CN113973403B - Temperature-induced strain field redistribution intelligent sensing method based on structure discrete measurement point topology - Google Patents
Temperature-induced strain field redistribution intelligent sensing method based on structure discrete measurement point topology Download PDFInfo
- Publication number
- CN113973403B CN113973403B CN202111326559.XA CN202111326559A CN113973403B CN 113973403 B CN113973403 B CN 113973403B CN 202111326559 A CN202111326559 A CN 202111326559A CN 113973403 B CN113973403 B CN 113973403B
- Authority
- CN
- China
- Prior art keywords
- temperature
- induced strain
- cluster
- data
- strain
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 59
- 238000005259 measurement Methods 0.000 title claims abstract description 19
- 238000012544 monitoring process Methods 0.000 claims abstract description 16
- 238000009833 condensation Methods 0.000 claims abstract description 9
- 230000005494 condensation Effects 0.000 claims abstract description 9
- 238000007689 inspection Methods 0.000 claims abstract description 6
- 239000013598 vector Substances 0.000 claims description 46
- 238000012549 training Methods 0.000 claims description 30
- 238000012360 testing method Methods 0.000 claims description 28
- 239000011159 matrix material Substances 0.000 claims description 16
- 238000004364 calculation method Methods 0.000 claims description 10
- 230000008569 process Effects 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 7
- 238000010606 normalization Methods 0.000 claims description 5
- 238000005070 sampling Methods 0.000 claims description 4
- 238000001228 spectrum Methods 0.000 claims description 4
- 238000010183 spectrum analysis Methods 0.000 claims description 4
- 230000009466 transformation Effects 0.000 claims description 4
- 238000001914 filtration Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000009529 body temperature measurement Methods 0.000 claims 1
- 230000009471 action Effects 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000002776 aggregation Effects 0.000 description 2
- 238000004220 aggregation Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 229910000831 Steel Inorganic materials 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 239000000919 ceramic Substances 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000007429 general method Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001373 regressive effect Effects 0.000 description 1
- 239000010959 steel Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000002277 temperature effect Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W84/00—Network topologies
- H04W84/18—Self-organising networks, e.g. ad-hoc networks or sensor networks
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01D—MEASURING NOT SPECIALLY ADAPTED FOR A SPECIFIC VARIABLE; ARRANGEMENTS FOR MEASURING TWO OR MORE VARIABLES NOT COVERED IN A SINGLE OTHER SUBCLASS; TARIFF METERING APPARATUS; MEASURING OR TESTING NOT OTHERWISE PROVIDED FOR
- G01D21/00—Measuring or testing not otherwise provided for
- G01D21/02—Measuring two or more variables by means not covered by a single other subclass
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
Abstract
The invention discloses a temperature induced strain field redistribution intelligent sensing method based on a structure discrete measurement point topology, which comprises the following steps: extracting temperature and temperature induced strain time sequence data of each discrete measuring point at each moment, performing condensation hierarchical clustering on temperature and strain measuring point coordinates, calculating average values of the temperature and temperature induced strain data at each moment in each cluster, calculating distribution topology characteristic values, constructing a two-way long-short-term memory regression network, establishing a big data fuzzy relation model from the temperature field distribution topology characteristic values to the temperature induced strain field distribution topology characteristic values, inputting the regression network model by converting the temperature field distribution topology characteristic values of real-time temperature monitoring data after the model passes the inspection, comparing the model regression results with the temperature induced strain field distribution topology characteristic values converted by the real-time strain monitoring data, and sensing structural temperature induced strain redistribution by difference of the model regression results and the temperature induced strain field distribution topology characteristic values. The method of the invention realizes the intelligent sensing of the redistribution of the structural temperature induced strain field based on discrete and universal temperature and strain sensor data.
Description
Technical Field
The invention belongs to the field of intelligent monitoring, detection and evaluation of structures, and particularly relates to an intelligent temperature-induced strain field redistribution sensing method based on a structure discrete measuring point topology.
Background
The engineering structure is continuously acted by the ambient temperature from the beginning of construction, the non-uniformity of the temperature action is more obvious along with the increase of the structure size, the structure temperature induced strain field is the most visual reflection of the non-uniform temperature field action, and the structure temperature field and the temperature induced strain field are accurately and effectively described as the premise of grasping the current state of the structure. With the development of testing and transmission technology, it is not difficult to obtain the temperature and strain of a large-scale engineering structure in real time through monitoring and detecting means. However, since the testing range of the mainstream temperature and strain sensor is too small relative to the size of the engineering structure, it is generally only possible to obtain discrete temperatures and strains at several points or areas, and the distribution characteristics of the engineering structure temperature and temperature induced strain are difficult to evaluate effectively, so that the temperature induced strain redistribution of the sensing structure in the whole life cycle is not considered. The rise and development of artificial intelligence makes it possible to simulate and perceive the time-varying distribution pattern of structural temperature-induced strain fields by adopting clustering and deep learning methods.
