Improving System Models Using Monte Carlo Techniques on Plant Models

Model-Based Design has become a standard in the aerospace industry. A system model is at the center of the development process, supporting the design of complex algorithms combined with physical hardware. In addition to the advantages that come from modeling control algorithms, modeling plants can lead to more robust designs. Using commercial-off-the-shelf (COTS) software allows both the controller and hardware to be modeled and simulated in a single environment. Modeling plants in the same simulation environment as an embedded controller enables engineers to test a controller with multiple plant parameters, as well as simulating with nominal or ideal values. Modeling variable physical parameters provides a better representation of what will happen in real hardware. Monte Carlo analysis is a standard method of simulating variability that occurs in real physical parameters. In aerospace applications, Monte Carlo techniques can be used to ensure highquality, robust designs. Even using a shared COTS environment, fully testing or optimizing a design can take thousands of simulation iterations and days to complete. Depending upon the complexity of the system and fidelity of the model, each iteration could take hours to run. Simulation time can become a critical bottleneck in the development process. Being able to run multiple, independent scenarios in parallel can lessen this time significantly. In this paper, we will discuss techniques for modeling, optimizing, and testing plant models to build better system models in MATLAB® and Simulink® from The MathWorks. We will also present new techniques for speeding up Monte Carlo techniques by using high-performance computing clusters.