A measurement data fusion method

Abstract

The invention discloses a measurement data fusion method, which is applied to a large-size measured part and belongs to the field of data fusion technology. The method comprises: classifying a plurality of measuring instruments according to measurable parameters; combining the numbers of the plurality of measuring instruments into an instrument matrix; measuring the measurement values ​​of each characteristic parameter of a standard part; combining the measurement values ​​of the characteristic parameters into a measurement matrix; normalizing the column vectors of the measurement matrix to obtain a normalized matrix; establishing an augmented matrix according to the instrument matrix and the normalized matrix; calculating the correlation coefficient between the column vectors of two augmented matrices, and judging whether the correlation coefficient is greater than a preset value; when the correlation coefficient is greater than the preset value, performing dimensionality reduction processing on the column vectors of the two augmented matrices; inputting the column vectors of each dimensionality reduction matrix into a data fusion model for training; and terminating the training of the data fusion model when the deviation between the output value and the standard value of each characteristic parameter of the standard part is within a preset range.

Description

Technical Field

[0001] The present invention belongs to the technical field of data fusion, and in particular relates to a measurement data fusion method. Background Art

[0002] In industrial manufacturing, large domestic aircraft, missiles, rockets, trains, ships, turbines, and generators all rely on large-scale geometric measurement technology. Large-scale measurement refers to the measurement of geometric parameters such as the position, coordinates, size, shape, and motion trajectory of the measured object within a spatial range of several meters to several hundred meters. Compared with conventional dimensional measurement, large-scale measurement is mainly used in industrial manufacturing sites with large measurement spaces, relatively complex environments, and many interference factors. The measuring instruments used include theodolites, total stations, laser trackers, lidars, scanners, interferometers, etc. These measuring instruments can measure one or more parameters, and the types and accuracy of the measured parameters vary.

[0003] For example, electronic theodolites require manual aiming, resulting in low measurement efficiency and low coordinate measurement accuracy. However, they play an irreplaceable role in the alignment of cubic mirrors in satellite and spacecraft angle measurement. LiDAR offers the advantages of high measurement efficiency, non-contact measurement, and high measurement accuracy, but cannot perform dynamic measurement. Laser trackers offer high measurement accuracy and can perform both static and dynamic measurements, but they require the use of corner cube prisms. For large and complex test pieces, a single large-scale measuring instrument is insufficient, requiring a combination of multiple instruments.

[0004] When using multiple measuring instruments to measure large-sized test pieces, the measurement data are often different due to the differences in the measuring instruments. In the process of processing and fusing the measurement data obtained by different measuring instruments, a lot of manpower is wasted, and the efficiency, accuracy and error are large. Summary of the Invention

[0005] The purpose of the embodiments of the present invention is to provide a measurement data fusion method that can solve the existing technical problems of wasting a lot of human resources, low efficiency, low precision and large errors in the process of processing and fusing measurement data measured by different measuring instruments.

[0006] In order to solve the above-mentioned technical problems, the present invention is achieved as follows:

[0007] An embodiment of the present invention provides a measurement data fusion method applied to a large-sized device under test, comprising:

[0008] S101: Classify multiple measuring instruments according to measurable parameters;

[0009] S102: Encoding the measuring instruments, combining the numbers of multiple measuring instruments into an instrument matrix;

[0010] S103: measuring the standard part using multiple measuring instruments to obtain measurement values ​​of characteristic parameters of the standard part, wherein the standard values ​​of the characteristic parameters of the standard part are known quantities, and the characteristic parameters include spatial coordinates, length, and angle;

[0011] S104: combining the measured values ​​of the characteristic parameters into a measurement matrix according to the measurable parameters of each measuring instrument;

[0012] S105: performing normalization processing on the column vectors of the measurement matrix to obtain a normalized matrix;

[0013] S106: establishing an augmented matrix according to the instrument matrix and the normalized matrix;

[0014] S107: performing a correlation analysis on the column vectors of any two augmented matrices, calculating a correlation coefficient between the column vectors of the two augmented matrices, and determining whether the correlation coefficient is greater than a preset value;

[0015] S108: When the correlation coefficient between the column vectors of the two augmented matrices is greater than a preset value, performing dimensionality reduction processing on the column vectors of the two augmented matrices to obtain a reduced-dimensional matrix;

[0016] S109: Build a data fusion model based on machine learning algorithms;

[0017] S110: Inputting the column vectors of each dimensionality reduction matrix into the data fusion model for training;

[0018] S111: The data fusion model outputs the output values ​​of each characteristic parameter of the standard component;

[0019] S112: When the deviation between the output value of each characteristic parameter of the standard part and the standard value of each characteristic parameter of the standard part is within a preset range, the training of the data fusion model is ended.

