A Fault Tolerant Control Method for Aero-Electric Fuel Pump Based on Intelligent Instruction Prediction

Abstract

The present invention discloses a fault-tolerant control method for an aviation electric fuel pump based on intelligent instruction prediction, comprising the following steps: Step 1, establish a non-linear dynamic model of the aviation electric fuel pump with multiple faults and uncertainties within the full flow range; Step 2, establish an intelligent speed instruction inverse model of the aviation electric fuel pump based on a combined radial basis function neural network; Step 3, design a robust fault-tolerant controller based on variable-speed grey wolf multi-objective optimization. The method of the present invention can ensure that the aviation electric fuel pump supplies fuel as required quickly, accurately, safely and reliably for the aeroengine within the full flow range in the face of the influence of multiple faults and multiple uncertainties, the speed of the gear pump does not exceed the limit, the motor current does not exceed the limit, and the output quantities such as fuel flow and speed are robust and stable. At the same time, multi-objective optimizations such as high steady-state accuracy, fast dynamic response, low fuel consumption rate, and reduction of carbon dioxide emissions are achieved.

Description

Technical Field

[0001] The present invention belongs to the field of aero-engine control, and particularly relates to a design method for an intelligent robust fault-tolerant control system of an aero-electric fuel pump. Background Art

[0002] Intelligentization is an important frontier direction for the development of aero-electric fuel pumps. By means of the idea of robust fault-tolerant control and combined with intelligent optimization algorithms, designing an intelligent controller to improve the fuel supply quality and enhance the reliability of aero-electric fuel pumps is a very worthy approach to explore. This can not only reduce the hardware upgrade and transformation costs of electric fuel pumps, but also fully tap the fuel supply potential of fuel pumps, make up for the performance degradation during the use of fuel pumps, and minimize the adverse effects of faults on the overall system as much as possible.

[0003] At present, there are few research schemes on the intelligent control of aero-electric fuel pumps at home and abroad. First, in terms of system control schemes, Reference [1] proposed a scheme for an aero-electric fuel pump based on distributed control, and the fault-tolerant performance of the control system can be ensured by dual-redundancy electronic channels, backup sensors, backup actuators, etc. in the digital controller. Reference [2] developed a root locus parameter design method suitable for the PI controller of the electric fuel pump control loop according to the characteristics of the engine, further improving the robust control performance of the aero-engine fuel control loop. Reference [3] designed a cascade PI pressure controller based on the nonlinear mathematical model of the electric fuel pump, and verified through experiments that the control method still has good robustness under extreme conditions. Reference [4] proposed a scheme for the flow control system of an electric fuel pump based on finite set prediction. This control strategy realizes more stable and faster current control, further reduces the motor speed fluctuation, and thus makes the fuel supply of the fuel pump more stable and accurate. Second, for various faults that may occur in the system, Reference [5] took a four-phase fault-tolerant permanent magnet motor as the research object. Each phase of the motor has independent units with magnetic, electrical, thermal, and mechanical isolation. Internal winding faults are detected by changes in relevant parameters, and then the fault phase of the power converter is short-circuited to handle the faults, realizing the fault-tolerant control of the electric fuel pump. Reference [6] carried out research on robust fault-tolerant control to ensure the reliable and stable operation of an aero-electric fuel pump under the conditions of uncertainty and actuator faults. Reference [7] proposed a robust control method based on the combination of a linear integral sliding surface and a quadratic integral sliding surface on the basis of sliding mode control, which still has good stability in the case of mismatched uncertainties in the aero-electric fuel pump system. In addition, the acquisition and processing of signals are also one of the links that need to be considered in the intelligent control process. Reference [8] considered that the fuel demand signal sent by the pilot may be interfered during the process, and by collecting the interfered signal and using the average processing method, the robustness of the fuel system was improved.

[0004] The present invention is directed to an aviation electric fuel pump with a structure of a six-phase permanent magnet synchronous motor directly driving an external gear pump. Among them, the external gear pump is the working machine, and the six-phase permanent magnet synchronous motor is the power machine. The pump and the motor are coaxially connected, so the rotational speed of the pump is the same as that of the motor. The structure of the direct-drive aviation electric fuel pump is shown in Figure 1 , for this type of aviation electric fuel pump, a novel intelligent robust fault-tolerant control system design method is proposed by combining intelligent optimization algorithms to supply fuel to the aeroengine quickly, accurately, safely and reliably.

[0005] [1] Gao Yijun, Huang Jinquan, Tang Shijian. Research on the Scheme of an Electric Fuel Pump for Aeroengine Based on Distributed Control [J]. Gas Turbine Experiment and Research, 2012, 25(S1): 36-40.

[0006] [2] Xu Jian, Yang Gang, Hu Wenfei. Application of Root Locus Method in the Parameter Design of PI Controller for Fuel Control Loop [J]. Aeroengine, 2016, 42(04): 17-20.

[0007] [3] Nils T, Stump P B, Thielecke F. A Robust Pressure Controller for a Variable Speed AC Motor Pump—Application to Aircraft Hydraulic Power Packsges [C]. Proceedings of the BATH / ASME 2018 Symposium on Fluid Power and Motion Control, September 12-141, 2018, Bath, UK, FPMC2018-8868.

[0008] [4] Xu Min. Research on Cascade Predictive Control of Aviation Electric Fuel Pump [D]. Nanjing: Nanjing University of Aeronautics and Astronautics, 2020.

[0009] [5] Mecrow B C, Jack A G, Atkinson D J, et al. Design and Testing of a Four-Phase Fault-Tolerant Permanent-Magnet Machine for an Engine Fuel Pump [J]. IEEE Transactions on Energy Conversion, 2004, 19(4): 671-678.

[0010] [6] Ding Runze. Research on Sliding Mode Fault Tolerant Control of Aero Electric Fuel Pump [D]. Nanjing University of Aeronautics and Astronautics, 2018.

[0011] [7] Ding R, Xiao L, Jin X. Robust Control for Electric Fuel Pump with Variant Nonlinear Loads Based on a New Combined Sliding Mode Surface [J]. International Journal of Control, Automation and Systems, 2019, 17(3): 716 - 728.

[0012] [8] Pooja K K, Preethi R, Sumatha V P, et al. Closed Loop Fuel Control of Aero Engine [J]. International Journal of Advanced Research in Electronics and Communication Engineering, 2015, 4(5): 1404 - 1410. Summary of the Invention

[0013] The object of the present invention is to provide a fault - tolerant control method for aero electric fuel pump based on intelligent command prediction, which ensures that the aero electric fuel pump can supply fuel on demand quickly, accurately, safely and reliably for aero engines within the full flow range when facing multiple faults and multiple uncertainties, the speed of the gear pump does not exceed the limit, the motor current does not exceed the limit, and the output quantities such as fuel flow and speed are robust and stable. At the same time, multi - objective optimizations such as high steady - state accuracy, fast dynamic response, low fuel consumption rate, and reduction of carbon dioxide emissions are achieved.

