open menu

Analytics Tools in modeFRONTIER

System level optimization often involves numerous design variables and can be computationally demanding. Even when an appropriate knowledge of complex design spaces is achieved, spotting the final solution from a set of best designs may be challenging.

Analytics tools within modeFRONTIER allow designers to explore the Design Space by investigating interrelated effects of input parameters and extracting relevant information about sensitivities and local influence of variables. Understanding problem characteristics a priori identifies the most significant input variables and the best strategy to initialize the algorithm. The Sensitivity Analysis tools and charts in modeFRONTIER® help make the optimization process more efficient by reducing the problem complexity and thus the number of computations required.

Multi-Variate Analysis

The powerful analytics features support designers in the key steps of the final choice of the best design among the set of candidates resulting from the optimization cycles. Multi-Variate Analysis conveys to engineers relevant insights on the interaction effects between variables, the local behaviors which may differ from global ones and lead to the best final balance among conflicting design attributes (objectives and design variables).

modeFRONTIER offers a sophisticated collection of MVA tools that provide an immediate visualization of patterns and relationships picturing the response of the system at a particular combinations of the variables: Multi-Dimensional Scaling (MDS), Self-Organizing Maps (SOM), Hierarchical and Partitive Clustering, Principal Component Analysis (PCA) and Computer Aided Principle (CAP).


NEW Sensitivity Analysis Tool

With the new Sensitivity Analysis tool in modeFRONTIER, using the Smoothing Spline ANOVA (SSANOVA) proprietary algorithm, users can now perform a variable screening to exclude variables from optimization or RSMs projects.

With a new dedicated table the most important factors and their contributions are highlighted immediately, taking into account both main and interaction effects.

SOM | Self-Organizing Maps

  • Self-Organizing Maps (SOM) are a powerful visual scheme for multi-variate data analysis.
  • Multi-variate data are “projected” into a bi-dimensional regular grid, offering a directly interpretable representation of high-dimensional datasets.
  • The self-organized tool effectively highlights local correlation of multiple variables and hidden cluster structures.

PCA | Principal Component Analysis

  • PCA provides decision makers with a robust guideline for reducing the number of variables of a complex dataset and revealing the simplified structure underlying it.
  • The goal of PCA is to re-express a noisy data set by computing and ordering the most meaningful basis, called principal components.

MDS | Multidimensional Scaling

    • MDS is a statistical technique useful to produce a low-dimensional representation of multivariate datasets.
    • MDS arranges data in the low-dimensional space in order to preserve the observed distances among input points and reveal hidden similarities or dissimilarities.
    • MDS supports designers’ analytics tasks with exploratory data mining allowing for multidimensional data visualization and graphical identification of clusters.

Clustering | Hierarchical and Partitive

    • Clustering techniques help analysts in dividing complex datasets in subgroups which are “similar” among them and “dissimilar” with respect to data belonging to other clusters.
    • Hierarchical Clustering iteratively merges smaller data sets into larger ones and provides a useful tree-like representation of clusters (dendrogram); users may then decide the proper number of clusters suitable for his analysis by cutting the dendrogram structure.
    • Partitive Clustering (K-means clustering) consider Euclidean distance from centroids and groups samples trying to minimize the variation within clusters. Each group represent a different local response of the system and the number of clusters is defined in advance by the user.


Sign in