An artificial intelligence-based corner in-situ calibration device and method

Abstract

The application discloses a kind of based on artificial intelligence's corner in situ calibration device and method.The device is characterized in that it includes loading mechanism (1), adjusting mechanism (2), calibration connecting piece (3), to be calibrated shaft (4), to be calibrated angle sensor (5), adjusting mechanism (2) upper and lower ends are loading mechanism (1) and calibration connecting piece (3) respectively, loading mechanism (1) is connected with adjusting mechanism (2) and calibration connecting piece (3) respectively, calibration connecting piece (3) is connected with to be calibrated shaft (4), and to be calibrated angle sensor (5) is fixed on to be calibrated shaft (4).The method is characterized in that it includes constructing bidirectional long short-term memory network BLSTM model;Generation loss function;Optimization model;Get function relationship.The device method of the application can eliminate the influence of online calibration interference factor, and solve the problem that sensor cannot be directly calibrated, with the advantages of high precision, easy operation.

Description

Technical Field

[0001] This invention belongs to the field of testing and calibration technology, specifically relating to an in-situ corner calibration device and method based on artificial intelligence. Background Technology

[0002] With the development of my country's equipment manufacturing industry, more and more equipment cannot calibrate the angle sensor alone due to limited space or complex installation. In addition, the application of angle sensors is becoming more and more widespread. Inventing a reliable in-situ calibration method for angle has become one of the key issues that urgently need to be solved.

[0003] In recent years, my country's artificial intelligence field has developed rapidly, with an increasing number of neural network types and higher training efficiency. Compared with manual methods, neural networks are much more efficient in processing repetitive data and monotonous tasks. Therefore, using neural network models to fit some nonlinear complex function relationships is a good solution.

[0004] Patent CN106352973A is titled "An In-situ Calibration Method for a Sensor." This method involves installing a standard sensor on top of the sensor under test and using an internal excitation signal to generate output signals from both the standard sensor and the angle sensor to be calibrated. The calibration is completed by comparing the two output signals. This method achieves in-situ calibration, but it requires the additional installation of an excitation unit and the removal of the standard sensor once, making the operation cumbersome. Furthermore, the zero-point error of the two sensors and other influencing factors caused by the installation cannot be well eliminated, so the calibration effect needs to be improved.

[0005] The patent CN112539876A is entitled "A Vertical Reference Torque Calibration Method and Device for Extreme Environments". In this method, the difference between the measured standard torque value and the torque value to be calibrated is used for calibration. However, it does not fit the function curve of the torque sensor under test, nor does it consider other influencing factors. The calibration effect needs to be improved. Summary of the Invention

[0006] The purpose of this invention is to perform in-situ calibration of the angle sensor based on artificial intelligence when it cannot be disassembled, which has the advantages of high precision and convenient operation.

[0007] The technical solution adopted by this invention to solve its technical problem is:

[0008] An in-situ angle calibration device based on artificial intelligence is characterized in that it includes a loading mechanism 1, an adjustment mechanism 2, a calibration connector 3, a shaft to be calibrated 4, and an angle sensor to be calibrated 5. The upper and lower ends of the adjustment mechanism 2 are the loading mechanism 1 and the calibration connector 3, respectively. The loading mechanism 1 is connected to the adjustment mechanism 2 and the calibration connector 3, respectively. The calibration connector 3 is connected to the shaft to be calibrated 4, and the angle sensor to be calibrated 5 is fixed on the shaft to be calibrated 4.

[0009] Furthermore, the loading mechanism 1 includes a circular grating encoder, a bearing 8, an internal shaft 9, and a torque motor 10. The circular grating disk 6 is connected to the reading head 7 to form a circular grating encoder. The circular grating encoder is fixedly connected to the internal shaft 9 through the bearing and measures the signal generated by the rotation of the internal shaft 9. The torque motor 10 is used to generate power to make the internal shaft 9 rotate.