[0014] To achieve the above object, the present invention adopts the following technical solutions:

[0015] A fault - tolerant control method for aero electric fuel pump based on intelligent command prediction, comprising the following steps:

[0016] Step 1, establish a nonlinear dynamic model of the aero electric fuel pump with multiple faults and uncertainties within the full flow range;

[0017] Step 2, establish an intelligent speed command inverse model of the aero electric fuel pump based on a combined radial basis function neural network;

[0018] Step 3, design a robust fault - tolerant controller based on variable - speed grey wolf multi - objective optimization.

[0019] Step 1 includes:

[0020] Step 11, first conduct an analysis of the fuel flow characteristics of the aviation electric fuel pump:

[0021] When the motor and the pump of the aviation electric fuel pump are coaxially connected, the rotational speed of the pump is equal to that of the motor. For a given external gear pump, the fuel flow rate Q is a function of the rotational speed n of the pump and the volumetric efficiency η v function

[0022] Q = η v Q T (1)

[0023]

[0024] Among them, Q T is the theoretical output flow rate of the gear pump, n is the rotational speed of the pump, b is the gear thickness, R a is the addendum circle pressure angle, R c is the pitch circle radius, t j is the base pitch length;

[0025] The voltage equation of the six-phase permanent magnet synchronous motor consists of two parts: the stator phase resistance voltage drop and the magnetic flux linkage change rate, that is

[0026]

[0027] Among them, u s = [u A u B u C u D u E u F T is the stator voltage vector, R s = diag[R s R s R s R s R s R s is the resistance vector, i s = [i A i B i C i D i E i F T is the stator current vector; ψ s represents the permanent magnet flux linkage;

[0028] The flux linkage equation of the six-phase permanent magnet synchronous motor consists of two parts: the armature winding flux linkage and the permanent magnet flux linkage, that is ​​

[0029] ψ s = L s i s + ψ m (4)

[0030] where L s is the phase inductance matrix and ψ m is the permanent magnet flux linkage;

[0031]

[0032] ψ m = ψ f [cosθ e cos(θ e -α) cos(θ e -2α) cos(θ e -3α) cos(θ e -4α) cos(θ e -5α)] T

[0033] where L 11 is the self-inductance of the first set of windings, L 22 is the self-inductance of the second set of windings, M 12 and M 21 are the mutual inductances between the two sets of windings, ψ f is the amplitude of the permanent magnet flux linkage, θ e is the electrical angle of the rotor, α = 2π / 3; T represents the transpose of the matrix;

[0034] According to the principle of electromechanical energy conversion, the co-energy can be written as

[0035]

[0036] Therefore, the torque equation of the six-phase permanent magnet synchronous motor is

[0037]

[0038] where θ m is the mechanical angular displacement of the rotor, p is the number of pole pairs;

[0039] The motion equation of the six-phase permanent magnet synchronous motor is

[0040]

[0041] where J is the moment of inertia, T e is the motor torque, T L is the load torque, B is the viscous friction coefficient, ω m is the mechanical angular velocity of the motor;

[0042] Step 12: Construct an adaptive combined nonlinear dynamic model of an aviation electric fuel pump in which a steady-state nonlinear model described by a polynomial function form is connected in parallel with a dynamic linear model under typical operating conditions;

[0043] The adaptive combined nonlinear dynamic model of the aviation electric fuel pump consists of three parts: a nonlinear steady-state model, an adaptive dynamic gain, and a linear dynamic model;

[0044] Based on the nonlinear steady-state model, by constructing an adaptive dynamic gain and using it to adjust some parameters of the linear dynamic model, an adaptive combined nonlinear dynamic model of the aviation electric fuel pump for control system analysis and design research that takes into account both steady-state accuracy and dynamic response speed is finally obtained;

[0045] The input quantity X of the model = [u, d T T , where u is the control quantity given by the electronic controller to the electric fuel pump, and u forms the voltage supplied to the motor after passing through the power converter, d is a column vector composed of external parameters such as flight altitude, flight Mach number, ambient temperature, and inlet pressure, X s is the steady-state input value corresponding to X, and W fbs is the steady-state fuel flow value; the output quantity of the model where W fb is the fuel flow, n is the speed of the electric fuel pump, and y3 is a column vector composed of measurable parameters other than fuel flow and speed such as fuel temperature, fuel pressure, and inlet and outlet pressure difference, and y s is y the corresponding steady-state output value;

[0046] Within the full flow range, according to the fuel demand command under different operating conditions of the engine Construct an adaptive strategy characterized by the adaptive dynamic gain K, and independently select the corresponding nonlinear steady-state model and linear dynamic model under large flow or small flow requirements;

[0047] The adaptive dynamic gain K in the combined model consists of two parts, namely The first part K0(X s ) is obtained based on the nonlinear steady-state model of the electric fuel pump using the numerical differentiation algorithm;

[0048]

[0049] where, δX is a very small input increment value; the second part is based on the fuel demand command ​The designed dynamic gain adaptive compensation term is used to make the dynamic gain more conform to the different flow characteristics under large flow demand and small flow demand when adjusting the linear steady-state model of the electric fuel pump, so as to obtain an adaptive combined nonlinear dynamic model of the electric fuel pump applicable in the full flow range that takes into account both steady-state accuracy and dynamic response speed.

[0050] The adaptive combined nonlinear dynamic model of the aviation electric fuel pump constructed in step 12 is as follows:

[0051]

[0052] Among them, f(·) is a one-dimensional dynamic nonlinear function describing the system speed; h1(·) is a one-dimensional function describing the system output n. Since n is also a state variable of the system, in the absence of uncertainties and faults, h1(·) = n; h2(·) is a one-dimensional nonlinear function describing the system output W fb and h3(·) is a ρ vector of one-dimensional nonlinear functions describing the measurable output y3 of the system other than fuel flow and speed, where ρ is a non-zero positive integer; Δf(·), Δh1(·), Δh2(·), Δh3(·) respectively represent the uncertainties corresponding to f(·), h1(·), h2(·), h3(·), and φ0(t), φ1(t), φ2(t) and φ j (t), (j = 3, 4,..., ρ + 2) are nonlinear function descriptions of multiple faults occurring in the electric fuel pump, and γ0(t - T 0f ), γ1(t - T 1f ), γ2(t - T 2f ), γ j (t - T jf ), (j = 3, 4,…, ρ + 2) represent the development types of faults acting on the system, including abrupt faults and gradual faults, and T 0f , T 1f , T 2f , T jf , (j = 3, 4,…, ρ + 2) represent the occurrence times of faults;

[0053]

[0054] Among them, l j > 0 (j = 0, 1,…, ρ + 2) represents the development type of the fault.

