Strain gauge calibration method, device, equipment, system and storage medium

Abstract

The invention provides a strain gauge calibration method, device, equipment and system and a storage medium, and relates to the technical field of calibration of measuring instruments. The method comprises the following steps: calculating a difference value between a displacement measurement value measured by a displacement measurement device and a strain target value of a measured strain gauge; performing multi-scale feature extraction and one-dimensional feature extraction on the difference value by using an intelligent control model formed based on a Res2Net block and a Flatten layer, and obtaining a predicted duty ratio and a predicted pulse number according to a multi-scale feature extraction result and a one-dimensional feature extraction result; controlling the servo motor according to the predicted duty ratio and the predicted pulse number so as to update the displacement measurement value; and acquiring a strain output value of the measured strain gauge based on the strain acquisition device when the updated displacement measurement value is equal to the strain target value, and calibrating the measured strain gauge according to the strain output value. According to the invention, the problems of insufficient precision and limited intelligent control degree, such as overshoot and jitter, possibly existing in the calibration process can be effectively avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of calibration of measuring instruments for building deformation monitoring, and particularly relates to a calibration method, device, equipment, system and storage medium for a strain gauge. Background Art

[0002] The vibrating wire strain gauge is an important measuring instrument widely used in the field of deformation monitoring of building structures such as bridges, tunnels, and slopes, and can accurately measure the strain and stress changes of the structure. Therefore, the metrological calibration and value traceability of such instruments are very crucial, and the importance of developing their calibration devices is highlighted.

[0003] Currently, the existing strain gauge calibration devices are divided into the following four categories according to their driving methods: The first category: manually driven screw transmission type; The second category: motor-driven lead screw nut transmission type; The third category: piezoelectric ceramic direct drive type; The fourth category: hydraulic component direct drive lead screw nut transmission type. The first category of manual drive uses a threaded screw or a lever device to manually rotate or adjust the calibration force loading system. This method is highly dependent on the technical level of the operator. In addition, the manual calibration time is long and the efficiency is low, which is not suitable for large-scale calibration work. Moreover, after collecting the calibration data, additional time is required to import the data into relevant data processing software for advanced data processing. The second, third, and fourth calibration methods drive the loading mechanism through a stepping motor or a hydraulic mechanism. Although manual operation can be reduced to a certain extent, the calibration usually relies on a control algorithm with fixed parameters and cannot dynamically optimize the calibration process according to the characteristics of the measured strain gauge. Problems such as overshoot, jitter, or inconsistent displacement may occur during the calibration process. At the same time, such systems usually lack intelligent control means and cannot dynamically adjust the calibration strategy according to the characteristics of the calibrated strain gauge. Summary of the Invention

[0004] Embodiments of the present invention provide a calibration method, device, equipment, system and storage medium for a strain gauge to solve the problems of insufficient accuracy such as overshoot and jitter and limited intelligent control degree that may exist in the existing strain gauge calibration system during the calibration process.

[0005] In a first aspect, an embodiment of the present invention provides a calibration method for a strain gauge, which is applied to a strain gauge calibration system. The strain gauge calibration system includes a clamping device, a sliding device, a power device, a displacement measuring device, and a strain acquisition device. The clamping device is used to fix the measured strain gauge and the displacement measuring device and connect the sliding device. The power device drives the sliding device to generate a displacement through a servo motor, so as to generate a micro-strain on the measured strain gauge. The method includes:

[0006] Obtain the displacement measurement value measured by the displacement measuring device, and calculate the difference between the displacement measurement value and the strain target value of the measured strain gauge;

[0007] Using an intelligent control model composed of a Res2Net block and a Flatten layer, perform multi-scale feature extraction and one-dimensional feature extraction on the difference respectively, and obtain a predicted duty cycle and predicted pulse number according to the multi-scale feature extraction result and the one-dimensional feature extraction result;

[0008] Control the servo motor according to the predicted duty cycle and the predicted pulse number to update the displacement measurement value so that the updated displacement measurement value is equal to the strain target value;

[0009] Based on the strain acquisition device, obtain the strain output value of the measured strain gauge when the updated displacement measurement value is equal to the strain target value, and calibrate the measured strain gauge according to the strain output value.

[0010] In a possible implementation manner, the using an intelligent control model composed of a Res2Net block and a Flatten layer, performing multi-scale feature extraction and one-dimensional feature extraction on the difference respectively, and obtaining a predicted duty cycle and predicted pulse number according to the multi-scale feature extraction result and the one-dimensional feature extraction result includes:

[0011] Expand the difference into a high-dimensional feature vector based on the input layer;

[0012] Perform multi-scale feature extraction on the high-dimensional feature vector based on the Res2Net block to obtain a multi-scale feature extraction result;

[0013] Perform one-dimensional feature extraction on the high-dimensional feature vector based on the Flatten layer to obtain a one-dimensional feature extraction result;

[0014] Obtain a predicted duty cycle and predicted pulse number according to the multi-scale feature extraction result and the one-dimensional feature extraction result.

[0015] In a possible implementation manner, the performing multi-scale feature extraction on the high-dimensional feature vector based on the Res2Net block to obtain a multi-scale feature extraction result includes:

[0016] Perform multi-scale feature extraction on the high-dimensional feature vector based on the Res2Net block to obtain high-dimensional features;

[0017] Perform weighted fusion on the features in different subspaces of the high-dimensional features based on the multi-head attention mechanism and spatial feature attention to obtain fused features;

[0018] Add Gaussian noise to the fused features based on the first Gaussian noise generator for Gaussian smoothing to obtain smoothed fused features;

[0019] Perform dimensionality reduction on the smoothed fusion feature based on the first fully connected layer to obtain an initial multi-scale feature extraction result;

[0020] Screen the initial multi-scale feature extraction result based on the first regularization layer to obtain a multi-scale feature extraction result.

[0021] In a possible implementation manner, the performing one-dimensional feature extraction on the high-dimensional feature vector based on the Flatten layer to obtain a one-dimensional feature extraction result includes:

[0022] Flatten the high-dimensional feature vector into a one-dimensional vector based on the Flatten layer to obtain a one-dimensional feature;

[0023] Perform a non-linear transformation on the one-dimensional feature based on the second fully connected layer to obtain a non-linear transformation feature;

[0024] Perform feature fusion on the non-linear transformation feature based on the feature fusion layer to obtain a fused transformation feature;

[0025] Enhance the fused transformation feature based on the residual connection layer with channel and spatial attention mechanisms to obtain an enhanced fused feature;

[0026] Map the enhanced fused feature to the target dimension based on the third fully connected layer to obtain a target feature;

[0027] Screen the target feature based on the second regularization layer to obtain a screened target feature;

[0028] Add Gaussian noise to the screened target feature based on the second Gaussian noise generator for Gaussian smoothing to obtain a smoothed target feature;

[0029] Screen the smoothed target feature based on the third regularization layer to obtain a one-dimensional feature extraction result.

[0030] In a possible implementation manner, obtaining a predicted duty cycle and predicted pulse number according to the multi-scale feature extraction result and the one-dimensional feature extraction result includes:

[0031] Connect the multi-scale feature extraction result and the one-dimensional feature extraction result based on the fourth fully connected layer to obtain an aggregated feature;

[0032] Perform adaptive normalization on the aggregated feature based on the adaptive normalization layer to obtain a normalized feature;

[0033] Screen the normalized feature based on the fourth regularization layer to obtain a predicted duty cycle and predicted pulse number.

