A high-dimensional and high-precision onboard modeling method for adaptive variable cycle engine

Abstract Adaptive variable cycle engines, capable of switching seamlessly between subsonic and supersonic modes, have become a central focus in propulsion research. Accurate and real-time onboard models are crucial for predicting performance in flight and supporting advanced functions such as control optimization, health monitoring, and fault-tolerant control. Yet, reconciling accuracy with computational efficiency remains challenging due to the engines’ inherent high dimensionality and strong nonlinearity. Here we present an NN-ΔNN-based modeling framework, where a neural network captures nonlinear dynamics, a ΔNN component represents high-dimensional features, and a Kalman filter enhances adaptability. Simulations show that this approach improves accuracy of key parameters by 0.17–1.6 times compared with NN-PSM across a wide flight envelope. It also achieves rapid thrust-tracking under single- and multi-component degradation within seconds, with low steady-state error, demonstrating strong adaptability and real-time capability.