[0055] Step 2 includes:

[0056] Step 21, constructing an intelligent speed command inverse model based on a combined radial basis function neural network;

[0057] Based on the steady-state historical data of the fuel flow of the aviation electric fuel pump, an intelligent rotational speed command inverse model is established based on the combined radial basis function neural network (CRBFNN). Using this inverse model, the actual temperature, pressure, pressure difference, and fuel demand command of the current electric fuel pump can be obtained. The corresponding required rotational speed n a ;

[0058] Step 22, rotational speed command compensator

[0059] According to the fuel demand command of the aeroengine Feed back the output fuel flow W of the current aviation electric fuel pump fb , using and the fuel flow error ΔW fb , aiming at the influence of the motor inertia, and considering the respective characteristics of the fuel flow under different working conditions such as starting, idle, intermediate, maximum, and acceleration / deceleration, a fuel pump rotational speed command regulator is designed to generate a rotational speed command compensation amount Δn ;

[0060] Through the rotational speed command compensation amount Δn Adjust the required rotational speed n a Finally, obtain the rotational speed command n of the aviation electric fuel pump at the current moment within the full flow range 0 ;

[0061] Step 23, multi-step predictive reference trajectory model of rotational speed command

[0062] Through the historical information memory, based on the rotational speed command n at the current moment 0 , design a reference trajectory model as shown in Equation (10), combined with the dynamic characteristics of the electric fuel pump, select the rise time of the system as the prediction time domain P, and form a multi-step predictive rotational speed command reference trajectory vector

[0063]

[0064] where, n 0 (k) represents the value of n 0 at the current moment k, n 0 (k±i) represents the value of n 0 at the moment k±i, (i = 1, 2,..., P), α j , is the weighting coefficient.

[0065] In the said Step 2

[0066] Based on the intelligent rotational speed command estimation of the combined radial basis function neural network (CRBFNN), during the forward training process, the steady-state fuel quantity and the steady-state measured value y3s As the input quantity, the steady-state speed n of the motor s As the output quantity, where the subscript s represents the steady-state value; during the reverse test process, the actual fuel demand command and the measured value y3 are used as the input quantities, and the required speed n a is used as the output quantity;

[0067] The overall input-output mapping relationship of the combined radial basis function neural network (CRBFNN) is shown in Equation (14)

[0068]

[0069] where is the input vector; n a is the output quantity; b j is the j-th central unit of the basis function hidden layer, which has the same dimension as x; h is the number of radial basis function units; v j is the weight coefficient between the hidden layer and the output layer; φ j is the width of the j-th basis function; ||·|| is the Euclidean norm of the vector; and are the weight coefficients respectively.

[0070] In step 3 described above

[0071] In the multi-objective optimization problem, minimizing the steady-state error, minimizing the dynamic response time, minimizing the fuel consumption rate, and minimizing the carbon dioxide emission are involved. The constraint conditions are shown in Equation (11), including that the speed does not exceed the limit, the motor current does not exceed the maximum value, the fuel pressure difference does not exceed the limit, and the fuel temperature does not exceed the limit

[0072]

[0073] where J j (j = 1, 2, 3, 4,...) represents the performance index, Ω j (j = 1, 2, 3, 4,...) represents the sliding mode surface prediction terminal domain that satisfies the input constraints, t s represents the time to reach the steady state, η SFC represents the fuel consumption rate, Q CO2 represents the carbon dioxide emission, n min and n max represent the lower limit and upper limit of the speed respectively, i A,B,C,D,E,F represents the phase current of the six-phase permanent magnet synchronous motor, i min and i max represent the lower limit and upper limit of the current respectively, T pump represents the fuel temperature, and respectively represent the lower and upper limits of the fuel temperature, and Δp represents the fuel pressure difference, Δp min and Δp max respectively represent the lower and upper limits of the fuel pressure difference.

[0074] Step 3 includes:

[0075] Step 31, adopting the corresponding discrete-time linearization model as the prediction model;

[0076] Step 32, on the basis of designing a non-singular terminal sliding mode surface, according to the state estimation error, study the sliding mode switching term including the adaptive law, so as to construct an adaptive non-singular terminal sliding mode observer for diagnosing multiple faults and signal reconstruction;

[0077] Step 33, rolling optimization solution based on the variable-speed grey wolf optimization algorithm

[0078] Generate the initial population using the chaos mapping, and introduce the velocity component based on the particle swarm optimization algorithm (PSO) into the GWO to form the variable-speed grey wolf optimization algorithm for online solution of the robust fault-tolerant controller.

[0079] Step 32 includes:

[0080] Step 321, design of the adaptive non-singular terminal sliding mode observer

[0081] For the measurement signal of the j-th sensor that may contain faults, the form of the designed sliding mode observer is

[0082]

[0083] where represents the first derivative of the rotational speed estimated value, represents the rotational speed estimated value, u represents the input of the system, f(·) is a non-linear function describing the pre-input of the rotational speed estimated value, k j represents the adaptive parameter; s j is the non-singular terminal sliding mode surface parameter, υ j (·) is the adaptive switching term with respect to the time-varying adaptive parameter k j (t) and the sliding mode surface parameter s j ; is the estimated value for the measurement signal of the j-th sensor, F j is a parameter that can be designed;

[0084] The non-singular terminal sliding mode surface function is designed as

[0085]

[0086] where is the difference between the j-th measured value and the estimated value of the j-th sliding mode observer output, λ j and σ j are positive odd numbers and satisfy 1 < λ j / σ j < 2; β j is a constant;

[0087] The time-varying adaptive parameter k j (t) has the structure shown in Equation (17)

[0088]

[0089] where, a j is the adaptive law gain coefficient;

[0090] Step 322, multi-fault diagnosis strategy

[0091] When there are τ sensor output parameters, τ sliding mode observers are designed correspondingly, that is, each observer observes one sensor signal. Using the measured output value of the.j.-th sensor and the estimated value of the j-th observer, a structured residual ξ j is generated. In the case of no sensor faults, the state estimation values of each observer will converge to the true state of the system. If the j-th sensor fails and the remaining sensors are normal, the state estimation value obtained by the j-th observer will be affected by the output signal of the sensor containing fault information. Therefore, the estimation result will deviate from the actual situation. This method is applied for the diagnosis of multi-fault signals;

[0092] Step 323, signal reconstruction

[0093] When a fault is detected, comprehensively using the fault-free measurement signals and the output signals of the aviation electric fuel pump model, a non-singular terminal sliding mode surface and an adaptive law are designed. Based on the sliding mode observer, the fault signal reconstruction is realized and the fault signal is replaced for the design of the robust fault-tolerant controller.

[0094] Beneficial effects: The present invention proposes a design method for an intelligent robust fault-tolerant control system of an aviation electric fuel pump. Compared with the existing technologies, the advantages of the present invention are as follows:

[0095] (1) Based on the new method for establishing the non-linear dynamic model of the aviation electric fuel pump within the full flow range given by the present invention, a new method for robust fault-tolerant control of the aviation electric fuel pump based on intelligent rotational speed command multi-step prediction is proposed.

[0096] (2) During the multi-step prediction process of the speed command, a combined radial basis function neural network is constructed by combining the Gaussian basis function with good non-linear fitting ability and the cubic basis function with good linear fitting ability, and is used to establish an intelligent inverse model of the speed command to obtain the main part of the speed command. In addition, according to the respective characteristics of fuel flow under different working conditions such as starting, idle, cruise, maximum, acceleration and deceleration, and considering the influence of the motor inertia, a suitable fuel pump speed command regulator is designed to generate a speed command compensation amount to dynamically compensate the main part of the speed command. Further, by using the current and historical information of the speed command, a reference trajectory model is constructed, and a prediction time domain is selected to form a multi-step predicted speed command reference trajectory vector.

