A temperature measurement error correction method based on hemispherical resonator
By establishing a three-dimensional model library and deep learning technology of hemispherical tooth oscillator, the problem of inconsistent drift compensation parameters caused by temperature error of hemispherical tooth oscillator is solved, and efficient and low-cost compensation parameters are achieved, and production efficiency is improved .
Patent Information
- Application Number
- CN202411322310.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-23
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-09-23
AI Technical Summary
The production process of hemispherical toothed oscillators is complicated, which leads to inconsistent drift compensation parameters caused by temperature errors, which increases the cost of use. Each oscillator needs to measure the compensation parameters separately, which is inefficient.
Establish a three-dimensional model library of hemispherical tooth-shaped oscillators, establish a drift compensation model through deep learning, and use DGCNN and Lasso regression models to reduce information complexity, improve learning efficiency, realize accurate drift compensation parameters acquisition, and reduce costs.
The accurate shape and size recording of the hemispherical toothed oscillator is achieved, which reduces subsequent drift compensation experiments, reduces production costs, improves efficiency, and ensures the accurate compensation parameters of each oscillator without repeated experimental calculations.
Smart Images

Figure CN118836901B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of inertial navigation and control technology, and in particular to a temperature measurement error correction method based on a hemispherical resonator. Background Art
[0002] Hemispherical Resonator Gyroscope (HRG) is a high-precision inertial sensor. Its core is to use the vibration characteristics of a hemispherical resonator to measure angular velocity. HRG is mainly used for high-precision navigation, attitude control and stabilization systems, and is widely used in aerospace, marine, ground and other fields.
[0003] The resonators in the hemispherical resonator gyroscope (HRG) are of different types according to different usage scenarios. Common types of resonators include hemispherical resonators and hemispherical toothed resonators. The hemispherical toothed resonator has a toothed notch on the hemispherical lip of the hemispherical resonator. This setting makes it easier for the resonator to vibrate, and when the power is off, the vibration duration is 5-10 times that of a hemispherical resonator of the same specification.
[0004] However, the direct reason why hemispherical tooth-shaped resonators are not widely used is that their production process is very complicated, resulting in inconsistency in the compensation parameters of errors caused by high temperature even for hemispherical tooth-shaped resonators in the same processing batch. This directly leads to a significant increase in the cost of using hemispherical tooth-shaped resonators, and each hemispherical tooth-shaped resonator needs to have its compensation parameters of drift caused by temperature errors measured separately.
[0005] Therefore, how to obtain the temperature error correction parameters of the hemispherical toothed resonator with low cost and high efficiency has become an urgent problem to be solved. Summary of the invention
[0006] The present invention aims to provide a temperature measurement error correction method based on a hemispherical resonator to solve the above problems.
[0007] The present invention is achieved through the following technical solutions:
[0008] A temperature measurement error correction method based on a hemispherical resonator comprises the following steps:
[0009] Step 1: Establish a three-dimensional model library of hemispherical toothed resonator;
[0010] All the hemispherical tooth-shaped resonators with the same size and design that have been processed are modeled one by one through a solid three-dimensional modeling device, and each hemispherical tooth-shaped resonator is individually numbered.
[0011] Since the production process of the hemispherical tooth-shaped resonator is relatively complicated, the error precision requirement required for the hemispherical tooth-shaped resonator is very high, and even if the accuracy requirement under normal conditions is met, when drift compensation is required, an extra dimensional error will cause the drift compensation model to be unable to be shared. Therefore, the present invention establishes a three-dimensional model library of the hemispherical tooth-shaped resonator, and accurately records the shape and size of each hemispherical tooth-shaped resonator produced, so as to solve the problem that each hemispherical tooth-shaped resonator needs to be individually tested for its drift compensation parameters when in use.
[0012] Step 2: Select the sample;
[0013] Step 2.1: Select standard samples;
[0014] The error requirement of the standard sample is: in all size designs of the hemispherical toothed resonator, all sizes are the same as the design values, and the number of standard samples is 10-20.
[0015] Step 2.2: Select learning samples;
[0016] The error requirement of the learning sample is: among all the size designs of the hemispherical toothed resonator, more than half of the sizes are the same as the design values, and the errors of the remaining sizes meet the error requirements determined during the design. The number of learning samples is 5-10 times the number of all size parameters of the hemispherical toothed resonator.
