Main Concepts

Discipline

Define an input-output discipline to interface a model.

Interfaces: analytic expressions, executable, surrogate model, much more.

Design space

Define a set of parameters, typically design parameters.

Types: deterministic parameter space, uncertain (or mixed) parameter space

Scenario

Define an evaluation process over a design space for a set of disciplines and a given objective.

Types: DOE scenario, MDO scenario

Saving & Storing Data

Store disciplinary evaluations in a cache, either in memory or saved in a file. Use a dataset to store many kinds of data and make them easy to handle for visualization, display and query purposes.

Features

Study Prototyping

An intuitive tool to discover MDO without writing any code, and define the right MDO problem and process. From an Excel workbook, specify your disciplines, design space, objective and constraints, select an MDO formulation and plot both coupling structure (N2 chart) and MDO process (XDSM), even before wrapping any software.

Optimization

Define, solve and post-process an optimization problem from an optimization algorithm.

Algorithms: BFGS, BOBYQA, COBYLA, L-BFGS-B, MMA, NEWUOA, ODD, SLSQP, TNC

based on nlopt, scipy and snopt.

DOE & Trade-off

Define, solve and post-process a trade-off problem from a DOE (design of experiments) algorithm.

Design of experiments: axial, bilevel full-factorial, Box-Behnken, central-composite, composite, custom, diagonal, Faure, full factorial, Halton, Haselgrove, LHS, Monte-Carlo, Plackett-Burman, reverse Halton, Sobol

based on OpenTURNS and pyDOE

MDO formulations

Define the way as the disciplinary coupling is formulated and managed by the optimization or DOE algorithm.

Formulations: bilevel, IDF, MDF, standard optimization

MDA

Find the coupled state of a multidisciplinary system using a Multi-Disciplinary Analysis.

Algorithms: Gauss-Seidel, Jacobi, MDA chain, Newton-Raphson, Quasi-Newton, Gauss-Seidel/Newton

Surrogate models

Replace a discipline by a surrogate one relying on a machine learning regression model.

Surrogate models: Gaussian process regression (kriging), linear model, radial basis regression, polynomial chaos expansion and surrogate quality measures.

based on scikit-learn and OpenTURNS

Scalable models

Use scalable data-driven models to compare MDO formulations and algorithms for different problem dimensions.

Features: scalability study, scalable problem, scalable discipline, diagonal-based, ...

Machine learning

Apply clustering, classification and regression methods from the machine learning community.

Features: clustering, classification, regression, quality measures, calibration, data transformation.

based on scikit-learn and OpenTURNS

Uncertainty

Define, propagate, analyze and manage uncertainties.

Features: distribution, uncertain space, empirical and parametric statistics, distribution fitting, sensitivity analysis, ...

based on OpenTURNS