Artificial intelligence wine picking device and wine picking method

Abstract

The invention discloses an artificial intelligence wine picking device, which belongs to the technical field of wine picking devices, and comprises a wine retort, wherein a pressure sensor is arranged at the lower end of the wine retort, a pipeline is arranged at the lower end of the side of the wine retort, a wine picking bowl is arranged below the pipeline, an online component analysis device is arranged inside the wine picking bowl, and a solenoid valve is arranged at the lower end conduit of the side of the wine picking bowl. The invention monitors and analyzes factors affecting the quality of wine such as alcohol concentration, fermentation time, chemical composition, container pressure, etc., to avoid the traditional manual wine picking process of picking wine by watching the flowers, which cannot guarantee the quality of the picked wine and has a strong dependence on the individual wine pickers. Based on data, the data is deeply mined, and the data is learned through a neural network to find out the implicit rules and correlations therein, and the grade of the original wine is judged according to the current input data, thereby improving the efficiency and accuracy of wine picking and avoiding human errors.

Description

Technical Field

[0001] The present invention belongs to the technical field of wine picking devices, and in particular relates to an artificial intelligence wine picking device and a wine picking method thereof. Background Art

[0002] At present, the traditional way of picking liquor in the liquor industry is to use the method of picking liquor by observing the size, shape and duration of the hops after the liquor flows out to judge the alcohol content of the flowing liquor and distinguish the three sections of the head, body and tail.

[0003] The existing technology has the following problems: 1. Picking wine by looking at flowers is completely dependent on the experience of operators, which is greatly influenced by the subjective influence of operators, and it is impossible to form a unified evaluation method and standard. In addition, the inheritance method depends entirely on the master to teach the apprentice, which is difficult to inherit and there is a shortage of talents;

[0004] 2. Hops are affected by temperature and pressure, and the higher the alcohol content, the faster the hops dissipate. Therefore, they are easily affected by the surrounding environmental factors and the detection speed of the operator. The efficiency and accuracy of picking wine are not high, and the disadvantage of human error is inevitable.

[0005] 3. It is impossible to accurately judge the wine by the single factor of hops. Other factors such as alcohol concentration, fermentation time, chemical composition, container pressure, etc. also affect the quality of the wine.

[0006] In summary, the current traditional wine picking process is manual operation to pick the wine according to the flowers, which cannot guarantee the quality of the picked wine, has a long production cycle, low efficiency, and is highly dependent on the individual wine pickers. Summary of the invention

[0007] In order to solve the problems raised in the above background technology, the present invention provides an artificial intelligence wine picking device, which has the characteristics of simple structure, convenient wine picking and high efficiency.

[0008] Another object of the present invention is to provide a method for picking wine by relying on an artificial intelligence wine picking device.

[0009] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: an artificial intelligence wine picking device, comprising a wine retort, a pressure sensor is arranged at the lower end of the wine retort, a pipe is arranged at the lower end of the side of the wine retort, a wine picking bowl is arranged below the pipe, an online component analysis device is arranged inside the wine picking bowl, a solenoid valve is arranged on the conduit at the lower end of the side of the wine picking bowl, an industrial computer is arranged on the side of the pipe, an online thermal imager is arranged on the side of the industrial computer and the lower end of the pipe, and a camera is arranged at the lower end of the online thermal imager.

[0010] Furthermore, a PLC controller is arranged on the side of the industrial computer.

[0011] Furthermore, the pressure sensor, the online thermal imager, the camera, the online component analysis device, the solenoid valve and the industrial computer are all electrically connected to the PLC controller.

[0012] Furthermore, a plurality of raw wine collectors are arranged below the solenoid valve.

[0013] Furthermore, the camera uses a laser camera to automatically emit laser without the need for a light source.

[0014] Furthermore, a timer for timing is arranged inside the industrial computer.