[0034] In a possible implementation manner, the training process of the intelligent control model is:

[0035] Obtain a training data set, where the training data set includes multiple groups of training data. Each group of training data includes the historical difference between the historical displacement measurement value measured by the displacement measurement device and the historical strain target value of the measured strain gauge during the historical calibration process of the strain gauge calibration system, and the actual duty cycle and actual number of pulses required for the servo motor to move when the servo motor is used to control the historical displacement measurement value to be equal to the historical strain target value;

[0036] Construct a loss function based on the error dynamic constraint model;

[0037] Train the intelligent control model based on the training data set and the loss function;

[0038] The error dynamic constraint model includes:

[0039]

[0040] where T loss is the error dynamic constraint model, δ min (epoch) is the dynamic weighting factor, epoch is the current training cycle number, max_epoch is the number of training cycles, f error is the kinematic error function, b is the bias coefficient of the kinematic error function, w1 to w 14 are the coefficients of the kinematic error function, x1 is the target movement distance of the servo motor, x2 is the speed of the servo motor, x3 is the load force of the measured strain gauge, x4 is the ambient temperature at which the measured strain gauge is currently located, γ is the weight of the physical constraint, P constraint is the physical constraint term, α and β are error constraint weight factors, x is the actual number of pulses required for the servo motor, is the predicted number of pulses corresponding to the servo motor, v is the actual duty cycle required for the servo motor, is the predicted duty cycle corresponding to the servo motor.

[0041] In a second aspect, an embodiment of the present invention provides a strain gauge calibration device, which is applied to a strain gauge calibration system. The strain gauge calibration system includes a clamping device, a sliding device, a power device, a displacement measurement device, and a strain acquisition device. The clamping device is used to fix the measured strain gauge and the displacement measurement device, and is connected to the sliding device. The power device drives the sliding device to generate displacement through a servo motor, so that the measured strain gauge generates micro-strain. The device includes:

[0042] The first processing module is configured to obtain the displacement measurement value measured by the displacement measurement device and calculate the difference between the displacement measurement value and the strain target value of the strain gauge to be measured;

[0043] The second processing module is configured to respectively perform multi-scale feature extraction and one-dimensional feature extraction on the difference by using an intelligent control model composed of a Res2Net block and a Flatten layer, and obtain a predicted duty cycle and a predicted number of pulses according to the multi-scale feature extraction result and the one-dimensional feature extraction result;

[0044] The third processing module is configured to control the servo motor according to the predicted duty cycle and the predicted number of pulses to update the displacement measurement value so that the updated displacement measurement value is equal to the strain target value;

[0045] The calibration module is configured to, based on the strain acquisition device, obtain the strain output value of the strain gauge to be measured when the updated displacement measurement value is equal to the strain target value, and calibrate the strain gauge to be measured according to the strain output value.

[0046] In a third aspect, an embodiment of the present invention provides a strain gauge calibration device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the method in the first aspect or any possible implementation manner of the first aspect above is implemented.

[0047] In a fourth aspect, an embodiment of the present invention provides a strain gauge calibration system, including the strain gauge calibration device in the third aspect above, and further including a clamping device, a sliding device, a power device, a displacement measurement device, and a strain acquisition device; the clamping device is configured to fix the strain gauge to be measured and the displacement measurement device and connect the sliding device, and the power device drives the sliding device to generate a displacement through a servo motor so that the strain gauge to be measured generates a micro-strain.

[0048] In a fifth aspect, an embodiment of the present invention provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the method in the first aspect or any possible implementation manner of the first aspect above is implemented.

[0049] In the embodiments of the present invention, a servo motor drives a sliding device to generate displacement, so that the measured strain gauge generates micro-strain, and strain can be accurately loaded within a small displacement range to ensure high-precision calibration. By using an intelligent control model composed of a Res2Net block and a Flatten layer, multi-scale feature extraction and one-dimensional feature extraction are performed on the difference between the displacement measurement value of the sliding device and the strain target value of the measured strain gauge. Then, a predicted duty cycle and predicted pulse number are obtained based on the multi-scale feature extraction result and the one-dimensional feature extraction result. The servo motor can be controlled with high precision by using the predicted duty cycle and predicted pulse number to achieve precise movement within the micro-strain range. Moreover, the intelligent control model can dynamically optimize the motion parameters of the servo motor according to real-time feedback data, effectively avoiding problems such as overshoot and jitter, ensuring the stability and accuracy of the motion, and improving the intelligent control level of the strain gauge calibration system during the calibration process. Description of the Drawings

[0050] Figure 1 is a schematic diagram of a strain gauge calibration system provided by an embodiment of the present invention;

[0051] Figure 2 is a flowchart of the implementation of a strain gauge calibration method provided by an embodiment of the present invention;

[0052] Figure 3 is a schematic diagram of the middle layer of an intelligent control model provided by an embodiment of the present invention;

[0053] Figure 4 is a flowchart of the execution of the host computer system provided by an embodiment of the present invention;

[0054] Figure 5 is a schematic structural diagram of a strain gauge calibration device provided by an embodiment of the present invention;

[0055] Figure 6 is a schematic diagram of a strain gauge calibration device provided by an embodiment of the present invention;

[0056] Figure 7 is a schematic diagram of the calibration process of a strain gauge calibration system provided by an embodiment of the present invention. Detailed Embodiments

[0057] The following will describe the embodiments of the present invention in detail with reference to the drawings.

[0058] The strain gauge calibration method provided by the embodiments of the present invention is applied to a strain gauge calibration system. Combining Figure 1As shown, the strain gauge calibration system may include a clamping device, a sliding device, a power device, a displacement measuring device, and a strain acquisition device (not shown in the figure). The clamping device is used to fix the strain gauge 18 under test and the displacement measuring device 19, and is connected to the sliding device. The power device drives the sliding device to generate displacement through a servo motor, so as to generate micro-strain on the strain gauge 18 under test.

[0059] Among them, the clamping device may include: a support plate 1, a first sliding block 2, and a clamping block 8.

[0060] The sliding device may include: a first sliding block 1, a second sliding block 3, and a small optical axis 7.

[0061] The power device may include: a servo motor 14 and a motor reducer 15.

[0062] The transmission device may include: a high-precision lead screw 5, a large gear 11, and a small gear 12.

[0063] The connecting device may include: a fixed vertical plate 4, a large optical axis 6, a front bottom plate 9, a rear bottom plate 10, a support block 13, a motor bottom plate 16, and a bracket 17.

[0064] The above-mentioned clamping device, sliding device, power device, transmission device, and connecting device can form a mechanical device for strain gauge calibration. During assembly, both ends of the strain gauge 18 under test can be respectively installed on the support plate 1 and the first sliding block 2 through the clamping block 8. There are three large optical axes 6, which are respectively connected to the support plate 1, the first sliding block 2, the second sliding block 3, and the fixed vertical plate 4. The sliding device is connected to the first sliding block 2 and the second sliding block 3 respectively through two small optical axes 7. The displacement measuring device 19 (such as a micrometer) is installed on a gauge stand, and the gauge stand is installed on the lowermost large optical axis 6. The servo motor 14 and the motor reducer 15 are closely matched and installed on the bracket 17 together. The transmission device is composed of the large gear 11 meshing with the small gear 12, and the large gear 11 is installed on the high-precision lead screw 5; the small gear 12 is installed on the motor reducer 15, thereby transmitting the power of the servo motor 14 to the sliding device. Among them, the specification of the large gear 11 can be 1.25 module, 80 teeth, and tooth thickness 19, and the specification of the small gear 12 can be 1.25 module, 20 teeth, and tooth thickness 19. The model of the servo motor 14 can be MS1H1-10B30CB-A330Z of Inovance Technology, and the reduction ratio of the motor reducer 15 can be 50.