[0017] Step 2.3: Verify the sample;
[0018] The error requirement of the verification sample is: among all the size designs of the hemispherical toothed resonator, more than half of the sizes are different from the design values, but all errors meet the error requirements determined during the design. The number of verification samples is several.
[0019] Step 2 of the present invention is to classify a number of produced hemispherical tooth-shaped resonators, among which samples that fully meet the design standards and have zero errors are used as standard samples to obtain comparison indicators for the remaining hemispherical tooth-shaped resonators. In addition, samples with certain errors but smaller numbers are selected as learning samples, which can reduce the information complexity in the subsequent learning process and increase learning efficiency.
[0020] Step 3: Establish a drift compensation model. Through experiments and calculations, obtain the drift compensation model of the standard sample, the drift compensation model of the learning sample, and the drift compensation models of several verification samples.
[0021] The present invention obtains drift compensation models for various samples through experiments and calculations, establishes raw data for subsequent deep learning, and the samples that have been tested can be directly put into use without incurring additional costs.
[0022] Step 4: Use deep learning to establish a correlation model between the three-dimensional image of the hemispherical toothed resonator and its drift compensation model;
[0023] Among them, in deep learning, the three-dimensional image information of the learning sample is used as the input layer, the three-dimensional image information of the standard sample and the drift compensation model of the standard sample are used as the hidden layer, and the drift compensation model of the learning sample is used as the output layer to obtain a preliminary model of the correlation between the three-dimensional image of the hemispherical tooth resonator and its drift compensation model. Then, by setting a verification layer, the three-dimensional image information of the verification sample is used as the input of the model, and the drift compensation model of the verification sample is output through the deep learning model and compared with the drift compensation model of the verification sample obtained from the experiment. If the verification layer fails, deep learning is continued to optimize the correlation model between the three-dimensional image of the hemispherical tooth resonator and its drift compensation model.
[0024] Since the amount of information contained in three-dimensional image information is huge and traditional algorithms cannot take it into account, a deep learning algorithm is used to obtain the most accurate drift compensation model, so that the subsequently processed hemispherical toothed resonators do not need to undergo drift compensation experiments one by one.
[0025] Step 5: Establish a hemispherical toothed resonator database;
[0026] Databases are established for hemispherical toothed resonators of different designs.
[0027] The present invention establishes a database of hemispherical toothed resonators of different designs, so that all similar products can directly call the database content in subsequent production without repeated experimental calculations, thereby greatly reducing costs.
[0028] Preferably, in step 1, the collected data is stored in the form of a point cloud, and through reconstruction and feature extraction, key features in the point cloud are obtained, including edges, curvature, horizontality, verticality, and dimensional length.
[0029] The present invention stores data through point clouds, which has the advantage that point clouds are the basic recording method for three-dimensional modeling, and the present invention does not need to convert these data into visual data. Therefore, all requirements can be met using basic data. In addition, key features are features with heavier weights and are not limited to only these features. In the present invention, features such as edges, curvatures, horizontality, verticality, and dimensional lengths are selected to cover the most important features affecting the hemispherical toothed resonator.
[0030] Preferably, in step 3, the drift data of each standard sample is detected in the range of 0-1500 degrees, and then the drift data of all standard samples are subjected to a linear regression algorithm to obtain a drift compensation model of the standard samples.
[0031] Since the material of the hemispherical toothed resonator is usually high-quality quartz stone, its material expansion rate at low temperature is extremely low and does not fall into the compensation category. Therefore, the present invention selects the range of 0-1500 degrees to detect its drift data, covering most of the temperatures that may occur during use.
[0032] Preferably, in step 4, the deep learning model specifically used is a DGCNN model (Dynamic Graph CNN), in which in the feature learning and aggregation of the point cloud, the fully connected layer is passed through the local features to determine the specific values of the edges, curvature, horizontality, verticality, and size length in the disordered point cloud, and the key values of the above parameters are obtained based on the segmentation neural network.
[0033] The deep learning model used in the present invention is specifically the DGCNN model (Dynamic Graph CNN), which is a model specifically for 3D view data analysis. It can accurately identify all feature data in the point cloud model. DGCNN effectively overcomes the limitations of traditional CNN in processing point cloud data through its dynamic graph structure and graph convolution operation.