[0015] Furthermore, the wine picking method of the artificial intelligence wine picking device is as follows:

[0016] 1. Data collection: select several different wine retorts, dynamically collect photos of hops during the entire wine picking process, wine retort pressure, fermentation time, temperature data, component data (including acetaldehyde, ethyl acetate, methanol, fusel oil, ethyl linoleate, ethyl oleate, ethyl palmitate, etc.), and the corresponding raw wine grade under each state;

[0017] 2. Classify the data into neural network learning part and test part;

[0018] 3. Model training: The collected hop state, temperature, composition, and grade are imported into deep learning for training to build a model. Deep learning has multiple hidden neural networks and integrates the feature extraction process to automatically learn the characteristics of the data set.

[0019] a) Fuzzy set data mining through artificial intelligence deep learning of hop state allows the definition of fuzzy thresholds or boundaries. Fuzzy logic uses a truth value of 0.0-1.0 to indicate the degree to which a specific value is a member given, rather than using a class or set to accurately truncate, abstracting the data, similar to experience, without a definite quantitative standard,

[0020] b) The temperature, pressure, fermentation time, ingredients, mined hop status data, and manually determined raw wine grade data are learned through the forward propagation and back propagation algorithms of the neural network. The neural network includes an input layer, an output layer, and a hidden layer. The mined hop status data, temperature, pressure, fermentation time, and ingredients are used as nodes of the input layer.

[0021] First, forward propagation learning is performed. The neurons in the hidden layer are connected to the nodes in the previous layer as nodes. The parameters of the nodes in the previous layer are linearly combined through a certain weight ratio to obtain a new number. Then a ReLU activation function is applied to this new number to obtain an output value. Multiple neurons run in parallel. The weight formulas of the linear combinations of neurons in the same layer are different. The results are passed to the next layer as new variables. By performing multi-layer abstraction on the combined low-level features and multiple nonlinear transformations, more abstract high-level representation attribute categories or features are formed to discover data features and finally obtain output values.

[0022] After forward propagation learning obtains weights and biases, back propagation learning is performed based on the manually determined raw wine grade data and the original input layer node data. The error can be calculated as the output changes, and the weights and biases can be corrected.

[0023] A discard regularization algorithm is also used in the deep learning process. This algorithm can effectively reduce the dependency between neurons, allowing the neural network to learn more robust features with other neurons. It can not only reduce overfitting, but also improve accuracy.

[0024] In the process of discard regularization, a discard probability adjustment function is added to the discard function to adjust the neurons with the largest weight value, but not the first and last layers.

[0025] Each neuron expression hi = Φ(∑x i w 1i +b i )

[0026] Each hidden layer takes the coefficient determinant

[0027] Calculate w in the determinant ni Maximum value, select the corresponding hi to adjust the discard probability,

[0028] x Drop probability:

[0029]

[0030] The rest are stretched accordingly, which enhances the strength of regularization.

[0031] The whole process includes accurate data such as hop status (fuzzy judgment criteria), retort pressure, fermentation time, temperature data, component data, etc. The combination of the two can more effectively and accurately determine the grade of the original wine, avoiding the inaccuracy caused by individual differences in traditional methods. Use test data for offline testing and make corrections if there is any deviation.

[0032] 4. Verification: Verify the obtained model, select a whole wine retort process for verification, and compare the machine inspection results with the wine picker's evaluation results. If there is a large deviation, make corrections;

[0033] 5. Data accumulation: The data is constantly accumulated and expanded during the use process. The more data, the more information it contains. The more accurate the results given by the neural network after learning. By collecting more data, mining can also help improve production processes and increase production efficiency.