At present, the method for establishing a temperature field-strain field distribution model by considering the topological characteristics of discrete measuring points of a structure in the civil and mechanical fields is less, and the case of sensing the temperature-induced strain field redistribution of the structure in real time by adopting a method of combining clustering and a neural network is more attractive. The following methods are commonly used: (1) The method can only measure the strain distribution in a small area, and cannot distinguish the load effect from which the strain originates, so that the method has limited application scenes in large civil engineering; (2) Based on the array type piezoelectric ceramic sensing element, the stress strain distribution of the structure is measured, the physical quantity measured by each sensing element is stored as a matrix in real time, so that the stress strain distribution characteristic of a certain area is obtained. In summary, the prior art is based on the more advanced novel material to realize the measurement of the structural strain distribution field, and has high technical requirements, high cost, immature application and no direct correlation with the temperature effect.
Therefore, it is necessary to develop a method with low cost and based on mature sensing technology, to invert the discrete and general real-time distribution mode of the data characteristic values of the temperature and strain sensors by machine learning and artificial intelligence theory, and to establish a fuzzy relation model between the temperature field and the temperature-induced strain field, so as to realize the intelligent sensing of the structural temperature-induced strain field redistribution.
Disclosure of Invention
The invention aims to: aiming at the problems, the invention provides an intelligent sensing method for temperature-induced strain field redistribution based on a structure discrete measurement point topology, which can realize the redistribution of the temperature-induced strain field of the structure based on discrete and general temperature and strain sensor data intelligent sensing.
The technical scheme is as follows: in order to achieve the purpose of the invention, the technical scheme adopted by the invention is as follows: a temperature-induced strain field redistribution intelligent sensing method based on a structure discrete measurement point topology specifically comprises the following steps:
step 1, for a structure to be tested, extracting discrete temperature time sequence data of each temperature measuring point and temperature induced strain time sequence data of each strain measuring point of the structure within a certain time, preprocessing, and storing the preprocessed temperature time sequence data of each temperature measuring point and temperature induced strain time sequence data into a plurality of data sets corresponding to each time one by one;
step 2, clustering temperature measuring point position coordinates and strain measuring point position coordinates based on a condensation hierarchical clustering method to obtain clustering clusters containing measuring point position topology information, and calculating average values of temperature and temperature induced strain data in the respective clustering clusters at each moment;
step 3, obtaining distribution topology characteristic value data of a temperature field and a temperature induced strain field by using average values of clustering clusters of the temperature and the temperature induced strain at each moment respectively;
step 4, dividing the topological characteristic value data of the temperature field and the temperature induced strain field distribution, which are in one-to-one correspondence with all the moments, into a training set and a checking set according to a fixed proportion, and carrying out normalization processing;
step 5, constructing a two-way long-short-term memory regression network model, setting network parameters and training parameters, taking temperature field distribution topology characteristic value data as input and temperature induced strain field distribution topology characteristic value data as output, and training the regression network model by using a training set; when training reaches a set value, verifying the validity of the regression network model by using a test set; when the difference definition test index between the output result of the regression network model and the data of the distribution topology characteristic values of the strain field caused by the test set temperature meets the requirement, training is completed, and a trained regression network model is obtained; otherwise, adjusting the network parameters and the training parameters, and retraining;
step 6, continuously acquiring discrete temperature time sequence data of each temperature measuring point and strain time sequence data of a strain measuring point, which are monitored in real time in preset time, of a structure to be detected in the structure monitoring process, and calculating distribution topology characteristic values of a temperature field and a temperature strain field at each moment by using the method in the steps 1-3; taking the distribution topology characteristic values of the temperature fields at all moments as the input of the trained regression network model to obtain the output result of the regression network model at the corresponding moment; calculating and judging whether the difference definition inspection indexes of the regression network model output result and the monitored temperature induced strain field distribution topology characteristic value at the same moment meet the requirements, and when the difference definition inspection indexes which do not meet the requirements in the preset time are larger than the preset proportion, representing that the temperature induced strain mode of the structure is redistributed; otherwise, it means that no redistribution of the temperature induced strain modes of the structure occurs.
