Random impact load positioning method based on reference database

Abstract

The invention provides a random impact load positioning method based on a reference database, and the method is characterized in that the method comprises the following steps: S1, collecting a plurality of groups of reference data; s2, collecting multiple groups of impact data through a sensor; s3, obtaining a data set according to the reference data and the impact data; s4, calculating a cosine distance according to the data set; s5, constructing a main feature matrix according to the cosine distance; s6, interpolating the main feature matrix to obtain a middle feature matrix; s7, performing superposition operation on the intermediate feature matrix to obtain a superposition matrix; s8, performing logarithmic operation on the superposition matrix to obtain a final feature matrix; step S9, obtaining a preferred point according to the final feature matrix; s10, obtaining a weight value according to the optimal selection point; and S11, obtaining the position of the impact point according to the weight value and the optimal point coordinate. In a word, the method can obtain an accurate impact point position prediction result through limited grid points.

Description

Technical Field

[0001] The invention relates to load impact positioning, and in particular to a random impact load positioning method based on a reference database. Background Art

[0002] For engineering structures, life cycle management that ensures their health status is crucial, and to achieve this, accurate identification of random impact loads imposed on the structure is vital.

[0003] In the prior art, the intelligent methods for locating random impact loads can be divided into three categories: feature extraction method based on impact response waveform, dynamic optimization method and reference database method. The feature extraction method based on waveform is applicable to beams and plates, the dynamic optimization method is applicable to limited degree of freedom systems, the reference database method is applicable to complex structures, and the reference database method can adjust the positioning accuracy by changing the density of the reference database to meet the different requirements of accuracy for various engineering applications, so the reference database method is widely used.

[0004] The existing reference database methods include reference database methods based on correlation, root mean square error or determination coefficient, reference database methods based on spline interpolation, etc. Among them, the reference database methods based on correlation, root mean square error or determination coefficient are easily affected by the density of reference points, and their positioning accuracy needs to be improved; the reference database method based on spline interpolation can reduce the influence of reference point density on the accuracy to a certain extent, but it does not consider the influence of multiple existing reference points in the interpolation process, and there is a certain interpolation error. At the same time, its positioning area is limited to the area surrounded by the reference points, and the impact point position outside the area cannot be located. In addition, most of the current reference database methods combine multiple machine learning algorithms, but because the accuracy and efficiency of the machine learning algorithm are affected by the selected features, and its internal derivation process is not transparent, the further development of the reference database method is limited.

[0005] Therefore, there is still much room for improvement in the positioning accuracy and efficiency of the existing reference database method. Summary of the invention

[0006] The present invention is made to solve the above-mentioned problem, and aims to provide a random impact load positioning method based on a reference database.

[0007] The present invention provides a random impact load positioning method based on a reference database, which is used to determine the position of the impact point when a composite material plate provided with k sensors is impacted by a random load, and has the following characteristics, comprising the following steps: step S1, setting m*n grid points on the composite material plate, and collecting the impact response time domain data of each grid point when impacted by the k sensors, to obtain m*n*k groups of reference data; step S2, collecting the impact response time domain data when the random load is impacted by the k sensors, to obtain 1*k groups of impact data; step S3, normalizing the reference data and the impact data, and then performing fast Fourier transform processing on the two, to obtain a pre-processed reference data set and a pre-processed impact data set respectively; step S4, calculating m*n*k cosine distances according to the pre-processed reference data set and the pre-processed impact data set; step S5 , according to the positions of the grid points corresponding to the m*n*k cosine distances on the composite material plate, k main feature matrices of size m*n are constructed; step S6, each main feature matrix is ​​interpolated by the bicubic interpolation method to obtain the corresponding k intermediate feature matrices; step S7, all the intermediate feature matrices are superimposed and processed to obtain a superimposed matrix; step S8, the superimposed matrix is ​​logarithmically processed to obtain a final feature matrix; step S9, each element in the final feature matrix is ​​sorted in descending order, and the grid points corresponding to the first several elements are selected as the preferred points; step S10, the elements corresponding to the preferred points in the superimposed matrix are used as weight values; step S11, according to the weight value of the preferred point and the horizontal and vertical coordinate values ​​of the preferred point on the composite material plate, the predicted horizontal and vertical coordinate values ​​of the impact point are calculated as the position of the impact point.