[0034] Preferably, in step 4, the deep learning model also includes a Lasso regression model, wherein the independent variable data is provided by the DGCNN model and the dependent variable data is provided by the drift compensation model of the standard sample.
[0035] The deep learning model of the present invention also uses the Lasso regression model, which is mainly used for linear regression models with higher complexity. It improves the performance and generalization ability of the model through appropriate feature selection, regularization technology, data preprocessing and model tuning, and can effectively cope with complex data patterns and achieve accurate model construction.
[0036] Preferably, the model assumptions of the Lasso regression model are as follows:
[0037] Suppose the dependent variable y is related to multiple independent variables x1, x2, ..., x n There is a linear relationship between them, which is as follows:
[0038]
[0039] in:
[0040] y is the dependent variable (target);
[0041] X i=( x1, x2, ..., x n ) is the independent variable;
[0042] =( ) are the parameters of the model, which represent the influence of the independent variable on the dependent variable;
[0043] is the intercept term, that is, the expected value of y when all independent variables xi are 0;
[0044] is the error term, which represents the random part that cannot be explained by the model.
[0045] Preferably, the Lasso regression model also introduces regularization, and its optimization goal is to obtain the optimal parameter estimation by minimizing the following loss function:
[0046] Loss function:
[0047] in:
[0048] Indicates that it is looking for parameters The value of makes the loss function reach the minimum;
[0049] It means to sum all sample points, where N is the number of samples;
[0050] y i is the actual value of the dependent variable of the ith sample;
[0051] is the intercept term;
[0052] is the residual square of the i-th sample;
[0053] is the weight of the regularization term, which is used to control the contribution of the regularization term to the loss function;
[0054] is the sum of the absolute values of all feature coefficients, which is the L1 regularization term, where p is the total number of all feature coefficients;
[0055] This is to simplify the calculation when taking derivatives;
[0056] is the jth coefficient in the model;
[0057] is the jth eigenvalue of the ith sample;
[0058] The regularized variable parameters are provided by the drift compensation model of the validation sample.
[0059] Preferably, the key features in the point cloud also include the surface quality of the hemispherical toothed resonator.
[0060] Preferably, when selecting verification samples, start with those with the least number of errors.
[0061] Preferably, the database described in step 5 includes the produced three-dimensional model of the hemispherical resonator and a correlation model between its three-dimensional image and its drift compensation model.
[0062] Compared with the prior art, the present invention has the following advantages and beneficial effects:
[0063] 1. Since the production process of the hemispherical tooth-shaped resonator is relatively complicated, the error precision requirement required for the hemispherical tooth-shaped resonator is very high, and even if the accuracy requirements under normal conditions are met, when drift compensation is required, an extra size error will cause the drift compensation model to be unable to be shared. Therefore, the present invention establishes a three-dimensional model library of the hemispherical tooth-shaped resonator, and accurately records the shape and size of each hemispherical tooth-shaped resonator produced, so as to solve the problem that each hemispherical tooth-shaped resonator needs to be individually tested for its drift compensation parameters when in use.
[0064] 2. Step 2 of the present invention is to classify a number of produced hemispherical tooth-shaped resonators, among which samples that fully meet the design standards and have zero errors are used as standard samples to obtain comparison indicators for the remaining hemispherical tooth-shaped resonators. In addition, samples with certain errors but smaller numbers are selected as learning samples, which can reduce the information complexity in the subsequent learning process and increase learning efficiency.
[0065] 3. The present invention obtains drift compensation models for each sample through experiments and calculations, establishes raw data for subsequent deep learning, and the samples that have been tested can be directly put into use without incurring additional costs.
[0066] 4. As the amount of information contained in three-dimensional image information is huge and traditional algorithms cannot take it into account, a deep learning algorithm is used to obtain the most accurate drift compensation model, so that the hemispherical toothed resonators processed later do not need to undergo drift compensation experiments one by one.
[0067] 5. The present invention establishes a database of hemispherical toothed resonators of different designs, so that all similar products can directly call the database content in subsequent production without repeating experimental calculations, which greatly reduces costs.