[0065] Among them, the two support blocks 13, the rear base plate 10, the front base plate 9, the servo motor 14, the motor reducer 15, the motor base plate 16 and the bracket 17 can be assembled into a whole by screw connection. The rear base plate 10 can be fixed together with this whole by screws. Six waist-shaped threaded holes can be designed on the rear base plate 10, and the position of the above-mentioned whole can be adjusted according to the meshing degree of the large gear 11 and the small gear 12. The servo motor 14 and the motor reducer 15 can be fixed on the bracket 17 by four M4 bolts. The bracket 17 can be fixed on the motor base plate 16 by four M6 bolts. The motor base plate 16 can be fixed together with the two support blocks 13 by six M8 bolts. The two support blocks 13 can be fixed on the rear base plate 10 by six M8 bolts.

[0066] When the servo motor 14 rotates, the cooperating motor reducer 15 will drive the connected small gear 12 to rotate, and the small gear 12 will drive the large gear 11 meshing with it to rotate. The large gear 11 will also drive the high-precision lead screw 5 to rotate synchronously, and the high-precision lead screw 5 will drive the first sliding block 2 and the second sliding block 3 to move, so as to generate micro-strain on the measured strain gauge 18. Then, the displacement measuring device 19 can collect the displacement of the sliding device, and obtain a displacement measurement value to represent the micro-strain generated by the measured strain gauge 18.

[0067] Exemplarily, the reduction ratio of the motor reducer 15 can be 50, the tooth number ratio of the large gear 11 and the small gear 12 can be 4:1, and the pitch of the high-precision lead screw 5 can be 0.05 mm.

[0068] Exemplarily, the measured strain gauge 18 can be a vibrating wire strain gauge, an inductive displacement strain gauge, a resistive strain gauge, a magnetostrictive displacement strain gauge, etc. The specific type of the measured strain gauge 18 is not limited in this embodiment.

[0069] It should be noted that although the specific type of the measured strain gauge 18 is not limited in this embodiment, the strain acquisition device needs to correspond to the type of the measured strain gauge 18. For example, when the measured strain gauge 18 is a vibrating wire strain gauge, the strain acquisition device is the corresponding frequency receiver. [[ID=X]] [[ID=Y]]

[0070] Exemplarily, according to actual needs, the displacement measuring device 19 can be selected from a dial indicator, a grating scale, a laser interferometer, etc. The specific displacement measuring device 19 is not limited in this embodiment.

[0071] Based on the above device, see Figure 2 , which shows the implementation flowchart of the strain gauge calibration method provided by the embodiment of the present invention. This method can be executed by the strain gauge calibration device in the strain gauge calibration. The strain gauge calibration device can be divided into an intelligent control system and a host computer system, which are described in detail as follows:

[0072] Step S201: Obtain the displacement measurement value measured by the displacement measurement device, and calculate the difference between the displacement measurement value and the strain target value of the strain gauge to be measured.

[0073] Exemplarily, in combination with Figure 1 , the displacement distance of the sliding device can be measured by the displacement measurement device 19 to obtain the displacement measurement value, and then calculate the difference between the current displacement measurement value and the strain target value that the strain gauge 18 to be measured expects to calibrate.

[0074] Step S202: Use the intelligent control model composed of a Res2Net block and a Flatten layer to perform multi-scale feature extraction and one-dimensional feature extraction on the difference respectively, and obtain the predicted duty cycle and the predicted number of pulses according to the multi-scale feature extraction result and the one-dimensional feature extraction result.

[0075] In this embodiment, the above difference is input into the intelligent control model, and the intelligent control model can output the predicted duty cycle time (i.e., the predicted duty cycle) and the predicted number of pulses that match the difference. The predicted values (including the predicted duty cycle and the predicted number of pulses) are input into the operation control system of the servo motor 14, so as to achieve precise control of the speed and displacement of the servo motor 14.

[0076] As Figure 3 shown, an intelligent control model can be constructed and trained using a Res2Net block and a Flatten layer. The intelligent control model is a multi-scale residual feature fusion network (MBRFF-Net) model, which can include an input layer, an intermediate layer (i.e., Figure 3 ), and an output layer connected in sequence.

[0077] Among them, the intermediate layer includes a multi-scale adaptive feature fusion network (MSAFF-Net) layer, a feature transformation layer (FTL), and a robust regularization layer (RRL).

[0078] The MSAFF-Net layer includes: a multi-scale feature extraction layer (Res2Net), a dynamic weighted fusion layer (Multi-HeadAttention), and a spatial and feature joint attention mechanism (Spatial Feature Attention).

[0079] The feature transformation layer (FTL) includes: a flatten layer (Flatten Layer), a second fully connected layer using the Relu activation function (ReLU-Connected Layer), a feature fusion layer (Feature Fusion Layer), a residual connection layer combined with CBAM (attention mechanism) (CBAM-Enhanced Residual Layer), a third fully connected layer, and a second regularization (Dropout) layer.

[0080] The Robust Regularization Layer (RRL), which is also the Gaussian Smoothing Layer, is divided into the first Gaussian Smoothing Layer and the second Gaussian Smoothing Layer. Among them, the first Gaussian Smoothing Layer includes the first Gaussian Noise Generator, the first ReLU-Connected Layer, and the first Dropout layer; the second Gaussian Smoothing Layer includes the second Gaussian Noise Generator and the third Dropout layer.

[0081] The output layer includes: the Adaptive Feature Aggregation Layer (AFAL), and the Adaptive Feature Aggregation Layer includes: the fourth ReLU-Connected Layer, the Adaptive Normalization Layer, and the fourth Dropout layer based on the siLU activation function.

[0082] The MSAFF-Net (Multi-Scale Adaptive Feature Fusion Network) layer, Feature Transformation Layer (FTL), Robust Regularization Layer (RRL), and Adaptive Feature Aggregation Layer (AFAL) included in the intermediate layer and the output layer are constructed using a multi-level series - multi-path parallel - dynamic hierarchical fusion structure.

[0083] Among them, multi-level series means connecting the MSAFF-Net layer and the first Gaussian Smoothing Layer in series to form the MSAFF-Net-RRL feature extraction network, and connecting the FTL and the second Gaussian Smoothing Layer in series to form the FTL-RRL feature extraction network. Multi-path parallel means extracting and optimizing features from two paths of the MSAFF-Net-RRL feature extraction network and the FTL-RRL feature extraction network, and dynamic hierarchical fusion means dynamically fusing the features output by the multi-path parallel network through the AFAL for each path feature.

[0084] Among them, the MSAFF-Net (Multi-Scale Adaptive Feature Fusion Network) layer includes a multi-scale feature extraction layer, a dynamic weighted fusion module, and a spatial and channel joint weighted network. This network uses a unique Iterative Feedback Fusion Mechanism (IFFM). This mechanism first extracts initial multi-scale features through the multi-scale feature extraction layer. The dynamic weighted fusion module makes a preliminary weight adjustment to the extracted multi-scale features. The spatial and channel joint weighted module further processes the fused features. The finally output features are passed back to the multi-scale feature extraction layer again to form a feedback loop, gradually strengthening the representation ability of the multi-scale features.

[0085] It should be noted that the number and size of the Feature FusionLayer in the multi-scale residual feature fusion network (MBRFF-Net) model, as well as the parameter settings of the Dropout layer, can all be optimized according to actual application requirements, and after iterative training, a stable result that can be directly applied to a high-precision intelligent control model can be obtained.

[0086] In one embodiment, the training process of the intelligent control model can be as follows:

[0087] Obtain a training data set, where the training data set includes multiple groups of training data. Each group of training data includes the historical difference between the historical displacement measurement value measured by the displacement measurement device and the historical strain target value of the measured strain gauge during the historical calibration process of the strain gauge calibration system, as well as the actual duty cycle and actual pulse number required for the servo motor to move when using the servo motor to control the historical displacement measurement value to be equal to the historical strain target value.