[0010] Furthermore, the adjustment mechanism 2 includes a disc 11 with a protruding coaxial adjustment ring 16. The coaxial adjustment ring 16 has openings on its side, and long screws 15 pass through these openings to contact the loading mechanism 1. Rotating each long screw 15 can adjust the position of the loading mechanism 1.

[0011] Furthermore, an adjusting block 12 is installed on the edge of the disc 11. The adjusting block 12 is connected to the support shaft 17. The adjusting block 12 has a spherical groove inside. By rotating the adjusting block 12, the level of the adjusting mechanism 2 can be adjusted, thereby adjusting the level of the entire mechanism and the angle of the internal shaft 9, thus ensuring the coaxiality requirement of the shaft to be calibrated 4 and the internal shaft 9.

[0012] Furthermore, the calibration connector 3 is used to connect the loading mechanism 1 and the shaft to be calibrated 4 to rotate together. It includes a circular block 13 and a curved thin plate 14. The upper end of the circular block 13 is connected to the loading mechanism 1, and the lower end is connected to the curved thin plate 14. The curved thin plate 14 is connected to the shaft to be calibrated 4. The circular block 13 is connected to the internal shaft 9 through a bearing. The rotation of the internal shaft 9 drives the curved thin plate 14 to rotate, which in turn drives the shaft to be calibrated 4 to rotate.

[0013] This invention also provides an artificial intelligence-based in-situ corner calibration method, characterized in that it uses the aforementioned artificial intelligence-based in-situ corner calibration device and includes the following steps:

[0014] Step 1, Model Building: First, construct a bidirectional long short-term memory network (BLSTM) model. The data collected by the angle sensor 5 to be calibrated, the forward and return marks, the zero-position angle difference between the angle sensor 5 to be calibrated and the circular encoder, and the angle difference between the angle sensor 5 to be calibrated and the circular encoder under a fixed sampling time are used as the model inputs. The output value is then fitted by an artificial intelligence algorithm.

[0015] Step 2, generate loss function: Perform static step measurement at the corner, use the data collected by the circular grating encoder as the standard value, and construct the loss function using the difference between the output value of the BLSTM model and the standard value;

[0016] Step 3, optimize the model: perform dynamic continuous measurement of the turning angle, minimize the loss function using gradient descent, and optimize the BLSTM model parameters in reverse.

[0017] Step 4, obtain the functional relationship: establish the functional relationship between the angle sensor 5 to be calibrated and the circular grating encoder.

[0018] Furthermore, in step 1, a bidirectional long short-term memory network model, namely the BLSTM model, is first established. This BLSTM model is input through an input layer, and the middle of the model consists of BLSTM layers, fully connected layers, and hidden layers between the two layers. The BLSTM model outputs the calculation results as output values ​​through an output layer.

[0019] The zero-position angle difference between the angle sensor 5 to be calibrated and the circular grating encoder is obtained as follows:

[0020] After connecting the loading mechanism 1 and the adjusting mechanism 2, when the connection between the calibration connector 3 and the internal shaft 9 is disconnected, the servo of the torque motor 10 is activated, and the circular encoder searches for the zero position.

[0021] Connect the calibration connector 3 to the internal shaft 9, and use a level to adjust the level of the loading mechanism 1 by adjusting the adjusting block 12 of the adjusting mechanism 2, so that the internal shaft 9 in the center of the loading mechanism 1 is coaxial with the shaft 4 to be calibrated below.

[0022] Turn off the torque motor 10 servo and control the motor below the shaft to be calibrated 4 to rotate. The shaft to be calibrated 4 drives the internal shaft 9 to rotate, which in turn drives the circular grating encoder above to rotate, causing the angle sensor 5 to be calibrated to return to zero. At this time, the difference between the measured values ​​of the circular grating encoder, which serves as the standard sensor, and the angle sensor 5 to be calibrated is the interval angle between the zero positions of the two sensors, i.e., the zero position angle difference.

