An internal wind direction monitoring system and method for a combine harvester cleaning device

Abstract

The application provides a combine harvester cleaning device internal wind direction monitoring system and method, comprising a prediction system and a cleaning wind direction monitoring device; the cleaning wind direction monitoring device comprises a supporting mechanism and a plurality of air pressure sensor chips; the plurality of air pressure sensor chips are installed above a fish scale sieve through the supporting mechanism; the air pressure sensor chips are connected with the prediction system, the air pressure sensor chips are used for detecting air flow pressure information above the fish scale sieve and transmitting the air flow pressure information to the prediction system; a neural network model is arranged in the prediction system, the neural network model takes the air flow pressure information as input and takes combine harvester cleaning device internal wind speed and wind direction angle prediction values as output. The application can accurately reflect the change of the combine harvester cleaning device internal wind direction, provide prediction information for a combine harvester control system, reduce the grain loss rate, reduce subsequent control lag, and improve the cleaning performance.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of air flow monitoring of a combine harvester cleaning device, and particularly relates to a combine harvester cleaning device internal wind direction monitoring system and method. BACKGROUND

[0002] The cleaning device is an important component of the combine harvester, and its operation quality directly affects the operation performance of the combine harvester. The existing cleaning device structure and motion parameters can only be adjusted in stages by stopping, manually, and according to experience, and there is a lack of cleaning device operation state monitoring and working parameter control system, which cannot adaptively adjust the working parameters of the cleaning device according to the changes in the environment to ensure the cleaning performance, and the harvesting adaptability is poor. However, the air flow speed and direction in the cleaning chamber are the main factors that determine the cleaning effect. Accurate monitoring of the air flow speed and direction changes at the key point position is a prerequisite for obtaining good cleaning performance. At the present stage, the CFD numerical simulation and wind speed measurement method are used to study the air flow field in the cleaning chamber, but the air flow direction under the existing technology cannot be verified, and the existing research on the cleaning device has not yet had a method for monitoring the air flow speed and direction in the cleaning chamber. Moreover, the existing neural network is used to predict the cleaning air flow parameters, which is mainly through the traditional BP neural network using working parameters such as fish scale sieve opening, fan speed and air distribution plate angle for prediction, but there are problems such as initial weight sensitivity and long training time. SUMMARY

[0003] In view of the above technical problems, the present application provides a combine harvester cleaning device internal wind direction monitoring system and method, which can accurately reflect the changes of the internal wind direction of the combine harvester cleaning device, provide prediction information for the combine harvester control system, reduce the grain loss rate, reduce the subsequent control lag, and improve the cleaning performance.

[0004] Note that the description of these objects does not hinder the existence of other objects. One embodiment of the present application does not need to achieve all the above-mentioned objects. The objects other than the above-mentioned objects can be extracted from the description, drawings, and claims.

[0005] The present application is achieved by the following technical means.

[0006] A combine harvester cleaning device internal wind direction monitoring system, comprising a prediction system and a cleaning wind direction monitoring device;

[0007] The cleaning wind direction monitoring device comprises a supporting mechanism and a plurality of air pressure sensor chips; the plurality of air pressure sensor chips are installed above the fish scale sieve through the supporting mechanism; the air pressure sensor chips are connected with a prediction system, and are used to detect air flow pressure information above the fish scale sieve and transmit the information to the prediction system; the prediction system is provided with a neural network model, and the neural network model takes the air flow pressure information as input and takes the joint harvester cleaning device internal wind speed and wind direction angle prediction value as output.

[0008] In the above scheme, the supporting mechanism comprises a supporting frame, a supporting plate, a supporting rod and a mounting plate

[0009] One side of the supporting frame is provided with a supporting plate, the supporting plate is provided with a supporting rod, the supporting rod is provided with a mounting plate, and the air pressure sensor chips are arranged on the mounting plate; the other side of the supporting frame is installed above the fish scale sieve.

[0010] In the above scheme, the air pressure sensor chip is a MEMS air pressure sensor chip.

[0011] In the above scheme, the prediction system comprises an STM32 single-chip microcomputer and a prediction unit, the STM32 single-chip microcomputer is connected with the air pressure sensor chip, is used to collect data of the air pressure sensor chip and transmit the data to the prediction unit of the prediction system, and the prediction unit is provided with a neural network model.

