Long-period continuous moving impact load identification method and system based on combination of time slice and mode identification

Abstract

The invention belongs to the technical field of load identification, and discloses a long-period continuous moving impact load identification method and system based on time slice combined mode identification. Extracting and classifying the characteristic vectors of the load in the lateral impact process through numerical simulation or a model experiment, wherein a characteristic value discrimination function comprises flight time difference discrimination, response peak gradient discrimination and similarity classification discrimination; performing uniform slicing processing on the time domain by using a time slicing method based on the long-time-period time domain response signal, and arranging all slices according to a time sequence; classifying and calibrating the response of the load through a pattern recognition method, and screening, classifying and re-aggregating time slices by adopting a similarity search method and taking cosine similarity as a similarity criterion and a discrimination function of pattern recognition; and finally, performing inversion and positioning by using a traditional Tikhonov regularization method in combination with a GCV operator, and reconstructing a time history containing load impact, migration and movement and a movement track under lateral impact. According to the method, the special calibration corresponding to the special type of load is achieved, the morbidity of matrix accumulation during inversion is reduced, and the accuracy of load identification is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of load identification, and in particular relates to a method and a system for identifying long-term continuous moving impact loads based on time slicing combined with pattern recognition. Background Art

[0002] The identification of long-term continuous random moving impact loads caused by aircraft landing is a relatively complex problem in engineering, especially for deck structures with known areas but unknown positions, which is an inverse structural dynamics problem that needs to be solved urgently. The identification of long-term continuous moving impact loads refers to the continuous monitoring and identification of the characteristics of moving impact loads, such as size, frequency, and position, over a long period of time. This identification is particularly important in the fields of bridge tracks, shipbuilding, marine engineering, aerospace, etc., and can help evaluate the safety and durability of structures. Ideally, sensors can be arranged at the impact point to directly measure the time history of the load. However, in reality, the impact position is often unknown, which requires indirect measurement of the load. With the improvement of inversion technology and sensor accuracy, indirect identification and measurement of impact force has become more and more common. However, due to the complexity of the structure and the interference of noise and vibration in actual measurements, this type of inverse problem often causes ill-posedness (ill-conditioned), which affects the accuracy of the solution.

[0003] In recent years, a variety of methods have been developed to identify impact loads, such as the conjugate gradient method, basis function expansion method, regularization method, and artificial neural network, but they are not compatible with complex multi-type loads and long-term identification. Although the newly proposed methods can reduce errors more effectively than traditional methods in some aspects, the overall identification accuracy is still not high. Under numerical simulation, the simulated impact forces are often faced with similar shapes and similar amplitudes. And full time domain inversion is often used, resulting in the reconstruction of the load time history accumulating large pathological conditions, resulting in large errors. At present, the research on inversion methods based on long-term load identification is relatively rare, and the identification objects are mostly dynamic loads with simple waveforms such as sine and half-sine. There is still a lack of research on the inversion of long-term complex loads, especially long-term continuous moving impact loads. Summary of the invention

[0004] The present invention provides a method for identifying long-term continuous moving impact loads based on time slicing combined with pattern recognition, which aims at solving the problems related to long-term complex load inversion, especially long-term continuous moving impact loads in the prior art.

[0005] The present invention provides a long-term continuous moving impact load identification system based on time slicing combined with pattern recognition, which is used to implement the present invention provides a long-term continuous moving impact load identification method based on time slicing combined with pattern recognition.

[0006] The present invention is achieved through the following technical solutions:

[0007] A method for identifying long-term continuous moving impact load based on time slicing combined with pattern recognition, the identification method comprising the following steps:

[0008] Step 1: Obtain a response signal for calibrating a short-term load;

[0009] Step 2: construct a pattern recognition discriminant function and obtain the characteristic value of the response signal for calibrating the short-term load, wherein the discriminant function includes: flight time difference discrimination, response peak gradient discrimination and similarity classification discrimination;

[0010] Step 3: Obtain the response signal of the long-term load to be identified;

[0011] Step 4: Preliminary positioning of the response signal based on the time difference between the load stress wave reaching each sensor;

[0012] Step 5: Extract the peak value of each sensor signal curve according to the response signal of the segment load, construct the peak gradient curve, and select the effective response slice according to the gradient descent value;

[0013] Step 6: Based on the time slice combined with the pattern recognition method, the response signal of the long-term load to be identified is time sliced ​​with the set time slice length, and the corresponding slice length mathematical model is constructed; the similarity search is performed on the slice feature value and the slice feature value according to the similarity measurement, the minimum error is taken, the corresponding type of load characteristics are screened, and the slices are re-aggregated according to the time series and feature classification;

[0014] Step 7: Invert the aggregate slices according to the regularization method combined with similarity search to obtain the load time history curve and movement trajectory, and complete the identification of long-term continuous moving impact loads.