[0008] The random impact load positioning method based on the reference database provided by the present invention may also have the following characteristics: wherein, in step S3, the normalized expression is: Where y norm is the normalized data, y is the impulse response time domain data before normalization, y min is the minimum value among all y, y max is the maximum value among all y.

[0009] In the random impact load positioning method based on the reference database provided by the present invention, it can also have the following characteristics: wherein the preprocessed impact data set of step S3 is k groups of first data groups corresponding to k sensors, each group of first data groups includes j impact response frequency domain data, the preprocessed reference data set of step S3 is m*n*k groups of second data groups corresponding to k sensors with m*n grid points, each group of second data groups includes j impact response frequency domain data, and in step S4, the calculation formula of cosine distance is: d cos(a,b)=1-CS(a,b), where a=(a1,a2,...,a j ) is the first data set corresponding to sensor a in the preprocessed impact data, b=(b1, b2, ..., b j ) is the second data set corresponding to sensor a at grid point b in the preprocessed reference data, d cos (a, b) is the cosine distance between grid point b and the impact point at sensor a.

[0010] The random impact load positioning method based on the reference database provided by the present invention may also have the following characteristics: wherein, in step S7, the superposition operation is performed by summing the elements at the same position of the k intermediate characteristic matrices.

[0011] The random impact load positioning method based on the reference database provided by the present invention may also have the following features: wherein, in step S8, the logarithmic operation is to logarithmize each element in the superposition matrix, and the logarithmic expression is: w xy =log 10 (v xy ), where v xy is the element at (x,y) in the superposition matrix, w xy is the element at (x,y) in the final feature matrix.

[0012] The random impact load positioning method based on the reference database provided by the present invention may also have the following characteristics: wherein, in step S11, the calculation expressions for the predicted abscissa value and the predicted ordinate value are: Where x predicted is the predicted horizontal coordinate value, y predicted is the predicted ordinate value, d cos (i) is the weight value of the i-th preferred point, x i is the horizontal coordinate value of the i-th preferred point, y i is the ordinate value of the i-th preferred point, and PPN is the number of preferred points.

[0013] The random impact load positioning method based on the reference database provided by the present invention may also have the following characteristics: wherein, in step S1, the grid points are intersection points on the m*n matrix grid, and the matrix grid is slightly smaller than the surface area of ​​the composite material plate.

[0014] Functions and Effects of the Invention

[0015] According to the random impact load positioning method based on the reference database involved in the present invention, on the one hand, the density of grid points is increased by the bicubic interpolation method, which improves the positioning accuracy while reducing the number of grid points required; on the other hand, the cosine distance corresponding to each preferred point is used as a weight to construct a calculation expression for calculating the impact point position, thereby reducing the error of the bicubic difference method, further improving the positioning accuracy, and can also expand the prediction range, increasing the adaptability of the method. Therefore, the random impact load positioning method based on the reference database of the present invention can obtain a more accurate prediction result of the impact point position through limited grid points. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a schematic diagram of a sensor layout in an embodiment of the present invention;

[0017] Figure 2 is a schematic flow chart of a random impact load positioning method based on a reference database in an embodiment of the present invention;

[0018] Figure 3 is a schematic diagram of a matrix grid in an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the following embodiments and accompanying drawings specifically illustrate the random impact load positioning method based on a reference database of the present invention.

[0020] The random impact load positioning method based on a reference database of the present invention is used to determine the position of the impact point when a composite material plate provided with k sensors is impacted by a random load. In this embodiment, k=4.

[0021] Figure 1 is a schematic diagram of the sensor layout in an embodiment of the present invention.

[0022] like Figure 1 As shown, four sensors are respectively arranged at the four corners of the composite material plate, so as to eliminate the influence of overload acceleration caused by the application of the hammer.

[0023] Figure 2 It is a schematic flow chart of a random impact load positioning method based on a reference database in an embodiment of the present invention.

[0024] like Figure 2 As shown, the random impact load positioning method based on the reference database of this embodiment includes the following steps:

[0025] Step S1, setting m*n grid points on the composite material plate, and collecting the impact response time domain data of each grid point when it is impacted through k sensors to obtain m*n*k groups of reference data. In this embodiment, the impact response time domain data corresponding to each grid point is 1*k groups of reference data.

[0026] The grid points are intersection points on the m*n matrix grid, and the matrix grid is slightly smaller than the surface area of ​​the composite material plate.

[0027] Figure 3 is a schematic diagram of a matrix grid in an embodiment of the present invention.