[0012] In the above scheme, the neural network model is a BP neural network comprising three hidden layers and an output layer, and residual networks are used to connect between each layer, and the forward transmission formula of each residual block of the residual network is represented as:

[0013] y = LeakyReLU (W2 * LeakyReLU (W1x + b1) + b2 + x)

[0014] Wherein x is input, y is output, W1 is the first layer hidden layer weight, W2 is the second layer hidden layer weight, b1 is the first layer hidden layer bias, b2 is the second layer hidden layer bias, and an initial value can be set according to experience.

[0015] In the above scheme, the neural network model takes the standardized relative air pressure value data as the input of the BP neural network model for training.

[0016] The calculation formula of the standardized relative air pressure value Δp' m is as follows:

[0017]

[0018] Wherein, Δp mμ i is the mean value of the data collected by the i-th pressure sensor chip, δ i is the standard deviation of the data collected by the i-th pressure sensor chip.

[0019] A monitoring method of the internal wind direction monitoring system of the combine harvester cleaning device, comprising the following steps:

[0020] The trained neural network model is loaded into the prediction system.

[0021] The cleaning wind direction monitoring device is installed at a predetermined position inside the cleaning device. Multiple pressure sensor chips are installed above the scale sieve through a support mechanism. The pressure sensor chips detect the airflow pressure information above the scale sieve and transmit it to the prediction system.

[0022] The neural network model in the prediction system takes the airflow pressure information as input and outputs the predicted values of the internal wind speed and wind direction angle of the combine harvester cleaning device.

[0023] In the above scheme, the neural network model is trained by the following steps:

[0024] Data collection: Collect multiple sets of horizontal wind direction angle and pitch wind direction angle under different wind speeds through calibrated pressure sensor chips, collect multiple sets of pressure data, and standardize the collected pressure data.

[0025] Building a neural network model: Based on the PyTorch framework and PyCharm Community Edition software, a BP neural network is built using Python language. The BP neural network includes three hidden layers and an output layer, and residual networks are used to connect between each layer. The activation function selects the variant function Leaky ReLU of ReLU, and the function optimizer selects the Adamp optimizer.

[0026] Neural network training: The standardized data in the data collection step is used as the input of the BP neural network model, and the BP neural network model is trained.

[0027] The trained BP neural network model is loaded into the prediction system to predict the newly collected data of the pressure sensor chip, and the predicted values of the internal wind speed and wind direction angle of the combine harvester cleaning device are obtained.

[0028] The scheme further includes the following steps: comparing the predicted value of the prediction system with the calculation results of the relative air pressure and wind speed relationship model and the relative air pressure and wind direction angle relationship model to verify the accuracy of the prediction system.

[0029] Further, the relative air pressure and wind speed relationship model is:

[0030]

[0031] In the formula, C A is a pressure value, p is air density, V is wind speed, and S is the cross-sectional area of the air pressure sensor chip.

[0032] The relative air pressure and wind direction angle relationship model is:

[0033] f (θ) = a0 + a1 × cos (ω × θ) + b1 × sin (ω × θ)

[0034] ω = (2π) / T

[0035] where θ is the wind direction angle, ω is the frequency of the curve, a0 is the amplitude of the direct current component, a1 is the amplitude of the sine component, b1 is the amplitude of the cosine component, and T is the period.

[0036] Compared with the prior art, the present application has the following advantages:

[0037] 1. The present application uses an air pressure sensor chip to monitor the wind direction inside the cleaning device of a combine harvester, which can output wind speed and wind direction angle, has low cost and small device size, is suitable for multiple-point arrangement inside the cleaning device of a combine harvester, and can also adapt to the harsh environment of high humidity and high dust inside the combine harvester.

[0038] 2. The present application solves the problem that the existing cleaning device of a combine harvester mainly uses numerical simulation and bench test to establish the best operation combination parameters, and less research is done on the internal airflow of the cleaning device during the harvesting process. The present application can directly monitor the wind speed and wind direction angle in real time during the harvesting process, which is beneficial to the accurate regulation and control of the working parameters of the cleaning device according to the wind speed and wind direction of the monitoring points, and can significantly improve the cleaning performance of the combine harvester.

[0039] Note that the description of these effects does not hinder the existence of other effects. One embodiment of the present application does not necessarily have all the above-mentioned effects. Effects other than the above-mentioned effects can be clearly seen and extracted from the description, drawings, claims, etc. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 Figure 1 is a schematic diagram of the installation of the combine harvester cleaning wind direction monitoring device according to an embodiment of the present application.

