Computing Sciences Adjunct Faculty Publishes Cutting Edge Research on Machine Learning
Dr. Iskrenova-Ekiert has presented a paper at the American Institute of Aeronautics and Astronautics (AIAA) SciTech 2024 Forum in Orlando, Florida. The title of the paper was “Machine Learning Model for an Aircraft Generator-Rectifier System” and was presented in the session “Machine Learning/Artificial Intelligence Applications to Design I” on January 9, 2024.
The digital transformation of the aerospace industry involves the adoption of advanced digital technologies in the development of digital representations, virtual prototyping, and the evaluation of aircraft power and thermal subsystem performance. Machine Learning techniques are well suited to produce models capable of making predictions based on large volumes of historical data.
In this work, a Temporal Fusion Transformer was trained on experimental hardware data for an aircraft generator-rectifier system. The resulting aircraft source model is well-suited to further trade space and performance analysis. The model was exported from Python to ONNX format and imported into MATLAB. Since MATLAB does not support all ONNX operators, a custom implementation of several ONNX operators was required. The model was then implemented in Simulink. The aim of this work was to expand the AFRL-developed Aircraft Power and Thermal Toolkit (APTT) to include machine learning tools for finding new insights into novel methods of traversing large system design spaces, discovering innovative system architectures, and better understanding what makes those architectures successful.