[0015] Furthermore, the steps 1 to 3 are performed through preliminary experiments to extract load response information, specifically,

[0016] The response characteristic parameters corresponding to the random impact load are obtained in advance through the response of the load, and a set of parameter combinations are selected as the feature vector according to the actual requirements of pattern recognition: after dividing the calibration area for the pre-impact position, the position characteristic space is constructed according to the response characteristic parameters of the load impact at different positions; and then the transfer matrix of the load time history is constructed according to the relationship between the load and the response, thereby completing the calibration of the entire area of ​​the random impact load.

[0017] Furthermore, the step 5 is specifically to determine and cut the invalid response period using the gradient interval of the response curve, that is, the corresponding no-load period in the response time domain.

[0018] Furthermore, the rough positioning in step 4 is as follows: since multiple sensors are arranged in the same area, the calibration area of ​​the initial impact can be roughly located by calculating the time difference of the initial signal transmission of different sensors; the subsequent loads will also be judged using the characteristic values ​​of this calibration area and its surrounding areas to improve the positioning accuracy.

[0019] Furthermore, the step 6 specifically utilizes the response similarity and uses similarity search to compare the response similarity of the calibration area, thereby locating the unknown load and determining the transfer matrix corresponding to the response, and then directly inverting the load curve to be identified.

[0020] The response similarity of the calibration area is compared using similarity search. The similarity measurement method is adopted, that is, when the two vectors have the same direction, the cosine similarity is 1; when the two vectors are 90°, the cosine similarity is 0;

[0021] In Cartesian coordinates, two vectors a = (a1, a2, a3, ..., an) and b = (b1, b2, b3, ..., bn), the cosine distance can be expressed as:

[0022]

[0023] Furthermore, the regularization method is combined with similarity search to construct the Green function to establish the transfer matrix G between the load matrix P and the response matrix Y, that is,

[0024] Y=GP

[0025] At this time, the general form of the regularization algorithm is constructed based on the least squares method combined with the regularization operator:

[0026]

[0027] Among them, the first is the least squares term; λ is the regularization parameter, whose value is always positive; Ω(P λ ) is a Tikhonov stable operator, which is taken as

[0028] Under the appropriate selection of regularization parameters, the solution P of the operator λ It can be expressed as:

[0029] P λ =(G T G+λ 2 I) -1 G T Y.

[0030] Furthermore, after constructing the transfer function, the method of calibration first and then search is adopted. Using similarity search, the similarity metric is selected to compare the response to be tested with the calibration response and the group with the smallest error is screened. The corresponding load-time history curve and positioning area of ​​the polymer slice to be tested can be obtained.

[0031] A long-term continuous moving impact load identification system based on time slicing combined with pattern recognition, the identification system uses the long-term continuous moving impact load identification method based on time slicing combined with pattern recognition as mentioned above, the identification system includes:

[0032] Short-term load response signal acquisition unit: acquires a response signal for calibrating a short-term load;

[0033] Characteristic value calibration unit: constructs a pattern recognition discriminant function and obtains characteristic values ​​of the response signal for calibrating the short-term load, wherein the discriminant function includes: flight time difference discrimination, response peak gradient discrimination and similarity classification discrimination;

[0034] Long-term load response signal acquisition unit: acquires the response signal of the long-term load to be identified;

[0035] Preliminary positioning unit: Preliminary positioning of the response signal based on the time difference between the load stress wave reaching each sensor;

[0036] Effective response slice screening unit: extract the peak value of each sensor signal curve according to the response signal of the segment load, construct the peak gradient curve, and screen the effective response slice according to the gradient descent value;

[0037] Aggregation slice calculation unit: according to the time slice combined with the pattern recognition method, the response signal of the long-term load to be identified is time sliced ​​with the set time slice length, and the corresponding slice length mathematical model is constructed; according to the similarity measurement, the slice feature value is searched for similarity with the slice feature value, the minimum error is taken, the corresponding type of load features are screened, and the slices are re-aggregated according to the time series and feature classification;

[0038] Aggregate slice inversion unit: The aggregate slices are inverted according to the regularization method combined with similarity search to obtain the load time history curve and movement trajectory, thus completing the identification of long-term continuous moving impact loads.

[0039] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the above method is implemented.

[0040] A computer-readable storage medium stores a computer program, and the computer program implements the above method when executed by a processor.

[0041] The beneficial effects of the present invention are:

[0042] The present invention is based on the time slicing method, combined with pattern recognition, and realizes the recognition process of cutting, classifying, and re-aggregating the long-term complex load response through a three-step discriminant function, and then uses the traditional regularization method combined with similarity search to realize the multi-type inversion corresponding to the complex load. Compared with the existing methods, the pathological accumulation of loads caused by long-term load inversion is greatly reduced through time slicing, and the trajectory inversion from coarse positioning to fine positioning is realized through pattern recognition to improve the positioning accuracy. The three-step discriminant function is used to realize load classification and eliminate invalid load response fragments, which improves the load curve inversion accuracy while increasing the recognition efficiency and reducing the time required for recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a flow chart of the method of the present invention.