[0097] (3) During the design process of the fault-tolerant controller, a novel design method of a robust fault-tolerant controller based on variable-speed grey wolf multi-objective optimization is proposed to ensure that the aviation electric fuel pump supplies fuel to the aero-engine on demand quickly, accurately, safely and reliably within the full flow range. Description of the Drawings

[0098] Figure 1 is the structural diagram of an aviation electric fuel pump with a six-phase permanent magnet synchronous motor directly driving an external meshing gear pump;

[0099] Figure 2 is the structural diagram of the design of a robust fault-tolerant control system for an aviation electric fuel pump;

[0100] Figure 3 is the equivalent physical model of a six-phase permanent magnet synchronous motor;

[0101] Figure 4 is the structural diagram of an adaptive combined non-linear dynamic model of an aviation electric fuel pump;

[0102] Figure 5 is the block diagram of the intelligent speed command multi-step prediction technology;

[0103] Figure 6 is the corresponding curve of various types of basis functions;

[0104] Figure 7 is the intelligent speed command estimation based on CRBFNN;

[0105] Figure 8 is the schematic diagram of the design of a robust fault-tolerant controller based on variable-speed grey wolf multi-objective optimization;

[0106] Figure 9 is the schematic diagram of the multi-fault diagnosis strategy;

[0107] Figure 10 is the flow chart of the variable-speed grey wolf optimization algorithm;

[0108] Figure 11It is the iteration curve of the cost function of the fast gray wolf optimization algorithm;

[0109] Figure 12 It is the flow response curve of the dynamic fuel pump;

[0110] Figure 13 It is the response curves of the d-axis current and q-axis current of the permanent magnet synchronous motor. Detailed implementation manners

[0111] The present invention will be further explained below with reference to the accompanying drawings.

[0112] The design structure diagram of the robust fault-tolerant control system for an aviation electric fuel pump is as Figure 2 shown. It can be seen from the figure that the fuel demand command of the aeroengine and the actual output flow rate W of the electric fuel pump fb as well as signals such as temperature, pressure, pressure difference, and rotational speed, after multi-step prediction of the intelligent rotational speed command, form a reference trajectory vector The robust fault-tolerant controller is based on and the model prediction output vector after feedback correction Based on variable-speed gray wolf multi-objective optimization solution, the control quantity u of the electric fuel pump is obtained; under the action of the control quantity u and the rotational speed n, the nonlinear discrete-time hybrid prediction model gives the model prediction output vector Y m ; through the rotational speed n, feedback correction is performed on Y m to form the corrected model prediction output vector

[0113] A design method for an intelligent robust fault-tolerant control system of an aviation electric fuel pump according to the present invention includes the following steps:

[0114] Step 1, Establishment of the nonlinear dynamic model of the aviation electric fuel pump within a certain range

[0115] One of the important bases for the analysis and synthesis research of the control system of the aviation electric fuel pump is to establish a suitable dynamic model of the control system. For the integrated aviation electric fuel pump studied in the present invention, the dynamic modeling steps are as follows:

[0116] Step 11, Analysis of the fuel flow characteristics of the aviation electric fuel pump

[0117] When the motor and the pump of the aviation electric fuel pump are coaxially connected, the rotational speed of the pump can be equivalent to the rotational speed of the motor. For a given external gear pump, the fuel flow rate Q is a function of the rotational speed n of the pump and the volumetric efficiency η v of the pump.

[0118] Q = η v Q T (1)

[0119]

[0120] Among them, Q T is the theoretical output flow rate of the gear pump, n is the rotational speed of the pump, b is the gear thickness, R a is the addendum circle pressure angle, R c is the pitch circle radius, t j is the base pitch length.

[0121] Figure 3 An equivalent physical model of a six-phase permanent magnet synchronous motor is given. The mathematical model of the six-phase permanent magnet synchronous motor in the natural coordinate system mainly includes the voltage equation, the flux linkage equation, the torque equation, and the motion equation.

[0122] The voltage equation of the six-phase permanent magnet synchronous motor consists of two parts: the stator phase resistance voltage drop and the rate of change of the flux linkage, that is

[0123]

[0124] Among them, u s =[u A u B u C u D u E u F T is the stator voltage vector, R s =diag[R s R s R s R s R s R s is the resistance vector, i s =[i A i B i C i D i E i F T is the stator current vector.

[0125] The flux linkage equation of the six-phase permanent magnet synchronous motor consists of two parts: the armature winding flux linkage and the permanent magnet flux linkage, that is

[0126] ψ s =L s i s +ψ m (4)

[0127] Among them, L s is the phase inductance matrix, ψ m is the permanent magnet flux linkage. Specifically,

[0128] ​​

[0129] ψ m = ψ f [cosθ e cos(θ e -α) cos(θ e -2α) cos(θ e -3α) cos(θ e -4α) cos(θ e -5α)] T

[0130] where L 11 is the self-inductance of the first set of windings, L 22 is the self-inductance of the second set of windings, M 12 and M 21 are the mutual inductances between the two sets of windings, ψ f is the amplitude of the permanent magnet flux linkage, θ e is the electrical angle of the rotor, and α = 2π / 3.

[0131] According to the principle of electromechanical energy conversion, the co-energy can be written as

[0132]

[0133] Therefore, the torque equation of the six-phase permanent magnet synchronous motor is

[0134]

[0135] where θ m is the mechanical angular displacement of the rotor, p is the number of pole pairs.

[0136] The motion equation of the six-phase permanent magnet synchronous motor is

[0137]

[0138] where J is the moment of inertia, T e is the motor torque, T L is the load torque, B is the viscous friction coefficient, and ω m is the mechanical angular velocity of the motor.

[0139] Step 12, full flow range adaptive dynamic modeling

[0140] In order to obtain an aviation electric fuel pump dynamic model that is applicable within the full flow range and takes into account both steady-state accuracy and dynamic response requirements, the present invention proposes a new method for constructing an adaptive combined nonlinear dynamic model of an aviation electric fuel pump, which is composed of a steady-state nonlinear model described by a polynomial function form and a dynamic linear model under typical operating conditions in parallel. The structural diagram of the adaptive combined nonlinear dynamic model of the aviation electric fuel pump is as shown in Figure 4 Shown. The combined model consists of three parts: a nonlinear steady-state model, an adaptive dynamic gain, and a linear dynamic model. Based on the nonlinear steady-state model, by constructing an adaptive dynamic gain and using it to adjust some parameters of the linear dynamic model, an adaptive combined nonlinear dynamic model of the aviation electric fuel pump for control system analysis and design research that takes into account both steady-state accuracy and dynamic response speed is finally obtained.

[0141] The input quantity X of the model = [u, d T T , where u is the control quantity given by the electronic controller to the electric fuel pump, and u forms the voltage supplied to the motor after passing through the power converter. d is a column vector composed of external parameters such as flight altitude, flight Mach number, ambient temperature, and inlet pressure. X s is the steady-state input value corresponding to X, is the steady-state fuel flow value; the output quantity of the model where W fb is the fuel flow, n is the rotational speed of the electric fuel pump, and y3 is a column vector composed of measurable parameters other than fuel flow and rotational speed, such as fuel temperature, fuel pressure, and differential pressure between the inlet and outlet. y s is y the corresponding steady-state output value.