[0068] 6. The present invention stores data through point clouds, which has the advantage that point clouds are the basic recording method for three-dimensional modeling, and the present invention does not need to convert these data into visual data. Therefore, all requirements can be met using basic data. In addition, key features are features with heavier weights, and are not limited to only these features. In the present invention, features such as edge, curvature, horizontality, verticality, and dimensional length are selected, covering the most important features affecting the hemispherical toothed resonator.
[0069] 7. Since the material of the hemispherical toothed resonator is usually high-quality quartz stone, its material expansion rate at low temperature is extremely low and does not fall into the compensation category. Therefore, the present invention selects the range of 0-1500 degrees to detect its drift data, covering most of the temperatures that may occur during use.
[0070] 8. The deep learning model used in the present invention is specifically the DGCNN model (Dynamic Graph CNN), which is a model specifically for 3D view data analysis. It can accurately identify all feature data in the point cloud model. DGCNN effectively overcomes the limitations of traditional CNN in processing point cloud data through its dynamic graph structure and graph convolution operation.
[0071] 9. The deep learning model of the present invention also uses the Lasso regression model, which is mainly used for linear regression models with higher complexity. It improves the performance and generalization ability of the model through appropriate feature selection, regularization technology, data preprocessing and model tuning, and can effectively cope with complex data patterns and achieve accurate model construction. BRIEF DESCRIPTION OF THE DRAWINGS
[0072] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:
[0073] Figure 1 is a flow chart of the method of the present invention;
[0074] Figure 2 It is a cross-sectional view of a hemispherical toothed resonator;
[0075] Figure 3 This is a top view of the hemispherical toothed resonator. DETAILED DESCRIPTION
[0076] In order to enable those skilled in the art to better understand the solution of the present application, the following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. 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 should fall within the scope of protection of this application.
[0077] It should be noted that if the description and claims of the present application and the above-mentioned drawings relate to the terms "first", "second", etc., they are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, if the terms "including" and "having" and any of their variations are involved, it is intended to cover non-exclusive inclusions, for example, the process, method, system, product or equipment that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or equipment.
[0078] In this application, if the terms "upper", "lower", "left", "right", "front", "back", "top", "bottom", "inside", "outside", "middle", "vertical", "horizontal", "lateral", "longitudinal", etc. are involved, the orientation or position relationship indicated is based on the orientation or position relationship shown in the drawings. These terms are mainly used to better describe the present application and its embodiments, and are not used to limit the indicated devices, elements or components to have a specific orientation, or to be constructed and operated in a specific orientation.
[0079] In addition, some of the above terms may be used to express other meanings in addition to indicating orientation or positional relationship. For example, the term "on" may also be used to express a certain dependency or connection relationship in some cases. For those of ordinary skill in the art, the specific meanings of these terms in this application can be understood according to specific circumstances.
[0080] In addition, in this application, the terms "installed", "set", "provided with", "connected", "connected", "socketed" and the like should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral structure; it can be a mechanical connection, or an electrical connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be an internal connection between two devices, elements or components. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to the specific circumstances.
[0081] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0082] Example
[0083] like Figures 1 to 3 As shown, this embodiment includes:
[0084] A temperature measurement error correction method based on a hemispherical resonator comprises the following steps:
[0085] Step 1: Establish a three-dimensional model library of hemispherical toothed resonator;
[0086] All the hemispherical tooth-shaped resonators with the same size and design that have been processed are modeled one by one through a solid three-dimensional modeling device, and each hemispherical tooth-shaped resonator is individually numbered. The shape of the hemispherical tooth-shaped resonator is as follows: Figure 2 and Figure 3 shown.
[0087] Since the production process of the hemispherical tooth-shaped resonator is relatively complicated, the error precision requirement required for the hemispherical tooth-shaped resonator is very high, and even if the accuracy requirement under normal conditions is met, when drift compensation is required, an extra dimensional error will cause the drift compensation model to be unable to be shared. Therefore, the present invention establishes a three-dimensional model library of the hemispherical tooth-shaped resonator, and accurately records the shape and size of each hemispherical tooth-shaped resonator produced, so as to solve the problem that each hemispherical tooth-shaped resonator needs to be individually tested for its drift compensation parameters when in use.
[0088] The three-dimensional model of this embodiment uses laser scanning technology to obtain point cloud data of the object surface.