[0041] Figure 2 The figure is a perspective view of the combined harvester cleaning wind direction monitoring device according to an embodiment of the present application;

[0042] Figure 3 The figure is a perspective view of the combined harvester cleaning wind direction monitoring device according to an embodiment of the present application;

[0043] Figure 4 The figure is a perspective view of the combined harvester cleaning wind direction monitoring device according to an embodiment of the present application;

[0044] Figure 5 The figure is a perspective view of the combined harvester cleaning wind direction monitoring device according to an embodiment of the present application;

[0045] Figure 6 The figure is a perspective view of the combined harvester cleaning wind direction monitoring device according to an embodiment of the present application;

[0046] Figure 7 The figure is a perspective view of the combined harvester cleaning wind direction monitoring device according to an embodiment of the present application.

[0047] In the figure, 1-prediction system, 2-cleaning wind direction monitoring device, 3-fish scale sieve, 4-support frame, 5-support plate, 6-support rod, 7-mounting plate, 8-air pressure sensor chip, 9-mounting hole, 10-cleaning room outer wall, 3-1-cleaning wind direction monitoring device A, 3-2-cleaning wind direction monitoring device B, 3-3-cleaning wind direction monitoring device C, 3-4-cleaning wind direction monitoring device D. DETAILED DESCRIPTION

[0048] Embodiments of the present application are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by reference to the accompanying drawings are exemplary and are intended to explain the present application, and cannot be understood as limiting the present application.

[0049] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "front," "rear," "left," "right," "upper," "lower," "axial," "radial," "vertical," "horizontal," "inner," and "outer," etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0050] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0051] like Figure 1 As shown, a preferred embodiment of the wind direction monitoring system inside the combine harvester cleaning device of the present invention is provided. The wind direction monitoring system inside the combine harvester cleaning device includes a prediction system 1 and a cleaning wind direction monitoring device 2. The cleaning wind direction monitoring device 2 includes a support mechanism and multiple air pressure sensor chips 8. The multiple air pressure sensor chips 8 are mounted above the fish scale screen 3 through the support mechanism. The air pressure sensor chips 8 are connected to the prediction system 1 and are used to detect the airflow pressure information above the fish scale screen 3 and transmit it to the prediction system 1. The prediction system 1 is equipped with a neural network model. The neural network model takes the airflow pressure information as input and the predicted values ​​of the wind speed and wind direction angle inside the combine harvester cleaning device as output.

[0052] In one specific embodiment of the present invention, the prediction system 1 is located inside the cab of the combine harvester, and the cleaning wind direction monitoring device 2 is located above the fish scale screen 3 inside the cleaning device. The cleaning wind direction monitoring device 2 transmits data with the prediction system 1 wirelessly to collect wind direction monitoring information inside the cleaning device. The prediction system 1 is used to provide the driver with real-time information on the airflow speed and wind direction angle inside the combine harvester's cleaning device.

[0053] like Figures 2-3 As shown, the support mechanism includes a support frame 4, a support plate 5, a support rod 6, and a mounting plate 7. The support frame 4 has a support plate 5 on one side, a support rod 6 on the support plate 5, and a mounting plate 7 on the support rod 6. The air pressure sensor chip 8 is arranged on the mounting plate 7. The other side of the support frame 4 is installed above the fish scale screen 3.

[0054] In one specific embodiment of the present invention, the other side of the support frame 4 is provided with a mounting hole 9, which is installed on the outer wall 10 of the cleaning chamber through the mounting hole 9 and located above the fish scale screen 3.

[0055] In one specific embodiment of the present invention, the number of cleaning wind direction monitoring devices 2 is four, namely cleaning wind direction monitoring device A3-1, cleaning wind direction monitoring device B3-2, cleaning wind direction monitoring device C3-3, and cleaning wind direction monitoring device D3-4; the cleaning wind direction monitoring devices A3-1, B3-2, C3-3, and D3-4 form an array of barometric pressure sensors.

[0056] Preferably, the air pressure sensor chip 8 is disposed on the four sides of the mounting plate 7; the four air pressure sensor chips 8 cooperate with each other to collect data and determine the airflow speed and airflow direction angle.

[0057] Preferably, the support frame 4 has a through groove at its center, and the bottom end of the support rod 6 passes through the support plate 5 and the through groove in sequence. The through groove can move within the through groove, thereby changing the monitoring position of the air pressure sensor chip 8. Preferably, the support frame 4 has a 20mm through groove at its center.

