A flow parameter identification method based on array-type artificial hair sensor

Abstract

The present invention discloses a flow parameter identification method based on an array-type artificial hair sensor, the method comprising the following steps: establishing a geometric model of an artificial hair sensor array consisting of a solid calculation domain and a fluid calculation domain; performing common node meshing on the geometric model, and importing the fluid mesh file into the CFD software and the solid mesh file into the finite element software; simulating a series of flow fields under the magnitude and angle of incoming flow velocities in the CFD software; applying the pneumatic pressure on the surface of the artificial hair sensor obtained by CFD simulation as the main load to its finite element model to obtain the corresponding tip displacement; training a BP neural network using the tip displacement as input and the magnitude and angle of the incoming flow velocity as output, and the trained BP neural network can be used to monitor regional incoming flow. The method may also include a verification step. The method of the present invention can monitor regional incoming flow conditions using artificial hair sensors and artificial neural networks.

Description

Technical Field

[0001] The invention belongs to the technical field of flow parameter identification, in particular to a flow parameter identification method based on an array type artificial hair sensor. Background Art

[0002] In aerodynamic experiments, traditional flow direction measurements are mainly carried out by multi-hole probes and hot wire anemometers. The hole and head shape of the probe are easily affected by the processing accuracy, so the calibration process is complicated and time-consuming. Similarly, the disadvantage of the hot wire anemometer is the interference between the support rod and the hot wire, and between the hot wires, which also requires tedious calibration. The emergence of hair sensors provides a new idea for flow direction measurement.

[0003] Hair sensors are micro-sensors that use bionic principles to imitate the micro-filaments on the surface of insects in nature to sense environmental cues such as wind speed and direction. They have the advantages of robustness and fault tolerance, as well as high sensitivity, high bandwidth, low power consumption, small footprint and low integration cost.

[0004] In recent years, many scholars at home and abroad have studied the application of artificial hair sensors in flow parameter identification. Professor Kaman Thapa Magar of the Institute of Wright State University in the United States used the vortex grid method to obtain the local flow variables of the airfoil and used the forward neural network to predict the airfoil angle of attack and incoming flow velocity. However, most research institutions currently rely on experiments and simplified calculation methods, and the design efficiency of array-type hair sensors is low and the accuracy is not high. Summary of the invention

[0005] The purpose of the present invention is to provide a flow parameter identification method based on an array-type artificial hair sensor to address the problems existing in the prior art. The flow parameter identification method identifies flow parameters through fluid-solid coupling numerical simulation calculation and neural network fitting, and has the advantages of high accuracy, robustness, simplicity and speed.

[0006] The purpose of the present invention is to be solved by the following technical solutions:

[0007] A flow parameter identification method based on an array type artificial hair sensor, characterized in that the steps of the identification method are:

[0008] A. Use 3D modeling software to construct a geometric model of an artificial hair sensor array consisting of a solid computational domain and a fluid computational domain;

[0009] B. Perform common-node hybrid meshing on the geometric model, import the fluid mesh file of the fluid calculation domain with a prismatic boundary layer into the CFD software, and import the solid mesh file of the solid calculation domain into the finite element software;

[0010] C. Establish a fluid flow analysis model in CFD software;

[0011] D. Establish a static structural analysis model in finite element software;

[0012] E. Perform unidirectional fluid-solid coupling calculation, set the flow field under the magnitude and angle of the inlet flow velocity in the CFD software, transfer the aerodynamic pressure on the surface of the artificial hair sensor simulated by the CFD software as the main load to the static structural analysis model in the finite element software, calculate the tip displacement of each artificial hair sensor and use it as the tip displacement response of the artificial hair sensor array;

[0013] F. Changing the magnitude and angle of the inlet flow velocity in the boundary conditions of the fluid flow analysis model in step C, repeating step E, and obtaining a series of tip displacement responses of the artificial hair sensor array with different magnitudes and angles of the inlet flow velocity;

[0014] G. Use the tip displacement response obtained in step F as input and the size and angle of the inlet flow velocity corresponding to the tip displacement response as output to train a multi-input and multi-output BP neural network. The trained BP neural network can be used to monitor the regional flow.