[0020] In an embodiment of the present invention, multiple measuring instruments are classified and encoded, and the measured data are gradually normalized and reduced in dimension. The column vectors of the obtained reduced-dimensional matrix are input into a data fusion model constructed using a machine learning algorithm for training, thereby gradually reducing the error of the measurement data fusion result. After the training is completed, the measurement data of the multiple measuring instruments are input into the data fusion model for fusion, and the fusion result can be obtained, which saves a lot of manpower input, greatly improves the fusion efficiency, and the fused measurement result has high accuracy and small error. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1It is a flowchart of a measurement data fusion method provided by an embodiment of the present invention.

[0022] The realization of the objectives, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0023] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0024] The measurement data fusion method provided by the embodiment of the present invention is described in detail below with reference to the accompanying drawings through specific embodiments and application scenarios.

[0025] Reference Figure 1 , which shows a flow chart of a measurement data fusion method provided by an embodiment of the present invention.

[0026] An embodiment of the present invention provides a measurement data fusion method, which is applied to a large-sized device under test.

[0027] Measurement data fusion methods include:

[0028] S101: Classify multiple measuring instruments according to measurable parameters.

[0029] The measurable parameters may be angles, lengths, and spatial coordinates. Multiple measuring instruments may be classified according to angles, lengths, and spatial coordinates. That is, among the multiple measuring instruments, it is possible to distinguish which ones measure angles, which ones measure lengths, and which ones measure spatial coordinates.

[0030] Optionally, the surveying instrument is a tracker, a theodolite, a laser scanner and a total station.

[0031] S102: Encode the measuring instruments, and combine the numbers of multiple measuring instruments into an instrument matrix.

[0032] It should be noted that encoding the measuring instrument can mark subsequent measurement data to indicate which measuring instrument the measurement data was measured by, so as to facilitate subsequent analysis.

[0033] In a possible implementation, there are four measuring instruments, namely a first measuring instrument, a second measuring instrument, a third measuring instrument, and a fourth measuring instrument. S102 specifically includes:

[0034] S1021: Represent each measuring instrument as a row matrix, where the first measuring instrument is represented as [1, 0, 0, 0], the second measuring instrument is represented as [0, 1, 0, 0], the third measuring instrument is represented as [0, 0, 1, 0], and the fourth measuring instrument is represented as [0, 0, 0, 1];

[0035] S1022: Obtain the number of times each measuring instrument measures the standard part, where the first measuring instrument measures the standard part i times, the second measuring instrument measures the standard part j times, the third measuring instrument measures the standard part k times, and the fourth measuring instrument measures the standard part g times.

[0036] S1023: Construct an instrument matrix based on the row matrix of each measuring instrument and the corresponding number of measurements:

[0037]

[0038] S103: Measure the standard part using multiple measuring instruments to obtain measurement values ​​of various characteristic parameters of the standard part.

[0039] Among them, standard parts refer to commonly used parts (components) whose structure, size, drawing method, marking, etc. have been completely standardized and produced by professional factories.

[0040] That is to say, the standard values ​​of each characteristic parameter of the standard part are known quantities, and the characteristic parameters include spatial coordinates, length, and angle.

[0041] It should be noted that by measuring the standard parts, the difference between the measurement results of the measuring instrument and the actual results can be intuitively perceived, which facilitates the calibration of the measuring instrument.

[0042] S104: According to the measurable parameters of each measuring instrument, the measured values ​​of the characteristic parameters are combined into a measurement matrix.

[0043] Among them, combining the chaotic and disordered measurement values ​​of characteristic parameters into an ordered measurement matrix and directly performing matrix operations in the subsequent analysis process can greatly improve the efficiency of data analysis.