[0034] 6. The pressure sensor sends the data to the PLC controller, which processes the data and sends it to the industrial computer. The data enters the input layer, and other data is directly processed by the industrial computer and enters the model input layer. After processing, the raw wine grade is obtained. The industrial computer feeds the data back to the PLC controller, and the PLC controller opens the corresponding solenoid valve and closes the unnecessary solenoid valves. The raw wine flows into the raw wine collector of the corresponding grade.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] The present invention monitors and analyzes factors that affect the quality of wine, such as alcohol concentration, fermentation time, chemical composition, container pressure, etc., to avoid the traditional manual wine picking process of picking wine by looking at the flowers, which cannot guarantee the quality of the picked wine and is highly dependent on the individual wine pickers. Based on data, the data is deeply mined, and the data is learned through neural networks to find out the implicit rules and correlations, and the grade of the original wine is determined according to the current input data, thereby improving the efficiency and accuracy of wine picking and avoiding human errors. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a structural schematic diagram of the present invention;

[0038] Figure 2 A neural network learning graph for the present invention;

[0039] In the figure: 1. Wine steamer; 2. Pressure sensor; 3. Pipeline; 4. Wine bowl; 5. Online thermal imager; 6. Camera; 7. Online component analysis device; 8. Solenoid valve; 9. Industrial computer; 10. PLC controller; 11. Raw wine collector. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] See also Figure 1-2 The present invention provides the following technical solutions: an artificial intelligence wine picking device, comprising a wine retort 1, a pressure sensor 2 is arranged at the lower end of the wine retort 1, a pipe 3 is arranged at the lower end of the side of the wine retort 1, a wine picking bowl 4 is arranged below the pipe 3, an online component analysis device 7 is arranged inside the wine picking bowl 4, a solenoid valve 8 is arranged on the lower end conduit of the side of the wine picking bowl 4, an industrial computer 9 is arranged on the side of the pipe 3, an online thermal imager 5 is arranged at the side of the industrial computer 9 and the lower end of the pipe 3, and a camera 6 is arranged at the lower end of the online thermal imager 5.

[0042] Furthermore, in the present invention, a PLC controller 10 is provided on the side of the industrial computer 9 .

[0043] Furthermore, in the present invention, the pressure sensor 2 , the online thermal imager 5 , the camera 6 , the online component analysis device 7 , the solenoid valve 8 and the industrial computer 9 are all electrically connected to the PLC controller 10 .

[0044] Furthermore, in the present invention, a plurality of raw wine collectors 11 are arranged below the solenoid valve 8 .

[0045] Furthermore, in the present invention, the camera 6 uses a laser camera to automatically emit laser light without the need for a light source.

[0046] Furthermore, in the present invention, a timer for timing is provided inside the industrial computer 9 .

[0047] In the present invention, the wine picking method of the artificial intelligence wine picking device is as follows:

[0048] 1. Data collection: select several different wine retorts 1, dynamically collect photos of hops during the entire wine picking process, wine retort pressure, fermentation time, temperature data, component data (including concentrations of acetaldehyde, ethyl acetate, methanol, fusel oil, ethyl linoleate, ethyl oleate, ethyl palmitate, etc.), and the corresponding raw wine grade under each state;

[0049] 2. Classify the data into neural network learning part and test part;

[0050] 3. Model training: The collected hop state, temperature, composition, and grade are imported into deep learning for training to build a model. Deep learning has multiple hidden neural networks and integrates the feature extraction process to automatically learn the characteristics of the data set.

[0051] a) Fuzzy set data mining through artificial intelligence deep learning of hop state allows the definition of fuzzy thresholds or boundaries. Fuzzy logic uses a truth value of 0.0-1.0 to indicate the degree to which a specific value is a member given, rather than using a class or set to accurately truncate, abstracting the data, similar to experience, without a definite quantitative standard,

[0052] b) The temperature, pressure, fermentation time, ingredients, mined hop status data, and manually determined raw wine grade data are learned through the forward propagation and back propagation algorithms of the neural network. The neural network includes an input layer, an output layer, and a hidden layer. The mined hop status data, temperature, pressure, fermentation time, and ingredients are used as nodes of the input layer.