[0015] The solid calculation domain in step A is composed of multiple cylinders of the same size in an array type, and the solid calculation domain is controlled by three parameters: the cylinder diameter, the height-to-diameter ratio of the cylinder, and the spacing between adjacent cylinders, wherein the height-to-diameter ratio of the cylinder is greater than 15, and the lateral spacing and longitudinal spacing between the cylinders are kept consistent; the fluid calculation domain in step A is a rectangular cavity minus the solid calculation domain.

[0016] The solid grid in the solid calculation domain in step B and the fluid grid in the fluid calculation domain are generated with common nodes; the solid calculation domain includes at least 10 units along the circumference of the hair sensor and at least 30 units along the axial direction of the hair sensor; the fluid calculation domain is an unstructured mixed grid, and there are at least 15 layers of prismatic grids near the wall to more accurately solve the flow in the boundary layer, and the total height of the prismatic layer is the boundary layer height of the preset incoming flow velocity.

[0017] The fluid flow analysis model in step C needs to set flow equations, space-time discretization methods, boundary conditions, monitoring variables, convergence requirements and iteration times; wherein, flow equations: use the steady incompressible Navier-Stokes equations, pressure-velocity coupling uses SIMPLE format, and gradient calculation uses the node-based Green-Gauss method; space-time discretization method: use the second-order format to spatially discretize the pressure, and use the second-order upwind format to discretize the momentum; boundary conditions: use velocity inlet, pressure outlet, no-slip wall and symmetric boundary types to set the relevant boundaries; monitoring variables: include the root mean square residual value of the mass of all grid cells after discretization and the momentum in three directions, the average pressure value of the artificial hair sensor surface and the pressure value at the top center of the artificial hair sensor; convergence requirements: the residuals of all monitoring variables need to be reduced to 10 -6 And the value of the monitored variable tends to be stable; number of iterations: not less than 3000 iterations.

[0018] The static structural analysis model in step D needs to set the basic unit type, material properties, constraints and load transfer interface of the solid domain; wherein, the basic unit type of the solid domain is an 8-node solid unit, each solid unit has 3 translational degrees of freedom; the material property is isotropic high-strength glass fiber; the constraint condition is to fix the support constraint on the bottom plane of each artificial hair sensor, that is, it cannot rotate or move; the load transfer interface is the side and top planes of the artificial hair sensor.

[0019] During the CFD software simulation calculation in step E: after the residual reaches the specified convergence requirement and the monitoring variable remains stable, it indicates that the calculation is completed; in step E, at the interface of fluid-solid coupling, a one-to-one complete mapping relationship is established between the solid grid unit nodes of the finite element model and the fluid grid unit nodes.

[0020] The tip displacement of the artificial hair sensor in step E refers to the average displacement of the circular plane at the top of the unconstrained end of the artificial hair sensor in the slender cylinder.

[0021] The magnitude of the inlet flow velocity in step E and step F is set to 5-35 m / s, and the angle of the inlet flow velocity is set to 0-27°.

[0022] The BP neural network in step G is a multi-input and multi-output double-layer BP neural network built using a neural network toolbox. The BP neural network includes an input layer unit, a single hidden layer unit and an output layer unit. The transfer function of the hidden layer unit adopts a sigmoid function, and the transfer function of the output layer unit adopts a linear function. The BP neural network includes two processes: signal forward calculation and error back propagation. It is trained using vectorized data form and the Levenberg-Marquardt back propagation algorithm. The input of the BP neural network needs to be normalized, and the mean square error is used during training to measure the degree of consistency between the predicted value and the true value.

