A method and system for constructing a multidimensional data set based on sensors and structural components
By preprocessing and selecting the initial data signals and screening, a cube data set is constructed, which solves the problem that multidimensional training data sets in the prior art that it is difficult to retain the intrinsic information of the structural vibration signal, and improves the performance of the damage recognition model of structural components and the ability to analyze damage characteristics.
Patent Information
- Application Number
- CN202210633166.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-06-06
- Publication Date
- 2025-04-18
- Estimated Expiration
- 2042-06-06
AI Technical Summary
When the prior art builds a multidimensional training data set for deep learning models, due to the number of sensors, it is difficult to retain the intrinsic information of the structural vibration signal, resulting in the absence of structural damage information, affecting the structural damage and fault diagnosis capabilities of CNN.
By preprocessing the initial data signal, the first eigenmodal function component is extracted, and the optimal filtering is performed according to its proportion. The second eigenmodal function component with high correlation is selected to construct a cube data set to retain the key information of the initial data signal.
It effectively improves the performance of the structural component damage recognition model, retains the intrinsic information of the structural vibration signal, enriches the structural component damage feature information of the cube data set, and improves the model's ability to analyze the damage situation of different structural components.
Smart Images
Figure CN115048993B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of constructing multidimensional data sets, and in particular to a method and system for constructing multidimensional data sets based on sensors and structural components. Background Art
[0002] When using deep learning technology for structural damage identification and mechanical component fault diagnosis, such as convolutional neural network (CNN), in many cases, the collected vibration data needs to be converted into an effective training data format that can be processed and learned by the deep learning model - image format, or a multidimensional text data format that can be processed by other neural networks. In real applications, the number of sensors arranged on the structure is small, and in many cases, insufficient training data can be collected. Moreover, the collected structural vibration data is mainly presented in one-dimensional form, which cannot meet the training data dimension requirements of the network model. Therefore, it is extremely important to construct an effective multidimensional training data set containing rich structural damage / fault information based on the small amount of one-dimensional vibration data collected.
[0003] At present, the method of constructing a data set using vibration data collected from structures or components is mainly based on the time method, that is, the collected one-dimensional vibration data of multiple sensors are arranged in parallel, and then truncated and spliced according to the time history information, and gradually construct two-dimensional data suitable for deep learning model processing. However, the dimension of this data set constructed by the time method is limited by the number of sensors on the structure or component. Sometimes, in order to adapt to the new network architecture, the original data is excessively truncated and spliced, and the final data set discards part of the information of the vibration signal of the structure or component, making the intrinsic information of the contained structural vibration signal unclear, resulting in further loss of structural or component damage information, which is not conducive to CNN processing of structural damage and component fault diagnosis information. Summary of the invention
[0004] The present invention provides a method and system for constructing a multidimensional data set based on sensors and structural components. On the basis of expanding the dimension of the data set, the characteristic information of the initial data signal is retained, thereby improving the performance of the trained structural component damage identification model.
[0005] In order to solve the above technical problems, an embodiment of the present invention provides a sensor-based multidimensional data set construction method, including:
[0006] Preprocessing the initial data signal to obtain a plurality of first intrinsic mode function components corresponding to the initial data signal; wherein the initial data signal is collected by a sensor on a structural component;
[0007] According to a preset algorithm, analyzing and calculating the proportion of each of the first intrinsic mode function components in the corresponding initial data signal, and selecting a plurality of the first intrinsic mode function components from all the first intrinsic mode function components according to the proportion as a plurality of second intrinsic mode function components corresponding to the initial data signal;
[0008] A multidimensional data set corresponding to the sensor is constructed according to the initial data signal and all corresponding second intrinsic mode function components; wherein the multidimensional data set is used to train a structural component damage identification model.
[0009] In implementing the embodiments of the present application, through proportion analysis, several first intrinsic mode function components corresponding to the initial data signal are preferentially screened to extract key components with a high correlation with the initial data signal, and the initial data signal and the extracted key components are used to construct a multidimensional data set for training a structural component damage identification model, retaining the key information in the initial data signal and avoiding the influence of non-critical components on the multidimensional data set on the model training effect, thereby improving the performance of the trained structural component damage identification model.