[0028] like Figure 3 As shown, the matrix grid set on the composite material plate includes 15*6 intersections, i.e., grid points, and are numbered from the 1st reference impact point to the 90th reference impact point from bottom to top and from left to right, and the distance between adjacent grid points is 60 mm. The area formed by the matrix grid is the reference area, the dotted rectangular area is the locatable area, and the position represented by the five-pointed star is the random impact point, i.e., the impact point during random load impact. In this embodiment, the sensor is arranged outside the locatable area.

[0029] Step S2, collecting the time domain data of the impact response during the random load impact through k sensors to obtain 1*k groups of impact data.

[0030] Step S3, normalize the reference data and the impact data respectively, and then perform fast Fourier transform on both to obtain a preprocessed reference data set and a preprocessed impact data set respectively. In this embodiment, the impact response time domain data is converted into corresponding impact response frequency domain data through fast Fourier transform.

[0031] Among them, the normalized expression is:

[0032]

[0033] Where y norm is the normalized data, y is the impulse response time domain data before normalization, y min is the minimum value among all y, y max is the maximum value among all y.

[0034] The preprocessed shock data set is k groups of first data sets corresponding to k sensors, and each group of first data sets includes j shock response frequency domain data. The preprocessed reference data set is m*n*k groups of second data sets corresponding to k sensors with m*n grid points, and each group of second data sets includes j shock response frequency domain data.

[0035] Step S4, calculating m*n*k cosine distances according to the preprocessed reference data set and the preprocessed impact data set.

[0036] Wherein, in step S4, the calculation formula of the cosine distance is:

[0037]

[0038] d cos (a,b)=1-CS(a,b),

[0039] Where a=(a1,a2,...,a j ) is the first data set corresponding to sensor a in the preprocessed impact data, b=(b1, b2, ..., b j ) is the second data set corresponding to sensor a at grid point b in the preprocessed reference data, d cos (a, b) is the cosine distance between grid point b and the impact point at sensor a.

[0040] Step S5, constructing k main feature matrices of size m*n according to the positions of the grid points corresponding to the m*n*k cosine distances on the composite material plate.

[0041] Step S6, interpolate each main feature matrix separately through the bicubic interpolation method to obtain the corresponding k intermediate feature matrices. In this embodiment, the spacing between elements in the main feature matrix is ​​reduced from 60 mm to 3 mm through the bicubic interpolation method, thereby increasing the density of the main feature matrix, avoiding the constraint of the number of grid points, and thus improving the positioning accuracy.

[0042] Step S7, performing a superposition operation on all intermediate feature matrices to obtain a superposition matrix.

[0043] The superposition operation is to sum the elements at the same position of the k intermediate feature matrices.

[0044] Step S8, performing logarithmic operation on the superposition matrix to obtain a final feature matrix.

[0045] The logarithmic operation is to logarithmize each element in the superposition matrix, and the logarithmic expression is:

[0046] w xy =log 10 (v xy ),

[0047] Where v xy is the element at (x,y) in the superposition matrix, w xy is the element at (x,y) in the final feature matrix.

[0048] Step S9: sort the elements in the final feature matrix in descending order, and select the grid points corresponding to the first several elements as the preferred points.

[0049] Step S10: taking the element corresponding to the preferred point in the superposition matrix as a weight value, the value of the element in the superposition matrix is ​​the superposition sum of the cosine distances corresponding to the preferred point.

[0050] Step S11, according to the weight value of the preferred point and the abscissa value and ordinate value of the preferred point on the composite material plate, the predicted abscissa value and predicted ordinate value of the impact point are calculated as the position of the impact point.

[0051] Among them, the calculation expressions of the predicted horizontal coordinate value and the predicted vertical coordinate value are:

[0052]

[0053] Where x predicted is the predicted horizontal coordinate value, y predicted is the predicted ordinate value, d cos (i) is the weight value of the i-th preferred point, x i is the horizontal coordinate value of the i-th preferred point, y i is the ordinate value of the i-th preferred point, PPN is the number of preferred points. In this embodiment, the cosine distance after the difference is used as the weight, and the role of each preferred point is comprehensively considered to further improve the positioning accuracy, and the position of the impact point in the locatable area can be predicted, rather than being limited to the reference area.