[0023] The trained BP neural network in step G needs to be verified, and the verification method is: select several different inlet flow velocity sizes and angles to repeat step CF or select some working conditions obtained in step F to verify the trained BP neural network. When the maximum error of the inlet flow velocity angle prediction does not exceed 6% and the maximum error of the inlet flow velocity size prediction does not exceed 1%, that is, the verification effect meets expectations, the trained BP neural network can be used for monitoring; if the verification effect does not meet expectations, it is necessary to return to step A and repeat steps AG to retrain the BP neural network that can be used for monitoring.

[0024] It is expected that the maximum error in predicting the angle of the inlet flow velocity (or the direction of the inlet flow velocity) does not exceed 6% and the maximum error in predicting the magnitude of the inlet flow velocity does not exceed 1%.

[0025] The technical solution of the present invention uses fluid-solid coupling calculation and artificial neural network to illustrate the use of array-type artificial hair sensors for fluid flow measurement. The fluid-solid coupling calculation method can obtain various behaviors such as the deformation of flexible solids under the action of fluid and the influence of deformation on the flow field. The unidirectional fluid-solid coupling calculation requires relatively few computing resources and is suitable for situations where the structure has small deformation and the amount of structural deformation has little effect on the flow field. Artificial neural network is a computational model composed of a large number of computing nodes (artificial neurons) connected together. It can learn from data and then change the internal structure (weights and thresholds), and can perform complex logical operations and nonlinear relationship realization. This model is widely used in pattern recognition, computer vision, automatic control and other fields due to its adaptive learning ability, high fault tolerance, parallel processing and other characteristics.

[0026] Compared with the prior art, the present invention has the following advantages:

[0027] (1) Convenient and quick implementation: Without the need for complicated wind tunnel experiment calibration, the performance of the artificial hair sensor array for flow parameter identification can be given through numerical simulation and neural network prediction;

[0028] (2) Strong scalability: The artificial hair sensor array can be adjusted according to the actual flow field environment, and its height distribution can also be further designed;

[0029] (3) Robustness and certain fault tolerance: It can give better prediction results when some hair sensors fail or have poor accuracy, and the method has a certain extrapolation capability for untrained working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor.

[0031] Attached Figure 1 is a flow chart of a flow parameter identification method based on an array-type artificial hair sensor of the present invention;

[0032] Attached Figure 2 A schematic diagram of a numerical simulation calculation area of ​​an array type artificial hair sensor used in an embodiment of the present invention;

[0033] Attached Figure 3 A schematic diagram of the structure of a BP neural network used in an embodiment of the present invention;

[0034] Attached Figure 4 A correlation coefficient curve of the BP neural network fitting results of the training set in the embodiment of the present invention;

[0035] Attached Figure 5 A correlation coefficient curve of the BP neural network fitting result for the verification set in the embodiment of the present invention;

[0036] Attached Figure 6 A correlation coefficient curve of the BP neural network fitting result for the test set in the embodiment of the present invention;

[0037] Attached Figure 7 The overall correlation coefficient curve after the BP neural network fitting results are merged for the training set, the validation set and the test set in the embodiment of the present invention;

[0038] Attached Figure 8 This is the verification situation of the BP neural network model under different inlet flow velocities and angles in the embodiment of the present invention. DETAILED DESCRIPTION

[0039] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0040] like Figure 1 The flowchart of a flow parameter identification method based on an array type artificial hair sensor is shown, and the specific steps of the identification method are:

[0041] A. A geometric model of an artificial hair sensor array consisting of a solid computational domain and a fluid computational domain is constructed using 3D modeling software; the solid computational domain in the geometric model is composed of multiple cylinders of the same size in an array type, and the solid computational domain is controlled by three parameters: the cylinder diameter, the height-to-diameter ratio of the cylinder, and the spacing between adjacent cylinders. The height-to-diameter ratio of the cylinder is greater than 15, and the lateral spacing and longitudinal spacing between cylinders are consistent; the fluid computational domain is a rectangular cavity minus the solid computational domain;

