基于神经网络时延预测的航空发动机内模控制器设计.pdf

2023 年 11 月第 44 卷 第 11 期推进技术JOURNAL OF PROPULSION TECHNOLOGYNov.2023Vol.44 No.11210903-1Design of Aero-Engine Internal Model Control System Based on Neural Network Time-Delay PredictionDU Xian1，2，MA Yan-hua2，3，WANG Xin-yue4，XU Ling1（1.School of Control Science and Engineering，Dalian University of Technology，Dalian 116024，China；2.Key Laboratory of Intelligent Control and Optimization for Industrial Equipment，Ministry of Education，Dalian University of Technology，Dalian 116024，China；3.School of Microelectronics，Dalian University of Technology，Dalian 116024，China；4.Beijing Power Machinery Institute，Beijing 100074，China）Abstract：Aiming at the uncertainty of time-delay in aero-engine distributed control system，an internal model multi-variable control system based on neural network time-delay prediction is proposed for the aero-engine control.Firstly，the causes and influencing factors of the network-induced time-delay in distributed control system are investigated.Then the internal model controller with time-delay is designed，which includes the delay prediction module based on neural network，the main internal model controller module，the actuator inner-loop control module，and the switching control module that combined with the open-loop control plan for the engine starting process.The system stability performance of internal model control with time-delay prediction is analyzed theoretically under ideal and disturbance conditions，as well as the permitted time-delay is then explained under the proposed control strategy.Finally，the full digital simulation and hardware-in-the-loop tests are carried out.The results show that the designed neural network delay prediction module owns the ability to predict current time-delay in high-precision，and the steady-state error of the internal model controller is less than 0.5%.In addition，the proposed controller has satisfactory anti-interference performance and meets the real-time requirements，which has the feasibility of engineering application.Key words：Aero-engine control；Neural network delay prediction；Internal model control；Multi-variable control；Hardware-in-the-loop simulationCLC number：V233.7 Document code：A Article number：1001-4055（2023）11-210903-13DOI：10.13675/ki.tjjs.210903基于神经网络时延预测的航空发动机内模控制器设计*杜宪1，2，马艳华2，3，王欣悦4，徐羚1（1.大连理工大学 控制科学与工程学院，辽宁 大连 116024；2.大连理工大学 工业装备智能控制与优化教育部重点实验室，辽宁 大连 116024；3.大连理工大学 微电子学院，辽宁 大连 116024；4.北京动力机械研究所，北京 100074）摘 要：针对航空发动机分布式控制系统中时延不确定问题，提出了一种基于神经网络时延预测的*收稿日期：2021-12-27；修订日期：2022-03-29。基金项目：国家自然科学基金（61903059；61903061；61890924）；航空科学基金（2019ZB063001）；辽宁省兴辽英才计划项目（XLYC1907070）；中央高校基本科研业务费（DUT22QN204）；国家科技重大专项（J2019-I-0019-0018）。作者简介：杜宪，博士，副教授，研究领域为航空发动机控制系统设计及应用。通讯作者：马艳华，博士，副教授，研究领域为航空发动机控制与故障诊断。E-mail：引用格式：杜宪，马艳华，王欣悦，等.基于神经网络时延预测的航空发动机内模控制器设计 J.推进技术，2023，44（11）：210903.（DU Xian，MA Yan-hua，WANG Xin-yue，et al.Design of Aero-Engine Internal Model Control System Based on Neural Network Time-Delay Prediction J.Journal of Propulsion Technology，2023，44（11）：210903.）Design of Aero-Engine Internal Model Control System Based on Neural Network Time-Delay Prediction第 44 卷 第 11 期2023 年210903-2航空发动机内模多变量控制设计方法。首先分析了分布式控制架构下网络时延产生的原因及影响因素。然后设计了内模控制器，包括基于神经网络的时延预测模块、内模主控制器模块、执行机构小闭环控制模块以及与发动机起动过程开环控制计划相结合的切换控制模块。在理想和扰动条件下，从理论上分析了基于预测时延的内模控制系统的稳定性能，并对所提控制策略下允许的最大时延进行了说明。最后进行了全数字仿真和硬件在环仿真试验。结果表明，所设计的神经网络时延预测模块具备高精度预测能力，内模控制器的稳态误差不超过0.5%，具有良好的抗干扰能力、并满足实时性要求，具有一定的工程应用价值。关键词：航空发动机控制；神经网络时延预测；内模控制；多变量控制；硬件在环仿真1 IntroductionWith the intelligent improvement of aero-engine control requirements，the traditional centralized control architecture has been difficult to meet the complex control tasks.In order to further improve engine performance，aero-engine control system is gradually transitioning from centralized control architecture to distributed control architecture，due to its advantages of light weight，high reliability and low maintenance cost1-2.Although the distributed control system（DCS）has been considered as an important direction of future intelligent aviation power，the time-delay caused by distributed system in abnormal condition will affect the safety of the aero-engine3.This is because the existence of time-delay directly affects the signal transmission.If the control signals from the controller and the signals collected by the sensors cannot be sent to the corresponding node in time，the control performance of the system will be worsen.In addition，the time-delay characteristics of distributed control system inevitably lead to the complexity of the control system4.On one hand，the characteristic equations of infinite-dimensional systems are generally transcendental equations that are difficult to solve.On the other hand，based on classical control theory，the situation that the control effect deteriorates and even becomes unstable in the presence of time-delay may occur5.Therefore，the solution of time-delay problem is one of the key issues related to the application of distributed control.At present，robust control methods such as state feedback and