[0088] Construct a loss function based on the error dynamic constraint model.

[0089] Train an intelligent control model based on the training data set and the loss function.

[0090] The error dynamic constraint model includes:

[0091]

[0092] Among them, T loss is the error dynamic constraint model, δ min (epoch) is the dynamic weighting factor. As the state of the servo motor changes, the weight of the error will be dynamically adjusted; epoch is the current number of training cycles, max_epoch is the number of training cycles, f error is the kinematic error function, b is the bias coefficient of the kinematic error function, w1 to w 14 are the coefficients of the kinematic error function, x1 is the target movement distance of the servo motor, x2 is the speed of the servo motor, x3 is the load force of the measured strain gauge, x4 is the ambient temperature at which the measured strain gauge is currently located, γ is the weight of the physical constraint, used to balance the influence between the physical constraint and the data error, P constraint is the physical constraint term, representing the dynamic constraint based on the servo motor kinematic model, α, β are the error constraint weight factors, which can be determined through experiments, x is the actual pulse number required for the servo motor, is the predicted pulse number corresponding to the servo motor, v is the actual duty cycle required for the servo motor, is the predicted duty cycle corresponding to the servo motor.

[0093] Among them, when obtaining the training data set, a high-quality training data set can be constructed. For each set of training data in the training data set, the difference between the displacement measurement value of the sliding device and the stress target value can be measured by the above displacement measurement device. The servo motor output power is controlled according to this difference. When the target position is reached without error (that is, when the displacement measurement value is equal to the stress target value), record the actual number of pulses and the actual duty cycle time required for the servo motor to move at this time.

[0094] After obtaining the training data set, the training data set can be preprocessed, that is, the data in the training data set is filtered to remove abnormal data.

[0095] Then, based on the preprocessed training data set, train the constructed multi-scale residual feature fusion network (MBRFF-Net) model, and use the Adam optimizer to perform backpropagation supervised training on the multi-scale residual feature fusion network (MBRFF-Net) model to determine the intelligent control model.

[0096] Among them, the loss function of the multi-scale residual feature fusion network (MBRFF-Net) model can fuse the error dynamic constraint model (EDC). By introducing physical constraint errors, the model can take into account complex factors in reality, rather than just the errors in the data itself. This new loss function design will help improve the non-linear relationship between sensor data and mechanical structure errors, thereby improving the accuracy and stability of mechanical devices.

[0097] Among them, the error dynamic constraint model (EDC) can construct a kinematic error model based on the mechanical device and power system calibrated by the strain gauge in the embodiment of the present invention, combined with a dynamic weighting strategy. The construction steps are as follows:

[0098] Construct a kinematic error function f according to the target movement distance, speed of the servo motor, the load force of the measured strain gauge, and the ambient temperature where the measured strain gauge is currently located error , where the kinematic error function f error The coefficients of are obtained by fitting experimental data. The experimental steps are as follows:

[0099] 1. Let the servo motor move to different target positions d m .

[0100] 2. Measure the actual movement distance d corresponding to each target value a .

[0101] 3. Collect speed V, load F, and temperature T data.

[0102] Using the above collected data, use polynomial regression fitting technology to calculate the coefficients w1 to w of the kinematic error function f error to 14and the offset coefficient b.

[0103] The fitted kinematic error function can reflect the physical constraint parameters of the servo motor, including the backlash error of the lead screw or gear, the motor control accuracy, the temperature, and the mechanical assembly error.

[0104] Construct an error dynamic constraint model T by combining the kinematic error function and the dynamic weighting strategy loss , and the dynamic weighting strategy can dynamically adjust the physical constraint weights, enabling the model to better focus on error elimination at different training stages.

[0105] Step S203: Control the servo motor according to the predicted duty cycle and the predicted number of pulses to update the displacement measurement value so that the updated displacement measurement value is equal to the strain target value.

[0106] Step S204: Based on the strain acquisition device, obtain the strain output value of the measured strain gauge when the updated displacement measurement value is equal to the strain target value, and calibrate the measured strain gauge according to the strain output value.

[0107] Among them, steps S201 to S203 can be executed by the intelligent control system, and step S204 can be executed by the host computer system. The intelligent control system can be integrated into the host computer system or can be executed independently. For example Figure 4 As shown, taking the calibration of the vibrating wire strain gauge as an example, the specific implementation steps executed by the host computer system can be:

[0108] The data receiving module S401, whose main function is to receive in real time the calibration data uploaded from the field device, including the vibrating wire frequency and displacement information, and perform integrity verification on the received original data to ensure the accuracy and reliability of data transmission.

[0109] The data correction module S402 can adopt a linear correction algorithm and fitting technology, and combine a preset error compensation model to perform real-time correction and optimization on the data received by the data receiving module S401. Through this module, the authenticity and accuracy of the data can be guaranteed, providing a reliable basis for subsequent calculations and analyses.

[0110] The calibration index calculation module S404 is the core of the entire host computer program, responsible for performing operations and evaluations of calibration indexes based on the corrected data. The module embeds a calibration model and completes the analysis and evaluation of performance indexes such as strain sensitivity, linearity, and repeatability according to the set calculation logic. The operation results provide data support for generating a calibration report.

[0111] The data after correction and calculation is archived and saved by the data storage module S405. The module adopts a hierarchical storage strategy to classify and save the original data, corrected data, and calculation results into the database to ensure the integrity and traceability of the data.

[0112] To improve the operation convenience of users, the host computer program designs an image data display module S403. The module can generate intuitive frequency-strain curves, linear fitting diagrams and other visualization charts based on the calibration data of the data receiving module S401 and the calibration index calculation module 404. Through the graphical interface, users can quickly understand the calibration results. At the same time, the module also supports dynamic updates of real-time monitoring data. Based on the analysis results of the calibration index calculation module S404, the program can generate a calibration report S406 containing device performance indicators, data curves, test conditions, etc. with one key. The report file can be in the national standard PDF format, which is convenient for users to archive or submit for review.

[0113] In this embodiment, the host computer system integrates functions of information collection, data processing and result calculation. After receiving the collected data, the host computer can automatically complete linear correction and calibration index calculation, greatly simplifying the calibration process and shortening the calibration time. The built-in data processing module of the system can deeply analyze the collected frequency and position information, including non-linear error correction and calibration curve fitting. Automatically generate a calibration report, which is convenient for users to evaluate the calibration effect. The embodiment of the present invention can be compatible with different models of vibrating wire strain gauges, and can adapt to the calibration requirements of different measured strain gauges by adjusting control parameters, broadening the application scope of the system.

[0114] In one embodiment, using an intelligent control model composed of a Res2Net block and a Flatten layer to perform multi-scale feature extraction and one-dimensional feature extraction on the difference respectively, and obtaining the predicted duty cycle and predicted pulse number according to the multi-scale feature extraction result and the one-dimensional feature extraction result may include:

[0115] Based on the input layer, expand the difference into a high-dimensional feature vector.

[0116] Based on the Res2Net block, perform multi-scale feature extraction on the high-dimensional feature vector to obtain the multi-scale feature extraction result.

[0117] Based on the Flatten layer, perform one-dimensional feature extraction on the high-dimensional feature vector to obtain the one-dimensional feature extraction result.

[0118] Obtain the predicted duty cycle and predicted pulse number according to the multi-scale feature extraction result and the one-dimensional feature extraction result.

[0119] Among them, the input layer can be a fully connected layer.