[0010] As a preferred solution, the method of analyzing and calculating the proportion of each of the first intrinsic mode function components in the corresponding initial data signal according to a preset algorithm, and selecting a plurality of the first intrinsic mode function components from all the first intrinsic mode function components according to the proportion as the second intrinsic mode function components corresponding to the initial data signal, is specifically as follows:
[0011] When the structural component is in several structural component damage conditions, obtaining a plurality of third eigenmode function components corresponding to each of the first eigenmode function components, and calculating variance contribution values of all the third eigenmode function components;
[0012] Calculating an average value of variance contribution values of a plurality of the third eigenmode function components corresponding to each of the first eigenmode function components, and using the average value as a proportion of each of the first eigenmode function components in the corresponding initial data signal;
[0013] All the first intrinsic mode function components are arranged in descending order according to the proportions, and the first M first intrinsic mode function components in the arrangement result are selected as the second intrinsic mode function components.
[0014] A preferred solution for implementing the embodiments of the present application is that, since the greater the variance contribution rate of an intrinsic mode function component, the greater the proportion of its variance value in the sum of the variances of all intrinsic mode function components, that is, the more important the intrinsic mode function component is among all the intrinsic mode function components, therefore, the average value of the variance contribution values of multiple third intrinsic mode function components under different structural component damage conditions is used to characterize the proportion of the corresponding first intrinsic mode function component in the corresponding initial data signal, which can avoid the loss of structural component damage information, enrich the structural component damage feature information of the multidimensional data set, and enable the trained model to have the ability to analyze and process different structural component damage conditions.
[0015] As a preferred solution, constructing a multidimensional data set corresponding to the sensor according to the initial data signal and all corresponding second intrinsic mode function components is specifically as follows:
[0016] When the multidimensional data set is a two-dimensional data set, the initial data signal and all the corresponding second intrinsic mode function components are sequentially arranged and spliced vertically to form a first two-dimensional data set corresponding to the sensor, thereby completing the construction of the multidimensional data set of the sensor;
[0017] When the multidimensional data set is a Q-dimensional data set, the initial data signal is used as the first side of the Q-dimensional data set, and all the second intrinsic mode function components are used in sequence as the second side to the Qth side of the Q-dimensional data set to form the Q-dimensional data set corresponding to the sensor, thereby completing the construction of the multidimensional data set of the sensor; wherein, Q≥3.
[0018] The preferred scheme for implementing the embodiments of the present application arranges the initial data signal and all corresponding second intrinsic mode function components into a two-dimensional, three-dimensional or even multi-dimensional data set according to a preset arrangement method, which is conducive to the effective and in-depth feature mining and learning of the structural component damage identification model, thereby improving the model's ability to resist noise interference.
[0019] As a preferred solution, the sensor-based multidimensional data set construction method further includes:
[0020] When the multidimensional data set is a two-dimensional data set, if the data volume of the first two-dimensional data set does not meet the preset requirements, the first two-dimensional data set is truncated and spliced according to the time information to obtain the corresponding second two-dimensional data set, and the second two-dimensional data set is used as the first two-dimensional data set for iterative optimization until the data volume of the first two-dimensional data set meets the preset requirements, and the current first two-dimensional data set is used as the multidimensional data set construction result of the sensor.
[0021] A preferred solution of implementing the embodiment of the present application is to further expand the two-dimensional data set to meet the training data requirements of different network models when the data volume of the two-dimensional data set is insufficient.
[0022] As a preferred solution, the initial data signal is preprocessed to obtain a plurality of first intrinsic mode function components corresponding to the initial data signal, specifically:
[0023] Performing empirical mode decomposition on the initial data signal to obtain a plurality of fourth intrinsic mode function components corresponding to the initial data signal;
[0024] All the fourth intrinsic mode function components are arranged in descending order of the frequencies corresponding to the fourth intrinsic mode function components, and the first N fourth intrinsic mode function components in the arrangement result are selected as the first intrinsic mode function components.
[0025] The preferred solution of the embodiment of the present application is to perform empirical mode decomposition on the initial data signal to obtain multiple fourth intrinsic mode function components related to the initial data signal, so as to preliminarily expand the initial data signal and reduce the limitation of the number of sensors on the training data set. In addition, according to the frequency of each component, N fourth intrinsic mode function components are preliminarily screened out from the decomposition results. Since the key components are usually concentrated in the high-frequency area, the preliminary screening of the decomposition results can not only improve the data processing rate, but also retain the structural component damage information collected by the sensor, thereby improving the effectiveness of the multidimensional data set.
