CN114961985B - A method and system for intelligently predicting the performance of a hydrogen fueled aviation rotary engine - Google Patents

Abstract

The invention discloses an intelligent prediction method and system for performance of a hydrogen fuel aviation rotor engine, which comprises the following specific steps: constructing a zero-dimensional performance simulation model of the hydrogen fuel aviation rotor engine based on the actual gas physical parameters and the laminar flame propagation speed of the hydrogen fuel combustion process; obtaining the engine indicated power, the indicated heat efficiency and the indicated oil consumption rate according to the zero-dimensional performance simulation model to obtain a rotor engine zero-dimensional performance simulation data set; constructing a hydrogen fuel rotor engine performance neural network model based on a Bayesian regularization algorithm, and training the hydrogen fuel rotor engine performance neural network model by adopting a simulation data set; predicting and obtaining the indicated power, the indicated heat efficiency and the indicated fuel consumption rate of the hydrogen fuel rotor engine by utilizing the hydrogen fuel rotor engine performance neural network model; the invention can realize quick and accurate prediction of the engine performance and can provide firm theoretical support for the control system design of the engine and the establishment of the optimal control strategy of the engine.

Description

A method and system for intelligently predicting the performance of a hydrogen fueled aviation rotary engine

Technical Field

The invention belongs to the field of aviation power, and in particular to a method and system for intelligently predicting the performance of a hydrogen fuel aviation rotor engine integrating a zero-dimensional model and a neural network.

Background technique

Distributed and clustered air combat operations rely on distributed and clustered combat platforms composed of multiple small drones. Compared with traditional single all-round drones, this puts forward more urgent demands on the power system of small drone cluster platforms with high efficiency, low cost and long flight time. For micro drones with thrust power requirements within 1kN/50kW, piston engines, rotary engines, turbojet engines and electric drive systems are generally used as their power sources. The rotary engine is the Wankel engine, which has outstanding performance in small drone power devices due to its simple and compact structure, high power-to-weight ratio, low vibration and noise, and relatively low fuel consumption.

With the demand for carbon reduction in the aviation industry, more environmentally friendly and low-carbon hydrogen is being used to replace traditional high-carbon emission gasoline, kerosene and other fuels. The performance prediction of hydrogen-fueled aviation rotary engines under complex working conditions is crucial to the safe operation of aviation rotary engines. At present, the performance prediction method of aviation rotary engines mainly uses three-dimensional CFD numerical simulation software, which has the disadvantages of slow simulation prediction speed, huge computing resources, and memory of more than GB. At present, the one-dimensional simulation commercial software AVL BOOST and GT-POWER are mainly suitable for traditional reciprocating piston engines. The prediction method of using a three-cylinder four-stroke piston engine model to replace the rotary engine has the disadvantage of low prediction accuracy. In addition, the current three-dimensional CFD numerical simulation software and one-dimensional simulation commercial software both use ideal gas properties in the calculation process. However, the proportion of water in the combustion products of hydrogen fuel is much greater than that of traditional hydrocarbon fuels, and its actual physical properties seriously deviate from the theoretical gas properties, so it is not suitable to use ideal gas properties to calculate the combustion products of hydrogen fuel.

In summary, there is an urgent need for a fast, high-precision prediction method and system for hydrogen-fueled aviation rotary engines.

Summary of the invention

In order to solve the problems existing in the prior art, the present invention provides a method and system for intelligent prediction of the performance of a hydrogen fuel aviation rotary engine, which integrates a zero-dimensional model based on actual gas physical properties and flame propagation speed and a neural network model based on a Bayesian regularization algorithm, so as to realize fast and accurate prediction of the indicated power, indicated thermal efficiency and indicated fuel consumption rate of the aviation rotary engine under various operating conditions given the core geometric parameters of the hydrogen fuel engine.

To achieve the above object, the present invention provides the following technical solution: a method for intelligently predicting the performance of a hydrogen fuel aviation rotary engine, the specific steps of which are as follows:

S1 builds a zero-dimensional performance simulation model of a hydrogen fuel aviation rotary engine based on transient mass conservation, energy conservation, actual gas physical parameters, and laminar flame propagation speed of hydrogen fuel combustion process;

S2 Given the core geometric parameters of the hydrogen fuel rotary engine, under any combination of hydrogen fuel injection amount, rotation speed and ignition advance angle, obtain the indicated power, indicated thermal efficiency and indicated fuel consumption rate of the zero-dimensional performance simulation model of the hydrogen fuel aviation rotary engine, and obtain the zero-dimensional performance simulation data set of the rotary engine;

S3 builds a hydrogen fuel rotary engine performance neural network model based on the Bayesian regularization algorithm, and uses the rotary engine zero-dimensional performance simulation data set to train the hydrogen fuel rotary engine performance neural network model;

S4 inputs any combination of hydrogen fuel injection amount, rotation speed and ignition advance angle into the hydrogen fuel rotary engine performance neural network model to predict the indicated power, indicated thermal efficiency and indicated fuel consumption rate of the hydrogen fuel rotary engine.