output feedback have become powerful tools for time-delay system control.Yedavalli et al.6analyzed robust stability for uncertain sampling time-delay systems，gave a stability criterion under uncertain time-delay，and designed a robust controller for a turbine engine with given time-delay parameters.Belapurkar and Yedavall et al.7-9 designed a distributed full authority digital engine control（Decentralized Distributed FADEC，D2FADEC）system based on a decentralized architecture.They carried out a research about the stability under time-delay and packet loss，and used the concept of packet loss margin（PDM）to describe the robustness of the control system in the case of packet loss.Kratz et al.10designed a robust full-state feedback fan speed control system with the goal of maximizing the stability margin during instantaneous packet loss.Jonathan and Belapurkar et al.11-12applied robust control techniques to the linear models of the T700 turbo-shaft engine and the C-MAPSS40K engine，in order to solve the network time-delay and packet loss problems in distributed control systems.In 2017，Seitz et al.13 proposed two new structural robust stability bounds，to improve multi-delay modeling techniques，and a control logic based on dynamic compensators and state estimators，to enhance the robust stability bounds.However，the research on aero-engine time-delay system mostly stays in the traditional uni-variate control，and most of them only consider fixed time-delay，lacking the investigations on time-varying time-delay and ultra-long time-delay14-15.Internal model control（IMC）was first proposed by Garcia and Morari16 in 1982，and improved on the basis of Smith predictor，which greatly enhanced the robustness of the control system.At the same time，internal model control is widely used in the field of time-delay system control because of its simple structure，small on-line adjustment parameters，strong anti-interference and good control effect for large time-delay systems17-18.Therefore，the aero-engine internal model 推进技术2023 年第 44 卷 第 11 期210903-3control strategy is preferred in view of the existence of time-delay.On this basis，considering the uncertainty of the actual system time-delay，neural network19-20 time-delay prediction method is then introduced，and thus the internal model control system based on time-delay prediction is proposed for the aero-engine control.2 Aero-engine control scheme with time-delay2.1 Time-delay analysis under distributed architectureIn a distributed control system，multiple intelligent nodes share a communication line，and each intelligent node needs to abide by the bus scheduling mechanism，and can occupy the bus for data transmission when the bus is idle and allowed to access.When the bus bandwidth is fixed，the waiting time for each node to compete for the bus will be prolonged with the increase of the number of nodes，which leads to non-negligible network time-delay，also known as network-induced time-delay.At the same time，due to the variety of data transmission paths in DCS，data packet loss and confusion occur from time to time，which will also cause network time-delay and further increase the uncertainty of the system.The existence of network-induced time-delay makes the future states of the control system not only related to the current states，but also related to the past states.In the field of network control，such systems with network delay are called time-delay systems21.The time-delay composition of typical distributed control system is shown in Fig.1.It can be seen from Fig.1 that the time-delay of distributed control system is mainly composed of transmission delay and execution delay.In the process of network transmission，the delay caused by queuing and competing to use bus is called the transmission delay，and the delay caused by each intelligent node in data processing and operation tasks is called the execution delay，which is also referred to as software delay.Total time-delay of distributed control system is described as=ca+sc+a+s+c（1）Where，is the total delay of distributed control system.ca is the transmission delay of data from controller node to actuator node.sc is the transmission delay of data from sensor node to controller node.a，c and s are the execution delays of actuator node，controller node，and sensor node respectively.The size of the execution delay is determined by the hardware and software of the system.Yu et al.22made an in-depth analysis of the software delay of the distributed control system，and concluded that with the condition of high-speed CPU and high-efficiency software coding，the system execution delay is about 100s，which almost exert no influence on the performance of the control system.Therefore，the execution delay can be ignored，and the network time-delay in DCS can be simplified as follows：=ca+sc（2）2.2 Time-delay influence on aero-engine control systemAssuming that the transfer functions of the controller and the aero-engine（controlled object）in Fig.2 are expressed as Gc（s）and