[0026] In order to solve the same technical problem, an embodiment of the present invention further provides a method for constructing a multidimensional data set based on structural components, comprising:
[0027] Collecting a plurality of initial data signals through a plurality of sensors on the structural component; wherein the sensors and the initial data signals correspond one to one;
[0028] According to the sensor-based multidimensional data set construction method, construct a multidimensional data set for the multiple initial data signals in sequence to obtain a multidimensional data set corresponding to each of the sensors;
[0029] The multidimensional data sets corresponding to all the sensors are used as the multidimensional data sets of the structural component to complete the construction of the multidimensional data set of the structural component; wherein the multidimensional data set of the structural component is used to train a structural component damage identification model.
[0030] In order to solve the same technical problem, an embodiment of the present invention further provides a sensor-based multi-dimensional data set construction system, including:
[0031] A preprocessing module, used for preprocessing the initial data signal to obtain a plurality of first intrinsic mode function components corresponding to the initial data signal; wherein the initial data signal is collected by a sensor on a structural component;
[0032] an analysis and calculation module, configured to analyze and calculate the proportion of each of the first intrinsic mode function components in the corresponding initial data signal according to a preset algorithm, and select a plurality of the first intrinsic mode function components from all the first intrinsic mode function components according to the proportion as a plurality of second intrinsic mode function components corresponding to the initial data signal;
[0033] A data set construction module is used to construct a multidimensional data set corresponding to the sensor based on the initial data signal and all corresponding second intrinsic mode function components; wherein the multidimensional data set is used to train a structural component damage identification model.
[0034] As a preferred solution, the analysis and calculation module specifically includes:
[0035] a calculation unit, configured to obtain, when the structural component is in several structural component damage conditions, a plurality of third eigenmode function components corresponding to each of the first eigenmode function components, and calculate variance contribution values of all the third eigenmode function components; calculate an average value of variance contribution values of the plurality of third eigenmode function components corresponding to each of the first eigenmode function components, and use the average value as a proportion of each of the first eigenmode function components in the corresponding initial data signal;
[0036] The first selection unit is used to arrange all the first intrinsic mode function components according to the order of their respective proportions from large to small, and select the first M first intrinsic mode function components in the arrangement result as the second intrinsic mode function components.
[0037] As a preferred solution, the data set construction module specifically includes:
[0038] A first construction unit is used for, when the multidimensional data set is a two-dimensional data set, arranging and splicing the initial data signal and all the corresponding second intrinsic mode function components in sequence in a longitudinal direction to form a first two-dimensional data set corresponding to the sensor, and completing the construction of the multidimensional data set of the sensor; when the multidimensional data set is a two-dimensional data set, if the data volume of the first two-dimensional data set does not meet the preset requirements, truncating and splicing the first two-dimensional data set according to the time history information to obtain the corresponding second two-dimensional data set, and taking the second two-dimensional data set as the first two-dimensional data set for iterative optimization until the data volume of the first two-dimensional data set meets the preset requirements, and then taking the current first two-dimensional data set as the result of the construction of the multidimensional data set of the sensor;
[0039] A second construction unit is used to, when the multidimensional data set is a Q-dimensional data set, use the initial data signal as the first side of the Q-dimensional data set, and use all the second intrinsic mode function components in sequence as the second side to the Q-th side of the Q-dimensional data set, to construct the Q-dimensional data set corresponding to the sensor, and complete the construction of the multidimensional data set of the sensor; wherein Q≥3.
[0040] As a preferred solution, the preprocessing module specifically includes:
[0041] a signal decomposition unit, configured to perform empirical mode decomposition on the initial data signal to obtain a plurality of fourth intrinsic mode function components corresponding to the initial data signal;
[0042] The second selection unit is used to arrange all the fourth intrinsic mode function components according to the order of frequencies corresponding to the fourth intrinsic mode function components from high to low, and select the first N fourth intrinsic mode function components in the arrangement result as the first intrinsic mode function components. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] Figure 1 : A flow chart of a method for constructing a multidimensional data set based on a sensor provided in the first embodiment of the present invention;
[0044] Figure 2 : A flowchart of a method for constructing a multidimensional data set based on structural components provided in Example 1 of the present invention;
[0045] Figure 3 : A structural diagram of a sensor-based multidimensional data set construction system provided in Example 1 of the present invention. DETAILED DESCRIPTION
[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in 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 creative work are within the scope of protection of the present invention.