Further, the method of the step 1 specifically comprises the following steps:
step 1.1, extracting discrete temperature time sequence data of each temperature measuring point in a certain time from a structure to be measured;
step 1.2, extracting time sequence data of each discrete strain measuring point in a certain time of a structure to be measured, and performing spectrum analysis on temperature time sequence data of temperature measuring points in a certain range of the structural strain measuring points to obtain frequency bands [0, f corresponding to high power density values of a power spectrum of the temperature time sequence data t ]Processing the data of each strain measuring point, including wavelet transformation and low-pass filtering to obtain a frequency less than f t And takes the strain signal as temperature-induced strain time sequence data of each strain measuring point;
and 1.3, adjusting the sampling frequency of the temperature time sequence data of the temperature measuring points and the temperature induced strain time sequence data to smaller values or random values lower than the smaller values, and storing the temperature of each measuring point and the temperature induced strain time sequence data into a plurality of data sets corresponding to each time one by one.
Further, in the step 2, the temperature measuring point position coordinates are clustered based on a condensation hierarchical clustering method to obtain a cluster containing the measuring point position topology information, and the method is as follows:
step 2.1, taking the position coordinates of each temperature measuring point as a cluster, taking the Euler distance of the measuring point as the distance between every two clusters, calculating the distance between every two clusters, and combining the two clusters closest to each other into a new cluster;
step 2.2, taking the sum of the squares of the increment of all the data point coordinates in the clusters as the distance between the clusters, calculating the distance between every two clusters, and combining the two clusters with the nearest distance to obtain a new cluster;
step 2.3, repeating the step 2.2 until the preset clustering number is reached;
and 2.4, judging whether the increment square sum of all data point coordinates in each cluster is larger than a preset distance in the clustering process, and outputting a cluster result that the increment square sum is larger than the preset distance and the absolute value of the difference between the increment square sum and the preset distance is the smallest.
Further, the calculation formula of the euler distance in step 2.1 is as follows:
wherein d pq Euler distance for coordinate points p and q, C p And C q One-dimensional or multidimensional coordinate vectors of coordinate points p and q respectively, j represents the dimension of the j-th coordinate vector, F represents the total dimension of the coordinate vectors, and c p,j And c p,j Respectively is vector C p And C q Is the value of the j-th coordinate vector dimension of (c).
Further, the calculation formula of the sum of squares of the increment of the coordinates of all the data points in the cluster in step 2.2 is as follows:
wherein D (r, s) is the sum of the squares of the increments of cluster r and cluster s, n r And n s The number of coordinate points within cluster r and cluster s, I.I 2 The euler distance is indicated as the sum of the euler distances,and->Centroid coordinates of the cluster r and the cluster s respectively; the calculation formula of the centroid coordinates of each cluster is as follows:
wherein C is ri Is the coordinate vector of the ith coordinate point in cluster r.
Further, in the step 2, the method for clustering the position coordinates of the strain measuring points based on the condensed hierarchical clustering method is consistent with the method for clustering the position coordinates of the temperature measuring points based on the condensed hierarchical clustering method.
Further, the distribution topology characteristic value of the temperature field in the step 3 is a pairwise difference vector of the average value of each cluster of the temperature at each moment, and the distribution topology characteristic value of the temperature induced strain field is a pairwise difference vector of the average value of each cluster of the temperature induced strain at each moment; the pairwise difference vectors are expressed as follows:
[d 1,2 ,d 1,3 ,…,d 1,M ,……,d u,v ,……,d M-1,M ]
wherein d u,v For the difference of the average value of the ith cluster minus the average value of the ith cluster, u.epsilon.1, M],v∈[1,M]And u+.v.
Further, the distribution topology characteristic value of the temperature field in the step 3 is a difference value adjacent matrix of the average value of each cluster of the temperature at each time, the distribution topology characteristic value of the temperature strain field is a difference value adjacent matrix of the average value of each cluster of the temperature strain at each time, and the difference value adjacent matrix is expressed as follows:
wherein d M,1 Subtracting the average value of the 1 st cluster from the average value of the M-th cluster, d 1,M Subtracting the Mth cluster from the average value of test data in the 1 st clusterDifferences, d, of the mean values of the test data u,v For the difference of the average value of the ith cluster minus the average value of the ith cluster, u.epsilon.1, M],v∈[1,M]And u+.v.
Further, the calculation formula of the difference definition inspection index is as follows:
wherein T is r Defining a test index for the difference, wherein R is the number of elements in the single data, k is the kth element in the single data, re k Te for outputting the result of the regression model k A is data of topological characteristic values of temperature-induced strain field distribution t For all Te k Maximum amplitude of the element value variation.