[0120] Combined with Figure 3As shown, in the intelligent control model, the middle layer is the core part for processing and optimizing data, responsible for performing in-depth analysis, feature extraction, and decision generation on the input data. The middle layer adopts a variety of advanced deep learning methods and network structures, and through hierarchical and modular processing, ensures precise control and optimization of the motor error by the system.

[0121] In one embodiment, performing multi-scale feature extraction on the high-dimensional feature vector based on the Res2Net block to obtain the multi-scale feature extraction result may include:

[0122] Performing multi-scale feature extraction on the high-dimensional feature vector based on the Res2Net block to obtain high-dimensional features.

[0123] Performing weighted fusion on the features in different subspaces of the high-dimensional features based on the multi-head attention mechanism and spatial feature attention to obtain the fused features.

[0124] Adding Gaussian noise to the fused features based on the first Gaussian noise generator for Gaussian smoothing to obtain the smoothed fused features.

[0125] Performing dimensionality reduction processing on the smoothed fused features based on the first fully connected layer to obtain the initial multi-scale feature extraction result.

[0126] Screening the initial multi-scale feature extraction result based on the first regularization layer to obtain the multi-scale feature extraction result.

[0127] In one embodiment, performing one-dimensional feature extraction on the high-dimensional feature vector based on the Flatten layer to obtain the one-dimensional feature extraction result may include:

[0128] Flattening the high-dimensional feature vector into a one-dimensional vector based on the Flatten layer to obtain one-dimensional features.

[0129] Performing non-linear transformation on the one-dimensional features based on the second fully connected layer to obtain the non-linearly transformed features.

[0130] Performing feature fusion on the non-linearly transformed features based on the feature fusion layer to obtain the fused transformed features.

[0131] Enhancing the fused transformed features based on the residual connection layer with channel and spatial attention mechanisms to obtain the enhanced fused features.

[0132] Mapping the enhanced fused features to the target dimension based on the third fully connected layer to obtain the target features.

[0133] Screening the target features based on the second regularization layer to obtain the screened target features.

[0134] Adding Gaussian noise to the screened target features based on the second Gaussian noise generator for Gaussian smoothing to obtain the smoothed target features.

[0135] Screen the smoothed target features based on the third regularization layer to obtain a one-dimensional feature extraction result.

[0136] In one embodiment, obtaining the predicted duty cycle and the predicted number of pulses according to the multi-scale feature extraction result and the one-dimensional feature extraction result may include:

[0137] Connect the multi-scale feature extraction result and the one-dimensional feature extraction result based on the fourth fully connected layer to obtain an aggregated feature.

[0138] Perform adaptive normalization on the aggregated feature based on the adaptive normalization layer to obtain a normalized feature.

[0139] Screen the normalized feature based on the fourth regularization layer to obtain the predicted duty cycle and the predicted number of pulses.

[0140] In this embodiment, the input data is: the difference (scalar) between the displacement measurement value of the sliding device and the stress target value. Therefore, input dimensionality increase is performed. Since the input is a scalar, it is first extended to a high-dimensional feature vector through a fully connected layer (FC). Then:

[0141] MSAFF-Net-RRL feature extraction path:

[0142] Res2Net groups the input feature maps (which can be 64-dimensional, i.e., high-dimensional feature vectors), and each group extracts features through convolutional kernels of different scales; the multi-scale features are fused through residual connections to generate a richer high-dimensional feature representation. Among them, Res2Net is a residual network, which can effectively improve the expression ability of the network through the extraction and fusion of multi-scale features. In this step, the Res2Net network performs multi-level and multi-scale feature extraction on the original data to obtain richer high-dimensional features. The multi-head attention mechanism divides the feature map into multiple subspaces, calculates the attention weights respectively, weights and fuses the features in different subspaces to enhance the representation of important features, calculates the spatial attention weights of the feature map, and highlights the features in important regions. This mechanism further weights and fuses the extracted features, can simultaneously focus on multiple subspaces in the data, and ensures that the processing of different features does not interfere with each other. After feature extraction, the system performs weighted processing on the features in a specific spatial region through spatial feature attention. Then, a first Gaussian noise generator is used to inject an appropriate amount of Gaussian noise into the network to further enhance the robustness of the model. By simulating the noise interference in the actual working environment, the system can maintain high-efficiency control performance when facing complex signal interference. Then, the first fully connected layer is used to reduce the dimension of the features, and the first Dropout layer randomly discards some neurons to prevent overfitting and enhance the generalization ability of the network. Through the introduction of this layer, the network can avoid over-relying on certain specific features during the training process, thereby improving the stability of the overall control system. Finally, the high-dimensional extracted features are output.

[0143] FTL-RRL feature extraction path:

[0144] The Flatten layer flattens the high-dimensional feature vector into a one-dimensional vector to adapt it to the subsequent fully connected layer. Through this layer, the model transforms the high-level features into a one-dimensional vector, facilitating more efficient processing and decision-making generation in the subsequent steps. The ReLU (Rectified Linear Unit) activation layer performs a non-linear transformation on the features to increase the non-linear representation ability of the network. This process effectively enhances the network's ability to handle complex control tasks and better adapt to changes in motor errors. The transformed features are fused through the feature fusion layer to generate a more comprehensive feature representation. By integrating various features, this layer enhances the system's ability to solve motor control problems in complex environments. The fused features are enhanced through channel and spatial attention mechanisms, effectively improving the feature selection ability and thus enhancing the expressive power of the residual network. This layer further improves the accuracy of motor error control by optimizing the residual learning process. Subsequently, the feature vector is mapped to the target dimension through the third fully connected layer and the second regularization layer. At this time, the one-dimensional feature passes through the one-dimensional Gaussian smoothing layer, injecting Gaussian noise into the network to enhance the robustness of the model. Finally, the one-dimensional extracted feature is output through the third regularization layer.

[0145] Output layer: Adaptive Feature Aggregation Layer (AFAL), which aggregates the above-mentioned high-dimensional features and one-dimensional features. It includes connecting the high-dimensional features and one-dimensional features through the fourth fully connected layer, passing through the adaptive normalization layer, and randomly discarding some neurons through the fourth Dropout layer based on the siLU activation function to prevent overfitting. Finally, the predicted values (including predicted duty cycle and predicted number of pulses) are output.

[0146] In the embodiments of the present invention, the servo motor drives the sliding device to generate displacement, causing the measured strain gauge to generate micro-strain. This can accurately load strain within a small displacement range, ensuring high-precision calibration. Using the intelligent control model composed of Res2Net blocks and Flatten layers, multi-scale feature extraction and one-dimensional feature extraction are performed on the difference between the displacement measurement value of the sliding device and the strain target value of the measured strain gauge. Then, the predicted duty cycle and predicted number of pulses are obtained based on the multi-scale feature extraction results and one-dimensional feature extraction results. The servo motor can be precisely controlled using the predicted duty cycle and predicted number of pulses to achieve precise movement within the micro-strain range. Moreover, the intelligent control model can dynamically optimize the motion parameters of the servo motor according to real-time feedback data, effectively avoiding problems such as overshoot and jitter, ensuring the stability and accuracy of the motion, and improving the intelligent control level of the strain gauge calibration system during the calibration process.

[0147] It should be understood that the magnitudes of the sequence numbers of the steps in the above embodiments do not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.

[0148] The following is an apparatus embodiment of the present invention. For details not described in detail herein, reference may be made to the corresponding method embodiment above.

[0149] Figure 5 FIG. shows a schematic structural diagram of a strain gauge calibration apparatus provided by an embodiment of the present invention. The strain gauge calibration apparatus is applied to a strain gauge calibration system (such as Figure 1 shown). The strain gauge calibration system includes a clamping device, a sliding device, a power device, a displacement measuring device, and a strain acquisition device. The clamping device is used to fix the strain gauge under test and the displacement measuring device, and is connected to the sliding device. The power device drives the sliding device to generate a displacement through a servo motor, so as to generate a micro-strain in the strain gauge under test. For the sake of convenience of description, only parts related to the embodiment of the present invention are shown and are described in detail as follows:

[0150] As Figure 5 shown, the strain gauge calibration apparatus includes: a first processing module 51, a second processing module 52, a third processing module 53, and a calibration module 54.

