Multi-factor torque measurement method based on selectable prior model identification

Abstract

The invention relates to a multi-factor torque measurement method based on selectable prior model identification, which comprises the following steps of: 1, acquiring a rotating speed parameter, a temperature parameter and a vibration parameter, researching the influence of the rotating speed on the torque measurement precision, and forming a multi-sensor parameter database; step 2, calculating the peak-to-peak value of the measured torque so as to conveniently judge whether the torque needs to be additionally corrected or not subsequently; and 3, based on the database in the step 1 and the peak-to-peak value calculated in the step 2, judging whether the torque needs to be additionally corrected or not. The method comprises the following steps: designing a multiple linear regression model, carrying out deep analysis on rotating speed, temperature, vibration and priori knowledge, adjusting model parameters to enable the model to be converged, verifying by using a test set, storing an optimal model, and inputting the actually measured rotating speed, temperature, vibration, priori knowledge and to-be-tested torque parameters into the optimal model. The target parameter high-precision torque can be obtained, and accurate calibration and optimization of the torque measurement deviation of the to-be-tested torque sensor are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical torque measurement, and in particular to a multi-factor torque measurement method based on optional priori model identification. Background Art

[0002] With the rapid development of industrial automation and intelligent manufacturing, torque sensors play an increasingly important role in various engineering applications, especially in the fields of automobiles, aerospace, robotics, energy, etc. Accurate torque measurement is crucial to product quality, system stability and safety. However, in practical applications, due to environmental factors and equipment complexity, traditional torque sensors face many challenges in high-precision measurement, especially when the correlation between vibration signals and torque signals is weak. How to ensure the accuracy of sensor data has become a technical problem that needs to be solved urgently.

[0003] Although the back propagation (BP) neural network can solve the problem of torque data correction to a certain extent, under multiple working conditions, the BP neural network has the problem of long training time. When it must be retrained for each working condition, the BP neural network will consume a lot of time and may have overfitting problems. In addition, the relationship between torque data and temperature, speed, vibration and other data is simple, and the BP neural network, as a powerful nonlinear model, is more likely to have overfitting problems when processing linear data.

[0004] In addition, previous studies have found that when the shaft vibrates within a certain frequency range, there is a significant deviation between the measured torque data and the standard torque data.

[0005] Traditional torque sensors face many challenges in high-precision measurement, especially when the correlation between vibration signals and torque signals is weak. How to ensure the accuracy of sensor data becomes a technical problem that needs to be solved urgently.

[0006] Under multiple working conditions, the BP neural network has the problem of long training time. If it must be retrained for each working condition, the BP neural network will consume a lot of time. In addition, since the relationship between torque data and temperature, speed, vibration and other data is simple, and the BP neural network is a powerful nonlinear model, it is more prone to overfitting problems when processing linear data.

[0007] Finally, when the shaft vibrates within a certain frequency range, there is a significant deviation between the measured torque data and the standard torque data. Summary of the invention

[0008] In order to solve the above technical problems, the present invention proposes a torque sensor performance evaluation method for a multi-factor torque measurement method with optional prior model identification. The method can solve the problems of long training time and poor fitting effect of BP neural network during torque measurement. At the same time, based on prior knowledge, the present invention solves the problem of increased deviation in shaft torque measurement within certain specific frequency ranges.

[0009] To achieve the above object, the present invention is implemented by the following technical solution: a multi-factor torque measurement method based on optional prior model identification, the steps of which are:

[0010] Step 1: Collect speed parameters, temperature parameters, and vibration parameters to study the influence of speed on torque measurement accuracy and form a multi-sensor parameter database;

[0011] Step 2: Calculate the peak-to-peak value of the measured torque to facilitate subsequent judgment on whether additional correction of the torque is required;

[0012] Step 3: Based on the database in step 1 and the peak-to-peak value calculated in step 2, determine whether the torque needs additional correction.