The beneficial effects are that: compared with the prior art, the technical scheme of the invention has the following beneficial effects:
(1) The method has strict logic. The invention provides a cluster topology model for establishing temperature measuring points and strain measuring points based on coordinate information of discretely arranged temperature and strain sensors, wherein in structural monitoring, the arrangement of the measuring points is deeply analyzed and researched by related practitioners, the information such as important focusing positions of structural safety and high probability action positions of extreme environmental loads is contained, and the temperature and strain test data are classified based on the cluster of the measuring point coordinates, so that the stress and strain distribution characteristics of the structure under the action of an uneven temperature field can be more effectively mastered.
(2) And the consideration is comprehensive. According to the invention, the fuzzy relation model between the topological characteristic values of the temperature field and the temperature induced strain field is established based on the deep learning network model, and the interference of high-frequency loads such as vehicles, wind and the like is eliminated in the temperature induced strain extraction process, so that the deep learning network model can effectively simulate the nonlinear fuzzy relation between the temperature field and the temperature induced strain field, and the randomness and variability of an actual structure under the action of a complex environment can be more fully considered based on the deep learning of the topological characteristic values of the temperature field and the temperature induced strain field. The method is relatively comprehensive in consideration, and the obtained result is more in accordance with engineering practice than the traditional method.
(3) The implementation is chapter-circulated. The implementation process of the invention is basically established on the processing and calculation of the sensor position information and the test data, the experience factors are few, and any technician with a certain mechanics, mathematics and computer foundation can realize the intelligent sensing of the structural temperature-induced strain field weight distribution based on the topological characteristics of the discrete measuring points of the universal sensor according to the patent. The method has strong replicability and is convenient to apply and popularize.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a schematic illustration of a temperature clustering result output by a hierarchical clustering method in an application example;
FIG. 3 is a schematic diagram of a result of clustering strain clusters output by a hierarchical clustering method in an application example;
FIG. 4 is a schematic diagram of a BiLSTM regression network designed in an application example;
FIG. 5 is a graph showing regression of cluster mean difference vector for temperature-induced strain field in redistribution versus actual measurement;
FIG. 6 is a graph showing regression of cluster mean difference adjacency matrix for temperature-induced strain field in redistribution versus actual measurement.
Detailed Description
The technical scheme of the invention is further described below with reference to the accompanying drawings and examples.
As shown in fig. 1, the temperature induced strain field redistribution intelligent sensing method based on the structure discrete measurement point topology mainly comprises the following steps:
step 1: discrete temperature measuring point time sequence data and strain measuring point temperature induced strain time sequence data are extracted, preprocessing is carried out, and the temperature and strain induced time sequence data of each measuring point after processing are stored as a plurality of data sets corresponding to each moment one by one, and the specific method is as follows:
extracting discrete time sequence data of each temperature measuring point;
extracting discrete onesThe time sequence data of the strain measuring points are subjected to spectrum analysis on the temperature measuring point data within a certain range of the structural strain measuring points to obtain frequency bands [0, f corresponding to high power density values of a power spectrum of the temperature data t ]Processing strain data of each measuring point by adopting general methods such as wavelet transformation, low-pass filtering and the like to obtain the strain data with the frequency less than f t And takes the strain signal as temperature-induced strain time sequence data of each strain measuring point;
and adjusting the sampling frequency of the temperature and temperature-induced strain time sequence data to be smaller or lower, and storing the temperature of each measuring point and the temperature-induced strain time sequence data into a plurality of data sets which are in one-to-one correspondence at each moment.
Step 2: clustering is carried out on temperature measuring point coordinates and strain measuring point coordinates based on a condensation hierarchical clustering method, and the distance between the measuring points is defined by Euler distance:
wherein C is p And C q One-dimensional or multidimensional coordinate vector of coordinate points p and q respectively, F is the dimension of the coordinate vector, c p,j And c p,j Respectively is vector C p And C q Is the j-th dimensional value of (c).
With the continuous aggregation of coordinate points, the distance between clusters containing a plurality of coordinate points is defined by the sum of squares of the increment of all the coordinates of data points in the clusters:
wherein I 2 Representing the calculated Euler distance, n r And n s The number of coordinate points within clusters r and s respectively,andthe centroid coordinates of the clusters r and s are calculated according to the following formula:
wherein C is ri Is the coordinate vector of the ith coordinate point in cluster r. After the clustering calculation is finished, outputting that D (r, s) of each cluster in the aggregation process of each level is larger than a preset distance D y And with D y Cluster results with the smallest absolute difference. Obtaining cluster clusters containing the position topology information of the measuring points, wherein the number of the output temperature cluster clusters and the number of the output strain cluster clusters are required to be kept consistent;
the temperature or temperature-induced strain average of the temperature and temperature-induced strain data within the respective cluster at each time instant is calculated.