[0023] Furthermore, in step 2, the static step measurement at the corner includes:

[0024] Select the rotation direction and angle gradient, with the step of the angle gradient being A. Control the torque motor 10 to rotate at the selected speed and angle A, then stop. Hold this position for a period of time, then control the motor to rotate to the next gradient A and hold for a period of time until the motor rotates one revolution. Hold this position for a period of time and then end the rotation. At the same time, record the angle value of the circular grating encoder, the angle value of the angle sensor 5 to be calibrated, and the progress or return mark.

[0025] Furthermore, in step 3, the dynamic continuous measurement of the rotation angle includes:

[0026] Select the rotation speed and direction, and control the torque motor 10 to rotate continuously at the selected speed. After one revolution, maintain the rotation for a period of time before stopping. Simultaneously record the angle value of the circular grating encoder, the angle value of the angle sensor 5 to be calibrated, and the progress or return mark.

[0027] Furthermore, its optimization model steps include:

[0028] 1) Establish a BLSTM (Bidirectional Long Short-Time Memory) model, and read the angle value of the angle sensor 5 to be calibrated at time t, the forward and return marks, the zero-position angle difference between the angle sensor 5 to be calibrated and the circular encoder, and the angle difference between the angle sensor 5 to be calibrated and the circular encoder at time t as the model input. ;

[0029] 2) Calculate the forgetting gate and select the information to be forgotten: ;

[0030] 3) Calculate the memory gate and select the information to be memorized:

[0031] ;in: For memory gate, This is a temporary cell state;

[0032] 4) Calculate the cell state at the current moment: ;

[0033] 5) Calculate the output gate and the hidden state at the current time: , ;in: For output gate, This represents the current state of the cell.

[0034] 6) Through forgetting and remembering information, useful information for subsequent time steps is passed on, while useless information is discarded, and the hidden state is output at each time step. Forgetting, remembering, and output are determined by the hidden state of the previous time step. and current input Calculated forget gate Memory Gate Output gate To control, of which: , where is the activation function for each node;

[0035] 7) Take the output gate data at time t, the data from the previous 20 times, and the data from the next 20 times, and obtain the corresponding output value at the current time. ;

[0036] 8) Set the loss function E: Where N is the number of output layer nodes. Output for the model Standard value;

[0037] 9) Backward calculation: Modify the weights layer by layer according to the gradient descent algorithm. and deviation : , Where: α and β are the learning efficiency and bias adjustment coefficients, respectively, and N is the number of nodes in layer j;

[0038] 10) Repeat the dynamic continuous measurement of the angle multiple times, reread the new time input data, and repeat steps 1) to 9) until all the recorded data is processed. Through continuous loop iteration, the connection weights and biases between neurons in each layer are optimized so that the model can finally fit the standard encoder with ideal accuracy, thereby completing the in-situ calibration of the angle based on artificial intelligence.

[0039] The apparatus and method of the present invention can eliminate the influence of online calibration interference factors and solve the problem of not being able to directly calibrate the sensor, and has the advantages of high precision and easy operation.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] 1) When the control motor drives the shaft to be calibrated to rotate, static step measurement and dynamic continuous measurement of the rotation angle are performed, which comprehensively reflects the measurement accuracy of the sensor during smooth operation and acceleration operation.

[0042] 2) A large amount of data was used when training the neural network. The input data was not only the angle value measured by the angle sensor 5 to be calibrated, but also the forward and return marks. Since there may be errors in a single measurement, the function fitted by the neural network from a large amount of data can weaken the effect of other influencing factors, or even make other influencing factors negligible, thus improving the reliability of the calibration effect.

[0043] 3) The BLSTM model controls the information to be forgotten and remembered by calculating the forget gate, memory gate, cell state, and hidden layer state, so that the model output is not only related to the input at the current time, but also to some time before and after, which improves the accuracy of the calibration effect. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of the structure of the artificial intelligence-based in-situ rotation calibration device of the present invention.

[0045] Figure 2 This is a schematic diagram of the loading mechanism of the present invention.

[0046] Figure 3 This is a schematic diagram of the adjustment mechanism of the present invention.

[0047] Figure 4 This is a physical diagram of an example of the loading mechanism and adjusting mechanism of the present invention.

[0048] Figure 5 This is a schematic diagram of the calibration connector of the present invention.