[0058] In a specific embodiment of the present invention, preferably, the cleaning wind direction monitoring device A3-1, cleaning wind direction monitoring device B3-2, cleaning wind direction monitoring device C3-3, and cleaning wind direction monitoring device D3-4 are specifically configured according to... Figure 3 The cleaning airflow monitoring devices A3-1 and B3-2 are positioned 510mm apart, located at the front and rear of one side of the fish scale screen 3, respectively; cleaning airflow monitoring devices C3-3 and D3-4 are located at the front and rear of the other side of the fish scale screen 3, respectively; cleaning airflow monitoring devices C3-3 and D3-4 are symmetrically arranged with cleaning airflow monitoring devices A3-1 and B3-2 along the center line of the fish scale screen 3. The cleaning airflow monitoring devices A3-1, B3-2, C3-3, and D3-4 are mounted on the outer wall 10 of the cleaning chamber using two M5 bolts through mounting holes 9.

[0059] In an embodiment of the present application, preferably, the air pressure sensor chip 8 is a MEMS air pressure sensor chip.

[0060] In an embodiment of the present application, preferably, the prediction system 1 comprises an STM32 single-chip microcomputer and a prediction unit, the STM32 single-chip microcomputer is connected with the air pressure sensor chip 8, used for collecting data of the air pressure sensor chip 8 and transmitting to the prediction unit of the prediction system, the prediction unit is provided with a neural network model, which can predict the data transmitted by the STM32 single-chip microcomputer and display the results.

[0061] As shown in Figure 5 , preferably, the neural network model is a BP neural network comprising three hidden layers and an output layer, and a residual network is used to connect between each layer, and the forward transmission formula of each residual block of the residual network is represented as:

[0062] y = LeakyReLU (W2 * LeakyReLU (W1x + b1) + b2 + x)

[0063] Wherein x is the input, y is the output, W1 is the weight of the first hidden layer, W2 is the weight of the second hidden layer, b1 is the bias of the first hidden layer, and b2 is the bias of the first hidden layer, which can be set to an initial value according to experience.

[0064] The neural network model takes the standardized relative air pressure value data as the input of the BP neural network model for training.

[0065] The calculation formula of the standardized relative air pressure value Δp' m is:

[0066]

[0067] Wherein, Δp m is the m relative air pressure values selected from the n relative air pressure values ΔPn, ΔPn is the relative air pressure value obtained by subtracting the reference air pressure from the value of the air pressure sensor chip 8 at the point, μ i is the mean value of the data collected by the i-th air pressure sensor chip 8, and δ i is the standard deviation of the data collected by the i-th air pressure sensor chip 8.

[0068] A monitoring method of the joint harvester cleaning device internal wind direction monitoring system, comprising the following steps:

[0069] The trained neural network model is carried to the prediction system 1;

[0070] The cleaning wind direction monitoring device 2 is installed at a predetermined position inside the cleaning device, and a plurality of air pressure sensor chips 8 are installed above the fish scale sieve 3 through a support mechanism. The air pressure sensor chip 8 detects the air flow pressure information above the fish scale sieve 3 and transmits it to the prediction system 1.

[0071] The neural network model in the prediction system 1 takes the air flow pressure information as input and outputs the predicted values of the wind speed and the wind direction angle inside the combine harvester cleaning device.

[0072] The neural network model is trained by the following steps:

[0073] Data collection: After calibration, the air pressure sensor chip 8 collects multiple sets of horizontal wind direction angle and pitch wind direction angle test under different wind speed, collects multiple sets of air pressure data, and standardizes the collected air pressure data;

[0074] Building a neural network model: based on the PyTorch framework and PyCharm Community Edition software, a BP neural network is built using Python language. The BP neural network includes three hidden layers and an output layer, and a residual network is used to connect between each layer. The activation function selects the variant function Leaky ReLU of ReLU, and the function optimizer selects the Adamp optimizer;

[0075] Neural network training: the standardized data in the data collection step is used as the input of the BP neural network model, and the BP neural network model is trained;

[0076] The trained BP neural network model is loaded into the prediction system 1 for data prediction of the newly collected data of the air pressure sensor chip 8, and the predicted values of the wind speed and the wind direction angle inside the combine harvester cleaning device are obtained.

[0077] The monitoring method of the joint harvester cleaning device internal wind direction monitoring system also includes the following steps: comparing the predicted values of the prediction system 1 with the calculation results of the relative air pressure and wind speed relationship model and the relative air pressure and wind direction angle relationship model to verify the accuracy of the prediction system.