[0042] B. Perform common-node hybrid meshing on the geometric model, import the fluid mesh file of the fluid calculation domain with a prismatic boundary layer into the CFD software, and import the solid mesh file of the solid calculation domain into the finite element software;

[0043] C. Establish a fluid flow analysis model in the CFD software and set the flow equation, time and space discretization method, boundary conditions, monitoring variables, convergence requirements and number of iterations;

[0044] D. Establish a static structural analysis model in the finite element software and set the basic unit type, material properties, constraints and load transfer interface of the solid domain;

[0045] E. Perform unidirectional fluid-solid coupling calculation, set the flow field under the size and angle of the inlet flow velocity in the CFD software (the size of the inlet flow velocity is set to 5-35 m / s, and the angle of the inlet flow velocity is set to 0-27°), and the calculation is completed after the residual reaches the specified convergence requirement and the monitoring variable remains stable; transfer the aerodynamic pressure on the surface of the artificial hair sensor simulated by the CFD software as the main load to the static structural analysis model in the finite element software, calculate the tip displacement of each artificial hair sensor and use it as the tip displacement response of the artificial hair sensor array, the tip displacement of the artificial hair sensor refers to the average displacement of the circular plane at the top of the unconstrained end of the artificial hair sensor of the slender cylinder;

[0046] F. Changing the magnitude and angle of the inlet flow velocity in the boundary conditions of the fluid flow analysis model in step C (the magnitude of the inlet flow velocity is set to 5-35 m / s, and the angle of the inlet flow velocity is set to 0-27°), repeating step E, and obtaining a series of tip displacement responses of the artificial hair sensor array with different magnitudes and angles of the inlet flow velocity;

[0047] G. A multi-input and multi-output double-layer BP neural network built using a neural network toolbox, wherein the BP neural network includes an input layer unit, a single hidden layer unit and an output layer unit. The transfer function of the hidden layer unit adopts a sigmoid function, and the transfer function of the output layer unit adopts a linear function. The BP neural network includes two processes: signal forward calculation and error back propagation. It is trained using vectorized data form and the Levenberg-Marquardt back propagation algorithm. The input of the BP neural network needs to be normalized, and the mean square error is used during training to measure the degree of agreement between the predicted value and the true value. The tip displacement response obtained in step F is used as input, and the size and angle of the inlet flow velocity corresponding to the tip displacement response are used as output to train the multi-input and multi-output BP neural network. The trained BP neural network can be used to monitor regional flow.

[0048] The trained BP neural network needs to be verified. The verification method is: select several different inlet flow velocity sizes and angles to repeat steps CF or select some working conditions obtained in step F to verify the trained BP neural network. When the maximum error of the inlet flow velocity angle prediction does not exceed 6% and the maximum error of the inlet flow velocity size prediction does not exceed 1%, that is, the verification effect meets expectations, the trained BP neural network can be used for monitoring; if the verification effect does not meet expectations, it is necessary to return to step A and repeat steps AG to retrain the BP neural network that can be used for monitoring.

[0049] Example

[0050] like Figure 1 The method flow chart shown is a flow parameter identification method based on an array type artificial hair sensor, the method specifically comprises the following steps:

[0051] A. Use 3D modeling software to build a simplified geometric model of the artificial hair sensor array. The geometric model consists of a solid calculation domain and a fluid calculation domain. Figure 2, this embodiment demonstrates a 3*3 form of artificial hair sensor array, the geometric parameters of the fluid calculation domain and the solid calculation domain of this embodiment are: diameter D is 9μm, length L is 25D, spacing Δ is 0.5L, entrance distance L1 is 7L, exit distance L2 is 11L, side distance L3 is 8L, top distance L4 is 2L, wherein the coordinate origin coincides with the bottom center of the artificial hair sensor at the center of the artificial hair sensor array, D is the diameter of each cylindrical artificial hair sensor, L is the length of each cylindrical artificial hair sensor, Δ is the lateral spacing and longitudinal spacing of each artificial hair sensor, L1 is the vertical distance from the entrance of the fluid calculation domain to the coordinate origin, L2 is the vertical distance from the exit of the fluid calculation domain to the coordinate origin, L3 is the vertical distance from the side wall of the fluid calculation domain to the coordinate origin, and L4 is the vertical distance from the top of the fluid calculation domain to the coordinate origin;