Further, in step S1, the zero-dimensional performance simulation model of the hydrogen fuel aviation rotary engine includes intake, compression, combustion, expansion, and exhaust process modules, and the above modules are embedded with geometric sub-models, thermodynamic sub-models, heat exchange loss sub-models, mass leakage sub-models, and combustion heat release sub-models, wherein the geometric sub-model is used to obtain the volume V of each working chamber in the rotary engine; the thermodynamic sub-model is used to obtain the working chamber pressure P through the actual gas physical property state equation; the heat exchange loss sub-model is used to obtain the convective heat exchange loss _Qw between the gas in the engine working chamber and the engine wall, and the combustion heat release sub-model is used to obtain the hydrogen fuel combustion heat release _QB according to the laminar flame propagation speed S; the mass leakage sub-model is used to obtain the leakage mass _mleak of the gas between adjacent chambers at the top of the rotor.

Furthermore, in step S1, the combustion heat release sub-model uses the Weber heat release model to predict the heat release Q _B of hydrogen fuel combustion, specifically:

Where: LHV is the lower heating value of the fuel; η _B is the combustion efficiency; is the combustion duration angle; /> is the ignition angle; m is the combustion quality coefficient.

Furthermore, in step S1, the laminar flame propagation speed S during the hydrogen combustion process determines the combustion duration angle The laminar flame propagation speed S under different temperature and pressure conditions is obtained by correcting the laminar flame propagation speed S _ref under standard conditions, as shown in formulas 6 to 8.

γ＝2.18-0.8(φ-1) (7)

σ＝-0.16+0.22(φ-1) (8)

Where: γ is the temperature correction coefficient; σ is the pressure correction coefficient; φ is the equivalence ratio.

Furthermore, in step S1, the combustion duration angle The calculation formula is:

Wherein, S _des is the flame propagation speed under rated conditions.

Furthermore, in step S1, the working chamber pressure P is obtained by the Benedict-Webb-Rubin actual gas physical property state equation, specifically:

In the formula: m _c is the mass of the working chamber working fluid; u is the internal energy of the chamber working fluid; Q _B is the heat released by combustion; min _is the mass of the inhaled air; h _in is the enthalpy value of the inhaled air; m _fuel is the mass of the inhaled fuel; h _fuel is the enthalpy value of the inhaled fuel; m _exh is the exhaust mass; h _exh is the exhaust enthalpy value; m _leak is the leakage mass; h _leak is the enthalpy value of the leaking working fluid; Q _w is the heat dissipation between the chamber gas and the cylinder body and the rotor wall; V _m is the molar volume; R is the gas constant; A ₀ , B ₀ , C ₀ , a, b, c, α, γ are all constants.

Furthermore, in step S3, the hydrogen fuel rotary engine performance neural network model includes 3 input layer network nodes, 20 hidden layer network nodes and 3 output layer network nodes; the input layer x = [x ₁ , x ₂ , x ₃ ] ^T , wherein x ₁ , x ₂ , x ₃ represent the hydrogen fuel injection amount, the rotation speed and the ignition advance angle x ₃ , respectively; the output layer y = [y ₁ , y ₂ , y ₃ ] ^T , wherein y ₁ , y ₂ , y ₃ represent the indicated power y ₁ , the indicated thermal efficiency y ₂ , and the indicated fuel consumption rate y ₃ , respectively.

Furthermore, in step S3, the performance neural network model of the hydrogen fuel rotary engine is specifically:

In the formula, _Si is the hidden layer node data at the i-th iteration, K is the activation function max(0.1x,x), w, U, v are weight coefficients, and b is the difference between the predicted output layer data _Y and the actual output layer data at the k-th iteration. Deviation; /> is the normalized input layer; is the normalized output layer.

Furthermore, in step S3, the predicted output layer data _Y is reverse normalized to obtain the predicted actual output layer node data Y = [y ₁ , y ₂ , y ₃ ] ^T , namely the indicated power y ₁ , indicated thermal efficiency y ₂ , and indicated fuel consumption rate y ₃ of the hydrogen fuel rotary engine, specifically:

Where: ps is the mapping of output layer data.