[0142] Within the full flow range, according to the fuel demand command under different operating conditions of the engine construct an adaptive strategy characterized by the adaptive dynamic gain K, and independently select the nonlinear steady-state model and linear dynamic model corresponding to the large flow or small flow requirements.

[0143] The adaptive dynamic gain K in the combined model consists of two parts, namely The first part K0(X s ) can be obtained based on the nonlinear steady-state model of the electric fuel pump using the numerical differentiation algorithm.

[0144]

[0145] Among them, δX is a very small input increment value. The second part is based on the fuel demand command ​A dynamically gain adaptive compensation term is designed to make the dynamic gain more conform to different flow characteristics under large flow demand and small flow demand when adjusting the linear steady-state model of the electric fuel pump, so as to obtain an adaptive combined nonlinear dynamic model of the electric fuel pump applicable in the full flow range that takes into account both steady-state accuracy and dynamic response speed.

[0146] Step 2, Multi-step prediction of intelligent speed command

[0147] The present invention proposes Figure 5 The multi-step prediction technology of intelligent speed command as shown. This technology consists of three main modules: an inverse model of intelligent speed command based on a combined radial basis function neural network, a speed command compensator, and a multi-step prediction reference trajectory model of speed command.

[0148] Step 21, Inverse model of intelligent speed command based on combined radial basis function neural network (CRBFNN)

[0149] Neural networks have been widely applied in system identification and controller design because they have many advantages, such as being able to achieve any non-linear mapping, simultaneously process a large number of different types of inputs, and solve the problems of complementarity and redundancy between input information, etc.

[0150] The radial basis function neural network has a three-layer feedforward structure with a single hidden layer and is usually used for function approximation and classification. Compared with the BP neural network, the radial basis function neural network has a faster learning speed. The basis functions of the radial basis function neural network include Gaussian basis functions, cubic basis functions, multiquadrics, and inverse multiquadrics. The advantage of the Gaussian function is its simple structure, smooth curve, and good analytical performance. The Gaussian function has a strong local fitting ability near the mean value.

[0151] The present invention uses a combination of a Gaussian basis function with good non-linear fitting ability and a cubic basis function with good linear fitting ability, which is called a combined radial basis function neural network (CRBFNN). Figure 6 The corresponding curves of various types of basis functions are shown.

[0152] Based on the CRBFNN, the present invention establishes an inverse model of intelligent speed command according to the steady-state historical data of the fuel flow of the aviation electric fuel pump. Using this inverse model, the required speed n corresponding to the actual temperature, pressure, pressure difference and other parameters and fuel demand command of the current electric fuel pump can be obtained. at a .

[0153] Step 22, Speed command compensator

[0154] Aviation electric fuel pumps have extremely high requirements for the fuel response speed. However, the moment of inertia of the motor is an important factor affecting the fuel supply response speed. Moreover, under the requirement of small fuel flow, the influence of the motor moment of inertia on the dynamic response of the electric fuel pump is more prominent than that under the requirement of large fuel flow, further increasing the complexity of the corresponding relationship between the rotational speed and the fuel flow.

[0155] The present invention is based on the fuel demand command of the aeroengine to feedback the output fuel flow rate W of the current aviation electric fuel pump fb , and uses and the fuel flow error ΔW fb . Aiming at the influence of the motor moment of inertia and considering the respective characteristics of the fuel flow under different working conditions such as starting, idle, intermediate, maximum, and acceleration / deceleration, a suitable fuel pump rotational speed command regulator is designed to generate a rotational speed command compensation amount Δn .

[0156] The demand rotational speed n a is adjusted through the rotational speed command compensation amount Δn, and finally the rotational speed command n 0 at the current moment of the aviation electric fuel pump within the full flow range is obtained.

[0157] Step 23, rotational speed command multi-step preview reference trajectory model

[0158] Making full use of the past and future information of the command signal is beneficial to improving the control quality of the aviation electric fuel pump. Therefore, the present invention designs a reference trajectory model as shown in Equation (10) through a historical information memory based on the rotational speed command n 0 at the current moment, and combines the dynamic characteristics of the electric fuel pump to select the rise time of the system as the prediction time domain P to form a multi-step preview rotational speed command reference trajectory vector

[0159]

[0160] where n 0 (k) represents the value of n 0 at the current moment k, n 0 (k±i) represents the value of n 0 at the moment k±i, (i = 1, 2,..., P), and α j , is the weighting coefficient.

[0161] Step 3, design of a robust fault-tolerant controller based on variable-speed grey wolf multi-objective optimization

[0162] As Figure 8As shown, drawing on the basic theory of predictive control, the design technology of a robust fault-tolerant controller based on variable-speed grey wolf multi-objective optimization consists of several main modules, namely, a prediction model, feedback correction, multi-fault diagnosis and signal reconstruction, and rolling optimization solution. Among them, the role of the feedback correction module is to use the rotational speed n and, by designing an appropriate correction coefficient vector, perform feedback correction on the prediction model output vector Y m to form a corrected model prediction output vector so as to further improve the model prediction accuracy.

[0163] In the multi-objective optimization problem of the present invention, it involves minimizing the steady-state error, shortening the dynamic response time, minimizing the fuel consumption rate, minimizing the carbon dioxide emissions, etc. The constraint conditions are as shown in Equation (11), including that the rotational speed does not exceed the limit, the motor current does not exceed the maximum value, the fuel pressure difference does not exceed the limit, the fuel temperature does not exceed the limit, etc.

[0164]

[0165] Among them, J j (j = 1, 2, 3, 4,...) represents the performance index, Ω j (j = 1, 2, 3, 4,...) represents the sliding mode surface prediction terminal domain that satisfies the input constraints, t s represents the time to reach the steady state, η SFC represents the fuel consumption rate, represents the carbon dioxide emissions, n min and n max respectively represent the lower limit and upper limit of the rotational speed, i A,B,C,D,E,F represents the currents of each phase of the six-phase permanent magnet synchronous motor, i min and i max respectively represent the lower limit and upper limit of the current, T pump represents the fuel temperature, and respectively represent the lower limit and upper limit of the fuel temperature, Δp represents the fuel pressure difference, Δp min and Δp max respectively represent the lower limit and upper limit of the fuel pressure difference.

[0166] The specific implementation process of the present invention is gradually carried out according to the foregoing technical solution, and the key implementation technologies involved are as follows:

[0167] (1) Nonlinear dynamic model of an aviation electric fuel pump with multiple faults and uncertainties in the full flow range

[0168] Aviation electric fuel pumps have extremely demanding reliability requirements. However, they often operate in environments with high temperature, high pressure, and strong vibration, which undoubtedly increases the likelihood of their failure. This also makes it necessary to pay high attention to possible motor faults such as permanent magnet demagnetization and pump body faults such as gear fracture and bearing wear even after designing an electric fuel pump with fault tolerance function. In addition, data such as fuel flow, rotational speed, temperature, and pressure of aviation electric fuel pumps are obtained through sensors, and these sensors become components prone to failure in environments with high temperature, high pressure, and strong vibration. Therefore, during the modeling process of aviation electric fuel pumps, the influence of multiple faults must be considered, and the problem that multiple faults may occur at different times should also be taken into account. In addition, there are inevitably uncertain factors such as assembly manufacturing tolerances, component aging, and performance degradation in aviation electric fuel pumps, and the influence of uncertainty is also an important content that cannot be ignored in the modeling of aviation electric fuel pumps.