[0052] B. Grid the geometric model, and use an unstructured hybrid grid with 15 layers of prism grids to discretize the fluid calculation domain to more accurately solve the flow in the boundary layer. The total height of the prism layer is the boundary layer height of the preset incoming flow velocity type; 12 units are arranged circumferentially and 40 units are arranged axially in the cylindrical artificial hair sensor. The internal solid grid of the artificial hair sensor is generated by the sweeping method, and the fluid grid file and the solid grid file are imported into the CFD software and the finite element software respectively;

[0053] C. Establish a fluid flow analysis model in the CFD software. Due to the low flow rate, the air compressibility is not considered to simplify the calculation, and the air density is kept at 1.225kg / m 3 , the dynamic viscosity coefficient is 1.7894×10 -5 N·s / m 2 , so the steady incompressible Navier-Stokes equations are solved here, the pressure-velocity coupling adopts SIMPLE format, and the gradient calculation adopts the node-based Green-Gauss method; the pressure is spatially discretized using the second-order format, and the momentum is discretized using the second-order upwind format; Figure 2 The "entrance" in the figure is the velocity entrance, and the velocity magnitude at infinity is U ∞ , the direction of the velocity is α, and the velocity formula used is: Where U(z) is the velocity distribution along the vertical direction of the ground, L is the length of the artificial hair sensor, z is the height of the artificial hair sensor from the ground, and e xis the direction vector of the X-axis; the “outlet” is the pressure outlet, the static pressure is 0, the “ground” and the hair surface are no-slip walls, the “side wall” and “top surface” boundaries are set as symmetric boundary conditions; in addition to the root mean square residual values ​​of the mass and momentum in three directions of all grid cells after discretization, the average pressure value on the surface of the additional artificial hair sensor and the pressure value at the center point of the sensor top are the monitoring variables, and the residuals of all monitoring values ​​need to be reduced to 10 -6 Or the monitored value is stable and reaches the specified 3000 iterations;

[0054] D. A static structural analysis model was established in the finite element software. The basic unit type of the solid domain used in the static analysis was an 8-node solid unit, and each solid unit had 3 translational degrees of freedom. The solid unit was given the material properties of high-strength glass fiber, i.e., the elastic modulus was 90,000 MPa and the Poisson's ratio was 0.22. The bottom plane of each artificial hair sensor was set as a fixed support constraint, i.e., it could not rotate or move. The side and top planes of the artificial hair sensor were set as pressure load transfer interfaces, and the average displacement value of the unconstrained tip plane was monitored.

[0055] E. Perform unidirectional fluid-solid coupling calculation. First, start CFD calculation. After the residual reaches the specified convergence requirement and the variables of the monitoring points remain stable, transfer the aerodynamic pressure on the surface of the artificial hair sensor obtained by CFD simulation as the main load to the static structural analysis model in the finite element software. Finally, the tip displacement of each artificial hair sensor is obtained by static structural calculation and used as the tip displacement response of the artificial hair sensor array.

[0056] F. The preset measurement conditions are: flow velocity 5-35 m / s, flow direction 0-27°, uniform sampling method, velocity sampling interval 2 m / s, flow direction sampling interval 3°, and a total of 160 cases need to be calculated; change the size and angle of the inlet flow velocity in step C, and repeat step E until all the tip displacement responses required for training are obtained;

[0057] G. The established BP neural network structure is shown in Figure 3 The input layer contains 9 neurons, the hidden layer contains 15 neurons, and the output layer contains 2 neurons. The transfer function of the hidden layer unit uses the sigmoid function, and the transfer function of the output layer unit uses the linear function. The Levenberg-Marquardt back-propagation algorithm is used for training. The mean square error is used in the calculation to measure the performance of the neural network. The training set, validation set, and test set are randomly divided, and the sample size ratio is 8:1:1. The maximum number of epochs is set to 1000, and the weights and thresholds are initialized.