The present invention also provides a hydrogen fuel aviation rotor engine performance intelligent prediction system, comprising

The simulation model building module is used to build a zero-dimensional performance simulation model of a hydrogen fuel aviation rotary engine based on transient mass conservation, energy conservation, actual gas physical parameters, and laminar flame propagation speed of the hydrogen fuel combustion process;

The simulation data set establishment module is used to obtain the indicated power, indicated thermal efficiency and indicated fuel consumption rate of the zero-dimensional performance simulation model of the hydrogen fuel aviation rotary engine under the conditions of any combination of hydrogen fuel injection amount, rotation speed and ignition advance angle, and obtain the zero-dimensional performance simulation data set of the rotary engine;

A network model building training module is used to build a hydrogen fuel rotary engine performance neural network model based on a Bayesian regularization algorithm, and use a rotary engine zero-dimensional performance simulation data set to train the hydrogen fuel rotary engine performance neural network model;

The prediction module is used to input any combination of hydrogen fuel injection amount, rotation speed and ignition advance angle into the hydrogen fuel rotary engine performance neural network model to predict the indicated power, indicated thermal efficiency and indicated fuel consumption rate of the hydrogen fuel rotary engine.

Compared with the prior art, the present invention has at least the following beneficial effects:

The invention discloses an intelligent prediction method for a hydrogen fuel aviation rotary engine, which integrates a zero-dimensional model based on actual gas physical property parameters and laminar flame propagation speed and a neural network model based on a Bayesian regularization algorithm. The method can quickly predict the indicated power, indicated thermal efficiency and indicated fuel consumption rate of the aviation rotary engine under conditions of different hydrogen fuel injection amounts, different rotation speeds and different ignition advance angles, thereby realizing rapid and accurate prediction of engine performance. The method can provide solid theoretical support for the design of engine control systems and the formulation of engine optimal control strategies. The predicted engine performance data can also be used to evaluate the performance degradation of the hydrogen fuel rotary engine by comparing it with actual test performance data.

The intelligent prediction method for hydrogen fuel aviation rotary engine proposed in the present invention obtains actual gas physical property parameters including internal energy u, enthalpy value h, specific heat capacity at constant pressure _cp , specific heat capacity at constant volume _CV , and adiabatic coefficient k through interpolation method based on a self-built working fluid thermophysical property software library. The actual gas state equation is used to replace the traditional ideal gas state equation to solve the working chamber pressure. The simulated working chamber temperature and pressure are closer to the actual state, the calculation accuracy is improved, and the determination coefficient ^R2 of the prediction result reaches above 0.98.

The intelligent prediction method for hydrogen fuel aviation rotary engines proposed in the present invention has strong scalability. The performance prediction of rotary engines with different structures can be achieved by only changing the core parameters of the rotary engines. In addition, the present invention transforms the iterative solution of multiple complex differential equations in the traditional simulation calculation method into intelligent calculation of a neural network model. By reducing the amount of calculation tasks, the prediction speed is improved and the calculation memory is reduced. The prediction speed can reach millisecond level, and the algorithm occupies little computer memory, only KB level.

BRIEF DESCRIPTION OF THE DRAWINGS

Figure 1 Intelligent prediction method for hydrogen fuel aviation rotary engine performance;

Fig. 2 is a schematic diagram of the structure of a rotary engine;

FIG3 is a zero-dimensional simulation model structure of a hydrogen fuel rotary engine;

FIG4 shows the variation trend of the volume V of each working chamber of the rotary engine with the rotation angle of the eccentric shaft;

FIG5 is a graph showing the variation trend of the heat release rate of hydrogen combustion in a rotary engine with the rotation angle of the eccentric shaft;

Figure 6 shows the variation trend of laminar flame propagation velocity S _ref of hydrogen combustion with equivalence ratio under standard conditions;

FIG7 is a comparison result of the simulated working pressure of the rotary engine zero-dimensional model and the experimental result;

FIG8 is a comparison of the simulation working temperature of the rotary engine zero-dimensional model and the experimental result;

Figure 9 shows the change trend of indicated fuel consumption rate of hydrogen fuel aviation rotary engine at different speeds;

Figure 10 shows the indicated power variation trend of the hydrogen fuel aviation rotary engine at different speeds;

Figure 11 shows the change trend of indicated thermal efficiency of hydrogen fuel aviation rotary engine at different speeds;

Figure 12 shows the change trend of the indicated fuel consumption rate of the hydrogen fuel aviation rotary engine under different hydrogen fuel injection amounts;

Figure 13 shows the change trend of the indicated power of the hydrogen fuel aviation rotary engine under different hydrogen fuel injection amounts;

FIG14 shows the change trend of the indicated thermal efficiency of the hydrogen fuel aviation rotary engine under different hydrogen fuel injection amounts;

Figure 15 shows the change trend of indicated fuel consumption rate of hydrogen fuel aviation rotary engine under different ignition advance angles;

Figure 16 shows the indicated power variation trend of the hydrogen fuel aviation rotary engine at different ignition advance angles;

Figure 17 shows the change trend of the indicated thermal efficiency of the hydrogen fuel aviation rotary engine at different ignition advance angles;

FIG18 is a schematic diagram of the structure of a neural network model;

Figure 19 shows the neural network model fitting results.