[0089] Step 2: Select the sample;
[0090] Step 2.1: Select standard samples;
[0091] The error requirement of the standard sample is: in all dimension designs of the hemispherical toothed resonator, all dimensions are the same as the design values, and the number of standard samples is 10-20, preferably 15.
[0092] Step 2.2: Select learning samples;
[0093] The error requirement of the learning sample is: among all the size designs of the hemispherical toothed resonator, more than half of the sizes are the same as the design values, and the errors of the remaining sizes meet the error requirements determined during the design. The number of learning samples is 5-10 times the number of all size parameters of the hemispherical toothed resonator, preferably 10 times;
[0094] Step 2.3: Verify the sample;
[0095] The error requirement of the verification sample is: among all the size designs of the hemispherical toothed resonator, more than half of the sizes are different from the design values, but all errors meet the error requirements determined during the design. The number of verification samples is several.
[0096] Step 2 of the present invention is to classify a number of produced hemispherical tooth-shaped resonators, among which samples that fully meet the design standards and have zero errors are used as standard samples to obtain comparison indicators for the remaining hemispherical tooth-shaped resonators. In addition, samples with certain errors but smaller numbers are selected as learning samples, which can reduce the information complexity in the subsequent learning process and increase learning efficiency.
[0097] Step 3: Establish a drift compensation model. Through experiments and calculations, obtain the drift compensation model of the standard sample, the drift compensation model of the learning sample, and the drift compensation models of several verification samples.
[0098] The present invention obtains drift compensation models for various samples through experiments and calculations, establishes raw data for subsequent deep learning, and the samples that have been tested can be directly put into use without incurring additional costs.
[0099] Step 4: Use deep learning to establish a correlation model between the three-dimensional image of the hemispherical toothed resonator and its drift compensation model;
[0100] Among them, in deep learning, the three-dimensional image information of the learning sample is used as the input layer, the three-dimensional image information of the standard sample and the drift compensation model of the standard sample are used as the hidden layer, and the drift compensation model of the learning sample is used as the output layer to obtain a preliminary model of the correlation between the three-dimensional image of the hemispherical tooth resonator and its drift compensation model. Then, by setting a verification layer, the three-dimensional image information of the verification sample is used as the input of the model, and the drift compensation model of the verification sample is output through the deep learning model and compared with the drift compensation model of the verification sample obtained from the experiment. If the verification layer fails, deep learning is continued to optimize the correlation model between the three-dimensional image of the hemispherical tooth resonator and its drift compensation model.
[0101] Since the amount of information contained in three-dimensional image information is huge and traditional algorithms cannot take it into account, a deep learning algorithm is used to obtain the most accurate drift compensation model, so that the subsequently processed hemispherical toothed resonators do not need to undergo drift compensation experiments one by one.
[0102] Step 5: Establish a hemispherical toothed resonator database;
[0103] Databases are established for hemispherical toothed resonators of different designs.
[0104] The present invention establishes a database of hemispherical toothed resonators of different designs, so that all similar products can directly call the database content in subsequent production without repeated experimental calculations, thereby greatly reducing costs.
[0105] In this embodiment, in step 1, the collected data is stored in the form of point cloud, and through reconstruction and feature extraction, key features in the point cloud are obtained, including edges, curvature, horizontality, verticality, and dimensional length.
[0106] The present invention stores data through point clouds, which has the advantage that point clouds are the basic recording method for three-dimensional modeling, and the present invention does not need to convert these data into visual data. Therefore, all requirements can be met using basic data. In addition, key features are features with heavier weights and are not limited to only these features. In the present invention, features such as edges, curvatures, horizontality, verticality, and dimensional lengths are selected to cover the most important features affecting the hemispherical toothed resonator.
[0107] In this embodiment, in step 3, the drift data of each standard sample is detected in the range of 0-1500 degrees, and then the drift data of all standard samples are subjected to a linear regression algorithm to obtain a drift compensation model of the standard sample.
[0108] Since the material of the hemispherical toothed resonator is usually high-quality quartz stone, its material expansion rate at low temperature is extremely low and does not fall into the compensation category. Therefore, the present invention selects the range of 0-1500 degrees to detect its drift data, covering most of the temperatures that may occur during use.
[0109] In this embodiment, in step 4, the deep learning model specifically used is the DGCNN model (Dynamic GraphCNN), in which in the feature learning and aggregation of the point cloud, the fully connected layer is passed through the local features to determine the specific values of the edges, curvature, horizontality, verticality, and size length in the disordered point cloud, and the key values of the above parameters are obtained based on the segmentation neural network.
