• What Role Does Analysis Play in Aircraft Design?

    Analysis can be loosely defied as the steps required to measure the performance of an individual design. The logical conclusion to draw from this definition leads us to a simple idea: build the design and test its performance. While in certain cases (generally involving cutting edge innovation) this is the only available approach, the cost of building and testing every design is prohibitive. We must therefore introduce analysis models that provide an approximation of the true aircraft performance.

    Analysis models differ from true performance numbers due to the series of simplifying assumptions that decrease the cost or time to obtain a result. For example, the most significant assumption for a wind tunnel test is that the non-dimensional scaling relationships hold between the wind tunnel and the free flight of the production aircraft, while aerodynamic methods such as panel methods make dozens of high impact assumptions, such as incompressible and inviscid flow. Each of these assumptions introduces some new deviation between the true flight performance and the performance predicted by the model. These deviations are called uncertainty in the design community. As we introduce more assumptions into the model, we generally say that the model fidelity decreases. IE, wind tunnel tests and CFD are considered "high" fidelity, while panel methods, 2D strip theory, and thin airfoil theory would be considered "low" fidelity.

    There is an assumed general trend between the fidelity of a model and the cost (both monetary and temporal) of running such a model:
    Quantitative curves for each discipline have not been well mapped out, but it is these curves qualitatively assume that low fidelity methods are extremely fast, but have high error. Then, some non-linear trend gives diminishing returns as you approach the full flight test (at which point the error is nominally zero, but measurement error still exists).

    At the early phases of the design process, where hundreds or thousands of design variables are being traded across a wide variety of configurations, only low fidelity models can be run in tractable time. But as we see above, these low fidelity models introduce significant modeling error into our predicted performance. The logical course is to therefore select some subset of design variables to lock in using these low fidelity models, and then trade in a smaller design space with the next higher fidelity method. In this way, the traditional 3 phase design process can be seen as an exercise in continuously minimizing uncertainty until a single design is selected to be built and tested.

    Selecting the correct analysis method for a given aircraft design program can make or break the design. Evolving a design to too high of fidelity (often called "analysis paralysis") can lead to significant delays, not to mention computing resources and engineering salaries. On the other hand, not evolving a design to sufficiently high fidelity can cause significant deviations between predicted and realized performance. Understanding the available analysis tools is therefore crucial to any aircraft designer.

  • What is the Current State of Analysis Methods for Aircraft Design?

    The research community surrounding analysis methods for aircraft design is one of the most diverse and decentralized bodies of literature in the field. I am of the belief that new research in analysis methods generally falls into one of the following categories:
    • Speeding up existing models, particularly high fidelity models (ie, shifting points on the error vs. time curve to the left)
    • Creating new models with decreased uncertainty when compared to an existing model (ie, shifting points on the error vs. time curve down)
    • Synthesis of multi-fidelity data into shared models
    • Propagating uncertainty from an individual model to a final performance metric
    • Integrating multiple disciplines into a common analysis framework (Multi-Disciplinary Analysis, MDA)
    • Surrogate modeling methods for a generalized analysis method
    Each of these fields improves the design process, through either direct speed ups, or by decreasing the number of analysis steps that must be taken for an individual design.

  • What Research do I Contribute to the Analysis Community?

    I contribute to the analysis community in a two key ways. First, I am a developer on the EnCAPS software, which enables me to consider multi-fidelity analysis modeling from a common geometry definition. Second, my work seeks to develop analysis methods and surrogate models which are conducive to optimization, particularly log-convex optimization. Work in these areas is ongoing and will be subject to future publication.