[0047] Embodiment 1:
[0048] Please refer to Figure 1 , is a flow chart of a method for constructing a multidimensional data set based on a sensor provided by an embodiment of the present invention, the method comprising steps S1 to S3, each step being specifically as follows:
[0049] Step S1, preprocessing an initial data signal to obtain a plurality of first intrinsic mode function components corresponding to the initial data signal; wherein the initial data signal is collected by a sensor on a structural component.
[0050] As a preferred solution, step S1 includes step S11 to step S12, and each step is specifically as follows:
[0051] Step S11, performing empirical mode decomposition on the initial data signal to obtain a plurality of fourth intrinsic mode function components corresponding to the initial data signal.
[0052] In this embodiment, due to the limitation of the number of sensors arranged on the structural components, the number of initial vibration signals collected is small, and the initial vibration signals are mainly presented in one-dimensional form, which usually cannot meet the model training requirements. Among them, the initial data signal can be any form of response signal such as acceleration, velocity, displacement, etc. Therefore, through empirical mode decomposition (EMD), the original signal x(t) can be decomposed into N intrinsic mode function (IMF) components with independent characteristics between components, and according to the EMD decomposition principle and constraints, the expression of the original signal x(t) can be seen in formula (1).
[0053]
[0054] Among them, C i (t) represents each IMF component, r N (t) represents the residual component after decomposition.
[0055] Step S12, arranging all the fourth intrinsic mode function components according to the order of frequencies corresponding to the fourth intrinsic mode function components from high to low, and selecting the first N fourth intrinsic mode function components in the arrangement result as the first intrinsic mode function components.
[0056] In this embodiment, since the initial data signals collected by the sensor under different structural damage conditions are different, the number of IMF components obtained after EMD decomposition may be different. At the same time, the key IMF components obtained after EMD decomposition are mainly concentrated in the high-frequency area, so in order to take into account the effective EMD decomposition of all sensor signals and ensure that each sensor signal can obtain effective IMF components, N fourth IMF components in the high-frequency area are preliminarily screened out from the decomposition results as the first IMF components.
[0057] As an example, the value of N is 4, that is, the decomposition results are preliminarily screened to obtain 4 first IMF components.
[0058] Step S2, according to a preset algorithm, analyze and calculate the proportion of each of the first intrinsic mode function components in the corresponding initial data signal, and based on the proportion, select multiple first intrinsic mode function components from all the first intrinsic mode function components as multiple second intrinsic mode function components corresponding to the initial data signal.
[0059] As a preferred solution, step S2 includes steps S21 to S23, and each step is specifically as follows:
[0060] Step S21, when the structural component is in several structural component damage conditions, obtaining multiple third eigenmode function components corresponding to each of the first eigenmode function components, and calculating variance contribution values of all the third eigenmode function components.
[0061] Step S22, calculating an average value of variance contribution values of a plurality of the third eigenmode function components corresponding to each of the first eigenmode function components, and using the average value as a proportion of each of the first eigenmode function components in the corresponding initial data signal.
[0062] Step S23: arranging all the first eigenmode function components according to the order of the proportions from large to small, and selecting the first M first eigenmode function components in the arrangement result as the second eigenmode function components.
[0063] In this embodiment, since the initial screening of IMF components by frequency is relatively rough, all first IMF components are screened a second time by analyzing the proportion of each first IMF component in the corresponding initial data signal, thereby further improving the validity of the multidimensional data set.
[0064] Specifically, the variance contribution rate (VCR) is introduced here to reflect the proportion of the first IMF component in the initial data signal.
[0065] Furthermore, in order to avoid the loss of structural component damage information and enrich the structural component damage feature information of the multidimensional data set, so as to improve the ability of the trained structural component damage identification model to process structural damage and component fault diagnosis information, multiple third IMF components under different structural component damage conditions are obtained, and the average value of the variance contribution rate of the multiple third IMF components corresponding to each first IMF component is calculated as the proportion of each first IMF component in the initial data signal, and then the first IMF component with a high proportion is selected as the second IMF component for constructing the multidimensional data set. Among them, the calculation method of the VCR of the multiple third IMF components corresponding to the i-th first IMF component is shown in formula (2), which reflects the proportion of the variance of the multiple third IMF components corresponding to the i-th first IMF component in the sum of the variances of the multiple third IMF components corresponding to each first IMF component.