Detailed ways

In order to make the purpose, technical effect and technical solution of the embodiment of the present invention clearer, the technical solution in the embodiment of the present invention is clearly and completely described below in conjunction with the drawings in the embodiment of the present invention; it is obvious that the described embodiment is a part of the embodiment of the present invention. Based on the embodiment disclosed in the present invention, other embodiments obtained by ordinary technicians in this field without making creative work should all fall within the scope of protection of the present invention.

As shown in FIG1 , a method for intelligently predicting the performance of a hydrogen fuel aviation rotary engine of the present invention integrates a zero-dimensional model based on actual gas physical parameters and laminar flame propagation velocity of the hydrogen fuel combustion process and a neural network model based on a Bayesian regularization algorithm, so as to achieve fast and accurate prediction of the indicated power, indicated thermal efficiency and indicated fuel consumption rate of the aviation rotary engine under various operating conditions by giving the core geometric parameters of the hydrogen fuel engine. The specific steps are as follows:

Step 1 M1: Construct a zero-dimensional performance simulation model of a hydrogen fuel aviation rotary engine based on transient mass conservation, energy conservation, actual gas physical state parameters, and laminar flame propagation speed of the hydrogen fuel combustion process.

The second step M2: Use experimental data to verify the zero-dimensional simulation model of the hydrogen fuel aviation rotary engine. Based on the zero-dimensional simulation model after experimental verification, simulate the given core geometric parameters of the hydrogen fuel rotary engine (shape coefficient K, eccentricity e, cylinder thickness B, combustion chamber pit volume _Vc ), and obtain the indicated power, indicated thermal efficiency, and indicated fuel consumption rate of the aviation rotary engine under no less than 10,000 groups of different hydrogen fuel injection amounts, different speeds, and different ignition advance angles to obtain the zero-dimensional performance simulation data set of the rotary engine.

The third step M3: construct a hydrogen fuel rotary engine performance neural network model based on the Bayesian regularization algorithm, and use the rotary engine zero-dimensional performance simulation data set to train the hydrogen fuel rotary engine performance neural network model.

Step 4 M4: Specify the core geometric parameters of the hydrogen fuel rotary engine (shape coefficient K, eccentricity e, cylinder thickness B, combustion chamber pit volume V _c ), input the hydrogen fuel injection amount, speed and ignition advance angle into the hydrogen fuel rotary engine performance neural network model, and the hydrogen fuel rotary engine performance neural network model can quickly and accurately predict the indicated power, indicated thermal efficiency and indicated fuel consumption rate of the hydrogen fuel rotary engine.

In step one, the zero-dimensional performance simulation model of the hydrogen-fueled aviation rotary engine includes intake, compression, combustion, expansion, and exhaust process modules. Each of the above process modules is embedded with a geometric sub-model, a thermodynamic sub-model based on actual gas properties, a heat transfer loss sub-model, and a mass leakage sub-model. The combustion module is additionally embedded with a combustion heat release sub-model based on the propagation speed of hydrogen laminar flames.

in:

The geometric sub-model is used to obtain the volume of the working chamber. The three chambers formed between the cylinder body and the triangular rotor of the rotary engine are all working chambers. The volume V of each working chamber is obtained by formula 1:

Where: K is the shape coefficient; e is the eccentricity; B is the cylinder thickness; _Vk is the volume of the combustion chamber pit.

The thermodynamic sub-model is used to obtain the working chamber pressure P and temperature T. The three working chambers undergo the same intake, compression, combustion, expansion and exhaust processes at different phases. For each rotation of the rotor, the engine performs work three times and outputs mechanical energy to the outside through the eccentric shaft. The working temperature T of the chamber in each working process is obtained by the transient energy (Formula 2) and the mass conservation equation (Formula 3). The working pressure of the chamber is obtained by the Benedict-Webb-Rubin actual gas physical property state equation (Formula 4). The physical property parameters involved in the thermodynamic sub-model, such as internal energy u, enthalpy value h, constant pressure specific heat capacity _cp , constant volume specific heat capacity _cv , and adiabatic coefficient k, are all interpolated and called from the self-built physical property database.

In the formula: m _c is the mass of the working fluid in the working chamber; u is the internal energy of the working fluid in the chamber; Q _B is the heat released by combustion; min _is the mass of the inhaled air; h _in is the enthalpy value of the inhaled air; m _fuel is the mass of the inhaled fuel; h _fuel is the enthalpy value of the inhaled fuel; m _exh is the exhaust mass; h _exh is the exhaust enthalpy value; m _leak is the leakage mass; h _leak is the enthalpy value of the leaking working fluid; Q _w is the heat dissipation between the chamber gas and the cylinder body and the rotor wall; V _m is the molar volume; R is the gas constant; A ₀ , B ₀ , C ₀ , a, b, c, α, γ are all constants.

The combustion heat release sub-model is used to obtain the heat release Q _B of hydrogen fuel combustion, and the Weber heat release model (Formula 5) is used to predict the heat release Q _B during the fuel combustion process.