Gp（s），respectively.Considering the time-delay ca and sc in formula（2），the closed-loop characteristic equation of the system can be expressed asFig.1Time-delay composition of typical distributed control systemsDesign of Aero-Engine Internal Model Control System Based on Neural Network Time-Delay Prediction第 44 卷 第 11 期2023 年210903-41+Gc(s)e-casGp(s)e-scs=0（3）It can be seen from formula（3）that the characteristic equation contains the time-delay exponential terms of e-cas and e-scs，which will have a great impact on the stability and control quality of the system.In order to illustrate the effects of time-delay on the stability of the control system more intuitively，simulation examples under different cases are given below.Here the pre-designed H multi-variable control algorithm with adjusted control parameters is utilized，referred to as the traditional method.In the first case，it is assumed that sc is a fixed value 500ms，and ca is changed to observe the influence of ca.The system output curves are shown in Fig.2.It is worth noting that the high-pressure rotor speed is the relative conversion speed and it is a dimensionless quantity，which is also applied for other figures.It can be seen that，when sc is constant，the overshoot of the system increases with the increases of ca，and the time required to reach stability grows accordingly.It is obvious that oscillations happen during transient state，which is not expected for the aero-engine.In the second case，total time-delay is considered as a fixed value 1000ms，and the simulation results with different time-delay combinations are shown in Fig.3.It is presented that，when the total time-delay is certain，the larger the value ca is，the slower the system s dynamic response will be.Therefore，the impact of forward channel time-delay ca on the system performance should be fully considered in the design of the aero-enFig.2Simulation examples with fixed time-delay scFig.3Simulation examples with fixed total time-delay推进技术2023 年第 44 卷 第 11 期210903-5gine control system.From the above two cases，the system dynamic performance with time-delay is not satisfying under the traditional control strategy，which has large overshoot，oscillations and long settling time.So the new control strategy with time-delay should be investigated to enhance the control performance of the aero-engine.2.3 Aero-engine internal model control structure with time-delayIn this paper，only the time-delay ca is considered，and the structure of internal model control system with time-delay for aero-engine is shown in Fig.4，which mainly includes a starting open-loop controller，an actuator PID controller，a switching controller and the internal model main controller.In the proposed control scheme，the control variables are main fuel flow Wfm and nozzle area A8，and the controlled variables are high-pressure rotor speed nHand turbine pressure drop ratio T.The throttle power lever angle（PLA）is taken as the control command，and the starting threshold PLAst is set as the basis for switching between the starting state and the non-starting state.When PLA is less than PLAst，the engine is in starting state，and the target value of the control quantity is directly given by the open-loop controller，and when PLA is larger than PLAst，the engine is at idle state or above.At this time，the system obtains the corresponding target value of high-pressure rotor speed and turbine pressure drop ratio through the interpolation table of control schedule，and then the closed-loop master controller（IMC controller）computes the control values Wfm and A8.As for the actuator part，the actuator controller receives the target values of the control quantity，calculates the PWM（Pulse-Width Modulation）signal to drive the actuator structure through PID control logic，and receives the feedback value of the actuator sensor to form an inner-loop control structure.The control period of the inner-loop control is 5ms.The aero-engine is driven by the actuator.The data of high-pressure rotor speed and turbine pressure drop ratio in the current state are collected by the speed and pressure sensors，and fed back to the main controller to form an outer-loop control loop.The control period of the outer-loop control is 10ms.At the same time，considering the high requirements of IMC controller on the accuracy of the model，the neural network time series prediction algorithm is proposed to predict the time-delay，and the predicted time-delay is sent to the IMC main controller，further optimizing the control effect.3 Design of internal model controller based on time-delay prediction3.1 Time-delay prediction based on neural networkBP neural network，also known as back propagation network，is a multi-layer forward network for weight training of nonlinear differentiable functions，including input layer，hidden layer and output layer.The structure diagram of typical BP neural network is shown in Fig.5.The nodes in BP network represent the data processing function of the current position，which is calle