[0169] Therefore, based on a full analysis of the fuel flow characteristics of aviation electric fuel pumps, this invention obtains a non-linear dynamic model of the full flow range of aviation electric fuel pumps with multiple faults and uncertainties as shown in Equation (12) according to the method for establishing an adaptive combined non-linear dynamic model of electric fuel pumps mentioned in the technical solution.

[0170]

[0171] Among them, f(·) is a one-dimensional dynamic non-linear function describing the rotational speed of the system; h1(·) is a one-dimensional function describing the output n of the system. Since n is also a state variable of the system, in the absence of uncertainty and faults, h1(·) = n; h2(·) is a one-dimensional non-linear function describing the output W fb of the system, and h3(·) is a ρ vector of one-dimensional non-linear function describing the measurable output y3 of the system other than fuel flow and rotational speed, where ρ is a non-zero positive integer. Δf(·), Δh1(·), Δh2(·), Δh3(·) respectively represent the uncertainties corresponding to f(·), h1(·), h2(·), h3(·), and φ0(t), φ1(t), φ2(t) and φ j (t), (j = 3, 4,..., ρ + 2) are non-linear function descriptions of multiple faults occurring in the electric fuel pump. γ0(t - T 0f ), γ1(t - T 1f ), γ2(t - T 2f ), γ j (t - T jf ), (j = 3, 4,..., ρ + 2) represent the development types of faults acting on the system, including sudden faults and gradual faults. T 0f , T 1f , T2f , T jf , (j = 3, 4, …, ρ + 2) represents the time when the fault occurs.

[0172]

[0173] Among them, l j > 0 (j = 0, 1, …, ρ + 2) represents the type of fault development. When l j is very small, it indicates that there is a gradual fault in the system; on the contrary, when l j is relatively large, it indicates that there is a sudden fault in the system.

[0174] (2) Intelligent speed command inverse model of aviation electric fuel pump based on combined radial basis function neural network (CRBFNN)

[0175] Based on the intelligent speed command estimation of CRBFNN, as Figure 7 shown. During the forward training process, the steady-state fuel quantity and the steady-state measurement value y 3s are used as input quantities, and the steady-state motor speed n s is used as the output quantity, where the subscript s represents the steady-state value; during the reverse test process, the actual fuel demand command and the measurement value y3 are used as input quantities, and the required speed n a is used as the output quantity.

[0176] As Figure 7 shown, the overall input-output mapping relationship of CRBFNN is shown in Equation (14)

[0177]

[0178] Among them, is the input vector; n a is the output quantity; b j is the j-th central unit of the basis function hidden layer, which has the same dimension as x; h is the number of radial basis function units; v j is the weight coefficient between the hidden layer and the output layer; φ j is the width of the j-th basis function; ||·|| is the Euclidean norm of the vector; and are the weight coefficients respectively.

[0179] (3) Acquisition of robust fault-tolerant controller based on variable-speed grey wolf multi-objective optimization

[0180] (31) Establish a discrete-time prediction model

[0181] Considering the requirements of the control system for real-time performance, directly using the non-linear dynamic model of an aviation electric fuel pump within the full flow range as the multi-step prediction model may result in too slow acquisition of the multi-step prediction vector of the rotational speed, thereby affecting the acquisition of the robust fault-tolerant controller. Therefore, the present invention uses the corresponding discrete-time linearized model as the prediction model.

[0182] (32) Multi-fault diagnosis and signal reconstruction based on an adaptive non-singular terminal sliding mode observer

[0183] Based on the design of a non-singular terminal sliding mode surface, the present invention studies the sliding mode switching term containing an adaptation law according to the state estimation error, so as to construct an adaptive non-singular terminal sliding mode observer for diagnosing multi-faults and signal reconstruction.

[0184] i) Design of the adaptive non-singular terminal sliding mode observer

[0185] For the measurement signal of the j-th sensor that may contain faults, the form of the designed sliding mode observer is

[0186]

[0187] Among them, represents the first derivative of the rotational speed estimated value, represents the rotational speed estimated value, u represents the input of the system, f(·) is a non-linear function describing the pre-input of the rotational speed estimated value, k j represents the adaptive parameter; s j is the non-singular terminal sliding mode surface parameter. υ j (·) is an adaptive switching term regarding the time-varying adaptive parameter k j (t) and the sliding mode surface parameter s j , is the estimated value for the measurement signal of the j-th sensor, F j is a parameter that can be designed.

[0188] The non-singular terminal sliding mode surface function is designed as

[0189]

[0190] Among them, is the difference between the j-th measured value and the estimated value of the output of the j-th sliding mode observer, λ j and σ j are positive odd numbers and satisfy 1 < λ j / σ j < 2; β j is a constant value;

[0191] The structure of the time-varying adaptive parameter k j (t) is shown in Equation (17)

[0192]

[0193] where a j is the adaptive law gain coefficient.

[0194] ii) Multi-fault diagnosis strategy

[0195] When there are τ sensor output parameters, τ sliding mode observers are designed correspondingly, that is, each observer observes one sensor signal. Using the measured output value of the j-th sensor and the estimated value of the j-th observer, a structured residual ξ j is generated. In the case of no sensor faults, the state estimation values of each observer will converge to the true state of the system. If the j-th sensor fails and the rest of the sensors are normal, the state estimation value obtained by the j-th observer will be affected by the output signal of the sensor containing fault information. Therefore, the estimation result will deviate from the actual situation, and this method is applied to diagnose multi-fault signals.

[0196] The principle of the multi-fault diagnosis strategy is as Figure 9 shown.

[0197] iii) Signal reconstruction approach

[0198] The present invention uses a sliding mode observer to realize the reconstruction of the fault signal. When a fault is detected, the fault-free measurement signal and the output signal of the aviation electric fuel pump model are comprehensively utilized to design a non-singular terminal sliding mode surface and an adaptive law. Based on the sliding mode observer, the fault signal reconstruction is realized and the fault signal is replaced for the design of the robust fault-tolerant controller.

[0199] (33) Rolling optimization solution based on the variable-speed grey wolf optimization algorithm

[0200] The grey wolf optimization algorithm (GWO) is a new swarm intelligence optimization algorithm proposed by S. Mirjalili et al. in 2014, which is based on the coordinated work of individuals in a population. Compared with other intelligent algorithms, GWO has a faster convergence speed and better optimization ability in solving function optimization problems. However, the conventional GWO has deficiencies such as random population uncertainty and being easily trapped in local optima. The present invention intends to use chaotic mapping to generate the initial population and introduce the velocity component based on the particle swarm optimization algorithm (PSO) into GWO to form a variable-speed grey wolf optimization algorithm. The algorithm flow is as Figure 10 shown, so as to weaken the random population uncertainty, avoid being trapped in local optima, and further improve the convergence accuracy and optimization speed of the GWO algorithm, providing an efficient optimization solution method with high solution accuracy, fast convergence speed and strong global search ability for the online solution of the robust fault-tolerant controller.