[0044] Figure 2 It is a schematic diagram of the experimental verification simulation of the present invention.

[0045] Figure 3 It is a flow chart of time slice feature extraction based on pattern recognition of the present invention.

[0046] Figure 4 It is a schematic diagram of the load response processing of the present invention.

[0047] Figure 5 This is the three-step discrimination process of the pattern recognition discrimination function of the present invention.

[0048] Figure 6 It is a schematic diagram of the time difference of each sensor receiving the load response signal of the present invention.

[0049] Figure 7 It is a schematic diagram of fitting the effective segment peak gradient curve between the load amplitude curve and the response amplitude signal of the present invention.

[0050] Figure 8 It is a flow chart of load inversion and positioning of the present invention.

[0051] Fig. 9 It is a schematic diagram of the response slice selection operation during the time slice calculation of the present invention.

[0052] Fig.10 It is a load amplitude curve of the simulated long-term continuous moving impact load and a schematic diagram of the identification result of the present invention.

[0053] Fig.11 It is a schematic diagram of the load movement trajectory and identification result of simulating a long-term continuous moving impact load of the present invention. DETAILED DESCRIPTION

[0054] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0055] It should be understood that when used in this specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0056] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, the singular forms "a", "an" and "the" are intended to include plural forms unless the context clearly indicates otherwise.

[0057] The following is attached to this application specification Figure 1-11 , the technical solutions in the embodiments of the present application are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application.

[0058] In the following description, many specific details are set forth to facilitate a full understanding of the present application, but the present application may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present application. Therefore, the present application is not limited to the specific embodiments disclosed below.

[0059] A method for identifying long-term continuous moving impact load based on time slicing combined with pattern recognition, the identification method comprising the following steps:

[0060] Step 1: Obtain a response signal for calibrating a short-term load;

[0061] Step 2: construct a pattern recognition discriminant function and obtain the characteristic value of the response signal for calibrating the short-term load, wherein the discriminant function includes: flight time difference discrimination, response peak gradient discrimination and similarity classification discrimination;

[0062] Step 3: Obtain the response signal of the long-term load to be identified;

[0063] Step 4: Based on the time difference of the load stress wave reaching each sensor, the response signal is initially positioned using the time-of-flight method, i.e., rough positioning;

[0064] Step 5: Extract the peak value of each sensor signal curve according to the response signal of the segment load, construct the peak gradient curve, and select the effective response slice according to the gradient descent value;

[0065] Step 6: Based on the time slice combined with the pattern recognition method, the response signal of the long-term load to be identified is time sliced ​​with the set time slice length, and the corresponding slice length mathematical model is constructed; the similarity search is performed on the slice feature value and the slice feature value according to the similarity measurement, the minimum error is taken, the corresponding type of load characteristics are screened, and the slices are re-aggregated according to the time series and feature classification;

[0066] Step 7: Invert the aggregate slices according to the regularization method combined with similarity search to obtain the load time history curve and movement trajectory, and complete the identification of long-term continuous moving impact loads.

[0067] Furthermore, the steps 1 to 3 are performed through preliminary experiments to extract load response information, specifically,

[0068] The response characteristic parameters corresponding to the random impact load are obtained in advance through the response of the load, and a set of parameter combinations are selected as the feature vector according to the actual requirements of pattern recognition; after dividing the calibration area for the pre-impact position, the position feature space is constructed according to the response characteristic parameters of the load impact at different positions; then, according to the relationship between the load and the response, the transfer matrix of the load time history is constructed to complete the calibration of the entire area of ​​the random impact load; in the actual experiment, after obtaining the response of the load, the response is processed by equal time slicing; the processed slices are subjected to feature recognition using a discriminant function, and the slices of the same feature type are classified into one category, and then re-aggregated and classified according to the time series; at this time, the response data has been classified into corresponding impact load segments, migration load segments and no-load periods in the time domain; at this time, the corresponding load can be inverted according to the transfer function calibrated in the early stage, so as to obtain the time history and motion trajectory of the load at the same time; the discriminant function includes: flight time difference discrimination, response peak gradient discrimination and similarity classification discrimination.

[0069] The specific practical requirements include the selection of the response gradient peak value and the subsequent feature calibration of the transfer matrix and response features.

[0070] Furthermore, the step 5 is specifically to determine and cut the invalid response period using the gradient interval of the response curve, that is, the corresponding no-load period in the response time domain.