[0049] Figure 6 This is a flowchart of the artificial intelligence calibration method of the present invention.

[0050] In the figure: 1-loading mechanism, 2-adjustment mechanism, 3-calibration connector, 4-shaft to be calibrated, 5-angle sensor to be calibrated, 6-circular grating disk, 7-reading head, 8-bearing, 9-internal shaft, 10-torque motor, 11-disc, 12-adjustment block, 13-circular block, 14-curved thin plate. Detailed Implementation

[0051] The present invention will be further described below with reference to the accompanying drawings.

[0052] Figure 1 This is a schematic diagram of the artificial intelligence-based in-situ corner calibration device of the present invention. Figure 1 As shown, the artificial intelligence-based in-situ angle calibration device of the present invention includes a loading mechanism 1, an adjusting mechanism 2, a calibration connector 3, a shaft to be calibrated 4, and an angle sensor to be calibrated 5. The upper and lower ends of the adjusting mechanism 2 are the loading mechanism 1 and the calibration connector 3, respectively. The loading mechanism 1 is connected to both the adjusting mechanism 2 and the calibration connector 3. The calibration connector 3 is connected to the shaft to be calibrated 4 by screws, and the angle sensor to be calibrated 5 is fixed on the shaft to be calibrated 4.

[0053] Figure 2 This is a schematic diagram of the loading mechanism of the present invention. Figure 2 As shown, the loading mechanism 1 of the present invention includes a circular grating encoder, a bearing 8, an internal shaft 9, and a torque motor 10. The circular grating disk 6 is connected to the reading head 7, forming a circular grating encoder. The circular grating encoder is used to acquire signals and output them to a host computer. The circular grating encoder is fixedly connected to the internal shaft 9 and measures the signal generated by the rotation of the internal shaft 9. The torque motor 10 is used to generate power to rotate the internal shaft 9. The loading mechanism 1 integrates a circular grating encoder (circular grating disk 6 and reading head 7) and is connected to the internal shaft 9 via a bearing, improving the reliability and stability of the circular grating encoder.

[0054] Figure 3 This is a schematic diagram of the adjustment mechanism of the present invention. Figure 3 The left side shows a perspective view, and the right side shows a non-perspective view. (Example) Figure 3 As shown, the adjustment mechanism 2 of the present invention includes a disk 11. The disk 11 has a protruding coaxial adjustment ring 16, and the coaxial adjustment ring 16 has openings on its side. Long screws 15 pass through these openings and contact the loading mechanism 1. Rotating each long screw 15 can adjust the position of the loading mechanism 1.

[0055] Figure 4 This is a physical diagram of an example of the loading mechanism and adjusting mechanism of the present invention. Figure 4The loading mechanism 1 is located on the left, and the adjusting mechanism 2 is on the right. Evenly distributed adjusting blocks 12 are mounted on the edge of the disc 11. The adjusting blocks 12 are located on the edge of the disc 11 and are connected to the support shaft 17 by screws. The adjusting blocks 12 have spherical grooves inside and hemispherical recesses inside, connecting reliably to the support shaft 17 and increasing rigidity. Rotating the adjusting blocks 12 adjusts the level of the adjusting mechanism 2, thereby adjusting the level of the entire mechanism and the angle of the internal shaft 9, thus ensuring the coaxiality requirement between the shaft to be calibrated 4 and the internal shaft 9. The adjusting blocks 12 are used for effective connection and adjustment of the positional relationship between the loading mechanism 1 and the shaft to be calibrated 4, ensuring their coaxiality requirements.

[0056] Figure 5 This is a schematic diagram of the calibration connector of the present invention. Figure 5 As shown, the calibration connector 3 of the present invention includes a circular block 13 and a curved thin plate 14. The upper end of the circular block 13 is connected to the loading mechanism 1, and the lower end is connected to the curved thin plate 14 by screws. The curved thin plate 14 is small in size, lightweight, and has a small moment of inertia. It has holes on both sides and is connected to the shaft 4 to be calibrated by screws, driving the shaft 4 to rotate. The circular block 13 is connected to the internal shaft 9 and the curved thin plate 14 by screws and bearings. The rotation of the internal shaft 9 drives the curved thin plate 14 to rotate. The calibration connector 3 is used to connect the loading mechanism 1 and the shaft 4 to be calibrated to rotate together, making the connection reliable, long-lasting, with low wear, and easy to adjust.