Where: LHV is the lower heating value of the fuel; η _B is the combustion efficiency; is the combustion duration angle; /> is the ignition angle; m is the combustion quality coefficient.

Burning duration angle Determined by the laminar flame propagation speed S during hydrogen combustion, the combustion duration angle under rated conditions is Through three-dimensional CFD numerical simulation calibration, the combustion duration angle under other working conditions/> Obtained by formula 9.

The laminar flame propagation speed S under different temperature and pressure conditions is obtained by correcting the laminar flame propagation speed S _ref under standard conditions, as shown in formulas 6 to 8.

γ＝2.18-0.8(φ-1) (7)

σ＝-0.16+0.22(φ-1) (8)

Where: γ is the temperature correction coefficient; σ is the pressure correction coefficient; φ is the equivalence ratio; S _des is the flame propagation speed under rated conditions; is the combustion duration angle under rated operating conditions.

The heat loss sub-model is used to obtain the convective heat loss Q _w between the gas in the working chamber and the engine wall, which is calculated by formulas 10 to 11:

In the formula: _Aw is the wall heat transfer area; _αc is the convection heat transfer coefficient; U is the average rotor speed.

The mass leakage submodel is used to obtain the leakage mass m _leak of the gas between adjacent chambers at the top of the rotor. The leakage loss is calculated by the Bernoulli equation. The mass of the subsonic leakage flow when the working chamber pressure is lower than the critical pressure (Formula 14) is obtained by Formula 12, and the mass of the sonic leakage flow when the working chamber pressure is higher than the critical pressure is obtained by Formula 13.

Where: k is the air flow adiabatic coefficient; A _leak is the leakage area; τ is the time; P _cr is the critical pressure; P ₀ is the ambient pressure.

In step 3, the hydrogen fuel rotary engine performance neural network model includes 3 input layer network nodes, 20 hidden layer network nodes and 3 output layer network nodes;

Input layer x = [x ₁ , x ₂ , x ₃ ] ^T network nodes represent the hydrogen fuel injection amount x ₁ , speed x ₂ , ignition advance angle x ₃ ;

Output layer y = [y ₁ , y ₂ , y ₃ ] ^{The network} nodes represent the indicated power y ₁ , the indicated thermal efficiency y ₂ , and the indicated fuel consumption rate y ₃ respectively.

First, read the training data set, normalize the data of each node in the input layer and output layer to data between 0 and 1, and obtain the normalized input layer and output layer/>

Where: ps is the mapping of output layer data.

Secondly, a neural network is created, and the total number of iterations is set to N. At the k-1th iteration, the hidden layer data S ^k-1 is stored in real time and fed back for the kth iteration calculation; the weight coefficients w and U are random numbers, and the weight coefficients w and U are the same in each iteration; at the kth iteration, the hidden layer data S ^k is calculated by formula 17, where the activation function K(x) is max(0.1x,x);

Where: b is the kth iteration prediction output layer data With the actual output layer data/> Deviation;

Further, the kth iteration predicts the output layer data Calculated by formula 18, the weight coefficient v is a random number, and the weight coefficient v is the same in each iteration;

Finally, after the output layer data is reversed, the predicted actual output layer node data Y = [y ₁ ,y ₂ ,y ₃ ] ^T can be obtained, that is, the indicated power y ₁ , indicated thermal efficiency y ₂ , and indicated fuel consumption rate y ₃ of the hydrogen fuel rotary engine;

Based on the neural network structure model constructed above, the Bayesian regularization algorithm is used to train the model. The training data is a randomly selected set of no less than 10,000 zero-dimensional performance simulation data sets, 80% of which are used for training the model and 20% for testing the model. The hydrogen fuel rotary engine performance neural network model can be obtained by iterating N times until the prediction error b meets the training target error.

The function corresponding to the neural network model that has passed the training, i.e., the determination coefficient ^R2 is not less than 0.98, is derived to obtain a trained neural network model of hydrogen fuel rotary engine performance.

The present invention also provides a hydrogen fuel aviation rotor engine performance intelligent prediction system, comprising:

The simulation model building module is used to build a zero-dimensional performance simulation model of a hydrogen fuel aviation rotary engine based on transient mass conservation, energy conservation, actual gas physical parameters, and laminar flame propagation speed of the hydrogen fuel combustion process;

The simulation data set establishment module is used to obtain the indicated power, indicated thermal efficiency and indicated fuel consumption rate of the zero-dimensional performance simulation model of the hydrogen fuel aviation rotary engine under the conditions of any combination of hydrogen fuel injection amount, rotation speed and ignition advance angle, and obtain the zero-dimensional performance simulation data set of the rotary engine;

A network model building and training module is used to build a hydrogen fuel rotary engine performance neural network model based on a Bayesian regularization algorithm, and use a rotary engine zero-dimensional performance simulation data set to train the hydrogen fuel rotary engine performance neural network model;

The prediction module is used to input any combination of hydrogen fuel injection amount, rotation speed and ignition advance angle into the hydrogen fuel rotary engine performance neural network model to predict the indicated power, indicated thermal efficiency and indicated fuel consumption rate of the hydrogen fuel rotary engine.