[0066]
[0067] in, It represents the variance of multiple third IMF components corresponding to the i-th first IMF component. Its calculation method is shown in formula (3).
[0068]
[0069] in, represents the mean square value of multiple third IMF components corresponding to the i-th first IMF component, E(IMF i ) represents the mean of multiple third IMF components corresponding to the i-th first IMF component.
[0070] Step S3, constructing a multidimensional data set corresponding to the sensor according to the initial data signal and all corresponding second intrinsic mode function components; wherein the multidimensional data set is used to train a structural component damage identification model.
[0071] As a preferred solution, step S3 includes step S31 to step S32, and each step is specifically as follows:
[0072] Step S31, when the multidimensional data set is a two-dimensional data set, the initial data signal and all corresponding second intrinsic mode function components are arranged and spliced vertically in sequence to form a first two-dimensional data set corresponding to the sensor, thereby completing the construction of the multidimensional data set of the sensor.
[0073] Step S32, when the multidimensional data set is a Q-dimensional data set, the initial data signal is used as the first side of the Q-dimensional data set, and all the second intrinsic mode function components are used in sequence as the second side to the Q-th side of the Q-dimensional data set to form the Q-dimensional data set corresponding to the sensor, thereby completing the construction of the multidimensional data set of the sensor; wherein, Q≥3.
[0074] In this embodiment, the initial data signal and all corresponding second intrinsic mode function components are arranged into two-dimensional, three-dimensional or even multi-dimensional data sets according to a preset arrangement method, which is conducive to effective and in-depth feature mining and learning of the structural component damage identification model, thereby improving the model's ability to resist noise interference.
[0075] As a preferred solution, the sensor-based multidimensional data set construction method further includes step S4, which is as follows:
[0076] Step S4: When the multidimensional data set is a two-dimensional data set, if the data volume of the first two-dimensional data set does not meet the preset requirements, the first two-dimensional data set is truncated and spliced according to the time information to obtain a corresponding second two-dimensional data set, and the second two-dimensional data set is used as the first two-dimensional data set for iterative optimization until the data volume of the first two-dimensional data set meets the preset requirements, and the current first two-dimensional data set is used as the multidimensional data set construction result of the sensor.
[0077] In this embodiment, when the data volume of the two-dimensional data set is insufficient, a time method is used to further expand the two-dimensional data set to meet the training data requirements of different network models.
[0078] Please refer to Figure 2 , is a flow chart of a method for constructing a multidimensional data set based on structural components provided by an embodiment of the present invention, the method comprising steps S5 to S7, each of which is specifically as follows:
[0079] Step S5, collecting a plurality of initial data signals through a plurality of sensors on the structural component; wherein the sensors and the initial data signals correspond one to one.
[0080] Step S6, according to the sensor-based multidimensional data set construction method described in Example 1, construct a multidimensional data set for the multiple initial data signals in sequence to obtain a multidimensional data set corresponding to each of the sensors.
[0081] Step S7, taking the multidimensional data sets corresponding to all the sensors as the multidimensional data sets of the structural components, and completing the construction of the multidimensional data sets of the structural components; wherein the multidimensional data sets of the structural components are used to train the structural component damage identification model.
[0082] Specifically, based on the 6-channel sensor on the structural component, the vibration data s1, s2, s3, s4_z in the vertical direction of the structural component and the vibration data s4_x, s4_y in the horizontal direction of the structural component are collected as the "main" data constituting the multidimensional data set. Then, according to the sensor-based multidimensional data set construction method described in Example 1, the 6 acceleration sensor signals are respectively subjected to EMD decomposition, and the decomposition results are preliminarily screened to obtain the first 4 IMF components with the highest frequency of each sensor signal.