[0053] First, forward propagation learning is performed. The neurons in the hidden layer are connected to the nodes in the previous layer as nodes. The parameters of the nodes in the previous layer are linearly combined through a certain weight ratio to obtain a new number. Then a ReLU activation function is applied to this new number to obtain an output value. Multiple neurons run in parallel. The weight formulas of the linear combinations of neurons in the same layer are different. The results are passed to the next layer as new variables. By performing multi-layer abstraction on the combined low-level features and multiple nonlinear transformations, more abstract high-level representation attribute categories or features are formed to discover data features and finally obtain output values.

[0054] After forward propagation learning obtains weights and biases, back propagation learning is performed based on the manually determined raw wine grade data and the original input layer node data. The error can be calculated as the output changes, and the weights and biases can be corrected.

[0055] A discard regularization algorithm is also used in the deep learning process. This algorithm can effectively reduce the dependency between neurons, allowing the neural network to learn more robust features with other neurons. It can not only reduce overfitting, but also improve accuracy.

[0056] In the process of discard regularization, a discard probability adjustment function is added to the discard function to adjust the neurons with the largest weight value, but not the first and last layers.

[0057] Each neuron expression hi = Φ(∑x i w 1i +b i )

[0058] Each hidden layer takes the coefficient determinant

[0059] Calculate w in the determinant ni Maximum value, select the corresponding hi to adjust the discard probability,

[0060] x Drop probability:

[0061]

[0062] The rest are stretched accordingly, which enhances the strength of regularization.

[0063] The whole process has fuzzy judgment standards for hop status, as well as precise data such as retort pressure, fermentation time, temperature data, and ingredient data. The combination of the two can more effectively and accurately determine the grade of the original wine, avoiding the inaccuracy caused by individual differences in traditional methods. The test data is used for offline testing, and if there is a deviation, it is corrected;

[0064] 4. Verification: Verify the obtained model, select a whole wine retort process for verification, and compare the machine inspection results with the wine picker's evaluation results. If there is a large deviation, make corrections;

[0065] 5. Data accumulation: The data is constantly accumulated and expanded during the use process. The more data, the more information it contains. The more accurate the results given by the neural network after learning. By collecting more data, mining can also help improve production processes and increase production efficiency.

[0066] 6. The pressure sensor 2 sends the data to the PLC controller 10, which processes the data and sends it to the industrial computer 9. The data enters the input layer, and other data is directly processed by the industrial computer 9 and enters the model input layer. After processing, the raw wine grade is obtained, and the industrial computer feeds the data back to the PLC controller 10. The PLC controller 10 opens the corresponding solenoid valve 8 and closes the unnecessary solenoid valve 8, and the raw wine flows into the raw wine collector 11 of the corresponding grade.

[0067] The working principle and use process of the present invention: When the present invention is used:

[0068] 1. Data collection: select several different wine retorts 1, dynamically collect photos of hops during the entire wine picking process, wine retort pressure, fermentation time, temperature data, component data (including concentrations of acetaldehyde, ethyl acetate, methanol, fusel oil, ethyl linoleate, ethyl oleate, ethyl palmitate, etc.), and the corresponding raw wine grade under each state;

[0069] 2. Classify the data into neural network learning part and test part;

[0070] 3. Model training: The collected hop state, temperature, composition, and grade are imported into deep learning for training to build a model. Deep learning has multiple hidden neural networks and integrates the feature extraction process to automatically learn the characteristics of the data set.

[0071] a) Fuzzy set data mining through artificial intelligence deep learning of hop state allows the definition of fuzzy thresholds or boundaries. Fuzzy logic uses a truth value of 0.0-1.0 to indicate the degree to which a specific value is a member given, rather than using a class or set to accurately truncate, abstracting the data, similar to experience, without a definite quantitative standard,

[0072] b) The temperature, pressure, fermentation time, ingredients, mined hop status data, and manually determined raw wine grade data are learned through the forward propagation and back propagation algorithms of the neural network. The neural network includes an input layer, an output layer, and a hidden layer. The mined hop status data, temperature, pressure, fermentation time, and ingredients are used as nodes of the input layer.