[0071] After the structure is impacted, the load will quickly return to zero, but the response will last for a long time. If the load curve is directly inverted at this time, the continuous response will produce an invalid load curve; it has been observed that the response containing the real load segment often has a large gradient drop at the waveform peak. When the curve slope Δk is less than a certain value C, it can be determined that the response period here is a period without numerical load; therefore, the secondary judgment is the peak gradient judgment, and the period with a smaller gradient is judged as an invalid response period and does not participate in the inversion calculation. It can save a lot of computing time and the later judgment process.

[0072] Furthermore, the rough positioning in step 4 is as follows: since multiple sensors are arranged in the same area, the calibration area of ​​the initial impact can be roughly located by calculating the time difference of the initial signal transmission of different sensors; the subsequent loads will also be judged using the characteristic values ​​of this calibration area and its surrounding areas, which can save a lot of calculation time and maximize the positioning accuracy.

[0073] Furthermore, the step 6 specifically utilizes the response similarity and uses similarity search to compare the response similarity of the calibration area, thereby locating the unknown load and determining the transfer matrix corresponding to the response, and then directly inverting the load curve to be identified.

[0074] There are many common similarity measures, such as Lagrange distance, cosine similarity, Pearson correlation coefficient, etc. Considering that the amplitude of the true load and the calibration load is quite different, cosine similarity is selected as the discriminant function. Cosine similarity is a unique similarity measure that measures the error between two vectors by the cosine of the angle between two non-zero vectors. Obviously,

[0075] The response similarity of the calibration area is compared using similarity search. The similarity measurement method is adopted, that is, when the two vectors have the same direction, the cosine similarity is 1; when the two vectors are 90°, the cosine similarity is 0; in addition, the cosine similarity excludes the influence of the vector size;

[0076] In Cartesian coordinates, two vectors a = (a1, a2, a3, ..., an) and b = (b1, b2, b3, ..., bn), the cosine distance can be expressed as:

[0077]

[0078] Furthermore, the regularization method is combined with similarity search to construct the Green function to establish the transfer matrix G between the load matrix P and the response matrix Y, that is,

[0079] Y=GP

[0080] At this time, the general form of the regularization algorithm is constructed based on the least squares method combined with the regularization operator:

[0081]

[0082] Among them, the first is the least squares term; λ is the regularization parameter, whose value is always positive; Ω(P λ ) is a Tikhonov stable operator, which has various forms and is generally taken as Under the appropriate selection of regularization parameters, the solution P of the operator λ It can be expressed as:

[0083] P λ =(G T G+λ 2 I) -1 G T Y.

[0084] Furthermore, after constructing the transfer function, the method of calibration first and then search is adopted. Using similarity search, the similarity metric is selected to compare the response to be tested with the calibration response and the group with the smallest error is screened. The corresponding load-time history curve and positioning area of ​​the polymer slice to be tested can be obtained.

[0085] Specifically, the data collection includes: dividing the area to be tested into a large area and a small area, applying a calibration load to the small area and recording the response.

[0086] The feature extraction includes: calibrating the transfer matrix of the small area according to the load and the response, extracting the eigenvalues ​​of the calibration matrix, and classifying the specific load type.

[0087] The time slicing includes: formal experiments, collecting long-term load responses to be tested, and uniformly refining the long-term time domain.

[0088] The feature discrimination includes: extracting sensor signal difference, using the time-of-flight method to achieve large-area coarse positioning and extracting the response peak drop gradient, screening and eliminating invalid slices.

[0089] The slice aggregation includes: extracting slice features, classifying the loads using similarity search, and re-aggregating the classified response slices in time series.

[0090] The regional positioning and force reconstruction include: performing similarity search between the re-aggregated large slice response and the small area response under the calibrated large area, and selecting the small area with the smallest error to complete regional positioning; using the calibrated transfer function of the corresponding area for the slice that has completed regional positioning to complete the load curve inversion of the slice; and re-aggregating the slice that has completed inversion and positioning to form a complete curve and movement trajectory.

[0091] Combination Figure 2As shown, this figure is a schematic diagram of an experimental verification simulation in an embodiment of the present invention, which is characterized by including a stiffened plate 1, a data collector 2, a hammer 3, a terminal and a medium 4.

[0092] Specifically, the area to be tested is divided into large areas and small areas as described above, that is, the stiffened plate 1 is divided into 6 large areas according to the reinforcing T-shaped materials under the plate, and each large area is divided into 8×6 small areas, a total of 48 small areas.

[0093] Specifically, as described above, a calibration load is applied to a small area and the response is recorded, that is, a specific type of load is applied to the calibration area of ​​the stiffened plate 1 using a force hammer 3, and the load type may be an impact load, a migration load, a moving load, etc. The corresponding load type is marked and the applied load action area, the time history curve, and the sensor response are recorded. The signal is collected by the data collector 2, and the terminal and the medium 4 perform calculations and storage.

