shape maxima to be most effective.

9. Mode Participation Factors and Effective Mass

Not all modes contribute equally to the dynamic response. The modal participation factor \(\Gamma_i\) for excitation direction \(\{D\}\) is:

\[ \Gamma_i = \{\phi_i\}^T [M] \{D\} \]

The effective mass for mode \(i\) is \(m_{\text{eff},i} = \Gamma_i^2\) (when modes are mass-normalised). A standard rule of thumb is to include enough modes to capture at least 90% of total effective mass in each excitation direction, guaranteeing that the response calculation is not truncated prematurely.

Residual vectors (also called static correction vectors or residual attachment modes) can be added to account for the contribution of truncated high-frequency modes to quasi-static response — all major solvers support this option.

10. Practical Modelling Tips

• Mesh density: As a rule, use at least 6 elements per wavelength for the highest mode of interest. Under-meshing artificially raises natural frequencies.
• Rigid body modes: For free-free modal analysis, expect 6 near-zero rigid body modes. Frequencies below ~0.1% of the lowest elastic mode are typically numerical noise. Use shift-invert spectral transformation to isolate elastic modes cleanly.
• Constraint modes vs. fixed-free: Fixed boundary conditions eliminate rigid body modes but introduce artificial stiffness. Free-free modal is more general and should be preferred when base-excitation analyses will follow.
• Contact in modal analysis: Nonlinear contact cannot be linearised in a standard modal step. Open contacts are typically either fully bonded (optimistic) or fully open (conservative) for the modal extraction, then a contact verification is done separately.
• Mass scaling: Never apply mass scaling (an explicit dynamics trick) to implicit modal models — it corrupts all natural frequencies proportionally to the scale factor.

Concept Q&A — Student & Professor