[0083] Then, taking each sensor signal as the benchmark, all data under 37 types of damage labels are counted. Among them, there are 200 groups of data under each type of damage label, and for each first IMF component of a single sensor signal, there are 7400 third IMF components. Therefore, taking the 4 first IMF components obtained by decomposing a single acceleration sensor signal as an example, the average value of the variance contribution value of the 7400 third IMF components corresponding to each first IMF component is calculated to characterize the proportion of the corresponding first IMF component in the single sensor signal, and the top two first IMF components are used as the key IMF components of the single sensor signal, namely the second IMF components, to achieve secondary screening of the IMF components. Furthermore, the same method is used to screen out the two second IMF components of the other five acceleration sensor signals, and the two second IMF components of the six acceleration sensor signals obtained by screening out constitute the two "slave" data of the multidimensional data set.
[0084] Among them, when the structural components are under the same excitation, the key IMF component combinations of the vibration signals in the vertical and horizontal directions are different. Therefore, when conditions permit, by deploying different types of sensors at different positions of the structural components and at different spatial dimensions at the same position, the diversity of structural damage information can be enriched, which is conducive to the effective identification of structural damage / fault diagnosis.
[0085] As an example, the "master" data and two "slave" data are arranged and spliced vertically in sequence to form a two-dimensional data set of the structural component.
[0086] As another example, the "master" data is used as the first side of the three-dimensional data set, and the two "slave" data are used as the second and third sides of the three-dimensional data set respectively, to form the three-dimensional data set of the structural component.
[0087] Please refer to Figure 3 , is a schematic diagram of a structure of a sensor-based multidimensional data set construction system provided by an embodiment of the present invention, the system includes a preprocessing module 1, an analysis and calculation module 2 and a data set construction module 3, and the specific details of each module are as follows:
[0088] A preprocessing module 1 is used to preprocess the initial data signal to obtain a plurality of first intrinsic mode function components corresponding to the initial data signal; wherein the initial data signal is collected by a sensor on a structural component;
[0089] an analysis and calculation module 2, configured to analyze and calculate the proportion of each of the first intrinsic mode function components in the corresponding initial data signal according to a preset algorithm, and select a plurality of the first intrinsic mode function components from all the first intrinsic mode function components according to the proportion as a plurality of second intrinsic mode function components corresponding to the initial data signal;
[0090] The data set construction module 3 is used to construct a multidimensional data set corresponding to the sensor according to the initial data signal and all corresponding second intrinsic mode function components; wherein the multidimensional data set is used to train a structural component damage identification model.
[0091] As a preferred solution, the analysis and calculation module 2 specifically includes:
[0092] a calculation unit, configured to obtain, when the structural component is in several structural component damage conditions, a plurality of third eigenmode function components corresponding to each of the first eigenmode function components, and calculate variance contribution values of all the third eigenmode function components; calculate an average value of variance contribution values of the plurality of third eigenmode function components corresponding to each of the first eigenmode function components, and use the average value as a proportion of each of the first eigenmode function components in the corresponding initial data signal;
[0093] The first selection unit is used to arrange all the first intrinsic mode function components according to the order of their respective proportions from large to small, and select the first M first intrinsic mode function components in the arrangement result as the second intrinsic mode function components.
[0094] As a preferred solution, the data set construction module 3 specifically includes:
[0095] A first construction unit is used for, when the multidimensional data set is a two-dimensional data set, arranging and splicing the initial data signal and all the corresponding second intrinsic mode function components in sequence in a longitudinal direction to form a first two-dimensional data set corresponding to the sensor, and completing the construction of the multidimensional data set of the sensor; when the multidimensional data set is a two-dimensional data set, if the data volume of the first two-dimensional data set does not meet the preset requirements, truncating and splicing the first two-dimensional data set according to the time history information to obtain the corresponding second two-dimensional data set, and taking the second two-dimensional data set as the first two-dimensional data set for iterative optimization until the data volume of the first two-dimensional data set meets the preset requirements, and then taking the current first two-dimensional data set as the result of the construction of the multidimensional data set of the sensor;
[0096] A second construction unit is used to, when the multidimensional data set is a Q-dimensional data set, use the initial data signal as the first side of the Q-dimensional data set, and use all the second intrinsic mode function components in sequence as the second side to the Q-th side of the Q-dimensional data set, to construct the Q-dimensional data set corresponding to the sensor, and complete the construction of the multidimensional data set of the sensor; wherein Q≥3.
