Application of machine learning and its effectiveness in performance model adaptation for a turbofan engine

Engine performance models are used in various fields, such as operation optimization, control, and diagnosis. In order to effectively apply an engine performance model in these fields, a high level of prediction accuracy is required. When improving the accuracy of an engine model through performance adaptation using measurement data, it is important to functionalize the component adaptation factors calculated at each measurement point to enable them to be utilized at other operating points. A method of curve fitting to find the adaptation factor for a single performance parameter was previously used to functionalize the component adaptation factor. In this paper, machine learning techniques are newly applied to functionalize component adaptation factors for several parameters, including engine operation characteristics, and their effectiveness is analyzed. The prediction accuracy of the adapted model for a two-spool mixed-flow turbofan engine is compared based on the fan exit total pressure, high-pressure compressor (HPC) exit static pressure, exhaust gas temperature (EGT), and high-pressure spool speed. For an engine model adapted using the existing functionalization method, the average absolute value of the relative errors in the fan exit total pressure, HPC exit static pressure, EGT, and high-pressure spool speed in the validation dataset were 2.91 %, 2.20 %, 5.32 %, and 1.84 %, respectively, whereas for an engine model adapted using the proposed machine learning-based functionalization method, these errors were 0.70 %, 1.00 %, 2.56 %, and 0.26 %, respectively, under the same conditions. As a result, it was confirmed that the performance prediction accuracy of the engine model adapted using the proposed method was improved compared to the existing method.