[0151] The first processing module 51 is configured to obtain a displacement measurement value measured by the displacement measuring device, and calculate a difference between the displacement measurement value and a strain target value of the strain gauge under test.

[0152] The second processing module 52 is configured to respectively perform multi-scale feature extraction and one-dimensional feature extraction on the difference by using an intelligent control model composed of a Res2Net block and a Flatten layer, and obtain a predicted duty cycle and a predicted pulse number according to the multi-scale feature extraction result and the one-dimensional feature extraction result.

[0153] The third processing module 53 is configured to control the servo motor according to the predicted duty cycle and the predicted pulse number to update the displacement measurement value, so that the updated displacement measurement value is equal to the strain target value.

[0154] The calibration module 54 is configured to, based on the strain acquisition device, obtain a strain output value of the strain gauge under test when the updated displacement measurement value is equal to the strain target value, and calibrate the strain gauge under test according to the strain output value.

[0155] In a possible implementation manner, the second processing module 52 may be configured to expand the difference into a high-dimensional feature vector based on an input layer; perform multi-scale feature extraction on the high-dimensional feature vector by using a Res2Net block to obtain a multi-scale feature extraction result; perform one-dimensional feature extraction on the high-dimensional feature vector by using a Flatten layer to obtain a one-dimensional feature extraction result; and obtain a predicted duty cycle and a predicted pulse number according to the multi-scale feature extraction result and the one-dimensional feature extraction result.

[0156] In a possible implementation, the second processing module 52 can be used to perform multi-scale feature extraction on the high-dimensional feature vector based on the Res2Net block to obtain high-dimensional features; perform weighted fusion on the features in different subspaces of the high-dimensional features based on the multi-head attention mechanism and spatial feature attention to obtain fused features; add Gaussian noise to the fused features based on the first Gaussian noise generator for Gaussian smoothing to obtain smoothed fused features; perform dimensionality reduction processing on the smoothed fused features based on the first fully connected layer to obtain an initial multi-scale feature extraction result; and screen the initial multi-scale feature extraction result based on the first regularization layer to obtain a multi-scale feature extraction result.

[0157] In a possible implementation, the second processing module 52 can be used to flatten the high-dimensional feature vector into a one-dimensional vector based on the Flatten layer to obtain one-dimensional features; perform a non-linear transformation on the one-dimensional features based on the second fully connected layer to obtain non-linearly transformed features; perform feature fusion on the non-linearly transformed features based on the feature fusion layer to obtain fused transformed features; enhance the fused transformed features based on the residual connection layer of the channel and spatial attention mechanism to obtain enhanced fused features; map the enhanced fused features to the target dimension based on the third fully connected layer to obtain target features; screen the target features based on the second regularization layer to obtain screened target features; add Gaussian noise to the screened target features based on the second Gaussian noise generator for Gaussian smoothing to obtain smoothed target features; and screen the smoothed target features based on the third regularization layer to obtain a one-dimensional feature extraction result.

[0158] In a possible implementation, the second processing module 52 can be used to connect the multi-scale feature extraction result and the one-dimensional feature extraction result based on the fourth fully connected layer to obtain aggregated features; perform adaptive normalization on the aggregated features based on the adaptive normalization layer to obtain normalized features; and screen the normalized features based on the fourth regularization layer to obtain a predicted duty cycle and predicted pulse number.

[0159] In a possible implementation, the training process of the intelligent control model is as follows:

[0160] Obtain a training data set. The training data set includes multiple groups of training data. Each group of training data includes the historical difference between the historical displacement measurement value measured by the displacement measurement device and the historical strain target value of the measured strain gauge during the historical calibration process of the strain gauge calibration system, and the actual duty cycle and actual pulse number required for the servo motor to move when the servo motor is used to control the historical displacement measurement value to be equal to the historical strain target value.

[0161] Construct a loss function based on the error dynamic constraint model.

[0162] Train an intelligent control model based on the training data set and the loss function.

[0163] The error dynamic constraint model includes:

[0164]

[0165] where T loss is the error dynamic constraint model, δ min (epoch) is the dynamic weighting factor, epoch is the current training cycle number, max_epoch is the training cycle number, f error is the kinematic error function, b is the bias coefficient of the kinematic error function, w1 to w 14 are the coefficients of the kinematic error function, x1 is the target movement distance of the servo motor, x2 is the speed of the servo motor, x3 is the load force of the measured strain gauge, x4 is the ambient temperature where the measured strain gauge is currently located, γ is the weight of the physical constraint, P constraint is the physical constraint term, α, β are the error constraint weight factors, x is the actual number of pulses required by the servo motor, is the predicted number of pulses corresponding to the servo motor, v is the actual duty cycle required by the servo motor, is the predicted duty cycle corresponding to the servo motor.

[0166] Figure 6 is a schematic diagram of the strain gauge calibration device provided by an embodiment of the present invention. As Figure 6 shown, the strain gauge calibration device 6 of this embodiment includes: a processor 60 and a memory 61. The memory 61 stores a computer program 62. When the processor 60 executes the computer program 62, the steps in the above-mentioned method embodiments are implemented. Alternatively, when the processor 60 executes the computer program 62, the functions of each module / unit in the above-mentioned device embodiments are implemented.

[0167] Exemplarily, the computer program 62 can be divided into one or more modules / units. The one or more modules / units are stored in the memory 61 and executed by the processor 60 to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, and these instruction segments are used to describe the execution process of the computer program 62 in the strain gauge calibration device 6.

[0168] The strain gauge calibration device 6 may include, but is not limited to, a processor 60 and a memory 61. Those skilled in the art can understand that Figure 6 this is only an example of the strain gauge calibration device 6 and does not constitute a limitation on the strain gauge calibration device 6. It may include more or fewer components than those shown in the figure, or combine certain components, or different components. For example, the strain gauge calibration device 6 may further include input / output devices, network access devices, buses, etc.

[0169] The processor 60 may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0170] The memory 61 may be an internal storage unit of the strain gauge calibration device 6, such as the hard disk or memory of the strain gauge calibration device 6. The memory 51 may also be an external storage device of the strain gauge calibration device 6, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc. equipped on the strain gauge calibration device 6. Further, the memory 61 may also include both the internal storage unit of the strain gauge calibration device 6 and the external storage device. The memory 61 is used to store the computer program 62 and other programs and data required by the strain gauge calibration device 6. The memory 61 may also be used to temporarily store the data that has been output or is to be output.

[0171] For the convenience and simplicity of description, only the above-mentioned division of each functional module / unit is used as an example for illustration. In actual applications, the above-mentioned functions can be allocated to different functional modules / units according to needs. The above-mentioned modules / units can be implemented in the form of hardware, or in the form of software, or in the form of a combination of hardware and software.

[0172] The embodiment of the present invention further provides a strain gauge calibration system, which includes the strain gauge calibration device in the above-mentioned embodiment, and further includes a clamping device, a sliding device, a power device, a displacement measurement device, and a strain acquisition device; the clamping device is used to fix the strain gauge under test and the displacement measurement device, and connect the sliding device, and the power device drives the sliding device to generate displacement through a servo motor, so as to cause the strain gauge under test to generate micro-strain.