Example 1

As shown in Figure 2, the geometric structural parameters of the rotor engine used in this embodiment include an eccentricity e of 15 mm, a shape coefficient K of 7, a cylinder thickness B of 70 mm, a rotor combustion pit volume of 50 ^cm3 , and a rotor top clearance leakage area _Aleak of 0.01 ^cm2 . The structural parameters are used to predict the performance parameters of the engine under different working conditions, and the rated speed of the engine is 7000 rpm, and the rated combustion ignition advance angle is 27°.

Step 1 M1: construct a zero-dimensional performance simulation model of a hydrogen fuel aviation rotary engine based on transient mass conservation, energy conservation, actual gas thermophysical property equations, and laminar flame propagation speed of hydrogen fuel combustion process, see Figure 3;

In the geometric sub-model, the three chambers between the cylinder body and the triangular rotor of the rotary engine are all working chambers. The volume V of each working chamber is obtained by formula (1).

Where: K is the shape coefficient, which is 7 in the example; e is the eccentricity, which is 15 mm in the example; B is the cylinder thickness, which is 70 mm in the example; V _k is the volume of the combustion chamber pit, which is 50 cm ³ in the example.

Please refer to Figure 4, the volume V of each working chamber of the rotary engine changes with the eccentric shaft rotation angle. The change value of .

The thermodynamic sub-model is used to obtain the working chamber pressure P and temperature T. The three working chambers undergo the same intake, compression, combustion, expansion and exhaust processes at different phases. For each rotation of the rotor, the engine performs work three times and outputs mechanical energy to the outside through the eccentric shaft. The working temperature T of the chamber in each working process is obtained by the transient energy (Formula 2) and the mass conservation equation (Formula 3). The working pressure of the chamber is obtained by the Benedict-Webb-Rubin actual gas physical property state equation (Formula 4). The physical property parameters involved in the thermodynamic sub-model, such as internal energy u, enthalpy value h, constant pressure specific heat capacity _cp , constant volume specific heat capacity _cv , and adiabatic coefficient k, are all interpolated and called from the self-built physical property database.

In the formula: m _c is the mass of the working chamber working fluid; u is the internal energy of the chamber working fluid; Q _B is the heat released by combustion; min _is the mass of the inhaled air; h _in is the enthalpy value of the inhaled air; m _fuel is the mass of the inhaled fuel; h _fuel is the enthalpy value of the inhaled fuel; m _exh is the exhaust mass; h _exh is the exhaust enthalpy value; m _leak is the leakage mass; h _leak is the enthalpy value of the leaking working fluid; Q _w is the heat dissipation between the chamber gas and the cylinder body and the rotor wall; V _m is the molar volume; R is the gas constant; A ₀ , B ₀ , C ₀ , a, b, c, α, γ are all constants.

The combustion heat release sub-model is used to obtain the heat release Q _B of hydrogen fuel combustion. The Weber heat release model (Formula 5) is used to predict the heat release Q _B during fuel combustion. Refer to Figure 5 for the heat release rate of hydrogen combustion.

Where: LHV is the lower heating value of the fuel; η _B is the combustion efficiency; is the combustion duration angle; /> is the ignition angle; m is the combustion quality coefficient, which is 3.

Burning duration angle Determined by the laminar flame propagation speed S during hydrogen combustion, the combustion duration angle under rated conditions is Through three-dimensional CFD numerical simulation calibration, the combustion duration angle under other working conditions/> Obtained by formula 9.

The laminar flame propagation speed S of hydrogen combustion under different temperature and pressure conditions is obtained by correcting the laminar flame propagation speed S _ref under standard conditions (see FIG. 6 ), as shown in formulas 6 to 8.

γ＝2.18-0.8(φ-1) (7)

σ＝-0.16+0.22(φ-1) (8)

Where: γ is the temperature correction coefficient; σ is the pressure correction coefficient; φ is the equivalence ratio; S _des is the flame propagation speed under rated conditions.

The heat loss sub-model is used to obtain the convective heat loss Q _w between the gas in the working chamber and the engine wall, which is calculated by formulas 10 to 11:

In the formula: _Aw is the wall heat transfer area; _αc is the convection heat transfer coefficient; U is the average rotor speed.

The mass leakage submodel is used to obtain the leakage mass m _leak of the gas between adjacent chambers at the top of the rotor. The leakage loss is calculated by the Bernoulli equation. The mass of the subsonic leakage flow when the working chamber pressure is lower than the critical pressure (Formula 14) is obtained by Formula 12, and the mass of the sonic leakage flow when the working chamber pressure is higher than the critical pressure is obtained by Formula 13.

