Research On Intelligent Aero-engine Performance Seeking Control Technology

This paper focuses on the performance optimization control of intelligent aero-engine.The key technologies are studied in detail,including integrated propulsion system model,on-board adaptive propulsion system model,steady state performance seeking control,acceleration optimization control and performance recovery control.The main contributions and innovations of the paper are as follows:First,an integrated propulsion system model including inlet and turbofan engine is established and improved.The model can accurately reflect the internal and external flow characteristics of inlet and nozzle,the variable guide vane characteristics of fan and compressor,and the tip clearance variable characteristics of turbine.In the calculation of turbine tip clearance,because the semi-infinite plane method commonly used in heat conduction equation is not suitable for use conditions,a method of extracting the external surface differential equation from the heat conduction equation is proposed to predict the outer surface temperature of the shroud and blade,which improves the calculation accuracy.To verify the feasibility of inlet bleeding regulation,two-dimensional flow field simulation of inlet bleeding process is carried out.The accuracy of the integrated propulsion system model is verified by simulation experiments.Secondly,research on on-board adaptive propulsion system model is carried out.For on-board propulsion system steady model part,a new similarity criterion is proposed,which compresses the base-point sample data more effectively and improves the output parameter precision of the model after similarity conversion.For the simplified engine model part of on-board propulsion system model,a Taylor remainder modeling modifying method is proposed.The second order remainder is added to the linear model expansion,and the propulsion system matrix containing second order remainder is obtained.The unconstrained optimization algorithm is used to optimize the propulsion system matrix and further improve its accuracy.For on-board propulsion system dynamic model part,an equilibrium manifold method with modified dynamic coefficients is proposed.By constructing coefficient polynomials in advance and optimizing them at each sample point,the coefficient polynomials of global optimization are obtained,which solves the problem of low accuracy of linear interpolation of conventional dynamic coefficients.In the parameter estimation of on-board adaptive model,a federated Kalman distributed filtering method is proposed,which has higher computational efficiency than the traditional centralized Kalman filtering method.Again,research on steady state performance seeking control is carried out.Three optimization models for maximum thrust,minimum fuel consumption and minimum temperature before turbine are established.An artificial bee colony optimization algorithm is proposed,which has the dual subpopulation evolution strategy of differential evolution algorithm,a random number satisfying Cauchy distribution and a more flexible neighborhood definition,so that the algorithm can easily jump out of the local optimal solution and finally find the optimal solution.It is applied to engine steady state performance seeking control.Simulation results show that the proposed method can achieve better performance than the conventional method while the engine satisfies the constraints.Then,research on acceleration optimization control is carried out.High pressure rotor speed and temperature before turbine are taken as the objective function,and the constrained optimization problem is converted to unconstrained acceleration optimization problem by penalty function.An unconstrained response surface algorithm is proposed with low complexity and fast convergence.The algorithm has the advantages of high real-time and global search ability.For real-time model of acceleration optimization,a tensor-product-simplex B spline modeling method is proposed,which has the advantages of high modeling accuracy,moreover the algorithm complexity is independent of sample size and only related to B spline coefficients.To solve the problem that the dimension of tensor-product-simplex B spline model is difficult to increase further,a neural network based on minimum batch gradient descent method is proposed.This method can model large sample data for large engine envelope,variable state and multi-variables.The effectiveness of the above method is verified by digital simulation.Finally,a design method of full envelope thrust estimator is proposed.Firstly,the input parameters are optimized based on leaving one method and sparse encoder,and then the samples are compressed reasonably based on Taylor principle.Finally,the thrust estimator is built based on the sparse automatic encoder and integrated learning is carried out.This method can improve the accuracy and generalization performance of thrust estimation.The inner loop traditional controller and outer loop thrust controller based on linear active disturbance rejection control are designed.The dual loop structure makes full use of the advantages of traditional control.The outer-loop control belongs to the slow time constant link,which requires less real time performance of the thrust estimator and is helpful to engineering implementation.Then a performance degradation recovery control law is designed,which can estimate the thrust performance degradation caused by the performance degradation of components and can recover thrust quickly and effectively.Simulation verifies the feasibility of the design method.