[0078] The relative air pressure and wind speed relationship model is:

[0079]

[0080] In the formula, C $ is the pressure value, p is the air density, V is the wind speed, and S is the cross-sectional area of the air pressure sensor chip 8;

[0081] The relative air pressure and wind direction angle relationship model is:

[0082] f(θ)=a0+a1×cos(ω×θ)+b1×sin(ω×θ)

[0083] ω=(2π) / T

[0084] Where θ is the wind direction angle, ω is the frequency of the curve, a0 is the amplitude of the DC component, a1 is the amplitude of the sine component, b1 is the amplitude of the cosine component, and T is the period.

[0085] The calculation of wind speed ν and wind direction angle θ using the relative air pressure-wind speed relationship model and the relative air pressure-wind direction angle relationship model specifically includes the following steps:

[0086] Step S1: Under laboratory conditions, test the cleaning wind direction monitoring device 2 at different wind speeds and angles to obtain relative air pressure values. Continuously read the air pressure values ​​using the cleaning wind direction monitoring device 2 in the absence of wind. Average the read air pressure values ​​and record them as the reference air pressure value, designated P0. Obtain the air pressure data from the air pressure sensor chip 8 in the wind field and record it as P1 to P2. n (n is the number of barometric pressure sensor chips 8); the relative pressure value is obtained by subtracting the value of the barometric pressure sensor chip 8 from the reference pressure, and denoted as ΔP. n ;

[0087] According to Bernoulli's theorem:

[0088] The model for the relationship between relative air pressure and wind speed at a point in the windward area, when the surface pressure is along the direction perpendicular to the wind speed (i.e., when the wind direction angle θ = 0), is as follows:

[0089]

[0090] In the formula, P n The pressure value collected by the barometric pressure sensor chip 8 is denoted by ΔPn, where ΔPn is the pressure difference at that point, ρ is the air density, V is the wind speed, and S is the cross-sectional area of ​​the barometric pressure sensor chip 8, i.e., the cross-sectional area of ​​the surface at the windward point. The formula shows that in the vertical direction, the relative air pressure and wind speed have a square relationship. This relationship can be used to derive the wind speed V from the air pressure, and the wind direction angle θ = 0.

[0091] When the wind direction angle θ is not equal to 0, the relationship model between relative air pressure and wind speed is as follows:

[0092]

[0093] ε=1-4sin 2 θ

[0094] ε is the pressure coefficient when the wind direction angle θ is not equal to 0.

[0095] like Figure 4The air flow wind direction angle collection method is shown: the wind direction is always from the front of the cleaning screen to the tail of the cleaning screen for the cleaning screen of the combine harvester cleaning device, so it is only necessary to divide the wind direction angle range of the horizontal windward surface 180 degrees and the pitch windward surface 90 degrees. In the horizontal windward surface 180 degrees and the pitch windward surface 90 degrees, the relative air pressure and the wind direction angle output by the sensor present a strong functional relationship, and according to the experimental results, a first-order Fourier function is used to fit the curve to obtain the air pressure and wind direction angle relationship model as follows:

[0096] f(θ)=a0+a1×cos(ω×θ)+b1×sin(ω×θ)

[0097] ω=(2π) / T

[0098] Wherein, θ is the wind direction angle, ω is the frequency of the curve, a0 is the amplitude of the direct current component, a1 is the amplitude of the sine component, b1 is the amplitude of the cosine component, and T is the period.

[0099] Relative air pressure and wind direction angle relationship modeling: under the premise of different wind speeds, the relative air pressure and wind direction angle are tested to obtain three groups of experimental data, two of which are used for modeling, and the other is used for verification. The first-order Fourier method with a sine relationship is used for fitting, and the correlation adjustment value R 2 of the fitting curve is observed.

[0100] Relative air pressure and wind speed relationship modeling: using the above experimental data, the relative air pressure and wind speed relationship under the same wind direction angle is modeled. From the principle analysis, it can be known that the relative air pressure and wind speed size fitting is a quadratic power relationship.

[0101] Step S2: screening the relative air pressure △Pn obtained in step S1, eliminating abnormal values, and screening m relative air pressure values, denoted as Δp m , wherein m≤n; the m screened data is standardized, and the calculation formula of the standardized relative air pressure value Δp′ m is:

[0102]

[0103] Wherein, Δp m is the m relative air pressure values screened from the n relative air pressures △Pn, μ i is the mean value of the data collected by the i-th air pressure sensor chip 8, and δ i is the standard deviation of the data collected by the i-th air pressure sensor chip 8.