[0013] Furthermore, when the peak-to-peak value of the torque is less than 0.5, it is judged that the torque does not need additional correction, and the speed signal, temperature signal, and vibration signal are directly used as independent variables of the multivariate linear regression; when the peak-to-peak value is greater than 0.5, it is judged that the torque needs additional correction.

[0014] Furthermore, in order to solve the problem of large deviation in torque measurement, the frequency domain information obtained when the torque deviation of the shaft is large is subjected to inverse Fourier transform to obtain time domain information, which is used as prior knowledge and input into the multivariate linear regression model.

[0015] Furthermore, a multivariate linear regression model is designed. First, the data is normalized. The normalization calculation formula is as follows:

[0016]

[0017] In the formula, mean(x) is the average operation, and std(x) is the variance operation;

[0018] Secondly, the measured torque signal, speed signal, temperature signal, vibration signal and prior knowledge are used as model inputs, and the regression coefficients are estimated using the least squares method. 2 The indicators are used to evaluate the fitting effect of the model. Through repeated experiments, the learning rate and regularization terms are adjusted to make the model achieve the best effect. The generalization ability of the model is evaluated through K-fold cross validation, and the fitted multivariate linear regression model is used to predict the new torque data.

[0019] The beneficial effects of the present invention are:

[0020] The present invention designs a multivariate linear regression model, performs in-depth analysis of the speed, temperature, vibration, and prior knowledge, adjusts the model parameters to achieve model convergence, and saves the best model after verification with a test set, thereby obtaining a multi-sensor torque measurement mathematical model that meets and covers multiple working conditions, and avoids the problems of long time consumption and easy overfitting that may be caused by traditional neural networks. The measured speed, temperature, vibration, prior knowledge, and the torque parameters to be tested are input into the optimal model, and the target parameter high-precision torque can be obtained, achieving accurate calibration and optimization of the torque measurement deviation of the torque sensor to be tested. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a flow chart of the torque measurement method of the present invention based on multi-factors of optional prior model identification. DETAILED DESCRIPTION

[0022] The following is a clear and complete description of the technical solutions in the embodiments of the present application in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0023] In the following description, many specific details are set forth to facilitate a full understanding of the present application, but the present application may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present application. Therefore, the present application is not limited to the specific embodiments disclosed below.

[0024] Example: Figure 1 As shown, a multi-factor torque measurement method based on optional prior model identification of the present invention comprises the following steps:

[0025] Step 1: Design a torque measurement experiment plan based on prior knowledge and rotational speed, integrate the rotational speed sensor, temperature sensor, acceleration sensor, high-precision torque sensor and the torque sensor to be measured, and study the influence of rotational speed on the torque measurement accuracy.

[0026] Step 2: Calculate the peak-to-peak value of the measured torque to facilitate subsequent judgment on whether additional correction of the torque is required.

[0027] Step 3: Establish a multivariate linear regression model. Based on the speed signal, temperature signal, vibration signal and prior knowledge, a multivariate linear regression model is designed to process the complex relationship between these data and torque measurement. The multivariate linear regression model can effectively process data through its structure, and is suitable for processing the linear dynamic relationship between speed signal, temperature signal, vibration signal and prior knowledge and torque. First, the data is normalized to eliminate the differences between different sensor dimensions and ensure the validity of the data input. The normalization operation can be processed using the mean-variance normalization formula, which is as follows:

[0028]

[0029] Then the model is designed. When designing the multivariate linear regression model, the feature data in the training set and the target torque signal are used to establish the multivariate linear regression model. The basic form of the regression model is:

[0030] y=β0+β1x1+β2x2+β3x3+β4x4+ε

[0031] Among them, y is the predicted torque signal, x1, x2, x3, x4 are input features such as vibration, temperature, and speed, β0 is the intercept, β1, β2, β3, β4 are regression coefficients, and ε is the error term.

[0032] The least squares method is used to optimize the regression coefficients to minimize the error between the predicted value and the true value. The optimization process of the regression coefficients is achieved by the gradient descent method. To avoid the overfitting problem, Ridge Regression is used to perform L2 regularization on the regression coefficients. The model complexity can be controlled by adjusting the regularization parameter λ.