[0044] In a possible implementation, S104 specifically includes:

[0045] S1041: Combine the measurement parameters into a row vector [α, l, x, y, z];

[0046] Among them, α represents the angle of the standard part, l represents the length of the standard part, and x, y, and z represent the spatial coordinates of the standard part.

[0047] S1042: Based on the measurable parameters of each measuring instrument, fill the measured values ​​of the characteristic parameters measured by each measuring instrument into the corresponding positions of the row vector, and fill the remaining positions with 0 to obtain the measurement matrix b:

[0048]

[0049] Among them, the first subscripts 1, 2, 3 and 4 of α, l, x, y and z represent the first measuring instrument, the second measuring instrument, the third measuring instrument and the fourth measuring instrument respectively, and the second subscripts of α, l, x, y and z represent the number of measurements.

[0050] S105: Normalize the column vectors of the measurement matrix to obtain a normalized matrix.

[0051] There are two main methods for normalization: converting numbers to decimals between (0 and 1), and converting dimensional expressions to dimensionless ones. Both methods can confine data to a certain range, thereby reducing the adverse effects of singular sample data and generalizing the statistical distribution of the sample.

[0052] In a possible implementation, taking the normalization process of converting a dimensional expression into a dimensionless expression as an example, S105 specifically includes:

[0053] S1051: Perform Max-Min normalization on each column of the measurement matrix to obtain a normalized matrix B:

[0054]

[0055] Among them, the value range of each column in the normalized matrix is ​​[0, 1] and dimensionless. The modulus of the column vector of the normalized matrix is ​​equal to 1, that is, |B f |=1, where f=1, 2, 3, 4 or 5.

[0056] Among them, Min-Max Normalization, also known as deviation normalization, is to perform a linear transformation on each column matrix in the measurement matrix.

[0057] S106: Establish an augmented matrix based on the instrument matrix and the normalized matrix.

[0058] It should be noted that by combining the instrument matrix and the normalized matrix into an augmented matrix and then performing subsequent analysis on the augmented matrix, both instrument features and data features can be processed simultaneously, thereby improving data processing efficiency.

[0059] In a possible implementation, S106 specifically includes:

[0060] S1061: Based on the initial matrix and the normalized matrix, establish the augmented matrix C:

[0061]

[0062] S107: performing a correlation analysis on the column vectors of any two augmented matrices, calculating a correlation coefficient between the column vectors of the two augmented matrices, and determining whether the correlation coefficient is greater than a preset value.

[0063] Among them, correlation analysis refers to the analysis of two or more correlated variable elements to measure the degree of correlation between the two variable factors.

[0064] Among them, those skilled in the art can determine the size of the preset value according to actual conditions, and the present invention does not limit it.

[0065] It can be understood that the larger the correlation coefficient is, the stronger the relationship between the two column vectors subjected to correlation analysis is.

[0066] In a possible implementation, S107 specifically includes:

[0067] S1071: Perform a covariance operation on the column vectors of any two augmented matrices to calculate the correlation coefficient between the column vectors of the two augmented matrices. The correlation coefficient is calculated as follows:

[0068]

[0069] Where cov(x, y) represents the covariance operation, represents the correlation coefficient, s and q represent the number of columns of the augmented matrix, that is, s,q∈[1,2,...,6] and s≠q, and r represents the number of rows of the augmented matrix;

[0070] S1072: Determine whether the correlation coefficient is greater than a preset value.

[0071] S108: When the correlation coefficient between the column vectors of the two augmented matrices is greater than a preset value, perform dimensionality reduction processing on the column vectors of the two augmented matrices to obtain a reduced-dimensional matrix.

[0072] It should be noted that when the correlation coefficient between the column vectors of the two augmented matrices is greater than a preset value, it can be understood that the correlation between the column vectors of the two augmented matrices is too low, and dimensionality reduction processing is required to eliminate the impact of singular data on subsequent processing efficiency.

[0073] It should be noted that after dimensionality reduction, the parameters will be reduced accordingly, which speeds up the calculation.