Where: k is the air flow adiabatic coefficient; A _leak is the leakage area; τ is the time; P _cr is the critical pressure; P ₀ is the ambient pressure.

The second step M2: Use experimental data to verify the zero-dimensional simulation model of the hydrogen fuel aviation rotary engine. Based on the zero-dimensional simulation model after experimental verification, simulate and obtain 10,000 sets of indicated power, indicated thermal efficiency, and indicated fuel consumption rate data sets of the aviation rotary engine under different hydrogen fuel injection amounts, different speeds, and different ignition advance angles when the core geometric parameters of the hydrogen fuel rotary engine (shape coefficient K, eccentricity e, cylinder thickness B, _and combustion chamber pit volume V c ) are given.

Please refer to Figures 7 and 8 for the comparison results between the simulated working pressure and working temperature of the zero-dimensional model of the aviation rotary engine and the experiment. The predicted relative error is less than 5%.

Please refer to Figures 9 to 11 for the changing trends of indicated power, indicated thermal efficiency, and indicated fuel consumption rate of a hydrogen fueled aviation rotary engine at different speeds predicted by the zero-dimensional model of the aviation rotary engine.

Please refer to Figures 12 to 14 for the changing trends of indicated power, indicated thermal efficiency and indicated fuel consumption rate of a hydrogen fuel aviation rotary engine at different hydrogen fuel injection amounts predicted by the zero-dimensional model of the aviation rotary engine.

Please refer to Figures 15 to 17 for the changing trends of indicated power, indicated thermal efficiency and indicated fuel consumption rate of a hydrogen fueled aviation rotary engine at different ignition advance angles predicted by the zero-dimensional model of the aviation rotary engine.

Step 3 M3: Construct a hydrogen fuel rotary engine performance neural network model based on the Bayesian regularization algorithm, and use the rotary engine zero-dimensional performance simulation data set to train the neural network model, see Figure 18.

As shown in Figure 18, the neural network model includes 3 input layer network nodes, 20 hidden layer network nodes, and 3 output layer network nodes;

The three input layer network nodes represent the hydrogen fuel injection amount, speed, and ignition advance angle respectively;

The number of 20 hidden layer network nodes is used to characterize the intrinsic logical relationship between input variables and output variables;

The number of network nodes in the three output layers represents indicated power, indicated thermal efficiency, and indicated fuel consumption rate respectively;

The neural network model uses the Bayesian regularization algorithm to train the model. The training data is 80% of the randomly selected 10,000 sets of zero-dimensional performance simulation data sets for training the model and 20% for testing the model. The function corresponding to the neural network model with qualified training, i.e., the determination coefficient ^R2 is not less than 0.98, is derived.

Please refer to Figure 19. The neural network model fitting result is good, and the overall determination coefficient ^R2 is 0.99621.

Step 4 M4: Specify the core geometric parameters of the hydrogen fuel rotary engine (shape coefficient K, eccentricity e, cylinder thickness B, combustion chamber pit volume V _c ), input any combination of hydrogen fuel injection amount, rotation speed and ignition advance angle within a reasonable range, and the neural network model can quickly and accurately predict the indicated power, indicated thermal efficiency and indicated fuel consumption rate of the hydrogen fuel rotary engine.

The shape coefficient K of the hydrogen fuel rotary engine is set to 7, the eccentricity e is 15 mm, the cylinder thickness B is 70 mm, and the volume of the combustion chamber pit V _c is 50 cm ^3. Any combination of hydrogen fuel injection amount 1.141x10 ^-5 kg/cycle, speed 6500 rpm and ignition advance angle 40° are input into the above neural network model. The neural network model can quickly and accurately predict the indicated power of the hydrogen fuel rotary engine as 38.04 kW, indicated thermal efficiency 21.99%, and indicated fuel consumption rate 0.117 kg/(kW·h). The engine performance data predicted by this intelligent prediction method can be used to evaluate the performance degradation of the hydrogen fuel engine by comparing it with the actual test performance data. It can also provide a solid theoretical support for the design of the engine control system and the formulation of the engine optimal control strategy.