[0097] As a preferred solution, the preprocessing module 1 specifically includes:
[0098] a signal decomposition unit, configured to perform empirical mode decomposition on the initial data signal to obtain a plurality of fourth intrinsic mode function components corresponding to the initial data signal;
[0099] The second selection unit is used to arrange all the fourth intrinsic mode function components according to the order of frequencies corresponding to the fourth intrinsic mode function components from high to low, and select the first N fourth intrinsic mode function components in the arrangement result as the first intrinsic mode function components.
[0100] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.
[0101] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:
[0102] The present invention proposes a method and system for constructing a multidimensional data set based on sensors and structural components. By analyzing the proportion of each first intrinsic mode function component in the corresponding initial data signal, the first intrinsic mode function components are preferentially screened to extract multiple second intrinsic mode function components with high correlation with the initial data signal, and the initial data signal and all the second intrinsic mode function components are used to construct a multidimensional data set for training a structural component damage identification model, so as to retain the key information in the initial data signal and further improve the performance of the trained structural component damage identification model.
[0103] Furthermore, the average of the variance contribution values of multiple third intrinsic mode function components under different structural component damage conditions is calculated as a basis for preferentially screening the second intrinsic mode function components used to construct the multidimensional data set, thereby enriching the structural component damage feature information of the multidimensional data set and avoiding the loss of structural component damage information, so that the trained model has the ability to analyze and process different structural component damage conditions.
[0104] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.
Claims
1. A sensor-based multidimensional dataset construction method, characterized in that: include: Preprocessing the initial data signal to obtain a plurality of first intrinsic mode function components corresponding to the initial data signal; wherein the initial data signal is collected by a sensor on a structural component; According to a preset algorithm, analyzing and calculating the proportion of each of the first intrinsic mode function components in the corresponding initial data signal, and selecting a plurality of the first intrinsic mode function components from all the first intrinsic mode function components according to the proportion as a plurality of second intrinsic mode function components corresponding to the initial data signal; Constructing a multidimensional data set corresponding to the sensor according to the initial data signal and all corresponding second intrinsic mode function components; wherein the multidimensional data set is used to train a structural component damage identification model; The method of analyzing and calculating the proportion of each of the first intrinsic mode function components in the corresponding initial data signal according to a preset algorithm is specifically as follows: When the structural component is in several structural component damage conditions, obtaining a plurality of third eigenmode function components corresponding to each of the first eigenmode function components, and calculating variance contribution values of all the third eigenmode function components; Calculating an average value of variance contribution values of a plurality of the third eigenmode function components corresponding to each of the first eigenmode function components, and using the average value as a proportion of each of the first eigenmode function components in the corresponding initial data signal; The step of constructing a multidimensional data set corresponding to the sensor according to the initial data signal and all corresponding second intrinsic mode function components is specifically as follows: When the multidimensional data set is a two-dimensional data set, the initial data signal and all the corresponding second intrinsic mode function components are sequentially arranged and spliced vertically to form a first two-dimensional data set corresponding to the sensor, thereby completing the construction of the multidimensional data set of the sensor; When the multidimensional data set is a Q-dimensional data set, the initial data signal is used as the first side of the Q-dimensional data set, and all the second intrinsic mode function components are used in sequence as the second side to the Qth side of the Q-dimensional data set to form the Q-dimensional data set corresponding to the sensor, thereby completing the construction of the multidimensional data set of the sensor; wherein, Q≥3.
2. A sensor-based multidimensional dataset construction method as claimed in claim 1, characterized in that: The selecting, according to the proportion, a plurality of the first intrinsic mode function components from all the first intrinsic mode function components as the second intrinsic mode function components corresponding to the initial data signal is specifically: All the first intrinsic mode function components are arranged in descending order according to the proportions, and the first M first intrinsic mode function components in the arrangement result are selected as the second intrinsic mode function components.
3. A sensor-based multidimensional dataset construction method as claimed in claim 1, characterized in that: Also includes: When the multidimensional data set is a two-dimensional data set, if the data volume of the first two-dimensional data set does not meet the preset requirements, the first two-dimensional data set is truncated and spliced according to the time information to obtain the corresponding second two-dimensional data set, and the second two-dimensional data set is used as the first two-dimensional data set for iterative optimization until the data volume of the first two-dimensional data set meets the preset requirements, and the current first two-dimensional data set is used as the multidimensional data set construction result of the sensor.