[0173] Combined with Figure 7 shown, taking the vibrating wire strain gauge as an example, the calibration process of the strain gauge calibration system is described as follows:

[0174] As Figure 7As shown in the figure, the automatic calibration system for vibrating wire strain gauges consists of four core components, namely: mechanical device, intelligent control system, information acquisition system, and host computer system. The system aims to achieve the automatic calibration of vibrating wire strain gauges through high-precision intelligent control and data processing technologies. The specific implementation steps are as follows:

[0175] Step 1: Initialize the control system. After the system starts, the central control board first performs initialization. This process includes connecting to the network (Wi-Fi) and ensuring the stability of communication with the host computer. The control system connects to the host computer system through the TCP / IP protocol to complete data exchange and subsequent information processing.

[0176] Step 2: After initialization, the control system turns on the host computer and starts collecting the first data through the network. The host computer system gives a response to the intelligent control system according to the response mechanism and performs preliminary processing on the received data.

[0177] Step 3: After receiving the data feedback from the host computer, the intelligent control system performs the following operations: Based on the microstrain target (i.e., the strain target value), the intelligent control system precisely controls the servo motor, enabling the mechanical device to push the measured strain gauge for microstrain adjustment. The mechanical device adjusts precisely to ensure that the microstrain of the measured strain gauge reaches the strain target value and avoid errors.

[0178] Step 4: When the microstrain reaches the set standard (i.e., the strain target value), the information acquisition system immediately obtains the actual position and frequency data of the measured strain gauge (representing the strain output value of the measured strain gauge). These data are transmitted to the intelligent control system through the RS-485 communication protocol. After the intelligent control system simply processes the data, it transmits the data to the host computer system through the local area network based on the TC / IP protocol to ensure the real-time and accuracy of the data.

[0179] Step 5: After receiving the position information and frequency data, the host computer performs linear correction on the collected raw data to eliminate systematic errors; uses a mathematical model to fit the data and displays it in a graph to observe the data fitting situation and further improve the data accuracy; automatically calculates the performance indicators of the device based on the corrected data; after processing, the system generates a calibration report containing device performance analysis, microstrain curve, and other key indicators. The report is visually presented and stored in the database for users to query and download conveniently.

[0180] Among them, the intelligent control system is the hub connecting the mechanical device and the host computer system, which can be composed of a minimum single-chip microcomputer system and an RS485-to-TTL module. The single-chip microcomputer system can use ESP32-WROOM-32UE (8MB) as the main control chip, and deploy the intelligent control model in the chip, which is mainly responsible for controlling the precise movement of the servo motor. In addition, the intelligent control system can summarize the data collected by the information acquisition system, judge whether the microstrain of the measured strain gauge reaches the set standard (i.e., the strain target value). If it reaches the set standard, it controls the information acquisition system to collect the position information and frequency data at this time, and after simply processing the collected data, sends it to the single-chip microcomputer.

[0181] Among them, the information acquisition system can include: the measured vibrating wire strain gauge, the frequency reading instrument (i.e., the strain acquisition device), the micrometer (i.e., the displacement measurement device), and the speed encoder. The frequency acquisition is connected to the frequency reading instrument through the wire led out by the measured strain gauge; the strain of the strain gauge is acquired by measuring the displacement of the sliding device with the micrometer to represent the microstrain generated by the measured strain gauge; the current speed data is acquired by measuring with the servo motor encoder; these three kinds of data are connected to the intelligent control system through the RS485 module to read the current data, and then send the data to the host computer according to the TCP / IP protocol.

[0182] The information acquisition device provided by the embodiment of the present application includes a micrometer, a frequency reading instrument, and a supporting RS-485 communication receiving and controlling device, aiming to collect the displacement data and frequency data of the vibrating wire strain gauge under the microstrain state in real time, and ensure the accuracy and stability of data collection.

[0183] Among them, the micrometer is used to collect the microstrain displacement data of the vibrating wire strain gauge in the calibration process in real time. Its high-precision measurement ability can capture the change of the tiny displacement of the measured strain gauge, and provide an accurate displacement reference value for subsequent calibration and data processing. The micrometer can be connected to the control board through the RS-485 interface, and transmit the measurement data to the main controller in real time.

[0184] The frequency reading instrument is used to collect the vibrating wire frequency data of the measured strain gauge in the microstrain state. The vibrating wire frequency data can reflect the dynamic characteristics of the strain gauge, and provide an important basis for calculating the performance indicators of subsequent equipment. The frequency reading instrument can also communicate with the control board through the RS-485 interface to achieve efficient and stable data transmission.

[0185] Among them, the communication receiving control device can be a highly integrated control board, which uses an ESP32 single-chip microcomputer as the main controller. The core functions of the communication receiving control device can include: receiving measurement data from dial indicators and frequency readers, and controlling the motor for precise control; parsing and processing the received data through the ESP32 single-chip microcomputer, and deploying the intelligent control model into the single-chip microcomputer to achieve precise control of the motor.

[0186] The host computer system provided by the embodiment of the present application may include: a data receiving module, a data correction module, a calibration index calculation module, an image data display module, and a calibration report generation module. The host computer system can be developed through the Python QT platform, integrating functions of data receiving, correction, analysis, visualization, and report generation, and is an important part of realizing high-precision calibration of vibrating wire strain gauges. Among them, the data receiving module can receive displacement data and frequency data transmitted by the information acquisition system through wireless communication technology based on the TC / IP protocol. The data receiving module can adopt a multi-threaded design to process real-time data of multiple sensors simultaneously, ensuring the timeliness and integrity of data reception. The data correction module can perform linear correction on the original collected data based on the linear interpolation algorithm. The calibration index calculation module can automatically calculate the performance indexes of the vibrating wire strain gauge based on the corrected data, including linearity, repeatability, sensitivity, basic error, and hysteresis error; the calculation results can be directly used to evaluate the performance of the measured strain gauge and support the extension of custom indexes. The image data display module can use the GUI function of the Python QT platform to visually present the data in the form of real-time charts and trend curves, and the real-time charts and trend curves display the dynamic changes of displacement, frequency, and related performance indexes. The calibration report generation module can automatically generate a standardized calibration report according to user requirements based on the collected, corrected, calculated data, and analysis results.

[0187] The strain gauge automatic calibration system provided by the embodiments of the present invention belongs to a high-precision calibration system. This system integrates four core components: a mechanical device for strain gauge calibration, an intelligent control system, an information acquisition system, and a host computer system. The mechanical device part can precisely clamp the strain gauge to be measured and can achieve the micro-strain of the strain gauge to be measured. The intelligent control system can achieve high-precision control of the servo motor through an intelligent control model, drive the calibration of the strain gauge to be measured, and ensure that the micro-strain of the strain gauge to be measured accurately reaches the target length. When the micro-strain of the strain gauge to be measured meets the standard, the information acquisition system immediately collects the position and frequency information and uploads it to the host computer. After receiving the data, the host computer sequentially performs linear correction and data fitting processing, calculates the device performance indicators with one key, stores the data in the database, and at the same time presents the data correlation in a visual chart and automatically generates a calibration report. The present invention is applicable to the calibration of various types of strain gauges. Compared with traditional manual and automatic calibration systems, it can greatly reduce the calibration time, significantly improve the work efficiency and calibration accuracy, and bring a new and efficient solution for the strain gauge calibration work.

[0188] The embodiments of the present invention also provide a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the methods in the above method embodiments are implemented.

[0189] The embodiments of the present invention also provide a computer program product, including a computer program. When the computer program is executed by a processor, the methods in the above method embodiments are implemented.

[0190] Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0191] In the above embodiments, the descriptions of each embodiment have their own emphases. For the parts not detailed or recorded in a certain embodiment, reference can be made to the relevant descriptions of other embodiments. Without special instructions and logical conflicts, the terms and / or descriptions between different embodiments are consistent and can be referenced to each other, and the technical features in different embodiments can be combined to form new embodiments according to their internal logical relationships.