[0094] Combination Figure 3 As shown, this figure is a flow chart of time slice feature extraction based on pattern recognition in one embodiment of the present invention, which is characterized by including transfer matrix calibration of small areas according to load and response, eigenvalue extraction of the calibration matrix, and classification of specific load types.

[0095] Specifically, the load and response calibrate the transfer matrix of the small area and extract the eigenvalue of the calibration matrix, and classify the specific load type, that is, construct the load and response matrix according to the Green function: load matrix P and response matrix Y, thereby constructing the transfer matrix G, and the following can be obtained:

[0096] Y=GP

[0097] From the above formula, we can get the transfer matrix G corresponding to the small area of ​​calibration: x-n , where x can represent the load type and n is the corresponding small area number. The response characteristic parameters corresponding to the random impact load are obtained in advance through the load response, and a set of parameter combinations are selected as feature vectors according to the actual requirements of pattern recognition. The position characteristic space is constructed according to the response characteristic parameters of load impacts at different positions. Then, according to the relationship between load and response, the transfer matrix G of the load time history is constructed. x-n , thus completing the calibration work of the entire area of ​​random impact load.

[0098] Combination Figure 4 As shown, this figure is a schematic diagram of load response processing using a time slicing method combined with pattern recognition in an embodiment of the present invention, which is characterized by including formal experiments, collecting long-term load responses to be tested, and uniformly refining and slicing the long-term time domain.

[0099] Specifically, at this point, we have entered the formal experiment stage. In the formal experiment, the load amplitude curve is unknown, and only the response signal collected at the sensor is collected. The signal will be evenly sliced ​​in time, and all small slices will be substituted into Figure 3 In the pattern recognition process, the slices are identified and marked, and then invalid slices are removed, that is, slices with load amplitudes that do not exist in the original load time domain corresponding to the response slice. Finally, slices of the same type are classified and sorted according to the time series and aggregated into complete segments, and can enter the final inversion calculation work.

[0100] Combination Figure 5 and Figure 6 , Figure 5 This is a three-step discrimination process of a pattern recognition discriminant function in one embodiment of the present invention. Figure 6 This is a schematic diagram of the time difference of each sensor receiving the load response signal. Its characteristics are the first step of the discriminant function: coarse positioning.

[0101] Specifically, when each sensor receives a response signal, a series of response vectors are formed: y1, y2, ..., and the vectors correspond to time series. The signals are detected at the same time, so when the stress wave diffuses to each sensor, a signal time difference will be generated. At this time, the corresponding time node is extracted: t 1,0 ,t 2,0 , ..., the time difference of the response signal can be obtained. Using the time-of-flight method, it can be determined in which large area the signal point occurs. Since each large area is equipped with multiple sensors, and the stress wave will lose energy and produce delays when crossing the reinforced T-profile, the accuracy of the initial identification can be ensured. Subsequent loads will also be identified using the characteristic values ​​of this calibration area and its surrounding areas, which can save a lot of computing time and maximize the positioning accuracy.

[0102] Combination Figure 5 and Figure 7 , Figure 5 This is a three-step discrimination process of a pattern recognition discriminant function in one embodiment of the present invention. Figure 7 The figure is a schematic diagram of the effective segment peak gradient curve fitting between the load amplitude curve and the response amplitude signal. Its characteristic is the second step of the discriminant function: screening the effective load segment.

[0103] Specifically, a plurality of sensors are set up in the experiment, and a plurality of sensor data sets are obtained: y1, y2, ... After achieving coarse positioning, the response peak data is extracted from the plurality of sensor data sets, and the peak drop gradient, that is, the slope of the peak curve, is calculated to obtain the corresponding Δk. According to the calibration data, if the slope Δk is continuously greater than a certain value C in a certain time slice, then this section of slice data is judged as valid data, that is, there is an impact load in this section of data. The value C is determined by the minimum slope of the peak curve of the impact response data in the calibration data.

[0104] Combination Figure 5 , Figure 5 This is a three-step discrimination process of the pattern recognition discriminant function in one embodiment of the present invention. The first two discrimination quantities, namely the time difference of the initial response and the descending gradient of the response peak, have been explained above. The third discrimination step, namely the response characteristics when the load acts on the calibration area, is now explained in detail.

[0105] Specifically, after the effective response slices are screened out by the slope, the slices of adjacent time are combined to obtain the response time periods Y1, Y2, ... At this time, based on the distribution of the calibration reference points and sensors, the division of the area, and the calculation method of similarity features, the characteristic values ​​of the early calibration area positioning can be unified as follows:

[0106] C i,j =d(t ij ,y ij )

[0107] Where i=1,2,… represents a plurality of small areas; j=1,2,… is the sensor number. ij is the time difference between the signals received by each sensor, and y ij is the acceleration response data.