[0058] H. The tip displacement of 9 artificial hair sensors is used as input, and the magnitude and angle of the inlet flow velocity are used as output to train the BP neural network. After the training, the overall mean square error value is 0.0055. At this time, the correlation coefficient R between the actual value and the predicted value of the training set, validation set, test set and the overall set is shown in Figure 2. Figure 4-Figure 7 , Figure 4-Figure 7 Data is the data point predicted by the neural network, Fit is the regression line, and Y=T represents the straight line where the predicted value is consistent with the target value.

[0059] In addition, different inlet flow velocities and inlet flow angles were used to verify the effectiveness of this method. The inlet flow velocities of 12m / s, 20m / s, 28m / s, and 36m / s and the inlet flow angles of 7°, 14°, 20°, and 28° were selected for verification. The results are shown in Table 1. Figure 8 ; Since the working conditions adopted did not participate in the construction of the neural network model, compared with the calculated value, the maximum relative error of the inlet flow velocity angle prediction was 5.7%, and the maximum relative error of the inlet flow velocity size prediction was 0.54%, which was in line with expectations and proved the effectiveness of this method. If the error of the verification result is unacceptable, it is necessary to return to the above steps and make adjustments in the calculation area, grid division, fluid and structure solution algorithm, neural network structure, etc. until the effectiveness verification is passed, and it can be used for fluid flow velocity measurement within the target working condition range.

[0060] The above embodiments are only for illustrating the technical idea of ​​the present invention, and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the present invention; any technology not involved in the present invention can be realized by existing technologies.

Claims

1. A flow parameter identification method based on an array type artificial hair sensor, characterized in that: The steps of the identification method are: A. Use 3D modeling software to construct a geometric model of an artificial hair sensor array consisting of a solid computational domain and a fluid computational domain; B. Perform common-node hybrid meshing on the geometric model, import the fluid mesh file of the fluid calculation domain with a prismatic boundary layer into the CFD software, and import the solid mesh file of the solid calculation domain into the finite element software; C. Establish a fluid flow analysis model in CFD software; D. Establish a static structural analysis model in finite element software; E. Perform unidirectional fluid-solid coupling calculation, set the flow field under the magnitude and angle of the inlet flow velocity in the CFD software, transfer the aerodynamic pressure on the surface of the artificial hair sensor simulated by the CFD software as the main load to the static structural analysis model in the finite element software, calculate the tip displacement of each artificial hair sensor and use it as the tip displacement response of the artificial hair sensor array; F. Changing the magnitude and angle of the inlet flow velocity in the boundary conditions of the fluid flow analysis model in step C, repeating step E, and obtaining a series of tip displacement responses of the artificial hair sensor array with different magnitudes and angles of the inlet flow velocity; G. Using the tip displacement response obtained in step F as input and the size and angle of the inlet flow velocity corresponding to the tip displacement response as output to train a multi-input multi-output BP neural network, the trained BP neural network can be used to monitor the regional flow; The fluid flow analysis model in step C requires setting flow equations, time and space discretization methods, boundary conditions, monitoring variables, convergence requirements and iteration times; wherein, Flow equations: The steady incompressible Navier-Stokes equations are used, the pressure-velocity coupling adopts the SIMPLE format, and the gradient calculation adopts the node-based Green-Gauss method; Space-time discretization method: use the second-order format to discretize the pressure in space, and use the second-order upwind format to discretize the momentum; Boundary conditions: Use velocity inlet, pressure outlet, no-slip wall and symmetry boundary types to set relevant boundaries; Monitoring variables: including the root mean square residual value of the mass and momentum of all grid cells in three directions after discretization, the average pressure on the surface of the artificial hair sensor, and the pressure value at the center point of the top of the artificial hair sensor; Convergence requirement: The residuals of all monitored variables need to drop to 10 -6 And the values ​​of the monitored variables tend to be stable; Number of iterations: no less than 3000 iterations; The static structural analysis model in step D needs to set the basic unit type, material properties, constraints and load transfer interface of the solid domain; wherein, The basic unit type of the solid domain is an 8-node solid unit, each of which has 3 translational degrees of freedom. Material properties: isotropic high-strength glass fiber; Constraint conditions: The bottom plane of each artificial hair sensor is constrained to be fixed and supported, that is, it cannot rotate or move; Load transfer interface: the side and top planes of the artificial hair sensor.