[0057] Figure 5 This is a flowchart of the artificial intelligence calibration method of the present invention. Figure 5 As shown, the in-situ calibration method for cornering based on artificial intelligence of the present invention uses the data collected by the circular grating encoder as the standard encoder as the standard value, and the error between the output value of the constructed BLSTM model and the standard value as the loss function. The loss function is minimized by the gradient descent method, and the parameters of the BLSTM model are optimized in reverse, thereby establishing the functional relationship between the angle sensor 5 to be calibrated and the circular grating encoder.

[0058] The artificial intelligence-based in-situ corner calibration method of the present invention includes the following specific contents:

[0059] Step 1, Model Building: First, construct a bidirectional long short-term memory network (BLSTM) model. The data collected by the angle sensor 5 to be calibrated, the forward and return marks, the zero-position angle difference between the angle sensor 5 to be calibrated and the circular encoder, and the angle difference between the angle sensor 5 to be calibrated and the circular encoder under a fixed sampling time are used as the model inputs. The output value is then fitted by an artificial intelligence algorithm.

[0060] First, a bidirectional long short-term memory network model (BLSTM model) is established. This model is input through an input layer, and the middle of the model consists of a BLSTM layer, a fully connected layer, and hidden layers between the two layers. The BLSTM model outputs the calculation results as output values ​​through an output layer.

[0061] Step 2, generate loss function: Perform static step measurement at the corner, use the data collected by the circular grating encoder as the standard value, and construct the loss function using the difference between the output value of the BLSTM model and the standard value;

[0062] A loss function is constructed using the difference between the output value of the BLSTM model and the standard value actually acquired by the circular grating encoder. The data acquired by the circular grating encoder is used as the standard value, and the error between the output value of the BLSTM model and the standard value actually acquired by the circular grating encoder is used as the loss function.

[0063] The model consists of a BLSTM layer, a fully connected layer, and hidden layers between the two layers. The BLSTM model outputs the computational results as output values ​​through the output layer. The output values ​​of the BLSTM model can be considered as standard values ​​fitted by artificial intelligence algorithms.

[0064] Step 3, optimize the model: perform dynamic continuous measurement of the turning angle, minimize the loss function using gradient descent, and optimize the BLSTM model parameters in reverse.

[0065] By minimizing the loss function using gradient descent and optimizing the model parameters in reverse, and through learning, training, and testing with a large number of samples, and through continuous iterative iteration, the connection weights and biases between neurons in each layer are optimized.

[0066] Step 4, obtain the functional relationship: establish the functional relationship between the angle sensor 5 to be calibrated and the circular grating encoder.

[0067] The BLSTM model can eventually fit the functional relationship between the angle sensor 5 to be calibrated and the circular grating encoder with ideal accuracy, thereby completing the in-situ calibration of the angle based on artificial intelligence.

[0068] The data testing method of this invention is as follows:

[0069] Test the zero-angle difference:

[0070] After connecting the loading mechanism 1 and the adjustment mechanism 2, when the connection between the calibration connector 3 and the internal shaft 9 is disconnected, turn on the servo of the torque motor 10 and control the torque motor 10 to rotate two revolutions, which drives the internal shaft 9 to rotate and causes the circular grating encoder to rotate. It is observed that the LED light of DSI changes from flashing green to solid green, indicating that the circular grating encoder has successfully found the zero position.

[0071] Connect the calibration connector 3 to the internal shaft 9, and use a level to adjust the level of the loading mechanism 1 by adjusting the adjusting block 12 of the adjusting mechanism 2, so that the internal shaft 9 in the center of the loading mechanism 1 is coaxial with the shaft 4 to be calibrated below.

[0072] Turn off the torque motor 10 servo, control the motor below the shaft to be calibrated 4 to rotate. The shaft to be calibrated 4 drives the internal shaft 9 to rotate, which in turn drives the circular encoder that has successfully found zero to rotate. Select the test result save path, start the acquisition, control the motor below the shaft to be calibrated 4 to rotate so that the angle sensor 5 to be calibrated returns to zero. After returning to zero, hold it at this position for a period of time. At this time, the difference between the measured value of the circular encoder (used as the standard sensor) and the angle sensor 5 to be calibrated is the interval angle between the zero positions of the upper and lower sensors (i.e., the zero position angle difference). Record it for later use in artificial intelligence calibration processing. Each subsequent rotation angle calibration experiment must start from the position after the motor returns to zero.

[0073] Static step measurement at corners:

[0074] Select the rotation direction and angle gradient, with the angle gradient step being A. Control the torque motor 10 to rotate at the selected speed and angle A, then stop. Hold this position for a period of time, then control the motor to rotate to the next gradient A and hold for a period of time, until the motor has rotated one full revolution. Hold this position for a period of time and then end the process. Simultaneously record the angle value of the circular grating encoder, the angle value of the angle sensor 5 to be calibrated, and the progress or return flag (progress flag is 1, return flag is -1) for later artificial intelligence calibration processing.

[0075] Dynamic continuous measurement of rotation angle:

[0076] Select the rotation speed and direction, and control the torque motor 10 to rotate continuously at the selected speed. After one revolution, maintain the rotation for a period of time to finish. At the same time, record the angle value of the circular grating encoder, the angle value of the angle sensor 5 to be calibrated, and the progress or return mark (progress mark is 1, return mark is -1) for later artificial intelligence calibration processing.

[0077] An example of the implementation steps of the optimization model in the artificial intelligence-based in-situ rotation calibration method of the present invention includes:

[0078] 1) Establish a BLSTM (Bidirectional Long Short-Time Memory) model, and read the angle value of the angle sensor 5 to be calibrated at time t, the forward and return marks, the zero-position angle difference between the angle sensor 5 to be calibrated and the circular encoder, and the angle difference between the angle sensor 5 to be calibrated and the circular encoder at time t as the model input. .

[0079] 2) Calculate the forgetting gate and select the information to be forgotten: ;

[0080] Input data Cell state C t Temporary cell state Output gate Hidden state Forgotten Gate Memory Gate W represents the weight of each node.

[0081] 3) Calculate the memory gate and select the information to be memorized:

[0082] ;in: For memory gate, This is a temporary cell state;

[0083] 4) Calculate the cell state at the current moment:

[0084] 5) Calculate the output gate and the hidden state at the current time: , ;in: For output gate, This represents the current state of the cell.

[0085] 6) Through forgetting and remembering information, useful information for subsequent time steps is passed on, while useless information is discarded, and the hidden state is output at each time step. Forgetting, remembering, and output are determined by the hidden state of the previous time step. and current input Calculated forget gate Memory Gate Output gate To control. Among them: , where is the activation function for each node.

[0086] 7) Take the output gate data at time t, the data from the previous 20 times, and the data from the next 20 times, and obtain the corresponding output value at the current time. .

[0087] 8) Set the loss function E: Where N is the number of output layer nodes. Output for the model Standard value.

[0088] 9) Backward calculation: Modify the weights layer by layer according to the gradient descent algorithm. and deviation : , Where: α and β are the learning efficiency and bias adjustment coefficients, respectively, and N is the number of nodes in layer j.

[0089] 10) Repeat the dynamic continuous measurement of the rotation angle multiple times, reread the new time input data, and repeat 1) to 9) until all the recorded data is processed. Through continuous loop iteration, the connection weights and deviations between neurons in each layer are optimized, so that the model can finally fit the functional relationship between the angle sensor 5 to be calibrated and the circular grating encoder as a circular grating encoder with ideal accuracy, thereby completing the in-situ calibration of the rotation angle based on artificial intelligence.

[0090] Therefore, this invention enables AI-based corner calibration without disassembly and in confined spaces.

[0091] The above embodiments are merely examples of applications of the present invention. Those skilled in the art can make various improvements and substitutions based on different design requirements and parameters without departing from the technical solution of the present invention, and these improvements and substitutions will also fall within the protection scope of the present invention.

Claims

1. An in-situ angle calibration device based on artificial intelligence, characterized in that, It includes a loading mechanism (1), an adjustment mechanism (2), a calibration connector (3), a shaft to be calibrated (4), and an angle sensor to be calibrated (5). The upper and lower ends of the adjustment mechanism (2) are the loading mechanism (1) and the calibration connector (3) respectively. The loading mechanism (1) is connected to the adjustment mechanism (2) and the calibration connector (3) respectively. The calibration connector (3) is connected to the shaft to be calibrated (4). The angle sensor to be calibrated (5) is fixed on the shaft to be calibrated (4). The loading mechanism (1) includes a circular grating encoder, a bearing (8), an internal shaft (9), and a torque motor (10). The circular grating disk (6) is connected to the reading head (7) to form a circular grating encoder. The circular grating encoder is fixedly connected to the internal shaft (9) through the bearing to measure the signal generated by the rotation of the internal shaft (9). The torque motor (10) is used to generate power to make the internal shaft (9) rotate. The adjustment mechanism (2) includes a disc (11) with a protruding coaxial adjustment ring (16) on the disc (11). The coaxial adjustment ring (16) has openings on its side, and long screws (15) pass through these openings to contact the loading mechanism (1). Rotating each long screw (15) can adjust the position of the loading mechanism (1). An adjusting block (12) is installed on the edge of the disc (11). The adjusting block (12) is connected to the support shaft (17). The adjusting block (12) has a spherical groove inside. By rotating the adjusting block (12), the level of the adjusting mechanism (2) can be adjusted, thereby adjusting the level of the entire mechanism and adjusting the angle of the internal shaft (9), thus ensuring the coaxiality requirement of the shaft (4) to be calibrated and the internal shaft (9). The calibration connector (3) is used to connect the loading mechanism (1) and the shaft to be calibrated (4) to rotate together. It includes a circular block (13) and a curved thin plate (14). The upper end of the circular block (13) is connected to the loading mechanism (1), and the lower end is connected to the curved thin plate (14). The curved thin plate (14) is connected to the shaft to be calibrated (4). The circular block (13) is connected to the internal shaft (9) through a bearing. The rotation of the internal shaft (9) drives the curved thin plate (14) to rotate, thereby driving the shaft to be calibrated (4) to rotate.

2. An in-situ corner calibration method based on artificial intelligence, characterized in that, It uses the artificial intelligence-based in-situ corner calibration device as described in claim 1, and includes the following steps: Step 1, Model building: First, construct a bidirectional long short-term memory network (BLSTM) model. The data collected by the angle sensor to be calibrated (5), the forward and return marks, the zero angle difference between the angle sensor to be calibrated (5) and the circular encoder, and the angle difference between the angle sensor to be calibrated (5) and the circular encoder under a fixed sampling time are used as the model inputs. The output value is then fitted by an artificial intelligence algorithm. Step 2, generate loss function: Perform static step measurement at the corner, use the data collected by the circular grating encoder as the standard value, and construct the loss function using the difference between the output value of the BLSTM model and the standard value; Step 3, optimize the model: perform dynamic continuous measurement of the turning angle, minimize the loss function using gradient descent, and optimize the BLSTM model parameters in reverse. Step 4, obtain the functional relationship: establish the functional relationship between the angle sensor (5) to be calibrated and the circular grating encoder.