[0108] In the experiment, each small calibration area corresponds to multiple eigenvalues, namely C i,1 ,C i,2 ,…The number of eigenvalues ​​depends on the number of sensors. Assuming there are m small areas and n sensors, there are m·n eigenvectors in the experiment. Therefore, the vector space K can be obtained:

[0109]

[0110] In actual measurement, the signal to be measured will obtain n groups of signals through n sensors, and the n groups of signals will be processed to obtain the corresponding Y1, Y2, .... Each signal includes the same type of feature vector t in the calibration signal ij and ij Based on the obtained feature vector, the region positioning is completed using the similarity classifier of cosine similarity.

[0111] Combination Figure 8 As shown in the figure, this is a flow chart of load inversion and positioning. As mentioned above, the process of slicing, identifying, screening, classifying and re-aggregating long-term load responses has been completed through time slicing combined with pattern recognition. Now it is necessary to invert the time history of the load from the aggregated response fragments.

[0112] Specifically, as mentioned above, based on the Green function, the load matrix P and the response matrix Y, the transfer function can be constructed: Y = GP. Converting it into a discrete model can be obtained:

[0113]

[0114] Among them, y(t) is the structural response under the impact load, p(t) is the dynamic load acting on the structure, and g(t) is the corresponding transfer function.

[0115] Because inverse problems are often accompanied by ill-posedness (ill-posedness), the Tikhonov regularization method is introduced. The regularization method can obtain a stable optimization solution by modifying the matrix singular values ​​and adding a regularization operator. It introduces functionals on the basis of the least squares method to realize the construction of the regularization operator, so that the ill-posed problem is transformed into a well-posed problem, and the solution of the transformed well-posed problem can better approach the solution of the original problem. The general form of the regularization method is:

[0116]

[0117] Where, the first term is the least squares term. λ is the regularization parameter, which is always positive. λ ) is a Tikhonov stable operator, which has various forms. In general, Under appropriate regularization parameters, the solution of the regularization equation can be expressed as:

[0118] P λ =(G T G+λ 2 I) -1 G T Y

[0119] Where I is the unit matrix.

[0120] The selection of regularization parameters will directly affect the accuracy of the solution. When the error level σ is unknown, generalized cross validation based on cross validation is a very popular method for selecting regularization parameters. It originates from the PRESS criterion for selecting the best model in statistical estimation theory, but its robustness is better than the latter. Since no prior information on the noise level is required and the accuracy is acceptable, the GCV method can be successfully used to calculate regularization parameters in various fields, especially discrete load identification problems under random noise. The regularization parameter λ can be determined by minimizing the GCV function, which can be expressed as:

[0121]

[0122] In the formula, G # =(G T G+λ 2 I) -1 G T ; trace() represents the trace of the matrix, that is, the sum of the diagonal elements of the matrix; I N is the identity matrix.

[0123] Depend on Figure 8 Middle process: First, the area is divided into multiple areas to be calibrated, and a simulated impact load P is applied to the calibration area. C , the simulated response Y is obtained through the pre-arranged sensors C Then, the transfer matrix G1, G2, ...G of each calibration area is obtained by combining the Tikhonov regularization algorithm with the GCV operator. n Then, the impact load to be tested is applied to obtain the measured corresponding value Y λ , Y λ Substitute all calibration transfer matrices G1, G2, ... G n In the above equation, we can obtain the loads to be identified P1, P2, ... P from the inversion of each transfer matrix. n , and then search for load similarity with the simulated impact load of each calibration area to determine the closest load value, that is, to obtain the time history closest to the true value. In this way, the inversion of all slices is completed.

[0124] Combination Fig. 9 As shown in the figure, this is a schematic diagram of the response slice selection operation during time slice calculation. Its characteristics are that load inversion belongs to the second type of inverse problem in dynamics, in which matrix inversion will inevitably cause pathological accumulation, which is specifically manifested in that a large disturbance will inevitably occur at the end of the inverted time history curve.

[0125] Specifically, the front part of the inverted curve is basically fitted with the true load amplitude, but it will gradually produce large fluctuations and the fluctuations will gradually expand. To avoid this situation, when calculating the time slice, a certain number of slices will be spliced ​​on the classified slices. When the inversion calculation is completed, the inversion curve will be trimmed of the equal-length time slices added before, so as to reduce unnecessary disturbances, reduce the pathological state of the matrix, and improve the recognition accuracy of the curve.