[0074] In a possible implementation, S108 specifically includes:

[0075] S1081: Based on the augmented matrix, construct the covariance matrix Cov:

[0076] Cov=CC T Formula 6;

[0077] S1082: Calculate the eigenvalues ​​and corresponding eigenvectors of the covariance matrix;

[0078] S1083: Sort the eigenvectors in descending order according to the eigenvalues, and construct a mapping matrix based on the eigenvectors;

[0079] S1084: Mapping the original data of the augmented matrix to a low-dimensional feature subspace through a mapping matrix;

[0080] S1085: Calculate the variance explanation rate of the eigenvalues, select the eigenvalues ​​whose variance explanation rate is greater than a preset ratio into the model, and obtain the dimensionality reduction matrix.

[0081] Optionally, the preset ratio is 90%.

[0082] S109: Build a data fusion model based on machine learning algorithms.

[0083] Optionally, the machine learning algorithm is a BP neural network algorithm.

[0084] Among them, the BP (Back Propagation) neural network algorithm is a supervised learning algorithm that mainly uses input training samples and the back propagation algorithm to repeatedly adjust the weights and biases to make the output value of the measured data as close to the standard value as possible.

[0085] S110: Inputting the column vectors of each dimensionality reduction matrix into the data fusion model for training.

[0086] It can be understood that the column vectors of the dimensionality reduction matrix are input into the data fusion model as training samples. The more training samples there are and the more representative they are, the more accurate the fusion results of the trained data fusion model will be.

[0087] S111: The data fusion model outputs the output values ​​of each characteristic parameter of the standard part.

[0088] It should be noted that the output value at this time can be understood as the measurement result given by the data fusion model, and further comparison with the actual results of the standard parts is required to evaluate the measurement accuracy of the data fusion model.

[0089] S112: When the deviation between the output value of each characteristic parameter of the standard part and the standard value of each characteristic parameter of the standard part is within a preset range, the training of the data fusion model is ended.

[0090] Those skilled in the art can select the size of the preset range based on actual circumstances. It is understood that a larger preset range indicates lower requirements for the measurement accuracy of the data fusion model, while a smaller preset range indicates higher requirements for the measurement accuracy of the data fusion model. In actual applications, a smaller preset range does not necessarily lead to higher measurement accuracy for the data fusion model. This is because higher requirements for the measurement accuracy of the data fusion model increase the model training time required.

[0091] It's understandable that the training process of a data fusion model involves a continuous convergence of output values. Training stops when the error between the output value and the standard value falls within a preset range. After training, measurement data from multiple measuring instruments can be fed into the data fusion model as input for data fusion.

[0092] In an embodiment of the present invention, multiple measuring instruments are classified and encoded, and the measured data are gradually normalized and reduced in dimension. The column vectors of the obtained reduced-dimensional matrix are input into a data fusion model constructed using a machine learning algorithm for training, thereby gradually reducing the error of the measurement data fusion result. After the training is completed, the measurement data of the multiple measuring instruments are input into the data fusion model for fusion, and the fusion result can be obtained, which saves a lot of manpower input, greatly improves the fusion efficiency, and the fused measurement result has high accuracy and small error.

[0093] The foregoing is merely an embodiment of the present invention and is not intended to limit the present invention. It will be apparent to those skilled in the art that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention are intended to be included within the scope of the claims of the present invention.

Claims

1. A measurement data fusion method, applied to large-sized test pieces, characterized in that: include: S101: Classify multiple measuring instruments according to measurable parameters; S102: Encoding the measuring instruments, and combining the serial numbers of a plurality of the measuring instruments into an instrument matrix; S103: measuring the standard part using the plurality of measuring instruments to obtain measurement values ​​of characteristic parameters of the standard part, wherein the standard values ​​of the characteristic parameters of the standard part are known quantities, and the characteristic parameters include spatial coordinates, lengths, and angles; S104: combining the measured values ​​of the characteristic parameters into a measurement matrix according to the measurable parameters of the respective measuring instruments; S105: performing normalization processing on the column vectors of the measurement matrix to obtain a normalized matrix; S106: establishing an augmented matrix according to the instrument matrix and the normalized matrix; S107: performing a correlation analysis on any two column vectors of the augmented matrices, calculating a correlation coefficient between the column vectors of the two augmented matrices, and determining whether the correlation coefficient is greater than a preset value; S108: When the correlation coefficient between the column vectors of the two augmented matrices is greater than the preset value, performing dimensionality reduction processing on the column vectors of the two augmented matrices to obtain a reduced-dimensional matrix; S109: Build a data fusion model based on machine learning algorithms; S110: Inputting the column vectors of each of the dimensionality reduction matrices into the data fusion model for training; S111: the data fusion model outputs the output values ​​of each characteristic parameter of the standard component; S112: If the deviation between the output value of each characteristic parameter of the standard component and the standard value of each characteristic parameter of the standard component is within a preset range, the training of the data fusion model is terminated.