2. The flow parameter identification method based on array-type artificial hair sensor according to claim 1, characterized in that: The solid calculation domain in step A is composed of multiple cylinders of the same size in an array type, and the solid calculation domain is controlled by three parameters: the cylinder diameter, the height-to-diameter ratio of the cylinder, and the spacing between adjacent cylinders, wherein the height-to-diameter ratio of the cylinder is greater than 15, and the lateral spacing and longitudinal spacing between the cylinders are kept consistent; the fluid calculation domain in step A is a rectangular cavity minus the solid calculation domain.

3. The flow parameter identification method based on array-type artificial hair sensor according to claim 1 or 2, characterized in that: The solid grid in the solid calculation domain in step B and the fluid grid in the fluid calculation domain are generated with common nodes; the solid calculation domain includes at least 10 units along the circumference of the hair sensor and at least 30 units along the axial direction of the hair sensor; the fluid calculation domain is an unstructured mixed grid, and there are at least 15 layers of prismatic grids near the wall to more accurately solve the flow in the boundary layer, and the total height of the prismatic layer is the boundary layer height of the preset incoming flow velocity.

4. The flow parameter identification method based on array-type artificial hair sensor according to claim 1, characterized in that: During the CFD software simulation calculation in step E: after the residual reaches the specified convergence requirement, it indicates that the calculation is finished; in step E, at the interface of fluid-solid coupling, a one-to-one complete mapping relationship is established between the solid grid unit nodes of the finite element model and the fluid grid unit nodes.

5. The flow parameter identification method based on array-type artificial hair sensor according to claim 1, characterized in that: The tip displacement of the artificial hair sensor in step E refers to the average displacement of the circular plane at the top of the unconstrained end of the artificial hair sensor in the slender cylinder.

6. The flow parameter identification method based on array-type artificial hair sensor according to claim 1, characterized in that: The magnitude of the inlet flow velocity in step E and step F is set to 5~35m / s , the angle of the inlet flow velocity is set to 0～27° .

7. The flow parameter identification method based on array-type artificial hair sensor according to claim 1, characterized in that: The BP neural network in step G is a multi-input and multi-output double-layer BP neural network built using a neural network toolbox. The BP neural network includes an input layer unit, a single hidden layer unit and an output layer unit. The transfer function of the hidden layer unit adopts a sigmoid function, and the transfer function of the output layer unit adopts a linear function. The BP neural network includes two processes: signal forward calculation and error back propagation. It is trained using vectorized data form and the Levenberg-Marquardt back propagation algorithm. The input of the BP neural network needs to be normalized, and the mean square error is used during training to measure the degree of consistency between the predicted value and the true value.

8. The flow parameter identification method based on array-type artificial hair sensor according to claim 1 or 7, characterized in that: The trained BP neural network in step G needs to be verified, and the verification method is: select several different inlet flow velocity sizes and angles to repeat step CF or select some working conditions obtained in step F to verify the trained BP neural network. When the maximum error of the inlet flow velocity angle prediction does not exceed 6% and the maximum error of the inlet flow velocity size prediction does not exceed 1%, that is, the verification effect meets expectations, the trained BP neural network can be used for monitoring; if the verification effect does not meet expectations, it is necessary to return to step A and repeat steps AG to retrain the BP neural network that can be used for monitoring.

Images