[0126] This example introduces the identification result of a long-term continuous moving impact load based on time slicing combined with pattern recognition. This example uses a force hammer to hit the reinforced plate laterally. The size of the reinforced plate is 2×1.2m, and the material is steel. The reinforcement ribs at the bottom of the plate are divided into 4×3, a total of 12 large areas, and each large area is divided into 8×6, a total of 48 small areas. The hammer head is made of nylon, which allows the hammer to complete continuous bounces in a short period of time, simulating a long-term continuous moving impact load. The load amplitude curve and identification results of the simulated long-term continuous moving impact load are shown in Fig.10 , where the black line is the original load amplitude curve and the red line is the inverted load amplitude curve. It can be clearly seen that the red line fits the black line very well. Although the amplitude of the later curve still has some errors over time, the curvature of the curve is basically fitted. The load movement trajectory and identification results of the simulated long-term continuous moving impact load are shown in Fig.11The upper shear head represents the moving trajectory of the hammer, the lower point represents the corresponding area of ​​the identified effective load slice, and the line between the points constitutes the identification trajectory. From the recognition results, it can be seen that this method has good adaptability for the identification problem of long-term continuous moving impact loads, the recognition results are good, the curve fit is high, and the error is within the acceptable range of engineering.

[0127] Implementation Method 2

[0128] The embodiment of the present invention provides a long-term continuous moving impact load identification system based on time slicing combined with pattern recognition. This embodiment uses the long-term continuous moving impact load identification method based on time slicing combined with pattern recognition as described in embodiment 1. The identification system includes:

[0129] Short-term load response signal acquisition unit: acquires a response signal for calibrating a short-term load;

[0130] Characteristic value calibration unit: constructs a pattern recognition discriminant function and obtains characteristic values ​​of the response signal for calibrating the short-term load, wherein the discriminant function includes: flight time difference discrimination, response peak gradient discrimination and similarity classification discrimination;

[0131] Long-term load response signal acquisition unit: acquires the response signal of the long-term load to be identified;

[0132] Preliminary positioning unit: Based on the time difference between the load stress wave reaching each sensor, the time-of-flight method is used to perform preliminary positioning of the response signal, i.e., rough positioning;

[0133] Effective response slice screening unit: extract the peak value of each sensor signal curve according to the response signal of the segment load, construct the peak gradient curve, and screen the effective response slice according to the gradient descent value;

[0134] Aggregation slice calculation unit: according to the time slice combined with the pattern recognition method, the response signal of the long-term load to be identified is time sliced ​​with the set time slice length, and the corresponding slice length mathematical model is constructed; according to the similarity measurement, the slice feature value is searched for similarity with the slice feature value, the minimum error is taken, the corresponding type of load features are screened, and the slices are re-aggregated according to the time series and feature classification;

[0135] Aggregate slice inversion unit: The aggregate slices are inverted according to the regularization method combined with similarity search to obtain the load time history curve and movement trajectory, thus completing the identification of long-term continuous moving impact loads.

[0136] Implementation Method 3

[0137] An embodiment of the present invention provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the memory is used to store software programs and modules, and the processor executes various functional applications and data processing by running the software programs and modules stored in the memory. The memory and the processor are connected via a bus. Specifically, the processor implements any step in the first embodiment above by running the computer program stored in the memory.

[0138] It should be understood that in the embodiments of the present invention, the processor referred to may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0139] The memory may include a read-only memory, a flash memory, and a random access memory, and provides instructions and data to the processor. A part or all of the memory may also include a nonvolatile random access memory.

[0140] It should be understood that if the above-mentioned integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned implementation method, and can also be completed by instructing the relevant hardware through a computer program. The above-mentioned computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method implementations when executed by the processor. Among them, the above-mentioned computer program includes computer program code, and the above-mentioned computer program code can be in source code form, object code form, executable file or some intermediate form. The above-mentioned computer-readable medium may include: any entity or device capable of carrying the above-mentioned computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the above-mentioned computer-readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction.

[0141] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest range consistent with the principles and novel features disclosed herein.

[0142] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the above-mentioned device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the implementation method can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method implementation method, which will not be repeated here.

[0143] It should be noted that the methods and detailed examples provided in the above embodiments can be combined with the devices and equipment provided in the embodiments, and references can be made to each other, and no further details will be given.

[0144] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0145] In the embodiments provided by the present invention, it should be understood that the disclosed device / terminal equipment and method can be implemented in other ways. For example, the device / equipment implementation described above is only illustrative, for example, the division of the above modules or units is only a logical function division, and in actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0146] The above-mentioned embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the above-mentioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above-mentioned embodiments, or replace some of the technical features therein by equivalents. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A method for identifying long-term continuous moving impact loads based on time slicing combined with pattern recognition, characterized in that: The identification method comprises the following steps: Step 1: Obtain a response signal for calibrating a short-term load; Step 2: construct a pattern recognition discriminant function and obtain the characteristic value of the response signal for calibrating the short-term load, wherein the discriminant function includes: flight time difference discrimination, response peak gradient discrimination and similarity classification discrimination; Step 3: Obtain the response signal of the long-term load to be identified; Step 4: Preliminary positioning of the response signal based on the time difference between the load stress wave reaching each sensor; Step 5: Extract the peak value of each sensor signal curve according to the response signal of the segment load, construct the peak gradient curve, and select the effective response slice according to the gradient descent value; Step 6: Based on the time slice combined with the pattern recognition method, the response signal of the long-term load to be identified is time sliced ​​with the set time slice length, and the corresponding slice length mathematical model is constructed; the similarity search is performed on the slice feature value and the slice feature value according to the similarity measurement, the minimum error is taken, the corresponding type of load characteristics are screened, and the slices are re-aggregated according to the time series and feature classification; Step 7: Invert the aggregate slices according to the regularization method combined with similarity search to obtain the load time history curve and movement trajectory, and complete the identification of long-term continuous moving impact loads.

