Spacecraft design is inherently difficult due to the complexity of the systems involved and the expense of testing hardware in a realistic environment. The number and cost of flight tests can be reduced by performing extensive simulation and analysis studies to understand vehicle operating limits and identify circumstances that lead to mission failure. A Monte Carlo simulation approach that varies a wide range of parameters is typically used to generate a large set of test scenarios. The results of these analyses bound the vehicle performance and eventually help certify a spacecraft for flight.
Identifying variables that drive the design is crucial to ensure safety and reliability of a spacecraft. The Monte Carlo simulation process is perhaps the most important, and also the most time-consuming, part of the design and analysis cycle of any space vehicle. Engineers seek to pinpoint influential variables that directly affect a particular system requirement in order to address the necessary changes in the design. The main objective of TRAM is to accelerate the data analysis process while providing engineers with more confidence in their analysis results than when the analysis process is done manually.
Monte Carlo data analysis for problems with a relatively small number of design variables has been addressed in a number of ways, but the analysis of data for fully integrated spacecraft has mostly been performed manually on an individual basis by a large number of people working simultaneously.
TRAM combines different pattern-recognition algorithms into an interactive analysis tool that allows a user to explore large data sets in a very efficient manner. TRAM automatically searches data sets for specific patterns and highlights critical design variables so engineers can focus their analysis efforts. This tool does not replace the analysts, but it can quickly point them to the design variables responsible for specific system failures. The tool streamlines the process of verifying performance requirements, making decisions on which design parameters must be updated, and reporting problems to other team members. Current results show that this tool can quickly and automatically identify individual design parameters, and most importantly, combinations of up to four design parameters that play a significant role in any specified system failures. TRAM was originally developed to analyze sets of flight dynamics Monte Carlo data, but the algorithms are applicable to any Monte Carlo data set. The inputs and outputs of TRAM have a very generic format, so the process can be applicable to any other engineering design problem with a large number of design parameters.
TRAM has two main pieces of code: A MATLAB graphical user interface (GUI) that contains some of the analysis algorithms, and a parallel code that runs on a graphical processing unit (GPU) located on a JSC server that contains the rest of the analysis algorithms. The MATLAB user interface takes the Monte Carlo data in the form of three MATLAB files. The GUI allows the analyst to select analysis variables and system performance metrics for a given analysis task. The MATLAB GUI sends the data to the GPU, where it runs through the analysis algorithms. The GPU sends the data back to the MATLAB GUI, where the user has the chance to display it graphically and further explore the results.
TRAM requires only three inputs in a very simple format so that any Monte Carlo data set can be quickly prepared for analysis. TRAM never manipulates the Monte Carlo data, it does not make any assumptions, it does not normalize the data, and it keeps original physical units throughout the analysis process. TRAM treats all design parameters as equals, and it does not require the analyst to categorize or group different types of variables. This allows a user to analyze a system as a whole rather than analyzing each subsystem separately.
The algorithms are based on two well-known pattern recognition algorithms: kernel density estimation and k-nearest neighbors. However, the results of these two algorithms are combined in a novel manner in order to rank the design variables and variable subspaces in order of importance. The cost function that represents the influence of a parameter on a specified failure was developed specifically for TRAM.