[0054] Functions and Effects of the Embodiments

[0055] According to the random impact load positioning method based on the reference database involved in this embodiment, on the one hand, the density of grid points is increased by the bicubic interpolation method, which improves the positioning accuracy while reducing the number of grid points required; on the other hand, the cosine distance corresponding to each preferred point is used as a weight to construct a calculation expression for calculating the impact point position, thereby reducing the error of the bicubic difference method, further improving the positioning accuracy, and also expanding the prediction range, increasing the adaptability of the method. In short, the method can obtain a more accurate prediction result of the impact point position through limited grid points.

[0056] The above-mentioned embodiments are preferred examples of the present invention and are not intended to limit the protection scope of the present invention.

Claims

1. A random impact load location method based on a reference database, used to determine the position of the impact point when a composite material plate provided with k sensors is impacted by a random load, characterized in that: The following steps are involved: Step S1, setting m*n grid points on the composite material plate, and collecting the impact response time domain data of each grid point when it is impacted by k sensors, to obtain m*n*k groups of reference data; Step S2, collecting the time domain data of the impact response during the random load impact by k sensors, and obtaining 1*k groups of impact data; Step S3, normalizing the reference data and the impact data, and then performing fast Fourier transform on the two to obtain a preprocessed reference data set and a preprocessed impact data set, respectively; Step S4, calculating m*n*k cosine distances according to the preprocessed reference data set and the preprocessed impact data set; Step S5, constructing k main feature matrices of size m*n according to the positions of the grid points corresponding to the m*n*k cosine distances on the composite material plate; Step S6, interpolating each of the main feature matrices by bicubic interpolation to obtain corresponding k intermediate feature matrices; Step S7, performing a superposition operation on all the intermediate characteristic matrices to obtain a superposition matrix; Step S8, performing logarithmic operation on the superposition matrix to obtain a final feature matrix; Step S9, sorting the elements in the final feature matrix in descending order, and selecting the grid points corresponding to the first several elements as the preferred points; Step S10, taking the element corresponding to the preferred point in the superposition matrix as a weight value; Step S11, calculating the predicted abscissa value and predicted ordinate value of the impact point as the position of the impact point according to the weight value of the preferred point and the abscissa value and ordinate value of the preferred point on the composite material plate.

2. The random impact load positioning method based on the reference database according to claim 1 is characterized in that: in, In step S3, the expression of the normalization process is: Where y norm is the normalized data, y is the impulse response time domain data before normalization, and y min is the minimum value among all y, y max is the maximum value among all y.

3. The random impact load positioning method based on the reference database according to claim 1 is characterized in that: in, The pre-processed shock data set of step S3 is k groups of first data sets corresponding to k sensors, each group of first data sets includes j shock response frequency domain data, The preprocessed reference data set of step S3 is m*n*k groups of second data sets corresponding to k sensors at m*n grid points, each group of second data sets includes j impulse response frequency domain data. In step S4, the calculation formula of the cosine distance is: d cos (a,b)=1-CS(a,b), Where a=(a1,a2,...,a j ) is the first data set corresponding to sensor a in the preprocessed impact data, b=(b1, b2, ..., b j ) is the second data set corresponding to sensor a at grid point b in the preprocessed reference data, d cos (a, b) is the cosine distance between grid point b and the impact point at sensor a.

4. The random impact load positioning method based on a reference database according to claim 1 is characterized in that: in, In step S7, the superposition operation is performed by summing the elements at the same position of the k intermediate feature matrices.

5. The random impact load positioning method based on a reference database according to claim 1 is characterized in that: in, In step S8, the logarithmic operation is to logarithmize each element in the superposition matrix, and the logarithmic expression is: w xy =log 10 (v xy ), Where v xy is the element at (x,y) in the superposition matrix, w xy is the element at (x,y) in the final feature matrix.

6. The random impact load positioning method based on a reference database according to claim 1, characterized in that: in, In step S11, the calculation expressions of the predicted abscissa value and the predicted ordinate value are: Where x predicted is the predicted horizontal coordinate value, y predicted is the predicted ordinate value, d cos (i) is the weight value of the i-th preferred point, x i is the horizontal coordinate value of the i-th preferred point, y i is the ordinate value of the i-th preferred point, and PPN is the number of the preferred points.

7. The random impact load positioning method based on a reference database according to claim 1 is characterized in that: in, In the step S1, the grid points are intersection points on an m*n matrix grid, and the matrix grid is slightly smaller than the board surface area of ​​the composite material board.

Images