[0073] First, forward propagation learning is performed. The neurons in the hidden layer are connected to the nodes in the previous layer as nodes. The parameters of the nodes in the previous layer are linearly combined through a certain weight ratio to obtain a new number. Then a ReLU activation function is applied to this new number to obtain an output value. Multiple neurons run in parallel. The weight formulas of the linear combinations of neurons in the same layer are different. The results are passed to the next layer as new variables. By performing multi-layer abstraction on the combined low-level features and multiple nonlinear transformations, more abstract high-level representation attribute categories or features are formed to discover data features and finally obtain output values.

[0074] After forward propagation learning obtains weights and biases, back propagation learning is performed based on the manually determined raw wine grade data and the original input layer node data. The error can be calculated as the output changes, and the weights and biases can be corrected.

[0075] A discard regularization algorithm is also used in the deep learning process. This algorithm can effectively reduce the dependency between neurons, allowing the neural network to learn more robust features with other neurons. It can not only reduce overfitting, but also improve accuracy.

[0076] In the process of discard regularization, a discard probability adjustment function is added to the discard function to adjust the neurons with the largest weight value, but not the first and last layers.

[0077] Each neuron expression hi = Φ(∑x i w 1i +b i )

[0078] Each hidden layer takes the coefficient determinant

[0079] Calculate w in the determinant ni Maximum value, select the corresponding hi to adjust the discard probability,

[0080] x Drop probability:

[0081]

[0082] The rest are stretched accordingly, which enhances the strength of regularization.

[0083] The whole process has fuzzy judgment standards for hop status, as well as precise data such as retort pressure, fermentation time, temperature data, and ingredient data. The combination of the two can more effectively and accurately determine the grade of the original wine, avoiding the inaccuracy caused by individual differences in traditional methods. The test data is used for offline testing, and if there is a deviation, it is corrected;

[0084] 4. Verification: Verify the obtained model, select a whole wine retort process for verification, and compare the machine inspection results with the wine picker's evaluation results. If there is a large deviation, make corrections;

[0085] 5. Data accumulation: The data is constantly accumulated and expanded during the use process. The more data, the more information it contains. The more accurate the results given by the neural network after learning. By collecting more data, mining can also help improve production processes and increase production efficiency.

[0086] 6. The pressure sensor 2 sends the data to the PLC controller 10, which processes the data and sends it to the industrial computer 9. The data enters the input layer, and other data is directly processed by the industrial computer 9 and enters the model input layer. After processing, the raw wine grade is obtained, and the industrial computer feeds the data back to the PLC controller 10. The PLC controller 10 opens the corresponding solenoid valve 8 and closes the unnecessary solenoid valve 8, and the raw wine flows into the raw wine collector 11 of the corresponding grade.