Step 3: calculating the pairwise difference vector of the average value of each cluster of the temperature and the temperature induced strain at each moment:
[d 1,2 ,d 1,3 ,…,d 1,M ,……,d m,m+1 ,d m,m+2 ,…,d m,M ,……,d M-1,M ]
based on the difference vector, let d m,M =-d M,m A differential adjacency matrix of the average value of each cluster of the assembly temperature and the temperature induced strain at each moment:
and taking a difference vector or a difference adjacency matrix of the average value of each cluster of the temperature and the temperature induced strain at each moment as a distribution topology characteristic value of the temperature field and the temperature induced strain field.
Step 4: dividing the topological characteristic value data of the temperature field and the temperature induced strain field distribution, which are in one-to-one correspondence at all moments, into a training set and a checking set according to a fixed proportion of 9:1 and the like; normalizing the training set and the checking set data according to the normalization parameters calculated by the training set data.
Step 5: designing a two-way long-short-term memory regression network, wherein the two-way long-short-term memory regression network model structure at least comprises: the input layer, the two-way long-short-term memory hidden layer and the regression output layer are used for setting network parameters such as an objective function optimization algorithm, hidden layer unit number and the like of the two-way long-short-term memory network; training parameters such as initial learning rate, minimum Batch Size, maximum Epoch and the like are set, temperature field distribution topological characteristic value data in a preprocessed (normalized and the like) training set is used as input, temperature induced strain field distribution topological characteristic value data is used as output, and training and learning are carried out on a regression network.
Step 6: after model training reaches a preset Epoch, checking the validity of the model by adopting normalized test set data, inputting the temperature field distribution topology characteristic values in the test set into a trained model, and defining a test index based on the difference between a training model regression result and test set temperature induced strain field distribution topology characteristic value data:
wherein R is the number of elements in single data, re k Te as model regression result k In order to test the data of the distribution topology characteristic values of the set of actual measurement temperature-induced strain fields, A t For a single Te k Maximum amplitude of the medium element value change; each data in the test set must be combined with the corresponding regression result to calculate a T r If T of the whole test set r If the average value is less than a predetermined value (e.g., 0.05), the regression network model is considered to be operational, and if not, the network parameters and training parameters are adjusted and the model is retrained.
Step 7: in the structure monitoring process, firstly, a regression model is built according to the steps 1 to 6 based on the existing monitoring big data with a certain time length (such as half a year); then calculating the topological characteristic value of the temperature field according to the steps 1 to 3 by using the temperature data monitored in real time, and inputting the topological characteristic value into the established regression network model for regression to obtain the distributed topological characteristic value of the temperature induced strain field; converting real-time regressive temperature induced strain field distribution topology characteristic value and real-time strain monitoring data to obtain temperature induced stressComparing the topological characteristic values of the variable field distribution, such as the difference between the variable field distribution and the topological characteristic values is in a larger state for a long time (such as T calculated by the variable field distribution and the topological characteristic values r An index greater than 0.1 or higher over time) indicates that a redistribution of the temperature-induced strain modes of the structure has occurred.
Example 1:
the specific implementation process of the invention is described below by taking the long-term monitoring data of the temperature and strain sensors of the midspan section of the structural health monitoring system of a certain large-span steel truss cable-stayed bridge in Anhui province as an example.
(1) Extracting time sequence data of 18 discrete temperature measuring points; extracting time sequence data of 20 discrete strain measuring points, and performing spectrum analysis on temperature measuring point data within a certain range of structural strain measuring points to obtain a frequency band which corresponds to a high power density value of a temperature data power spectrum and is 0 to 3 multiplied by 10 -3 Hz, adopting multi-layer wavelet transformation to process strain data of each measuring point to obtain the frequency less than 3 multiplied by 10 -3 The strain signal of Hz is used as temperature-induced strain time sequence data of each strain measuring point; and adjusting the sampling frequency of the temperature and temperature-induced strain time sequence data to 1/60Hz, and storing the temperature and temperature-induced strain time sequence data of each measuring point into a plurality of data sets which are in one-to-one correspondence at each moment.
(2) Clustering temperature measuring point coordinates and strain measuring point coordinates based on a condensation hierarchical clustering method respectively to obtain temperature and strain clustering clusters containing measuring point position topological information, outputting temperature and strain clustering cluster clustering results with cluster distance being respectively more than 610 cm and 1265 cm, wherein the number of the output temperature clustering clusters and the number of the strain clustering clusters are 8, and referring to the figures 2 and 3; the temperature or temperature-induced strain average of the temperature and temperature-induced strain data within the respective cluster at each time instant is calculated.
