Analyzing large data is the process of methodically and systematically making decisions to reduce incomprehensible datasets down to a manageable size that can be viewed and understood easily. These reductions are made by performing data analytics, producing metrics, identifying patterns, and/or producing any other criteria of comparison that can be mathematically modeled.

With the use of High Performance Computing (HPC) and the ever-shrinking limitations to storage and processing, the size of data on which analytics can be performed has been growing exponentially. Datasets that originally took years to manually process can now be interpreted and visualized in days or weeks.

TradeAnalyzer enables the user to readily analyze data through multiple methods. Custom R-scripts and objective functions can be applied to the tradespace in order to create additional metrics or reduce the tradespace to a more manageable size for future assessments. A tradespace is defined as the set of system and program parameters, properties, and facets needed to meet specific performance standards.

Custom R-scripts are the primary technique of analytics in the tool suite. Mathematical models, or machine learning algorithms written in R, can be applied to the tradespace and executed to create plots, metrics, or reduce the tradespace for further analysis. Objective functions can be written to link the interaction between the data and goals and used to reduce the tradespace for further analysis.

The implementation of set-based design (SBD) and the use of HPC are two related pillars of Engineered Resilient Systems (ERS) that require large tradespace visualization. The coupling of SBD and HPC generates large amounts of complex data that surpass conventional desktop-based data processing methods. As opposed to point-based design where the design cycle begins with a single design, SBD begins with a sufficient number of design alternatives to cover the feasible tradespace. This allows for analytical tradeoffs between system attributes and competing requirements.

Hundreds of thousands of design-alternatives are often necessary to fully cover the tradespace. Generating these large datasets are often beyond the capacity of the standard desktop computers, thus, HPC is integral to the ERS workflow. The data output from the HPC machines is often too large to be ingested into typical desktop programs. In addition to technology constraints, the computer science expertise and time required to comprehend this data are typically outside the purview of most potential users.

TradeAnalyzer enables the user to readily explore large tradespaces using a variety of interactive visualizations, including histograms, two-dimensional (2D) and three-dimensional (3D) scatterplots, mutual information (MI) diagrams, parallel coordinate plots, and platform alternative scoring diagrams. These capabilities enable the user to explore multi-dimensional tradespaces. Additionally, visualizations can be rendered using custom R-scripts. The scripts can help to detect linear and non-linear relationships across attributes, find feasible clusters, and identify optimal platform solutions. Lastly, users can filter the data to optimal solutions and compare them across many attributes, giving the user increased computational power and visualization capability to support the acquisition decision-making process.

This work was done by Timothy W. Garton, Willie H. Brown, Eric R. Mixon, and Joshua Q. Church for the Army Engineer Research and Development Center. ERDC-0009

This Brief includes a Technical Support Package (TSP).
Engineered Resilient Systems

(reference ERDC-0009) is currently available for download from the TSP library.

Don't have an account? Sign up here.

Aerospace & Defense Technology Magazine

This article first appeared in the October, 2019 issue of Aerospace & Defense Technology Magazine.

Read more articles from the archives here.