[0033] The trained model is evaluated using the test set, mainly using the mean square error (MSE) and the coefficient of determination (R 2 ) as a performance indicator. MSE is used to measure the difference between the model prediction value and the actual value, R 2 This reflects the model's ability to explain data variation. The K-fold cross-validation method is used to further validate the model to ensure the generalization ability of the model. By training and validating the model on different data subsets, the evaluation bias caused by a single partition can be avoided. Based on the evaluation results, the model is further optimized.

[0034] After the model training is completed, the model can be connected to new sensor data to predict the torque signal. The torque signal predicted by the model is compared with the actual collected torque signal to analyze the prediction accuracy of the model.

[0035] After sufficient training and verification, a high-precision multivariate linear regression model is obtained. This model can calibrate the measurement data of the torque sensor in practical applications, eliminate the measurement errors caused by environmental and operating conditions, and improve the accuracy of torque measurement. Finally, the optimized model will be saved and applied to the measured data to achieve accurate calibration and optimization of the torque measurement deviation of the torque sensor to be tested.

[0036] Specifically, the present invention provides necessary simulation experimental conditions by establishing a torque measurement device that takes into account rotational speed data, temperature data, vibration data and prior knowledge, and combining a rotational speed sensor, a temperature sensor, an acceleration sensor, a high-precision torque sensor and a torque sensor to be tested.

[0037] The speed data and prior knowledge are deeply analyzed. Through repeated experiments and model training, the multivariate linear regression model continuously learns and optimizes the linear relationship between speed data, temperature data, vibration data, prior knowledge and torque, and adjusts the model's hyperparameters, ultimately obtaining a high-precision torque measurement model that is suitable for different working conditions. This model can accurately calibrate and optimize the measurement deviation of the torque sensor to be measured.

[0038] After the model training is completed, the model is verified by using an independent test set to ensure that the model has good generalization ability and high accuracy. In practical applications, the model integrates the speed data and prior knowledge with the torque data to be measured, and outputs high-precision torque measurement results to achieve the purpose of accurate calibration and optimization of torque measurement deviation.

Claims

1. A multi-factor torque measurement method based on optional prior model identification, characterized in that: The steps are: Step 1: Collect speed parameters, temperature parameters, and vibration parameters to study the influence of speed on torque measurement accuracy and form a multi-sensor parameter database; Step 2: Calculate the peak-to-peak value of the measured torque to facilitate subsequent judgment on whether additional correction of the torque is required; Step 3: Based on the database in step 1 and the peak-to-peak value calculated in step 2, determine whether the torque needs additional correction.

2. The multi-factor torque measurement method based on optional prior model identification according to claim 1, characterized in that: When the peak-to-peak value of the torque is less than 0.5, it is judged that no additional correction is needed for the torque, and the speed signal, temperature signal, and vibration signal are directly used as independent variables for multivariate linear regression; when the peak-to-peak value is greater than 0.5, it is judged that additional correction is needed for the torque.

3. The multi-factor torque measurement method based on optional prior model identification according to claim 1, characterized in that: In order to solve the problem of large deviation in torque measurement, the frequency domain information obtained when the torque deviation of the shaft is large is subjected to inverse Fourier transform to obtain time domain information, which is used as prior knowledge and input into the multivariate linear regression model.

4. The multi-factor torque measurement method based on optional prior model identification according to claim 3 is characterized in that: To design a multivariate linear regression model, first, normalize the data. The normalization calculation formula is as follows: In the formula, mean(x) is the operation of taking the average value, and std(x) is the operation of taking the variance; Secondly, the measured torque signal, speed signal, temperature signal, vibration signal and prior knowledge are used as model inputs, and the regression coefficients are estimated using the least squares method. 2 The indicators are used to evaluate the fitting effect of the model. Through repeated experiments, the learning rate and regularization terms are adjusted to make the model achieve the best effect. The generalization ability of the model is evaluated through K-fold cross validation, and the fitted multivariate linear regression model is used to predict the new torque data.

Images