(3) Calculating the pairwise difference vectors of the average values of the clusters of the temperature and the temperature induced strain at each moment, wherein the temperature and the temperature induced strain are 230400 vectors with the length of 28; based on the difference vector, assembling a difference adjacent matrix of the average value of each cluster of temperature and temperature induced strain at each moment, and totally 2X 230400 matrices of 8X 8; and taking a difference vector and a difference adjacency matrix of the average value of each cluster of the temperature and the temperature induced strain at each moment as the distribution topology characteristic values of the temperature field and the temperature induced strain field.
(4) Dividing cluster mean value difference vector data of the temperature field and the temperature induced strain field which are in one-to-one correspondence at each moment into a training set and a checking set according to a fixed ratio of 9:1; normalizing the training set and the checking set data according to the normalization parameters calculated by the training set data.
(5) Designing a two-way long-short-term memory regression network, wherein the network structure is shown in figure 4, and setting an objective function optimization algorithm of the two-way long-short-term memory network as self-adaptive moment estimation and hidden layer unit number as 300; setting an initial learning rate of 0.01, minimum Batch Size of 4300 and a Maximum Epoch of 600, adopting cluster mean value difference vector data of a temperature field in a training set after normalization and the like as input, adopting cluster mean value difference vector data of a temperature induced strain field as output, and training and learning a regression network.
(6) After model training reaches a preset Epoch, cluster mean value difference vector data of a temperature field in the test set is input into the trained model to obtain a cluster mean value difference vector result of a temperature induced strain field regressed by the model, and the regression result precision of the model is tested by adopting the cluster mean value difference vector data of the temperature induced strain field in the normalized test set; each data in the test set must be combined with the corresponding regression result to calculate a T r If T of the whole test set r If the average value is smaller than the preset value of 0.05, the regression network model is considered to be available, otherwise, the network parameters and the training parameters are adjusted and the model is retrained.
(7) In the structure monitoring process, a regression model is established based on cluster mean value difference vector data (data within 160 days) of 230400 multiplied by 2 temperature fields and temperature induced strain fields; then, calculating cluster mean value difference vector real-time data of the temperature field by using the temperature data monitored in real time, and inputting the cluster mean value difference vector real-time regression data of the temperature-induced strain field obtained by regression in the established regression network model; comparing the cluster mean value difference vector of the temperature-induced strain field subjected to real-time regression with the cluster mean value difference vector of the temperature-induced strain field obtained by converting the real-time strain monitoring data, and if the difference between the cluster mean value difference vector and the cluster mean value difference vector is in a larger state for a long time, representing that the temperature-induced strain mode of the structure is redistributed. If engineering needs, the cluster mean value difference vector data of the temperature field and the temperature induced strain field can be converted into a cluster mean value difference adjacent matrix of the temperature field and the temperature induced strain field, and the cluster mean value difference adjacent matrix can more intuitively reflect the two-dimensional characteristics of the distribution field. Fig. 5 and fig. 6 are respectively a cluster mean value difference vector comparison diagram of a temperature-induced strain field obtained by converting a real-time regression of a cluster mean value difference vector of the temperature-induced strain field and a real-time strain monitoring data at a certain moment when redistribution occurs, and specifically refer to fig. 5, and a cluster mean value difference adjacency matrix of a temperature-induced strain field obtained by converting a real-time regression of a cluster mean value difference adjacency matrix of the temperature-induced strain field and a real-time strain monitoring data, and specifically refer to fig. 6.
The above embodiments are merely further specific illustrations of the solution of the present invention, and after reading the embodiments of the present invention, modifications and substitutions of various equivalent forms of the present invention by those skilled in the art are within the scope of protection defined by the claims of the present application.