2. The measurement data fusion method according to claim 1, characterized in that: There are four measuring instruments, namely a first measuring instrument, a second measuring instrument, a third measuring instrument and a fourth measuring instrument. S102 specifically includes: S1021: Represent each measuring instrument as a row matrix, wherein the first measuring instrument is represented as [1, 0, 0, 0], the second measuring instrument is represented as [0, 1, 0, 0], the third measuring instrument is represented as [0, 0, 1, 0], and the fourth measuring instrument is represented as [0, 0, 0, 1]; S1022: Obtain the number of times each measuring instrument measures the standard part, where the number of times the first measuring instrument measures the standard part is i times, the second measuring instrument measures the standard part a number of times j times, the third measuring instrument measures the standard part a number of times k times, the fourth measuring instrument measures the standard part a number of times g Second-rate; S1023: Construct the instrument matrix according to the row matrix of each measuring instrument and the corresponding number of measurements A :

3. The measurement data fusion method according to claim 2, characterized in that: The S104 specifically includes: S1041: Combining the measurable parameters into a row vector S1042: Fill the measured values ​​of the characteristic parameters measured by each measuring instrument into the corresponding positions of the row vector according to the measurable parameters of each measuring instrument, and fill the remaining positions with 0 to obtain the measurement matrix b : in, α Indicates the standard part angle, Indicates the length of the standard part, x , y , z represents the spatial coordinates of the standard part, α 、 、 x The first subscripts 1, 2, 3 and 4 of y and z represent the first measuring instrument, the second measuring instrument, the third measuring instrument and the fourth measuring instrument respectively. α 、 、 x 、 y and z The second subscript indicates the number of measurements.

4. The measurement data fusion method according to claim 3, characterized in that: The S105 specifically includes: S1051: Perform Max-Min normalization processing on each column matrix of the measurement matrix to obtain the normalized matrix B : The value range of each column in the normalized matrix is ​​[0, 1] and dimensionless, and the modulus of the column vector of the normalized matrix is ​​equal to 1, that is, , where f=1, 2, 3, 4 or 5.

5. The measurement data fusion method according to claim 4, characterized in that: The S106 specifically includes: S1061: Establish an augmented matrix based on the instrument matrix and the normalized matrix C :

6. The measurement data fusion method according to claim 5, characterized in that: The S107 specifically includes: S1071: Perform a covariance operation on any two column vectors of the augmented matrices to calculate the correlation coefficient between the column vectors of the two augmented matrices. The correlation coefficient is calculated as follows: in, cov ( x , y ) represents the covariance operation, represents the correlation coefficient, s and q represents the number of columns of the augmented matrix, that is, s , r represents the number of rows of the augmented matrix; S1072: Determine whether the correlation coefficient is greater than the preset value.

7. The measurement data fusion method according to claim 6, characterized in that: The S108 specifically includes: S1081: Construct a covariance matrix based on the augmented matrix Cov : S1082: Calculate the eigenvalues ​​and corresponding eigenvectors of the covariance matrix; S1083: Sort the eigenvectors in descending order according to the eigenvalues, and construct a mapping matrix according to the eigenvectors; S1084: Mapping the original data of the augmented matrix to a low-dimensional feature subspace through the mapping matrix; S1085: Calculate the variance explanation rate of the eigenvalues, select the eigenvalues ​​whose variance explanation rate is greater than a preset ratio to enter the model, and obtain the dimensionality reduction matrix.

8. The measurement data fusion method according to claim 7, characterized in that: The preset ratio is 90%.

9. The measurement data fusion method according to claim 1, characterized in that: The surveying instruments are trackers, theodolites, laser scanners and total stations.

10. The measurement data fusion method according to claim 1, characterized in that: The machine learning algorithm is a BP neural network algorithm.

Images