[0104] Step S3: Δp′ mThe relative air pressure and wind direction angle relationship model and the relative air pressure and wind speed relationship model are substituted respectively; a plurality of relative air pressure and wind direction angle relationship curves and relative air pressure and wind speed relationship curves are obtained; different relative air pressure data points of the screened air pressure sensor chip 8 at the same wind direction angle under the same wind speed are found out, so as to determine the wind speed v and the wind direction angle θ.

[0105] In one specific embodiment of the present application, the monitoring method of the internal wind direction monitoring system of the combine harvester cleaning device comprises the following steps:

[0106] Data acquisition: the air pressure sensor chip 8 after calibration collects five groups of horizontal wind direction angles under different wind speeds, which are set at an interval of 15° between 0-180° of the windward surface, and the pitch wind direction angles are set at an interval of 5° between-45° to 45° of the windward surface, and the test is performed for 40 s each time, the collected data is standardized, and the relative air pressure and wind direction angle relationship model is used to obtain the fitting curve of the relative air pressure and horizontal angle relationship under one group of wind speed Figure 6 is a fitting curve diagram of the relative air pressure and horizontal angle relationship under one group of wind speed, wherein:

[0107] Sensor (air pressure sensor chip) 1: General model Fourier1 (first-order Fourier):

[0108] f (θ) = 0.7107-1.586*cos (θ*0.03348) +0.3703*sin (θ*0.03348)

[0109] R 2 = 0.979

[0110] Sensor (air pressure sensor chip) 2: General model Fourier1 (first-order Fourier):

[0111] f (θ) = 30.13-27.81*cos (θ*0.00363) -13.49*sin (θ*0.00363)

[0112] R 2 = 0.953

[0113] Sensor (air pressure sensor chip) 3: General model Fourier1 (first-order Fourier):

[0114] f (θ) =-0.6238-0.2637*cos (θ*0.03253) +0.1393*sin (θ*0.03253)

[0115] R 2 = 0.8629

[0116] Sensor (barometric pressure sensor chip) 4: General model Fourier 1 (first-order Fourier):

[0117] f(θ)=0.8832-0.8578*cos(θ*0.01971)-1.632*sin(θ*0.01971)

[0118] R 2 =0.9595

[0119] Depend on Figure 6 It can be seen that the air pressure sensor chip 8 exhibits a strong sinusoidal relationship between the relative air pressure value and the wind direction angle within a 180-degree range from the area directly opposite the chip to the area blowing in from both sides. This relationship is based on the model of the relationship between relative air pressure and wind direction angle. Figure 7 This is a fitted curve of the relationship between relative air pressure and pitch angle at the same wind speed. Figure 7 Later, it was found that the correlation adjustment value R of the fitted mathematical model of curves 2 and 4 was... 2 Since the value is below 0.8, the fitting functions of curves 2 and 4 are not considered. Only curves 1 and 3 are needed to determine the pitch angle.

[0120] Sensor (barometric pressure sensor chip) 1: General model Fourier 1: (First-order Fourier)

[0121] f(θ)=1.415+0.9245*cos(θ*0.04493)+0.7474*sin(θ*0.04493)

[0122] R 2 =0.9862

[0123] Sensor (barometric pressure sensor chip) 3: General model Fourier 1 (first-order Fourier):

[0124] f(θ)=-0.6471+0.173*cos(θ*0.0433)-0.3213*sin(θ*0.0433)

[0125] R 2 =0.8663

[0126] Building a neural network model: Based on the PyTorch framework and PyCharm Community Edition software, a backpropagation (BP) neural network is built using the Python language.