[0201] Simulation experiment implementation and result analysis

[0202] Consider an aviation electric fuel pump with motor parameters and gear pump parameters as shown in Table 1 and Table 2.

[0203] Table 1 Parameters of Permanent Magnet Synchronous Motor

[0204]

[0205]

[0206] Table 2 Parameters of Gear Pump

[0207]

[0208] Figure 11 The fitness function iteration curve of the variable-speed grey wolf optimization algorithm is shown. The smaller the value of the fitness function, the better the control effect. The results show that the variable-speed grey wolf optimization algorithm provided by the present invention has better optimization ability compared with the traditional grey wolf optimization algorithm.

[0209] Figure 12 The flow rate response curve of the aviation electric fuel pump is shown. It can be seen that the control method provided by the present invention can make the flow rate output of the electric fuel pump quickly track the desired flow rate, and still has good tracking performance and high steady-state accuracy under a wide range of flow rate changes.

[0210] Figure 13 The variations of the d-axis current and q-axis current of the six-phase permanent magnet synchronous motor are shown. It can be seen that the d-axis current can be well maintained near the expected value; the q-axis current can change according to the change of fuel flow rate demand, and the response speed is fast.

[0211] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A fault-tolerant control method for an aviation electric fuel pump based on intelligent instruction prediction, characterized in that: It includes the following steps: Step 1, establish a nonlinear dynamic model of an aviation electric fuel pump with multiple faults and uncertainties within the full flow range; The said Step 1 includes: Step 11, first conduct an analysis of the fuel flow characteristics of the aviation electric fuel pump: When the motor of an aviation electric fuel pump is coaxially connected to the pump, the rotational speed of the pump is equal to that of the motor. For a given external gear pump, the fuel flow rate Q is a function of the pump rotational speed n and the volumetric efficiency η v function Q = η v Q T (1) Among them, Q T is the theoretical output flow rate of the gear pump, n is the rotational speed of the pump, b is the gear thickness, R a is the addendum circle pressure angle, R c is the pitch circle radius, t j is the base pitch length; The voltage equation of the six-phase permanent magnet synchronous motor consists of two parts: the stator phase resistance voltage drop and the magnetic flux change rate, that is where, u s = [u A u B u C u D u E u F T is the stator voltage vector, R s = diag[R s R s R s R s R s R s is the resistance vector, i s = [i A i B i C i D i E i F T is the stator current vector; ψ s represents the permanent magnet flux linkage;​​ The magnetic flux equation of the six-phase permanent magnet synchronous motor consists of two parts: the armature winding magnetic flux and the permanent magnet magnetic flux, that is ψ s = L s i s + ψ m (4) where, L s is the phase inductance matrix, and ψ m is the permanent magnet flux linkage; ψ m = ψ f [cosθ e cos(θ e - α) cos(θ e - 2α) cos(θ e - 3α) cos(θ e - 4α) cos(θ e - 5α)] T Among them, L 11 is the self-inductance of the first set of windings, L 22 is the self-inductance of the second set of windings, M 12 and M 21 are the mutual inductances between the two sets of windings, ψ f is the amplitude of the permanent magnet flux linkage, θ e is the electrical angle of the rotor, α = 2π / 3; T represents the transpose of a matrix; According to the principle of electromechanical energy conversion, the co-energy can be written as Therefore, the torque equation of the six-phase permanent magnet synchronous motor is where θ m is the mechanical angular displacement of the rotor, p and p is the number of pole pairs; The motion equation of the six-phase permanent magnet synchronous motor is Among them, J is the moment of inertia, T e is the motor torque, T L is the load torque, B is the viscous friction coefficient, ω m is the mechanical angular velocity of the motor; Step 12, construct an adaptive combined nonlinear dynamic model of the aviation electric fuel pump by paralleling a steady-state nonlinear model described by a polynomial function form and a dynamic linear model under typical working conditions; The adaptive combined nonlinear dynamic model of the aviation electric fuel pump consists of three parts: a nonlinear steady-state model, an adaptive dynamic gain, and a linear dynamic model; Based on the nonlinear steady-state model, by constructing an adaptive dynamic gain and using it to adjust some parameters of the linear dynamic model, finally obtain an adaptive combined nonlinear dynamic model of the aviation electric fuel pump for control system analysis and design research that takes into account both steady-state accuracy and dynamic response speed; The input quantity X of the model = [u, d T T , where u is the control quantity given by the electronic controller to the electric fuel pump, and u forms the voltage supplied to the motor after passing through the power converter. d is a column vector composed of external parameters such as flight altitude, flight Mach number, ambient temperature, and inlet pressure. X s is the steady-state input value corresponding to X, and W fbs is the steady-state fuel flow value; the output quantity of the model where W fb is the fuel flow, n is the rotational speed of the electric fuel pump, and y3 is a column vector composed of measurable parameters other than fuel flow and rotational speed, such as fuel temperature, fuel pressure, and differential pressure between the inlet and outlet. y s is y the corresponding steady-state output value;​ Within the full flow range, according to the fuel demand commands under different engine operating conditions Construct an adaptive strategy characterized by the adaptive dynamic gain K, and autonomously select the corresponding non-linear steady-state model and linear dynamic model under large flow or small flow requirements; The adaptive dynamic gain K in the combined model consists of two parts, namely The first part K0(X s ) is obtained by using the numerical differential algorithm based on the non-linear steady-state model of the electric fuel pump; Among them, δX is an extremely small input increment value; the second part is a dynamically gain adaptive compensation term designed according to the fuel demand command to make the dynamic gain more in line with the different flow characteristics under large flow demand and small flow demand when adjusting the linear steady-state model of the electric fuel pump, so as to obtain an adaptive combined nonlinear dynamic model of the electric fuel pump applicable in the full flow range that takes into account both steady-state accuracy and dynamic response speed; Step 2, establish an intelligent speed command inverse model of the aviation electric fuel pump based on a combined radial basis function neural network; The said Step 2 includes: Step 21, construct an intelligent speed command inverse model based on a combined radial basis function neural network; Based on the steady-state historical data of the fuel flow of the aviation electric fuel pump, an intelligent inverse model of the rotational speed command is established based on the combined radial basis function neural network (CRBFNN). Using this inverse model, the actual temperature, pressure, pressure difference and fuel demand command of the current electric fuel pump can be obtained. The corresponding required rotational speed n a ; Step 22, speed command compensator According to the fuel demand command of the aero-engine feedback the output fuel flow rate W of the current aero-electric fuel pump fb , and utilize and the fuel flow error ΔW fb , aiming at the influence of the motor inertia, and considering the respective characteristics of the fuel flow rate under different operating conditions such as start-up, idle, intermediate, maximum, and acceleration / deceleration, design a fuel pump speed command regulator to generate a speed command compensation amount Δn ; Through the rotational speed command compensation amount Δn Adjust the required rotational speed n a to finally obtain the current moment command n of the rotational speed of the aviation electric fuel pump within the full flow range 0 ; Step 23, multi-step predictive reference trajectory model of speed command Based on the current rotational speed command \(n\) at the current moment through the historical information memory 0 , a reference trajectory model as shown in Equation (10) is designed. Combining with the dynamic characteristics of the electric fuel pump, the rise time of the system is selected as the prediction horizon \(P\) to form a multi-step predictive rotational speed command reference trajectory vector where n 0 (k) represents the value of n 0 at the current moment k, and n 0 (k±i) represents the value of n 0 at the moments k±i, (i = 1, 2, …, P), and α j , is the weighting coefficient; Step 3, design a robust fault-tolerant controller based on variable-speed grey wolf multi-objective optimization; In the said Step 3, The multi-objective optimization problem involves minimizing the steady-state error, minimizing the dynamic response time, minimizing the fuel consumption rate, and minimizing the carbon dioxide emission. The constraint conditions are shown in Equation (11), including the speed not exceeding the limit, the motor current not exceeding the maximum value, the fuel pressure difference not exceeding the limit, and the fuel temperature not exceeding the limit, Among which J j (j = 1, 2, 3, 4,...) represents a performance index, Ω j (j = 1, 2, 3, 4,...) represents the sliding mode surface prediction terminal domain satisfying the input constraints, t s represents the steady-state arrival time, η SFC represents the fuel consumption rate, Q CO2 represents the carbon dioxide emission, n min and n max respectively represent the lower limit and the upper limit of the rotational speed, i A,B,C,D,E,F represents the phase currents of the six-phase permanent magnet synchronous motor, i min and i max respectively represent the lower limit and the upper limit of the current, T pump represents the fuel temperature, and respectively represent the lower limit and the upper limit of the fuel temperature, Δp represents the fuel pressure difference, Δp min and Δp max respectively represent the lower limit and the upper limit of the fuel pressure difference; The said Step 3 includes: Step 31, adopt the corresponding discrete-time linearized model as the prediction model; Step 32, based on the design of a non-singular terminal sliding mode surface, according to the state estimation error, study the sliding mode switching term including the adaptive law, so as to construct an adaptive non-singular terminal sliding mode observer for diagnosing multiple faults and signal reconstruction; Step 33, rolling optimization solution based on the variable-speed grey wolf optimization algorithm Use the chaos mapping to generate the initial population, and introduce the velocity component based on the particle swarm optimization algorithm (PSO) into the GWO to form the variable-speed grey wolf optimization algorithm for online solution of the robust fault-tolerant controller.