[0192] The above-described embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A strain gauge calibration method, characterized in that, Applied to a strain gauge calibration system, the strain gauge calibration system includes a clamping device, a sliding device, a power device, a displacement measuring device, and a strain acquisition device. The clamping device is used to fix the strain gauge under test and the displacement measuring device, and is connected to the sliding device. The power device drives the sliding device to generate a displacement through a servo motor, so that the strain gauge under test generates micro-strain. The method includes: Obtain the displacement measurement value measured by the displacement measuring device, and calculate the difference between the displacement measurement value and the strain target value of the strain gauge under test; Use an intelligent control model composed of a Res2Net block and a Flatten layer to perform multi-scale feature extraction and one-dimensional feature extraction on the difference respectively, and obtain a predicted duty cycle and predicted pulse number according to the multi-scale feature extraction result and the one-dimensional feature extraction result; Control the servo motor according to the predicted duty cycle and the predicted pulse number to update the displacement measurement value, so that the updated displacement measurement value is equal to the strain target value; Based on the strain acquisition device, obtain the strain output value of the strain gauge under test when the updated displacement measurement value is equal to the strain target value, and calibrate the strain gauge under test according to the strain output value.

2. The strain gauge calibration method according to claim 1, characterized in that, The using an intelligent control model composed of a Res2Net block and a Flatten layer to perform multi-scale feature extraction and one-dimensional feature extraction on the difference respectively, and obtain a predicted duty cycle and predicted pulse number according to the multi-scale feature extraction result and the one-dimensional feature extraction result includes: Expand the difference into a high-dimensional feature vector based on the input layer; Perform multi-scale feature extraction on the high-dimensional feature vector based on the Res2Net block to obtain a multi-scale feature extraction result; Perform one-dimensional feature extraction on the high-dimensional feature vector based on the Flatten layer to obtain a one-dimensional feature extraction result; Obtain a predicted duty cycle and predicted pulse number according to the multi-scale feature extraction result and the one-dimensional feature extraction result.

3. The strain gauge calibration method according to claim 2, characterized in that, The performing multi-scale feature extraction on the high-dimensional feature vector based on the Res2Net block to obtain a multi-scale feature extraction result includes: Perform multi-scale feature extraction on the high-dimensional feature vector based on the Res2Net block to obtain high-dimensional features; Perform weighted fusion on the features in different subspaces of the high-dimensional features based on the multi-head attention mechanism and the spatial feature attention to obtain a fused feature; Add Gaussian noise to the fused feature based on the first Gaussian noise generator for Gaussian smoothing to obtain a smoothed fused feature; Perform dimensionality reduction processing on the smoothed fused feature based on the first fully connected layer to obtain an initial multi-scale feature extraction result; Screen the initial multi-scale feature extraction result based on the first regularization layer to obtain a multi-scale feature extraction result.

4. The strain gauge calibration method according to claim 2, characterized in that, The performing one-dimensional feature extraction on the high-dimensional feature vector based on the Flatten layer to obtain a one-dimensional feature extraction result includes: Flatten the high-dimensional feature vector into a one-dimensional vector based on the Flatten layer to obtain a one-dimensional feature; Perform a non-linear transformation on the one-dimensional feature based on the second fully connected layer to obtain a non-linearly transformed feature; Feature fusion is performed on the non-linear transformation features based on the feature fusion layer to obtain fused transformation features; The fused transformation features are enhanced based on the residual connection layer with channel and spatial attention mechanisms to obtain enhanced fused features; The enhanced fused features are mapped to the target dimension based on the third fully connected layer to obtain target features; The target features are screened based on the second regularization layer to obtain screened target features; Gaussian noise is added to the screened target features based on the second Gaussian noise generator for Gaussian smoothing to obtain smoothed target features; The smoothed target features are screened based on the third regularization layer to obtain the one-dimensional feature extraction result.

5. The strain gauge calibration method according to claim 2, characterized in that, Obtaining the predicted duty cycle and predicted pulse number according to the multi-scale feature extraction result and the one-dimensional feature extraction result includes: The multi-scale feature extraction result and the one-dimensional feature extraction result are concatenated based on the fourth fully connected layer to obtain an aggregated feature; Adaptive normalization is performed on the aggregated feature based on the adaptive normalization layer to obtain a normalized feature; The normalized feature is screened based on the fourth regularization layer to obtain the predicted duty cycle and predicted pulse number.

6. The strain gauge calibration method according to claim 1, characterized in that, The training process of the intelligent control model is as follows: Obtain a training data set, where the training data set includes multiple groups of training data. Each group of training data includes the historical difference between the historical displacement measurement value measured by the displacement measurement device and the historical strain target value of the measured strain gauge during the historical calibration process of the strain gauge calibration system, and the actual duty cycle and actual pulse number required for the servo motor to move when the servo motor is used to control the historical displacement measurement value to be equal to the historical strain target value; Construct a loss function based on the error dynamic constraint model; The intelligent control model is trained based on the training data set and the loss function; The error dynamic constraint model includes: where, T loss is the error dynamic constraint model, δ min (epoch) is the dynamic weighting factor, epoch is the current number of training epochs, max_epoch is the number of training epochs, f error is the kinematic error function, b is the bias coefficient of the kinematic error function, w1 to w 14 are the coefficients of the kinematic error function, x1 is the target movement distance of the servo motor, x2 is the speed of the servo motor, x3 is the load force of the measured strain gauge, x4 is the ambient temperature at which the measured strain gauge is currently located, γ is the weight of the physical constraint, P constraint is the physical constraint term, α, β are the error constraint weight factors, x is the actual number of pulses required by the servo motor, is the predicted number of pulses corresponding to the servo motor, v is the actual duty cycle required by the servo motor, is the predicted duty cycle corresponding to the servo motor.

7. A strain gauge calibration device, characterized in that, Applied to a strain gauge calibration system, the strain gauge calibration system includes a clamping device, a sliding device, a power device, a displacement measurement device, and a strain acquisition device. The clamping device is used to fix the measured strain gauge and the displacement measurement device and connect the sliding device. The power device drives the sliding device to generate displacement through a servo motor, so that the measured strain gauge generates micro-strain. The device includes: A first processing module for obtaining the displacement measurement value measured by the displacement measurement device and calculating the difference between the displacement measurement value and the strain target value of the measured strain gauge; A second processing module for respectively performing multi-scale feature extraction and one-dimensional feature extraction on the difference using an intelligent control model composed of a Res2Net block and a Flatten layer, and obtaining the predicted duty cycle and predicted pulse number according to the multi-scale feature extraction result and the one-dimensional feature extraction result; A third processing module for controlling the servo motor according to the predicted duty cycle and the predicted pulse number to update the displacement measurement value so that the updated displacement measurement value is equal to the strain target value; A calibration module, configured to obtain, based on the strain acquisition device, the strain output value of the measured strain gauge when the updated displacement measurement value is equal to the strain target value, and calibrate the measured strain gauge according to the strain output value.

8. A strain gauge calibration device, characterized in that, It includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the method described in any one of claims 1 to 6 is implemented.

9. A strain gauge calibration system, characterized in that, It includes the strain gauge calibration device described in claim 8, and further includes a clamping device, a sliding device, a power device, a displacement measurement device, and a strain acquisition device; the clamping device is used to fix the measured strain gauge and the displacement measurement device, and connect the sliding device, and the power device drives the sliding device to generate a displacement through a servo motor, so as to cause the measured strain gauge to generate micro-strain.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in any one of claims 1 to 6 is implemented.

Images