4. The sensor-based multidimensional dataset construction method according to claim 1, characterized in that: The preprocessing of the initial data signal to obtain a plurality of first intrinsic mode function components corresponding to the initial data signal is specifically: Performing empirical mode decomposition on the initial data signal to obtain a plurality of fourth intrinsic mode function components corresponding to the initial data signal; All the fourth intrinsic mode function components are arranged in descending order of the frequencies corresponding to the fourth intrinsic mode function components, and the first N fourth intrinsic mode function components in the arrangement result are selected as the first intrinsic mode function components.
5. A method for constructing a multidimensional dataset based on structural components, characterized in that: include: Collecting a plurality of initial data signals through a plurality of sensors on the structural component; wherein the sensors and the initial data signals correspond one to one; According to the sensor-based multidimensional data set construction method according to any one of claims 1 to 4, the multidimensional data set construction is performed on the multiple initial data signals in sequence to obtain a multidimensional data set corresponding to each of the sensors; The multidimensional data sets corresponding to all the sensors are used as the multidimensional data sets of the structural component to complete the construction of the multidimensional data set of the structural component; wherein the multidimensional data set of the structural component is used to train a structural component damage identification model.
6. A sensor-based multidimensional dataset construction system, characterized in that: include: A preprocessing module, used for preprocessing the initial data signal to obtain a plurality of first intrinsic mode function components corresponding to the initial data signal; wherein the initial data signal is collected by a sensor on a structural component; an analysis and calculation module, configured to analyze and calculate the proportion of each of the first intrinsic mode function components in the corresponding initial data signal according to a preset algorithm, and select a plurality of the first intrinsic mode function components from all the first intrinsic mode function components according to the proportion as a plurality of second intrinsic mode function components corresponding to the initial data signal; A data set construction module, used to construct a multidimensional data set corresponding to the sensor according to the initial data signal and all corresponding second intrinsic mode function components; wherein the multidimensional data set is used to train a structural component damage identification model; Wherein, the analysis and calculation module includes: a calculation unit, configured to obtain, when the structural component is in several structural component damage conditions, a plurality of third eigenmode function components corresponding to each of the first eigenmode function components, and calculate variance contribution values of all the third eigenmode function components; calculate an average value of variance contribution values of the plurality of third eigenmode function components corresponding to each of the first eigenmode function components, and use the average value as a proportion of each of the first eigenmode function components in the corresponding initial data signal; The data set construction module specifically includes: A first construction unit is used for, when the multidimensional data set is a two-dimensional data set, arranging and splicing the initial data signal and all the corresponding second intrinsic mode function components in sequence in a longitudinal direction to form a first two-dimensional data set corresponding to the sensor, and completing the construction of the multidimensional data set of the sensor; when the multidimensional data set is a two-dimensional data set, if the data volume of the first two-dimensional data set does not meet the preset requirements, truncating and splicing the first two-dimensional data set according to the time history information to obtain the corresponding second two-dimensional data set, and taking the second two-dimensional data set as the first two-dimensional data set for iterative optimization until the data volume of the first two-dimensional data set meets the preset requirements, and then taking the current first two-dimensional data set as the result of the multidimensional data set construction of the sensor; A second construction unit is used to, when the multidimensional data set is a Q-dimensional data set, use the initial data signal as the first side of the Q-dimensional data set, and use all the second intrinsic mode function components in sequence as the second side to the Q-th side of the Q-dimensional data set, to construct the Q-dimensional data set corresponding to the sensor, and complete the construction of the multidimensional data set of the sensor; wherein Q≥3.
7. A sensor-based multi-dimensional dataset construction system as claimed in claim 6, characterized in that: The analysis and calculation module further includes: The first selection unit is used to arrange all the first intrinsic mode function components according to the order of their respective proportions from large to small, and select the first M first intrinsic mode function components in the arrangement result as the second intrinsic mode function components.
8. The sensor-based multi-dimensional dataset construction system according to claim 6, characterized in that: The preprocessing module specifically includes: a signal decomposition unit, configured to perform empirical mode decomposition on the initial data signal to obtain a plurality of fourth intrinsic mode function components corresponding to the initial data signal; The second selection unit is used to arrange all the fourth intrinsic mode function components according to the order of frequencies corresponding to the fourth intrinsic mode function components from high to low, and select the first N fourth intrinsic mode function components in the arrangement result as the first intrinsic mode function components.