2. The fault-tolerant control method for an aviation electric fuel pump based on intelligent instruction prediction according to claim 1, wherein: The adaptive combined nonlinear dynamic model of the aviation electric fuel pump constructed in the said Step 12 is: where, f(·) is a one-dimensional dynamic non-linear function describing the rotational speed of the system; h1(·) is a one-dimensional function describing the output n of the system. Since n is also a state variable of the system, in the case of no uncertainty and faults, h1(·) = n; h2(·) is a one-dimensional non-linear function describing the output W fb of the system, and h3(·) is a ρ vector of one-dimensional non-linear functions describing the measurable output y3 of the system other than fuel flow and rotational speed, where ρ is a non-zero positive integer; Δf(·), Δh1(·), Δh2(·), Δh3(·) respectively represent the uncertainties corresponding to f(·), h1(·), h2(·), h3(·), and φ0(t), φ1(t), φ2(t) and φ j (t), (j = 3, 4,..., ρ + 2) are non-linear function descriptions of multiple faults occurring in the electric fuel pump, and γ0(t - T 0f ), γ1(t - T 1f ), γ2(t - T 2f ), γ j (t - T jf ), (j = 3, 4,..., ρ + 2) represent the development types of faults acting on the system, including sudden faults and gradual faults, and T 0f , T 1f , T 2f , T jf , (j = 3, 4,..., ρ + 2) represent the occurrence times of the faults; where l j > 0 (j = 0, 1,..., ρ + 2) represents the development type of the fault.

3. The fault-tolerant control method for an aviation electric fuel pump based on intelligent instruction prediction according to claim 1, wherein: In the said Step 2, Intelligent speed command estimation based on the combined radial basis function neural network (CRBFNN). During the forward training process, the steady-state fuel quantity and the steady-state measurement value y 3s are used as input quantities, and the steady-state speed n of the motor s is used as the output quantity, where the subscript s represents the steady-state value; during the reverse testing process, the actual fuel demand command and the measurement value y3 are used as input quantities, and the required speed n a is used as the output quantity; The overall input-output mapping relationship of the combined radial basis function neural network (CRBFNN) is shown in Equation (14) Among them, is the input vector; n a is the output quantity; b j is the j-th central unit of the basis function hidden layer, having the same dimension as x; h is the number of radial basis function units; v j is the weight coefficient between the hidden layer and the output layer; φ j is the width of the j-th basis function; ||·|| is the Euclidean norm of the vector; and are the weight coefficients respectively.

4. The fault-tolerant control method for an aviation electric fuel pump based on intelligent instruction prediction according to claim 1, wherein: The said Step 32 includes: Step 321, design of the adaptive non-singular terminal sliding mode observer For the measurement signal of the j-th sensor that may contain faults, the designed sliding mode observer form is Among them, represents the first derivative of the rotational speed estimation value, represents the rotational speed estimation value, u represents the input of the system, f(·) is a nonlinear function describing the pre-input of the rotational speed estimation value, k j represents the adaptive parameter; s j is the non-singular terminal sliding mode surface parameter, υ j (·) is the adaptive switching term with respect to the time-varying adaptive parameter k j (t) and the sliding mode surface parameter s j ; is the estimation value for the measurement signal of the j-th sensor, F j is a designable parameter; The non-singular terminal sliding mode surface function is designed as wherein, is the difference between the j-th measured value and the estimated value output by the j-th sliding mode observer, λ j and σ j are positive odd numbers and satisfy 1 < λ j / σ j < 2; β j is a constant value; Time-varying adaptive parameter k j (t) has the structure shown in Equation (17). where a j is the adaptive law gain coefficient; Step 322, multi-fault diagnosis strategy When there are τ sensor output parameters, τ sliding mode observers are designed correspondingly, that is, each observer observes one sensor signal. Using the measured output value of the j-th sensor and the estimated value of the j-th observer, a structured residual ξ is generated. j , in the case of no sensor faults, the state estimation values of each observer will converge to the true state of the system. If the j-th sensor fails and the other sensors are normal, the state estimation value obtained by the j-th observer will be affected by the output signal of the sensor containing fault information. Therefore, the estimation result will deviate from the actual situation. This method is applied to diagnose multi-fault signals. Step 323, signal reconstruction When a fault is detected, comprehensively utilize the fault-free measurement signals and the output signals of the aeroelectric fuel pump model, design a non-singular terminal sliding mode surface and an adaptive law, and based on the sliding mode observer, realize the reconstruction of the fault signal and replace the fault signal for the design of the robust fault-tolerant controller.

Images