Subset Simulation Toolbox for MATLAB

This modular toolbox implements an algorithm for Subset Simulation (SuS) using Markov-chain Monte Carlo (MCMC) sampling. This is a more efficient way to estimate rare failure events than traditional Monte Carlo (MC) simulations, where you would need to simulate an exponential number of samples for diminishing event probabilities.

Features

Easy to use

The API is designed so that you can configure your case study with ease to get results quickly. Aside from the user manual, all functions and class definitions are annotated with help text to assist you pick the right options.

Modular object-oriented toolbox design

All key algorithm parts are defined as classes, so that you can either use our implementation for quick results, or you can customise individual parts by extending the abstract class interfaces.

Surrogate model acceleration

Slow-running simulations can easily be accelerated through surrogate models, such as moving-least-squares (MLS). Just specify algo.SurrogateModel = "MLS" to get started. Advanced users can even define their own surrogate models.

Parameter sensitivity analysis

Subset simulation results can be post-processed with out ReportGenerator tool to produce sensitivity rankings and plots. Plots can be aggregated in a template sensitivity report document.

Covariance and Copula dependency models

Stochastic variables from the sample distribution can either be treated as independent (default), or be correlated through a covariance matrix and even Vine Copula models.

Case Studies

Runway overrun risk estimation

In this study (link below), the authors studied the stop margin of B737 and B747 under various environmental conditions (modelled as causal chains) with uncertainties in the aircraft operation parameters. The operation parameters distributions where estimated from actual airline operations, then subset simulation drew (more extreme) samples from these models to obtain a failure probability in the magnitude of 10-9.

doi:10.2514/6.2019-2233

eVTOL Requirement Verification

In this study (link below), the authors analysed the influence of a new navigation solution on the hover performance of an eVTOL in the presence of environment, modelling and measurement uncertainties. Subset simulation was used to estimate the rare event probability of 10-8. The post processor was then used to rank and visualise the most influential parameters from the total set of 200+ uncertain distributions.

doi:10.2514/6.2021-0072

Contact

If you have questions, feedback, bug reports or issues regarding licensing, please contact us at: