Artificial Intelligence-Based Torque In-Situ Calibration Device and Method

Abstract

This invention provides an artificial intelligence-based torque in-situ calibration device and method, comprising: a calibration control unit, a loading mechanism, a loading arm, a standard torque sensor, a coupling, and a shaft to be calibrated; one end of the loading arm contacts the upper end of the loading mechanism to apply torque; the other end of the loading arm is equipped with a standard torque sensor, which is connected to the shaft to be calibrated via the coupling; a measured torque sensor is installed in the shaft to be calibrated; the calibration control unit is used to acquire first torque data collected by the measured torque sensor, input the torque data into a pre-trained torque in-situ calibration model to determine second torque data of the measured torque sensor, and perform compensation calibration on the measured torque sensor based on the second torque data. In this invention, when the torque motor drives the shaft to be calibrated to rotate, static torque step measurement and dynamic torque loading and unloading measurement are performed, enabling performance calibration under both short-term and long-term operation.

Description

Technical Field

[0001] This invention relates to the field of testing and calibration technology, and more specifically, to an artificial intelligence-based torque in-situ calibration device and method. Background Technology

[0002] With the development of my country's equipment manufacturing industry, more and more equipment cannot be calibrated by torque sensors alone due to limited space or complicated disassembly. In addition, the application of torque sensors is becoming more and more widespread. Inventing a reliable in-situ torque calibration method 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, and the training efficiency of neural networks has become increasingly higher. When processing some repetitive data, the efficiency is greatly improved compared with manual methods. Therefore, using neural network models to fit some nonlinear complex function relationships is a hot research direction for the future.

[0004] A search of existing technologies revealed a patent with publication number CN112539876A entitled "A Vertical Reference Torque Calibration Method and Device for Extreme Environments". This patent uses the difference between the measured standard torque value and the torque value to be calibrated for calibration, without using a large amount of training data to fit the function curve of the torque sensor to be tested. Since only the difference between the two is considered, the calibration effect needs to be improved. Summary of the Invention

[0005] In view of the deficiencies in the prior art, the purpose of this invention is to provide an artificial intelligence-based torque in-situ calibration device and method.

[0006] The artificial intelligence-based torque in-situ calibration device provided by the present invention includes: a calibration control unit, a loading mechanism, a loading arm, a standard torque sensor, a coupling, and a shaft to be calibrated;

[0007] One end of the loading arm contacts the upper end of the loading mechanism to achieve torque loading;

[0008] A standard torque sensor is installed at the other end of the loading arm, and the standard torque sensor is connected to the shaft to be calibrated via a coupling.

[0009] The shaft to be calibrated is equipped with a torque sensor.

[0010] The calibration control unit is used to acquire the first torque data collected by the torque sensor under test, input the torque data into a pre-trained torque in-situ calibration model to determine the second torque data of the torque sensor under test, and perform compensation calibration on the torque sensor under test based on the second torque data.

[0011] Preferably, the loading arm includes a spring assembly, a loading rod, and a flange;

[0012] One end of the loading rod is connected to the flange, and the other end is provided with a spring assembly;

[0013] The flange is connected to the upper end face of the standard torque sensor, and the spring assembly is in contact with the upper end of the loading mechanism.

[0014] Preferably, the spring assembly includes a ball-head push rod, a compression spring, a pressure cap, a screw, a pressure plate, and a base;

[0015] The base and pressure plate are fixed on the loading rod and are located at both ends of the screw;

[0016] A compression spring is provided between the ball-head push rod and the pressure cap, that is, the ball-head push rod is connected to the pressure cap through the compression spring; the ball-head push rod, the pressure cap and the compression spring are arranged in the base.

[0017] Preferably, the loading mechanism includes a drive shaft, bearings, a motor, a flange, and a cam;

[0018] A bearing and a motor are arranged axially on the drive shaft; the drive shaft is disposed in the inner ring of the bearing.

[0019] The outer end face of the drive shaft is fixed to a flange; a cam is provided on the flange; the cam and the ball head push rod on the loading arm are in point contact;

[0020] The motor sequentially applies a torque to the ball head push rod via a drive shaft, flange, and cam, thereby driving the force of the loading arm.

[0021] Preferably, the training of the torque in-situ calibration model includes the following steps:

[0022] Establish a DNN model;

[0023] The data collected by the torque sensor to be calibrated is input into the DNN model through the input layer, and the calculation results are output through the output layer.

[0024] The data collected by the standard torque sensor is used as the standard value. The error between the calculated result output by the model and the standard value is used as the loss function. The loss function is minimized by the gradient descent method to optimize the model parameters in reverse.

[0025] Through continuous iteration, the DNN model can eventually fit the functional relationship between the sensor to be calibrated and the standard sensor to generate the torque in-situ calibration model.

[0026] Preferably, the DNN model includes an input layer, a hidden layer, and an output layer;

[0027] The hidden layer consists of three fully connected layers and one dropout layer.

[0028] Preferably, the radial loading force of the loading mechanism is not less than 1363N.

[0029] The artificial intelligence-based in-situ torque calibration method provided by this invention includes the following steps:

[0030] Step M1: Establish a deep neural network model, read the value measured by the torque sensor to be calibrated at time t, and use it together with the infeed and return marks as the model input quantity x. t ;

[0031] Step M2: Forward calculation of the output results of each layer of the model: Among them, the output Output data for the j-th node in the i-th layer. Let the connection weight be the link between the j-th node in the i-th layer and the node in the (i-1)-th layer. Let j be the data vector of the j-th node in the (i-1)-th layer. Let f(x) be the bias vector of the j-th node in the i-th layer, and f(x) be the activation function of the neuron.

[0032] Step M3: Retrieve the data at time t, the data from the 20 times before and the 20 times after this time, and the corresponding output value y. t ;

[0033] Step M4: Set the loss function E: Where N is the number of output layer nodes. Output y for the model j Standard value;

[0034] Step M5: Reverse calculation, modifying the weights W layer by layer according to the gradient descent algorithm. j and deviation b j ; Where: α and β are the learning efficiency and bias adjustment coefficients, respectively, and N is the number of nodes in layer j;

[0035] Step M6: Reread the new input data at the new time and repeat steps M1 to M5 until all the recorded data has been processed. Through continuous iteration, the connection weights and biases between neurons in each layer are optimized, so that the model can fit the standard torque sensor with the preset accuracy. Then, the measured torque sensor is manually compensated, and the in-situ torque calibration based on artificial intelligence is completed.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] In this invention, when the torque motor drives the shaft to be calibrated to rotate, static torque step measurement and dynamic torque loading and unloading measurement are performed, which can realize performance calibration during short-term and long-term operation.

[0038] This invention uses a large amount of data when training the neural network. The input data is not only the angle value measured by the sensor to be calibrated, but also the forward and return marks. Since a single measurement may have errors, 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, thereby improving the reliability of the calibration effect. Attached Figure Description

[0039] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0040] Figure 1 This is a schematic diagram of the artificial intelligence-based torque in-situ calibration device in an embodiment of the present invention;

[0041] Figure 2 This is a schematic diagram of the loading arm structure in an embodiment of the present invention;

[0042] Figure 3 This is a schematic diagram of the spring assembly in an embodiment of the present invention;

[0043] Figure 4 This is a schematic diagram of the loading mechanism in an embodiment of the present invention;

[0044] Figure 5 This is a flowchart of the artificial intelligence calibration method in an embodiment of the present invention. Detailed Implementation

[0045] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention. These all fall within the scope of protection of the present invention.

[0046] Figure 1 This is a schematic diagram of the structure of the artificial intelligence-based torque in-situ calibration device in an embodiment of the present invention, as shown below. Figure 1 As shown, the artificial intelligence-based torque in-situ calibration device provided by the present invention includes a loading mechanism 1, a loading arm 2, a standard torque sensor 3, a coupling 4, and a shaft to be calibrated 5.

[0047] One end of the loading arm 2 contacts the upper end of the loading mechanism 1 to apply torque, and the other end is equipped with a standard torque sensor 3, which is connected to the shaft 5 to be calibrated via a coupling 4.

[0048] The shaft 5 to be calibrated is equipped with a torque sensor.

[0049] Figure 2 This is a schematic diagram of the loading arm structure in an embodiment of the present invention, as shown below. Figure 2 As shown, the loading arm 2 includes a spring assembly 201, a loading rod 202, and a flange 203;

[0050] One end of the loading rod 202 is connected to the flange 203 by bolts, and the other end is provided with a spring assembly 201;

[0051] The flange 203 is threaded to the upper end face of the standard torque sensor 3, and the spring assembly 201 is in contact with the upper end of the loading mechanism 1.

[0052] The length of the loading arm 2 is 2.2m.

[0053] Figure 3 This is a schematic diagram of the spring assembly in an embodiment of the present invention, as shown below. Figure 3 As shown, the spring assembly 201 includes a ball-head push rod 211, a compression spring 212, a pressure cap 213, a screw 214, a pressure plate 215, and a base 216;

[0054] The base 216 and the pressure plate 215 are fixed on the loading rod 202 and are located at both ends of the screw 214;

[0055] A compression spring 212 is provided between the ball head rod 211 and the pressure cap 213, that is, the ball head rod 211 is connected to the pressure cap 213 through the compression spring 212; the ball head rod 211, the pressure cap 213 and the compression spring 212 are arranged in the base 216.

[0056] Figure 4 This is a schematic diagram of the loading mechanism in an embodiment of the present invention, as shown below. Figure 4 , Figure 1 As shown, the loading mechanism includes a drive shaft 201, a bearing 202, a motor 203, a flange 204, and a cam 205;

[0057] A bearing 202 and a motor 203 are arranged axially on the drive shaft 201; the drive shaft 201 is disposed in the inner ring of the bearing 202.

[0058] The end face of the drive shaft 201 is fixed to the flange 204; a cam 205 is provided on the flange 204; the cam 205 and the ball head push rod 211 on the loading arm 2 are in point contact;

[0059] The motor 203 sequentially applies a loading torque to the ball head push rod 211 via the drive shaft 201, flange 204, and cam 205, thereby driving the loading arm 2.

[0060] In this embodiment of the invention, the radial loading force of the loading mechanism is not less than 1363N.

[0061] This invention enables point-to-point force application, integrates all components inside the loading mechanism, saves space, and is easy to operate;

[0062] The specific implementation steps of the artificial intelligence-based torque in-situ calibration device provided by this invention are as follows:

[0063] Step S1: Install the standard torque sensor and the torque sensor to be measured. Without connecting a load, control the motor to move and collect the torque values ​​of the two torque sensors at different points. Verify that the measured values ​​are independent of the installation angle and position and have good repeatability.

[0064] Step S2: Install the loading mechanism and loading arm. The loading mechanism and the indexing mechanism test bench rotate relative to each other to achieve torque change. Turn on the torque protection switch to prevent excessive torque from damaging the indexing mechanism.

[0065] Step S3: Select the rotation speed, determine the rotation direction, peak torque loading, and torque loading / unloading gradient. Zero both sensors and start data acquisition. Control the motor to rotate continuously at the selected speed. Stop loading whenever the loading gradient is reached, maintain for a period of time, and then continue loading until the torque reaches the determined peak value, which is then maintained for a period of time. Subsequently, control the motor to unload in reverse. Stop unloading whenever the unloading gradient is reached, maintain for a period of time, and then continue unloading until the torque value is zero. Maintain for a period of time and then end the experiment. Simultaneously record the torque value of the standard torque sensor, the torque value of the torque sensor under test, and the progress or return marker for later artificial intelligence calibration processing.

[0066] Step S4: Select the rotation speed, determine the rotation direction and peak torque loading, zero both sensors and start data acquisition; control the motor to rotate continuously at the selected speed until the torque reaches the determined peak value, stop the movement, maintain the peak value for a period of time, then control the motor to reverse and unload until the torque value is zero, maintain it for a period of time and end the experiment; at the same time, record the torque value of the standard torque sensor, the torque value of the torque sensor under test, and the progress or return mark for later artificial intelligence calibration processing.

[0067] The implementation steps of the artificial intelligence-based in-situ torque calibration method provided by this invention are as follows:

[0068] Step M1: Establish a deep neural network model, read the value measured by the torque sensor to be calibrated at time t, and use it together with the infeed and return marks as the model input quantity x. t ;

[0069] Step M2: Forward calculation of the output results of each layer: The output Output data for the j-th node in the i-th layer. Let the connection weight be the link between the j-th node in the i-th layer and the node in the (i-1)-th layer. The data vector of the j-th node in the (i-1)th layer (which is the original input x when it is in the first node) t ), Let f(x) be the bias vector of the j-th node in the i-th layer, and f(x) be the activation function of the neuron.

[0070] Step M3: Retrieve the output value y corresponding to the data at time t, the data from the previous 20 times, and the data from the next 20 times. t ;

[0071] Step M4: Set the loss function E: Where N is the number of output layer nodes. Output y for the model j Standard value;

[0072] Step M5: Reverse calculation, modifying the weights W layer by layer according to the gradient descent algorithm. j and deviation b j ; Where: α and β are the learning efficiency and bias adjustment coefficients, respectively, and N is the number of nodes in layer j.

[0073] Step M6: Reread the new input data at the new time and repeat steps M1 to M5 until all the recorded data has been processed. Through continuous iteration, the connection weights and biases between neurons in each layer are optimized, so that the model can fit the standard torque sensor with ideal accuracy, thereby achieving manual compensation and completing the in-situ torque calibration based on artificial intelligence.