[0110] The deep learning model used in the present invention is specifically the DGCNN model (Dynamic Graph CNN), which is a model specifically for 3D view data analysis. It can accurately identify all feature data in the point cloud model. DGCNN effectively overcomes the limitations of traditional CNN in processing point cloud data through its dynamic graph structure and graph convolution operation.
[0111] In this embodiment, in step 4, the deep learning model also includes a Lasso regression model, wherein the independent variable data is provided by the DGCNN model, and the dependent variable data is provided by the drift compensation model of the standard sample.
[0112] The deep learning model of the present invention also uses the Lasso regression model, which is mainly used for linear regression models with higher complexity. It improves the performance and generalization ability of the model through appropriate feature selection, regularization technology, data preprocessing and model tuning, and can effectively cope with complex data patterns and achieve accurate model construction.
[0113] In this embodiment, the model assumptions of the Lasso regression model are as follows:
[0114] Suppose the dependent variable y is related to multiple independent variables x1, x2, ..., x n There is a linear relationship between them, which is as follows:
[0115]
[0116] in:
[0117] y is the dependent variable (target);
[0118] X i=( x1, x2, ..., x n ) is the independent variable;
[0119] =( ) are the parameters of the model, which represent the influence of the independent variable on the dependent variable;
[0120] is the intercept term, that is, the expected value of y when all independent variables xi are 0;
[0121] is the error term, which represents the random part that cannot be explained by the model.
[0122] In this embodiment, the Lasso regression model also introduces regularization, and its optimization goal is to obtain the optimal parameter estimation by minimizing the following loss function:
[0123] Loss function:
[0124] in:
[0125] Indicates that it is looking for parameters The value of makes the loss function reach the minimum;
[0126] It means to sum all sample points, where N is the number of samples;
[0127] y i is the actual value of the dependent variable of the ith sample;
[0128] is the intercept term;
[0129] is the residual square of the i-th sample;
[0130] is the weight of the regularization term, which is used to control the contribution of the regularization term to the loss function;
[0131] is the sum of the absolute values of all feature coefficients, which is the L1 regularization term, where p is the total number of all feature coefficients;
[0132] This is to simplify the calculation when taking derivatives;
[0133] is the jth coefficient in the model;
[0134] is the jth eigenvalue of the ith sample;
[0135] The regularized variable parameters are provided by the drift compensation model of the validation sample.
[0136] In this embodiment, the key features in the point cloud also include the surface quality of the hemispherical toothed resonator.
[0137] In this embodiment, when selecting verification samples, the samples with the least number of errors are selected first.
[0138] In this embodiment, the database described in step 5 includes the produced three-dimensional model of the hemispherical resonator and a correlation model between its three-dimensional image and its drift compensation model.
[0139] Those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including a number of instructions to enable a single-chip microcomputer, a chip or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.
[0140] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.
Claims
1. A temperature measurement error correction method based on a hemispherical resonator, characterized in that: The steps include: Step 1: Establish a three-dimensional model library of hemispherical toothed resonator; All the hemispherical tooth-shaped resonators of the same size and design that have been processed are modeled one by one through a solid three-dimensional modeling device, and each hemispherical tooth-shaped resonator is individually numbered; Step 2: Select the sample; Step 2.1: Select standard samples; The error requirements of the standard sample are: in all dimension designs of the hemispherical toothed resonator, all dimensions are the same as the design values, and the number of standard samples is 10-20; Step 2.2: Select learning samples; The error requirement of the learning sample is: among all the size designs of the hemispherical toothed resonator, more than half of the sizes are the same as the design values, and the errors of the remaining sizes meet the error requirements determined during the design. The number of learning samples is 5-10 times the number of all size parameters of the hemispherical toothed resonator. Step 2.3: Verify the sample; The error requirements of the verification sample are: among all the dimension designs of the hemispherical toothed resonator, more than half of the dimensions are different from the design values, but all the errors meet the error requirements determined during the design; Step 3: Establish a drift compensation model. Through experiments and calculations, obtain the drift compensation model of the standard sample, the drift compensation model of the learning sample, and the drift compensation model of all verification samples. Step 4: Use a deep learning algorithm to establish a correlation model between the three-dimensional image of the hemispherical toothed resonator and its drift compensation model; Wherein, in the deep learning algorithm, the three-dimensional image information of the learning sample is used as the input layer, the three-dimensional image information of the standard sample and the drift compensation model of the standard sample are used as the hidden layer, and the drift compensation model of the learning sample is used as the output layer to obtain a preliminary model of the correlation between the three-dimensional image of the hemispherical tooth-shaped resonator and its drift compensation model, and then by setting a verification layer, the three-dimensional image information of the verification sample is used as the input of the model, and the drift compensation model of the verification sample is output through the model after deep learning, and compared with the drift compensation model of the verification sample obtained by experimental calculation in step 3. If it fails to pass the verification layer, the deep learning is continued to optimize the correlation model between the three-dimensional image of the hemispherical tooth-shaped resonator and its drift compensation model; Step 5: Establish a hemispherical toothed resonator database; Databases are established for hemispherical toothed resonators of different designs.