[0127] like Figure 5As shown, in one specific embodiment of the application, the BP neural network comprises three hidden layers and an output layer, and a residual network is used to connect between each layer. The main idea of the residual block is to introduce a "short circuit connection" to directly add the input to the output, which helps to solve the problems of gradient disappearance and gradient explosion in the neural network training process. Adding attention mechanism helps the model pay more attention to important parts of the input, Figure 6 and Figure 7 It can be seen that the R 2 of curve 1 has a strong fitting relationship and needs to be given a high weight, the R 2 of curve 3 has a low correlation and should be given a low weight. In the PyCharm Community Edition software, the weights of the attention layers of the neural network are set to 2, 1.5, 0.5, and 1.5 for curves 1-4, respectively. The activation function is selected as the variant function Leaky ReLU of ReLU, which solves the problem of gradient being 0 when the input value of ReLU is less than 0. The function optimizer is selected as the Adamp optimizer, and the "early stopping" mechanism is added. The early stopping threshold is set to 100 in the PyCharm Community Edition software. If the neural network does not improve within 100 training times, it will be stopped to prevent overfitting of the neural network. The training is stopped before overfitting to preserve the generalization ability of the model and reduce the waste of computing resources.

[0128] The forward transmission formula of each residual block of the BP neural network can be expressed as:

[0129] y = LeakyReLU (W2 * LeakyReLU (W1x + b1) + b2 + x)

[0130] where x is the input, y is the output, W1 is the weight of the first hidden layer, W2 is the weight of the second hidden layer, b1 is the bias of the first hidden layer, and b2 is the bias of the second hidden layer. An initial value can be set according to experience and updated by the optimizer Adamp.

[0131] The mathematical expression of the activation function Leaky ReLU is:

[0132] f(x) = max(0.01x, x);

[0133] The data standardized in the data acquisition step is taken as the input of the model of the BP neural network, and the training of the BP neural network model is started, the output values of the horizontal wind direction angle and the pitch wind direction angle of the BP neural network are fitted by using a first-order Fourier function, the fitting function of the prediction model is compared with the fitting function curve of the relative air pressure and wind direction angle relationship model in the data acquisition step, the output values of the wind speed of the BP neural network are fitted by using a proportional function, the fitting function of the prediction model is compared with the fitting curve of the relative air pressure and wind speed relationship model in the data acquisition step, so as to verify the accuracy of the BP neural network prediction model, and a trained BP neural network is obtained.

[0134] The trained BP neural network model is loaded into the prediction system 1, and the newly collected data of the monitoring device is predicted to obtain the prediction values of the wind speed and the wind direction angle in the cleaning device of the combine harvester.

[0135] The BP neural network can realize fast convergence and avoid falling into a local minimum value through a residual network, can help the model to better focus on important parts in the input by adding an attention mechanism, can prevent overfitting of the neural network by adding an 'early stop' mechanism, can stop training before overfitting, can retain the generalization ability of the model, and can also reduce the waste of computing resources. Therefore, according to the characteristics of the cleaning process of the cleaning device of the combine harvester, a model of the relationship among the relative air pressure, the wind direction angle and the wind speed is established, the neural network structure is optimized, and the neural network parameters are optimized, so that the speed and accuracy of the prediction task are improved. The air flow pressure in the cleaning device of the combine harvester is monitored by the air pressure sensor chip 8, the air flow pressure data is collected, and the air flow pressure data is converted into the measurement of the air flow speed and the air flow wind direction angle, so that the advantages of simple structure, convenient measurement and low cost are achieved. A feasible device is provided for measuring the wind direction in the cleaning device of the combine harvester, and the combine harvester can obtain the air flow in real time. The wind speed and the wind direction angle in the cleaning device of the combine harvester are predicted by using the neural network, the change of the wind direction in the cleaning device of the combine harvester is reflected, prediction information is provided for the control system of the combine harvester, the grain loss rate is reduced, the subsequent control lag is reduced, and the cleaning performance is improved.

[0136] It should be understood that although the present specification is described in terms of various embodiments, not every embodiment contains only one independent technical solution, and the description manner of the specification is only for the sake of clarity, and those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be combined appropriately to form other embodiments that can be understood by those skilled in the art.

[0137] The series of detailed descriptions listed above are only specific descriptions of the feasible embodiments of the present application, and are not used to limit the protection scope of the present application, and any equivalent embodiments or changes made without departing from the spirit of the present application should be included in the protection scope of the present application.