3. The artificial intelligence-based in-situ angle calibration method according to claim 2, characterized in that, In step 1, a bidirectional long short-term memory network model, namely the BLSTM model, is first established. This BLSTM model is input through an input layer, and the middle of the model consists of BLSTM layers, fully connected layers, and hidden layers between the two layers. The BLSTM model outputs the calculation results as output values ​​through an output layer. The zero-position angle difference between the angle sensor (5) to be calibrated and the circular grating encoder is obtained as follows: After connecting the loading mechanism (1) and the adjustment mechanism (2), when the connection between the calibration connector (3) and the internal shaft (9) is disconnected, the servo of the torque motor (10) is turned on, and the circular grating encoder searches for the zero position. Connect the calibration connector (3) to the internal shaft (9), and use a level to adjust the level of the loading mechanism (1) by adjusting the adjusting block (12) of the adjusting mechanism (2), so that the internal shaft (9) in the center of the loading mechanism (1) is coaxial with the shaft (4) to be calibrated below; Turn off the torque motor (10) servo and control the motor below the shaft to be calibrated (4) to rotate. The shaft to be calibrated (4) drives the internal shaft (9) to rotate, which causes the circular grating encoder to rotate. This drives the circular grating encoder above, which has successfully found zero, to rotate, so that the angle sensor (5) to be calibrated returns to zero. At this time, the difference between the measured values ​​of the circular grating encoder, which is the standard sensor, and the angle sensor (5) to be calibrated is the interval angle between the zero positions of the two sensors, i.e., the zero position angle difference.

4. The artificial intelligence-based in-situ angle calibration method according to claim 2, characterized in that, In step 2, the static step measurement of the rotation angle includes: selecting the rotation direction and the rotation angle gradient, the step of the rotation angle gradient is A, controlling the torque motor (10) to rotate at the selected speed and then stop, holding it at that point for a period of time, controlling the motor to rotate to the next gradient A and then holding it for a period of time until the motor rotates one revolution, holding it for a period of time and then ending the process, while recording the angle value of the circular grating encoder, the angle value of the angle sensor (5) to be calibrated, and the progress or return mark.

5. The artificial intelligence-based in-situ angle calibration method according to claim 4, characterized in that, In step 3, the dynamic continuous measurement of the rotation angle includes: selecting the rotation speed and rotation direction, controlling the torque motor (10) to rotate continuously at the selected speed, maintaining it for a period of time after one rotation, and recording the angle value of the circular grating encoder, the angle value of the angle sensor (5) to be calibrated, and the progress or return mark.

6. The artificial intelligence-based in-situ angle calibration method according to claim 2, characterized in that, Its optimization model steps include: 1) Establish a BLSTM (Bidirectional Long Short-Term Memory) model, and read the angle value of the angle sensor (5) to be calibrated at time t, the forward and return marks, the zero-position angle difference between the angle sensor (5) to be calibrated and the circular encoder, and the angle difference between the angle sensor (5) to be calibrated and the circular encoder at time t as the model input. ; 2) Calculate the forgetting gate and select the information to be forgotten: ; 3) Calculate the memory gate and select the information to be memorized: ;in: For memory gate, This is a temporary cell state; 4) Calculate the cell state at the current moment: ; 5) Calculate the output gate and the hidden state at the current time: , ;in: For output gate, This represents the current state of the cell. 6) Through forgetting and remembering information, useful information for subsequent time steps is passed on, while useless information is discarded, and the hidden state is output at each time step. Forgetting, remembering, and output are determined by the hidden state of the previous time step. and current input Calculated forget gate Memory Gate Output gate To control, of which: , where is the activation function for each node; 7) Take the output gate data at time t, the data from the previous 20 times, and the data from the next 20 times, and obtain the corresponding output value at the current time. ; 8) Set the loss function E: Where N is the number of output layer nodes. Output for model Standard value; 9) Backward calculation: Modify the weights layer by layer according to the gradient descent algorithm. and deviation : , Where: α and β are the learning efficiency and bias adjustment coefficients, respectively, and N is the number of nodes in layer j; 10) Repeat the dynamic continuous measurement of the angle multiple times, reread the new time input data, and repeat steps 1) to 9) until all the recorded data is processed. Through continuous loop iteration, the connection weights and biases between neurons in each layer are optimized so that the model can finally fit the standard encoder with ideal accuracy, thereby completing the in-situ calibration of the angle based on artificial intelligence.

Images