Claims (8)
1. The temperature-induced strain field redistribution intelligent sensing method based on the structure discrete measurement point topology is characterized by comprising the following steps of:
step 1, for a structure to be tested, extracting discrete temperature time sequence data of each temperature measuring point and temperature induced strain time sequence data of each strain measuring point of the structure within a certain time, preprocessing, and storing the preprocessed temperature time sequence data of each temperature measuring point and temperature induced strain time sequence data into a plurality of data sets corresponding to each time one by one;
step 2, clustering temperature measuring point position coordinates and strain measuring point position coordinates based on a condensation hierarchical clustering method to obtain clustering clusters containing measuring point position topology information, and calculating average values of temperature and temperature induced strain data in the respective clustering clusters at each moment;
step 3, obtaining distribution topology characteristic value data of a temperature field and a temperature induced strain field by using average values of clustering clusters of the temperature and the temperature induced strain at each moment respectively;
step 4, dividing the topological characteristic value data of the temperature field and the temperature induced strain field distribution, which are in one-to-one correspondence with all the moments, into a training set and a checking set according to a fixed proportion, and carrying out normalization processing;
step 5, constructing a two-way long-short-term memory regression network model, setting network parameters and training parameters, taking temperature field distribution topology characteristic value data as input and temperature induced strain field distribution topology characteristic value data as output, and training the regression network model by using a training set; when training reaches a set value, verifying the validity of the regression network model by using a test set; when the difference definition test index between the output result of the regression network model and the data of the distribution topology characteristic values of the strain field caused by the test set temperature meets the requirement, training is completed, and a trained regression network model is obtained; otherwise, adjusting the network parameters and the training parameters, and retraining;
step 6, continuously acquiring discrete temperature time sequence data of each temperature measuring point and strain time sequence data of a strain measuring point, which are monitored in real time in preset time, of a structure to be detected in the structure monitoring process, and calculating distribution topology characteristic values of a temperature field and a temperature strain field at each moment by using the method in the steps 1-3; taking the distribution topology characteristic values of the temperature fields at all moments as the input of the trained regression network model to obtain the output result of the regression network model at the corresponding moment; calculating and judging whether the difference definition inspection indexes of the regression network model output result and the monitored temperature induced strain field distribution topology characteristic value at the same moment meet the requirements, and when the difference definition inspection indexes which do not meet the requirements in the preset time are larger than the preset proportion, representing that the temperature induced strain mode of the structure is redistributed; otherwise, indicating that the temperature-induced strain mode of the structure is not redistributed;
in the step 2, the temperature measuring point position coordinates are clustered based on a condensation hierarchical clustering method to obtain a cluster containing measuring point position topology information, and the method comprises the following steps:
step 2.1, taking the position coordinates of each temperature measuring point as a cluster, taking the Euler distance of the measuring point as the distance between every two clusters, calculating the distance between every two clusters, and combining the two clusters closest to each other into a new cluster;
step 2.2, taking the sum of the squares of the increment of all the data point coordinates in the clusters as the distance between the clusters, calculating the distance between every two clusters, and combining the two clusters with the nearest distance to obtain a new cluster;
step 2.3, repeating the step 2.2 until the preset clustering number is reached;
and 2.4, judging whether the increment square sum of all data point coordinates in each cluster is larger than a preset distance in the clustering process, and outputting a cluster result that the increment square sum is larger than the preset distance and the absolute value of the difference between the increment square sum and the preset distance is the smallest.
2. The intelligent sensing method for temperature-induced strain field redistribution based on the structure discrete measurement point topology according to claim 1, wherein the method of step 1 is specifically as follows:
step 1.1, extracting discrete temperature time sequence data of each temperature measuring point in a certain time from a structure to be measured;
step 1.2, extracting time sequence data of each discrete strain measuring point in a certain time of a structure to be measured, and performing spectrum analysis on temperature time sequence data of temperature measuring points in a certain range of the structural strain measuring points to obtain frequency bands [0, f corresponding to high power density values of a power spectrum of the temperature time sequence data t ]Processing the data of each strain measuring point, including wavelet transformation and low-pass filtering to obtain a frequency less than f t And takes the strain signal as temperature-induced strain time sequence data of each strain measuring point;
and 1.3, adjusting the sampling frequency of the temperature time sequence data of the temperature measuring points and the temperature induced strain time sequence data to smaller values or random values lower than the smaller values, and storing the temperature of each measuring point and the temperature induced strain time sequence data into a plurality of data sets corresponding to each time one by one.
3. The intelligent sensing method for temperature-induced strain field weight distribution based on the structure discrete measurement point topology according to claim 1, wherein the calculation formula of the euler distance in step 2.1 is as follows:
wherein d pq Euler distance for coordinate points p and q, C p And C q One-dimensional or multidimensional coordinate vectors of coordinate points p and q respectively, j represents the dimension of the j-th coordinate vector, F represents the total dimension of the coordinate vectors, and c p,j And c q,j Respectively is vector C p And C q Is the value of the j-th coordinate vector dimension of (c).
4. The intelligent sensing method of temperature-induced strain field weight distribution based on the structure discrete measurement point topology according to claim 1, wherein the calculation formula of the increment square sum of all data point coordinates in the cluster in step 2.2 is as follows:
wherein D (r, s) is the sum of the squares of the increments of cluster r and cluster s, n r And n s The number of coordinate points within cluster r and cluster s, I.I 2 The euler distance is indicated as the sum of the euler distances,and->Centroid coordinates of the cluster r and the cluster s respectively; the calculation formula of the centroid coordinates of each cluster is as follows:
in the method, in the process of the invention,C ri is the coordinate vector of the ith coordinate point in cluster r.