2. The method according to claim 1, characterized in that Steps 1 to 3 are performed through preliminary experiments to extract load response information, specifically, The response characteristic parameters corresponding to the random impact load are obtained in advance through the response of the load, and a set of parameter combinations are selected as the feature vector according to the actual requirements of pattern recognition; after dividing the calibration area for the pre-impact position, the position characteristic space is constructed according to the response characteristic parameters of the load impact at different positions; and then, according to the relationship between the load and the response, the transfer matrix of the load time history is constructed, thereby completing the calibration of the entire area of ​​the random impact load.

3. The method according to claim 2, characterized in that The step 5 is specifically to determine and cut the invalid response period by using the gradient interval of the response curve, that is, the corresponding no-load period in the response time domain.

4. The method according to claim 1, characterized in that: Specifically, the step 4 of rough positioning is that, since multiple sensors are arranged in the same area, the calibration area of ​​the initial impact can be roughly located by calculating the time difference of the initial signal transmission of different sensors; the subsequent loads will also be judged using the characteristic values ​​of this calibration area and its nearby areas to improve the positioning accuracy.

5. The method according to claim 2, characterized in that: Specifically, step 6 is to utilize the response similarity and use similarity search to compare the response similarity of the calibration area, so as to locate the unknown load and determine the transfer matrix corresponding to the response, and then directly invert the load curve to be identified; The response similarity of the calibration area is compared using similarity search. The similarity measurement method is adopted, that is, when the two vectors have the same direction, the cosine similarity is 1; when the two vectors are 90°, the cosine similarity is 0; In Cartesian coordinates, two vectors a = (a1, a2, a3, ..., an) and b = (b1, b2, b3, ..., bn), the cosine distance can be expressed as:

6. The method according to claim 5, characterized in that The regularization method combined with similarity search is specifically to construct the Green function to establish the transfer matrix G between the load matrix P and the response matrix Y, that is, Y=GP At this time, the general form of the regularization algorithm is constructed based on the least squares method combined with the regularization operator: Among them, the first is the least squares term; λ is the regularization parameter, whose value is always positive; Ω(P λ ) is a Tikhonov stable operator, which is taken as Under the appropriate selection of regularization parameters, the solution P of the operator λ It is expressed as: P λ =(G T G+λ 2 I) -1 G T Y。 7. The method according to claim 5, characterized in that After constructing the transfer function, the method of calibration first and then search is adopted. Using similarity search, the similarity metric is selected to compare the response to be tested with the calibration response and screen the group with the smallest error. The corresponding load time history curve and positioning area of ​​the polymer slice to be tested can be obtained.

8. A long-term continuous moving impact load identification system based on time slicing combined with pattern recognition, characterized in that: The identification system uses the long-term continuous moving impact load identification method based on time slicing combined with pattern recognition as claimed in any one of claims 1 to 6, and the identification system includes: Short-term load response signal acquisition unit: acquires a response signal for calibrating a short-term load; Characteristic value calibration unit: constructs a pattern recognition discriminant function and obtains characteristic values ​​of the response signal for calibrating the short-term load, wherein the discriminant function includes: flight time difference discrimination, response peak gradient discrimination and similarity classification discrimination; Long-term load response signal acquisition unit: acquires the response signal of the long-term load to be identified; Preliminary positioning unit: Preliminary positioning of the response signal based on the time difference between the load stress wave reaching each sensor; Effective response slice screening unit: extract the peak value of each sensor signal curve according to the response signal of the segment load, construct the peak gradient curve, and screen the effective response slice according to the gradient descent value; Aggregation slice calculation unit: according to the time slice combined with the pattern recognition method, the response signal of the long-term load to be identified is time sliced ​​with the set time slice length, and the corresponding slice length mathematical model is constructed; according to the similarity measurement, the slice feature value is searched for similarity with the slice feature value, the minimum error is taken, the corresponding type of load features are screened, and the slices are re-aggregated according to the time series and feature classification; Aggregate slice inversion unit: The aggregate slices are inverted according to the regularization method combined with similarity search to obtain the load time history curve and movement trajectory, thus completing the identification of long-term continuous moving impact loads.

9. A computer device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images