Claims

1. An internal wind direction monitoring system for a combine harvester cleaning apparatus, characterized by, The prediction system (1) and the clean wind direction monitoring device (2) are included. The clean wind direction monitoring device (2) includes a support mechanism and a plurality of air pressure sensor chips (8); the plurality of air pressure sensor chips (8) are installed above the fish scale screen (3) through the support mechanism; the air pressure sensor chip (8) is connected with the prediction system (1), the air pressure sensor chip (8) is used for detecting the air flow pressure information above the fish scale screen (3) and transmitting to the prediction system (1); the prediction system (1) is provided with a neural network model, the neural network model takes the air flow pressure information as input, and takes the joint harvester cleaning device internal wind speed and wind direction angle prediction value as output; The air pressure sensor chip (8) is a MEMS air pressure sensor chip; The neural network model is a BP neural network including three hidden layers and an output layer, and a residual network is used for connection between each layer, and a forward transmission formula of each residual block of the residual network is represented as: y=LeakyReLU(W2*LeakyReLU(W1x+b1)+b2+x) Wherein x is input, y is output, W1 is the first layer hidden layer weight, W2 is the second layer hidden layer weight, b1 is the first layer hidden layer bias, b2 is the second layer hidden layer bias, and an initial value can be set according to experience; The neural network model takes the standardized relative air pressure value data as the input of the BP neural network model for training. The relative air pressure value Δp' after the standardization processing m The calculation formula is: wherein Δp m is the m relative air pressure values selected from the n relative air pressure values ΔPn, ΔPn is the relative air pressure value obtained by subtracting the reference air pressure from the value of the air pressure sensor chip (8) at the point, μ i is the mean value of the data collected by the i-th air pressure sensor chip (8), δ i is the standard deviation of the data collected by the i-th air pressure sensor chip (8).

2. The combine harvester cleaning system internal airflow direction monitoring system of claim 1, wherein, The support mechanism includes a support frame (4), a support plate (5), a support rod (6) and a mounting plate (7) One side of the support frame (4) is provided with the support plate (5), the support plate (5) is provided with the support rod (6), the support rod (6) is provided with the mounting plate (7), and the air pressure sensor chip (8) is arranged on the mounting plate (7); the other side of the support frame (4) is installed above the fish scale screen (3).

3. The combine harvester cleaning system internal airflow direction monitoring system of claim 1, wherein, The prediction system (1) includes an STM32 single-chip microcomputer and a prediction unit, the STM32 single-chip microcomputer is connected with the air pressure sensor chip (8), is used for collecting data of the air pressure sensor chip (8) and transmitting to the prediction unit of the prediction system, and the prediction unit is provided with a neural network model.

4. A method of monitoring an internal wind direction monitoring system of a combine harvester cleaning apparatus according to any one of claims 1 to 3, characterized in that, The steps include: The trained neural network model is carried to the prediction system (1); The clean wind direction monitoring device (2) is installed at a predetermined position in the cleaning device, a plurality of air pressure sensor chips (8) are installed above the fish scale screen (3) through the support mechanism, the air pressure sensor chip (8) detects the air flow pressure information above the fish scale screen (3) and transmits to the prediction system (1); The neural network model in the prediction system (1) takes the air flow pressure information as input, and takes the joint harvester cleaning device internal wind speed and wind direction angle prediction value as output; Further The steps include: comparing the prediction value of the prediction system (1) with the calculation results of the relative air pressure and wind speed relationship model and the relative air pressure and wind direction angle relationship model, and verifying the accuracy of the prediction system; The relative air pressure and wind speed relationship model is: In the formula, C $ is a pressure value, p is the air density, V is the wind speed, and S is the cross-sectional area of the air pressure sensor chip (8). The relative air pressure and wind direction angle relationship model is: f(θ)=a0+a1×cos(ω×θ)+b1×sin(ω×θ) ω = (2π) / T where θ is the wind direction angle, ω is the frequency of the curve, a0 is the amplitude of the direct current component, a1 is the amplitude of the sine component, b1 is the amplitude of the cosine component, and T is the period.

5. The method of claim 4, wherein the method further comprises: The neural network model is trained by the following steps: Data acquisition: Collect multiple sets of horizontal wind direction angles and pitch wind direction angles under different wind speeds through the calibrated air pressure sensor chip (8) for testing, collect multiple sets of air pressure data, and standardize the collected air pressure data; Building a neural network model: Based on the PyTorch framework and PyCharm Community Edition software, a BP neural network is built using Python language, which includes three hidden layers and an output layer, and a residual network is used to connect between each layer. The activation function selects the variant function Leaky ReLU of ReLU, and the function optimizer selects the Adamp optimizer; Neural network training: The standardized data in the data acquisition step is used as the input of the BP neural network model, and the BP neural network model is trained; The trained BP neural network model is loaded into the prediction system (1) to predict the new data collected by the air pressure sensor chip (8), and the predicted values of the wind speed and wind direction angle inside the combine harvester cleaning device are obtained.

Images