2. A temperature measurement error correction method based on a hemispherical resonator according to claim 1, characterized in that: In step 1, the collected data is stored in the form of point cloud, and through reconstruction and feature extraction, the key features in the point cloud are obtained, including edges, curvature, horizontality, verticality, and size length.
3. The temperature measurement error correction method based on a hemispherical resonator according to claim 1 is characterized in that: In step 3, the drift data of each standard sample is detected in the range of 0-1500 degrees, and then the drift data of all standard samples are used through a linear regression algorithm to obtain a drift compensation model of the standard sample.
4. The temperature measurement error correction method based on a hemispherical resonator according to claim 1 is characterized in that: In step 4, the deep learning model specifically used is the DGCNN model, in which in the feature learning and aggregation of the point cloud, the fully connected layer is passed through the local feature to determine the specific values of the edge, curvature, horizontality, verticality, and size length in the disordered point cloud, and according to the segmentation neural network, the key values of the edge, curvature, horizontality, verticality, and size length in the disordered point cloud are obtained.
5. A temperature measurement error correction method based on a hemispherical resonator according to claim 4, characterized in that: In step 4, the deep learning model also includes a Lasso regression model, in which the independent variable data is provided by the DGCNN model and the dependent variable data is provided by the drift compensation model of the standard sample.
6. A temperature measurement error correction method based on a hemispherical resonator according to claim 5, characterized in that: The model assumptions of the Lasso regression model are as follows: Suppose the dependent variable y is related to multiple independent variables x1, x2, ..., x n There is a linear relationship between them, which is as follows: in: y is the dependent variable; X i=( x1, x2, ..., x n ) is the independent variable; =( ) are the parameters of the model, which represent the influence of the independent variable on the dependent variable; is the intercept term, that is, the expected value of y when all independent variables xi are 0; is the error term, which represents the random part that cannot be explained by the model.
7. A temperature measurement error correction method based on a hemispherical resonator according to claim 6, characterized in that: The Lasso regression model also introduces regularization, and its optimization goal is to obtain the optimal parameter estimation by minimizing the following loss function: Loss function: in: Indicates that it is looking for parameters The value of makes the loss function reach the minimum; It means to sum all sample points, where N is the number of samples; y i is the actual value of the dependent variable of the ith sample; is the intercept term; is the square of the residual of the i-th sample; is the weight of the regularization term, which is used to control the contribution of the regularization term to the loss function; is the sum of the absolute values of all feature coefficients, which is the L1 regularization term, where p is the total number of all feature coefficients; This is to simplify the calculation when taking derivatives; is the jth coefficient in the model; is the jth eigenvalue of the ith sample; The regularized variable parameters are provided by the drift compensation model of the validation sample.
8. The temperature measurement error correction method based on a hemispherical resonator according to claim 2 is characterized in that: Key features in the point cloud also include the surface quality of the hemispherical toothed resonator.
9. The temperature measurement error correction method based on a hemispherical resonator according to claim 1, characterized in that: When selecting validation samples, start with those with the lowest number of errors.
10. The temperature measurement error correction method based on a hemispherical resonator according to claim 1, characterized in that: The database described in step 5 includes the three-dimensional model of the produced hemispherical resonator and the correlation model between its three-dimensional image and its drift compensation model.
