Multivariable Control For Aeroengines Based On Particle Swarm Optimization Algorithm

An aeroegine is a time-variant aerothermodynamic system, which has the featuresof nonlinearity, Multi-variations, and complicated structure. In recent years, therequirement of the aeroengine’s performance has raisen with the development ofaeronautical industry. In the flight envelopes, the model of an aeroengine’saerothermodynamic affects by the system’s environments and conditions. Theaeroengine might not achieve these requirements of performance and even could notwork a normal schedule, without a reasonable and effective control law. A control law isessential to ensure the safety and stability of an aeroengine in any case.In this article, an X biaxial turbofan aeroengine parts nonlinear system is used asthe controller plant. The aeroengine four-variable small perturbation state variablemodel (SVM) is built and analyzed, using the modified method of the Least SquaresFitting. An aeroengine controller is designed using the neural network inverse method,the PID neural network method, and the robust H∞control method. The first twomethods depend on the the neural network’s initial weights and the robust H∞controlmethod depend on the selection of weights array. The Particle Swarm OptimizationAlgorithm (PSO) is used to dynamic optimization for improvement of the system’sperformance. However, the PSO has many disadvantages, such as less sensitive toenvironmental variations, and easily immerging in partial minimum frequently. Anadaptive mutation-based thought is presented for improvement the search accuracy, theconvergence, and convergence rate of the PSO. The main contents of this article are asfollows:1The studied aeroengine parts nonlinear system is analyzed carefully. Thevariables combinations of the aeroengine system are chosen, according to the adjustableand obserable principles which are the chose rules of the variables of the aeroenginesystem. The aeroengine four-variable SVM is built in each equilibrium point in theflight envelope, using the modified method of the Least Squares Fitting.2An adaptive mutation-based thought is presented for improvement of the PSO.This algorithm has the features of the dynamic adjustment of the weights, update of theadaptive mutation locations and speeds, and limit of the update for the change is too big or too little. The dynamic adjustment of the weights improves the algorithm’sconvergence speed. The update of the adaptive mutation locations and speedsovercomes the disadvantage of easily immerging in partial minimum frequently. And itcan enhance particles’ activity and ensure the reasonable search speed. The results froma standard test function are presented to show the desired performance of the modifiedParticle Swarm Optimization Algorithm (MPSO).3The system identification of the aeroengine is established, using the BP neuralnetwork based on the MPSO. It is compared with the BP neural network based on theGenetic Algorithm (GA). An aeroengine controller is designed using neural networkinverse control method. Lots of experiments verify the high accuracy and desiredperformance.4The aeroengine multivariable decoupling controller is design using PID neuralnetwork method for the uncertainties caused by the SVM. However, the PID neuralnetwork depend on the network initial weights, the MPSO is used to optimize the PIDneural network controller’s weights. And the aeroengine decoupling controller based onthe PID neural network is established. Lots of experiments verify the desiredperformance of the system.5The aeroengine robust H∞controller based on the mixed sensitivity method isdesigned for the uncertainties caused by the SVM. However, the control effectly dependon the choice of the weights array. The MPSO is used to optimize the robustH∞controller’s weights to design the robust H∞controller, while the required aeroenginecontrol performance as the target function. Experiments results are presented to showthe desired performances. The qualities requirements are satisfied.