[0087] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A wine picking method of an artificial intelligence wine picking device is as follows: The invention comprises a wine retort (1), wherein a pressure sensor (2) is arranged at the lower end of the wine retort (1), a pipeline (3) is arranged at the lower end of the side of the wine retort (1), a wine-collecting bowl (4) is arranged below the pipeline (3), an online component analysis device (7) is arranged inside the wine-collecting bowl (4), a solenoid valve (8) is arranged at the lower end conduit of the side of the wine-collecting bowl (4), an industrial control computer (9) is arranged at the side of the pipeline (3), an online thermal imager (5) is arranged at the side of the industrial control computer (9) and the lower end of the pipeline (3), and a camera (6) is arranged at the lower end of the online thermal imager (5); 1. Data collection: select a number of different wine retorts (1), dynamically collect photos of hops during the entire wine picking process, wine retort pressure, fermentation time, temperature data, component data, and the corresponding raw wine grade under each state; 2. Classify the data into neural network learning part and test part; 3. Model training: The collected hop state, temperature, composition, and grade are imported into deep learning for training to build a model. Deep learning has multiple hidden neural networks and integrates the feature extraction process to automatically learn the characteristics of the data set. a) Fuzzy set data mining through artificial intelligence deep learning of hop state allows the definition of fuzzy thresholds or boundaries. Fuzzy logic uses a truth value of 0.0-1.0 to indicate the degree to which a specific value is a member given, rather than using a class or set to accurately truncate, abstracting the data, similar to experience, without a definite quantitative standard, b) The temperature, pressure, fermentation time, ingredients, mined hop status data, and manually judged raw wine grade data are learned through the forward propagation and back propagation algorithms of the neural network. The neural network includes an input layer, an output layer, and a hidden layer. The mined hop status data, temperature, pressure, fermentation time, and ingredients are used as nodes of the input layer. First, forward propagation learning is performed. The neurons in the hidden layer are connected to the nodes in the previous layer as nodes. The parameters of the nodes in the previous layer are linearly combined through a certain weight ratio to obtain a new number. Then a ReLU activation function is applied to this new number to obtain an output value. Multiple neurons run in parallel. The weight formulas of the linear combinations of neurons in the same layer are different. The results are passed to the next layer as new variables. By performing multi-layer abstraction on the combined low-level features and multiple nonlinear transformations, more abstract high-level representation attribute categories or features are formed to discover data features and finally obtain output values. After forward propagation learning obtains weights and biases, back propagation learning is performed based on the manually determined raw wine grade data and the original input layer node data. The error can be calculated as the output changes, and the weights and biases can be corrected. A discard regularization algorithm is also used in the deep learning process. This algorithm can effectively reduce the dependency between neurons, allowing the neural network to learn more robust features with other neurons. It can not only reduce overfitting, but also improve accuracy. In the process of discard regularization, a discard probability adjustment function is added to the discard function to adjust the neurons with the largest weight value, but not the first and last layers. Each neuron expression hi = Φ(∑x i w 1i +b i ) Each hidden layer takes the coefficient determinant Calculate w in the determinant ni Maximum value, select the corresponding hi to adjust the discard probability, x Drop probability: The rest are stretched accordingly, which enhances the strength of regularization. The whole process includes accurate data on hop status, retort pressure, fermentation time, temperature data, and component data. The combination of the two can more effectively and accurately determine the grade of the original wine, avoiding the inaccuracy caused by individual differences in traditional methods. Use test data for offline testing and make corrections if there is any deviation.

4. Verification: Verify the obtained model, select a whole wine retort process for verification, and compare the machine inspection results with the wine picker's evaluation results. If there is a large deviation, make corrections; 5. Data accumulation: The data is constantly accumulated and expanded during the use process. The more data, the more information it contains. The more accurate the results given by the neural network after learning. By collecting more data, mining can also help improve production processes and increase production efficiency.

6. The pressure sensor (2) sends data to the PLC controller (10), which processes the data and sends it to the industrial computer (9). The data enters the input layer. Other data are directly processed by the industrial computer (9) and enter the model input layer. After processing, the raw wine grade is obtained. The industrial computer feeds back the data to the PLC controller (10). The PLC controller (10) opens the corresponding solenoid valve (8) and closes the unnecessary solenoid valve (8). The raw wine flows into the raw wine collector (11) of the corresponding grade.

2. The wine picking method of the artificial intelligence wine picking device according to claim 1, characterized in that:A PLC controller (10) is arranged on the side of the industrial computer (9).

3. The wine picking method of the artificial intelligence wine picking device according to claim 1, characterized in that: The pressure sensor (2), the online thermal imager (5), the camera (6), the online component analysis device (7), the solenoid valve (8) and the industrial computer (9) are all electrically connected to the PLC controller (10).

4. The wine picking method of the artificial intelligence wine picking device according to claim 1, characterized in that: A plurality of raw wine collectors (11) are arranged below the solenoid valve (8).

5. The wine picking method of the artificial intelligence wine picking device according to claim 1, characterized in that: The camera (6) uses a laser camera, which automatically emits laser light without the need for a light source.

6. The wine picking method of the artificial intelligence wine picking device according to claim 1, characterized in that: A timer for timing is arranged inside the industrial computer (9).

Images