[0074] The specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various modifications or variations within the scope of the claims, which do not affect the essence of the present invention.

Claims

1. An artificial intelligence-based torque in-situ calibration device, characterized in that, include: The calibration control unit, loading mechanism, loading arm, standard torque sensor, coupling, and shaft to be calibrated; One end of the loading arm contacts the upper end of the loading mechanism to achieve torque loading; A standard torque sensor is installed at the other end of the loading arm, and the standard torque sensor is connected to the shaft to be calibrated via a coupling. The shaft to be calibrated is equipped with a torque sensor. The calibration control unit is used to acquire the first torque data collected by the torque sensor under test, input the torque data into a pre-trained torque in-situ calibration model to determine the second torque data of the torque sensor under test, and perform compensation calibration on the torque sensor under test based on the second torque data; the loading arm includes a spring assembly, a loading rod, and a flange. One end of the loading rod is connected to the flange, and the other end is provided with a spring assembly; The flange is connected to the upper end face of the standard torque sensor, and the spring assembly is in contact with the upper end of the loading mechanism; The spring assembly includes a ball-head push rod, a compression spring, a pressure cap, a screw, a pressure plate, and a base; The base and pressure plate are fixed on the loading rod and are located at both ends of the screw; A compression spring is provided between the ball-head push rod and the pressure cap, that is, the ball-head push rod is connected to the pressure cap through the compression spring; the ball-head push rod, the pressure cap and the compression spring are arranged in the base.

2. The artificial intelligence-based in-situ torque calibration device according to claim 1, characterized in that, The loading mechanism includes a drive shaft, bearings, a motor, a flange, and a cam; A bearing and a motor are arranged axially on the drive shaft; the drive shaft is disposed in the inner ring of the bearing. The outer end face of the drive shaft is fixed to a flange; a cam is provided on the flange; the cam and the ball head push rod on the loading arm are in point contact; The motor sequentially applies a torque to the ball head push rod via a drive shaft, flange, and cam, thereby driving the force of the loading arm.

3. The artificial intelligence-based in-situ torque calibration device according to claim 1, characterized in that, The training of the torque in-situ calibration model includes the following steps: Establish a DNN model; The data collected by the torque sensor to be calibrated is input into the DNN model through the input layer, and the calculation results are output through the output layer. The data collected by the standard torque sensor is used as the standard value. The error between the calculated result output by the model and the standard value is used as the loss function. The loss function is minimized by the gradient descent method to optimize the model parameters in reverse. Through continuous iteration, the DNN model can eventually fit the functional relationship between the sensor to be calibrated and the standard sensor to generate the torque in-situ calibration model.

4. The artificial intelligence-based in-situ torque calibration device according to claim 3, characterized in that, The DNN model includes an input layer, hidden layers, and an output layer; The hidden layer comprises three fully connected layers and one dropout layer.

5. The artificial intelligence-based in-situ torque calibration device according to claim 1, characterized in that, The radial loading force of the loading mechanism is not less than 1363N.

6. A torque in-situ calibration method based on the artificial intelligence-based torque in-situ calibration device according to any one of claims 1 to 5, characterized in that, Includes the following steps: Step M1: Establish a deep neural network model, read the value measured by the torque sensor to be calibrated at time t, and use it together with the infeed and return marks as the model input. Step M2: Forward calculation of the output results of each layer of the model: , , where the output For the first i The first layer j Each node outputs data. For the first i The first layer j The node and the first i-1 The connection weights between layer nodes, For the first i-1 Layer j Each node data vector For the first i Layer j Each node deviation vector, For neuron activation functions; Step M3: Get the current output t The data at this moment, along with the data from the 20 moments before and the 20 moments after this moment, and the corresponding output value at this moment. Step M4: Set the loss function E : ;in, N This represents the number of nodes in the output layer. Output for the model Standard value; Step M5: Reverse calculation, modifying the weights layer by layer according to the gradient descent algorithm. and deviation ; , ;in: α , β These are the learning efficiency and the deviation adjustment coefficient, respectively. N for j Number of layer nodes; Step M6: Reread the new input data at the new time and repeat steps M1 to M5 until all the recorded data has been processed. Through continuous iteration, the connection weights and biases between neurons in each layer are optimized, so that the model can fit the standard torque sensor with the preset accuracy. Then, the measured torque sensor is manually compensated, and the in-situ torque calibration based on artificial intelligence is completed.

Images