Adaptive Correction Of Aeroengine Model Based On Neural Network

Due to the difference in engine manufacturing and assembly levels,as well as performance degradation caused by wear,corrosion,fouling,etc.between various components during use,the aero-engine component level is established based on component characteristics and design point parameters of compressors,turbines,etc.It is often difficult to match the performance of an actual machine.Based on the "Vortex Fan Engine Control Law Research and Simulation Test" project of China Aerospace Turbine Research Institute,this paper proposes an adaptive correction method for aero-engine full-line model for a turbofan engine.First,a neural network-based dynamic parallel compensator is built to compensate the individual engine and the general mathematical model.Then,a multi-index decision algorithm with integrated evaluation is used to realize the selection of the health factor to be corrected.Finally,a modifier based on intelligent optimization algorithm is designed to obtain more accurate health factor correction values.The main research contents and innovations of this paper are as follows:(1)In terms of compensator,a dynamic parallel compensator is established for the typical time series of multi-input and multi-out of aeroengine,using fully connected neural network,recurrent neural network and Long Short-Term Memory neural network.The simulation results show that Long Short-Term Memory neurons cancompensate for the difference between the individual engine and the general mathematical model better.(2)In the selection of health factors,a decision-making method based on the entropy weight coefficient and TOPSIS integrated evaluation interval number is adopted.The attribute weights to be corrected are solved by solving the attribute weights for the median and length of the measurable parameter change intervals generated by the changes of the various health factors.(3)For the corrector,the particle swarm optimization and the firefly optimization algorithm are used to optimize the selected health factor.Aiming at the shortcomings of the two algorithms,such as premature convergence and local convergence,the improved inertia weight,adaptive dynamic learning factor,chaotic map and adaptive dynamic step factor are improved.Furthermore,in order to improve the adaptability of the single group optimization algorithm,a PSO-FA hybrid optimization algorithm is adopted.The simulation proves that the method has good effect in convergence speed and precision.(4)By combining the compensator with the corrector,an accurate estimation of the health factor in the aeroengine performance degradation state is achieved,and the method is adaptive and robust.On the one hand,the results contribute to the design of aero-engine control system,simplifying the health parameter correction process;on the other hand,it can improve the verification accuracy and reliability of the aeroengine digital simulation platform and the hardware-in-the-loop simulation platform,thus providing useful benefits for the verification of the project.