Claims (4)

1. An intelligent prediction method for performance of a hydrogen fuel aviation rotor engine is characterized by comprising the following specific steps:

S1, constructing a zero-dimensional performance simulation model of a hydrogen fuel aero-rotor engine based on transient mass conservation, energy conservation, actual gas physical parameters and laminar flame propagation speed in a hydrogen fuel combustion process;

S2, given the core geometric parameters of the hydrogen fuel rotor engine, under the condition of any combination of the hydrogen fuel injection quantity, the rotating speed and the ignition advance angle, the indicated power, the indicated heat efficiency and the indicated fuel consumption rate of a zero-dimensional performance simulation model of the hydrogen fuel aviation rotor engine are obtained, and a rotor engine zero-dimensional performance simulation data set is obtained;

S3, constructing a hydrogen fuel rotor engine performance neural network model based on a Bayesian regularization algorithm, and training the hydrogen fuel rotor engine performance neural network model by adopting a rotor engine zero-dimensional performance simulation data set;

s4, inputting the hydrogen fuel injection quantity, the rotating speed and the ignition advance angle which are arbitrarily combined into a hydrogen fuel rotor engine performance neural network model, and predicting to obtain the indicated power, the indicated heat efficiency and the indicated fuel consumption rate of the hydrogen fuel rotor engine;

In the step S1, a zero-dimensional performance simulation model of the hydrogen fuel aviation rotor engine comprises an air inlet process module, a compression process module, a combustion process module, an expansion process module and an exhaust process module, wherein a geometric submodel, a thermodynamic submodel, a heat exchange loss submodel, a mass leakage submodel and a combustion heat release submodel are embedded in the modules, and the geometric submodel is used for acquiring the volume V of each working chamber in the rotor engine; the thermodynamic submodel is used for obtaining the pressure P of the working chamber through an actual gas physical state equation; the heat exchange loss submodel is used for obtaining the heat convection loss Q _w of the gas in the working chamber of the engine and the wall surface of the engine, and the combustion heat release submodel is used for obtaining the combustion heat release Q _B of the hydrogen fuel according to the laminar flame propagation speed S; the mass leakage sub-model is used for acquiring leakage mass m _leak of gas between adjacent chambers at the top of the rotor;

In step S1, the combustion heat release model predicts the combustion heat release amount Q _B of the hydrogen fuel by using a weber heat release model, and specifically:

Wherein: LHV is the lower heating value of the fuel; Is combustion efficiency; /(I) Is the combustion sustaining angle; /(I)Is the firing angle; m is the combustion quality coefficient, m _fuel intake fuel mass,/>An eccentric shaft rotation angle;

In step S1, the laminar flame propagation speed S in the hydrogen combustion process determines the combustion duration angle The laminar flame propagation speed S under different temperature and pressure conditions is obtained by correcting the laminar flame propagation speed S _ref under the standard conditions, as follows;

Wherein: is a temperature correction coefficient; /(I) Is a pressure correction coefficient; /(I)Is equivalent ratio;

In step S1, combustion duration angle Is calculated according to the formula:

wherein S _des is the flame propagation speed under the rated working condition, Is the combustion continuous angle under the rated working condition;

in step S1, the working chamber pressure P is obtained by using a Benedict-Webb-Rubin actual gas physical state equation, specifically:

Wherein: m _c is the working medium mass of the working chamber; u is the internal energy of the working medium of the cavity; q _B is combustion heat release quantity; m _in intake air mass; h _in is the intake air enthalpy; h _fuel is the intake fuel enthalpy; m _exh is the exhaust mass; h _exh is the exhaust enthalpy; m _leak is the leakage mass; h _leak is the enthalpy value of the leaked working medium; q _w is the heat dissipation between the chamber gas and the cylinder and rotor wall; v _m is the molar volume; r is a gas constant; a ₀, B₀, C₀, a, b, c, , />Are all constant.

2. The intelligent prediction method for performance of a hydrogen-fueled aero-rotor engine according to claim 1, wherein in step S3, the hydrogen-fueled aero-rotor engine performance neural network model includes 3 input layer network node numbers, 20 hidden layer network node numbers, and 3 output layer network node numbers; an input layer x= [ x ₁,x₂,x₃]^T, wherein x ₁、x₂、x₃ represents the hydrogen fuel injection amount, the rotation speed and the ignition advance angle, respectively; output layer y= [ y ₁,y₂,y₃]^T, where y ₁、y₂、y₃ represents indicated power y ₁, indicated heat efficiency y ₂, and indicated fuel consumption y ₃, respectively.

3. The intelligent prediction method for performance of a hydrogen-fueled aircraft rotor engine according to claim 2, wherein in step S3, the hydrogen-fueled aircraft rotor engine performance neural network model is specifically:

Wherein: s ^k is hidden layer node data in the kth iteration, K is an activation function max (0.1 x, x), w, U and v are weight coefficients, and b is the K iteration prediction output layer data And actual output layer data/>Deviation of (2); /(I)Is the normalized input layer; /(I)Is the normalized output layer.

4. The intelligent prediction method for performance of hydrogen-fueled aero-rotor engine according to claim 3, wherein in step S3, the predicted output layer data is providedThe predicted actual output layer node data y= [ Y ₁,y₂,y₃]^T ] after inverse normalization, namely the indicated power Y ₁, the indicated thermal efficiency Y ₂ and the indicated fuel consumption Y ₃ of the hydrogen fuel rotor engine, specifically comprises:

wherein: ps is a mapping of output layer data.