Sondipon Adhikari · Glasgow

Research theme three of five

Uncertainty quantification and stochastic mechanics

Propagating uncertainty through a dynamic model is a different problem from propagating it through a static one, because the quantities of interest are spectra and resonances.

The quantity that matters in a reliability calculation is the tail, and the tail is exactly where a surrogate fitted to the bulk is least trustworthy.

The problem

General-purpose uncertainty quantification is well served by existing tools. The gap is the dynamic case, where the map from parameters to response is rough: eigenvalues cross and veer, resonances move through the observation band, and a small parameter change can reorder the modes. Polynomial approximations built for smooth maps degrade badly there.

The response of this theme is to work with the operators themselves: random eigenvalue problems, random matrix models of the system matrices themselves, and closed-form ensemble quantities wherever they can be obtained.

Why it matters

A safety case is an argument about probability, and for a dynamic structure the quantity it turns on is the location of a resonance or the chance of exceeding a limit. Both live in the tails, and both move discontinuously as parameters change, which is where the general-purpose surrogate machinery loses its guarantees. Composite aerospace structures are the sharpest case: the layup process produces scatter that the design has to absorb, and the aeroelastic response is exactly the quantity that scatter moves. Getting this right lets a manufacturer set a probabilistic margin from the process data it already collects, and the same argument sets the variance on the ten-year yield that finances a wind farm.

Current frontier

  • stochastic structural dynamics and random operators
  • random eigenvalue problems and random spectra
  • random matrix and Wishart system models
  • random fields and Karhunen-Loeve representations
  • stochastic Green functions and ensemble response
  • rare events and reliability in dynamic systems
  • stochastic metamaterials and stochastic localisation
  • surrogate modelling and operator learning

Signature concepts

  • random operators
  • ensemble response
  • rare events
  • stochastic localisation
  • measure transport

Open problems

  • Which ensemble of random matrices is the right model for a given class of manufactured structure?
  • How should a surrogate be trained when the quantity of interest is a failure probability?
  • What closed-form ensemble results exist beyond the Gaussian case?

Selected papers

  • Uncertainty quantification in inerter-based quasiperiodic lattices
    T. Chatterjee, D. Karličić, M. Cajić, S. Adhikari, M. I. Friswell · International Journal of Mechanical Sciences 249, 108258 · 2023
  • A comparative analysis of intrusive and non-intrusive PCE methods for random mode computation
    E. Jacquelin, S. Adhikari, D. Brizard · Probabilistic Engineering Mechanics 81, 103792 · 2025
  • Enhanced multi-fidelity modeling for digital twin and uncertainty quantification
    A. S. Desai, N. Navaneeth, S. Adhikari, S. Chakraborty · Probabilistic Engineering Mechanics 74, 103525 · 2023
  • Seismic reliability analysis of nonlinear structures by active learning-based adaptive sparse Bayesian regressions
    A. Roy, S. Chakraborty, S. Adhikari · International Journal of Non-Linear Mechanics 165, 104817 · 2024
  • Dynamic analysis of wind turbine towers on flexible foundations
    S. Adhikari, S. Bhattacharya · Shock and Vibration · 2012

Read more papers in this area

Foundations

Random vibration, Karhunen-Loeve expansions, stochastic finite elements, polynomial chaos, structural reliability, uncertainty propagation and stochastic model reduction, developed from 2001 onward and applied with Embraer to composite aerospace structures.