5. The intelligent sensing method for temperature-induced strain field weight distribution based on the structure discrete measurement point topology according to claim 1, wherein in the step 2, a method for clustering strain measurement point position coordinates based on a condensation hierarchical clustering method is consistent with a method for clustering temperature measurement point position coordinates based on a condensation hierarchical clustering method.
6. The intelligent sensing method for temperature-induced strain field weight distribution based on structure discrete measurement point topology according to claim 1, wherein in the step 3, the distribution topology characteristic value of the temperature field is a pairwise difference vector of the average value of each cluster of temperature at each moment, and the distribution topology characteristic value of the temperature-induced strain field is a pairwise difference vector of the average value of each cluster of temperature-induced strain at each moment; the pairwise difference vectors are expressed as follows:
[d 1,2 ,d 1,3 ,…,d 1,M ,……,d u,v ,……,d M-1,M ]
wherein d u,v For the difference of the average value of the ith cluster minus the average value of the ith cluster, u.epsilon.1, M],v∈[1,M]And u+.v.
7. The intelligent sensing method for temperature-induced strain field redistribution based on structure discrete measurement point topology according to claim 1, wherein in step 3, the distribution topology characteristic value of the temperature field is a difference value adjacent matrix of average values of temperature clusters at each time, the distribution topology characteristic value of the temperature-induced strain field is a difference value adjacent matrix of average values of temperature-induced strain clusters at each time, and the difference value adjacent matrix is expressed as follows:
wherein d M,1 For the M-th clusterThe difference of the average value minus the average value of cluster 1, d 1,M The difference of the average value of the test data in the 1 st cluster minus the average value of the test data in the M th cluster.
8. The intelligent sensing method for temperature-induced strain field redistribution based on structure discrete measurement point topology according to claim 1, wherein a calculation formula of the difference definition test index is as follows:
wherein T is r Defining a test index for the difference, wherein R is the number of elements in the single data, k is the kth element in the single data, re k Te for outputting the result of the regression model k A is data of topological characteristic values of temperature-induced strain field distribution t For all Te k Maximum amplitude of the element value variation.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111326559.XA CN113973403B (en) | 2021-11-10 | 2021-11-10 | Temperature-induced strain field redistribution intelligent sensing method based on structure discrete measurement point topology |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111326559.XA CN113973403B (en) | 2021-11-10 | 2021-11-10 | Temperature-induced strain field redistribution intelligent sensing method based on structure discrete measurement point topology |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113973403A CN113973403A (en) | 2022-01-25 |
CN113973403B true CN113973403B (en) | 2024-02-23 |
Family
ID=79589623
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111326559.XA Active CN113973403B (en) | 2021-11-10 | 2021-11-10 | Temperature-induced strain field redistribution intelligent sensing method based on structure discrete measurement point topology |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113973403B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116542146A (en) * | 2023-01-06 | 2023-08-04 | 中路高科交通检测检验认证有限公司 | Bridge monitoring temperature field-strain field space-time correlation model and health diagnosis method |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110224862A (en) * | 2019-05-20 | 2019-09-10 | 杭州电子科技大学 | Multi-agent system network appearance based on multilayer perceptron invades capability assessment method |
CN110533007A (en) * | 2019-09-13 | 2019-12-03 | 东南大学 | A kind of vehicle-mounted strain of bridge influences the identification of line feature intelligent and extracting method |
-
2021
- 2021-11-10 CN CN202111326559.XA patent/CN113973403B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110224862A (en) * | 2019-05-20 | 2019-09-10 | 杭州电子科技大学 | Multi-agent system network appearance based on multilayer perceptron invades capability assessment method |
CN110533007A (en) * | 2019-09-13 | 2019-12-03 | 东南大学 | A kind of vehicle-mounted strain of bridge influences the identification of line feature intelligent and extracting method |
Non-Patent Citations (1)
Title |
---|
高速铁路钢桁拱桥安全状态评估方法 研究.《中国博士学位论文全文数据库》.2021,全文. * |
Also Published As
Publication number | Publication date |
---|---|
CN113973403A (en) | 2022-01-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108344564B (en) | A kind of state recognition of main shaft features Testbed and prediction technique based on deep learning | |
CN105956216B (en) | Correction method for finite element model greatly across steel bridge based on uniform temperature response monitor value | |
CN109659933A (en) | A kind of prediction technique of power quality containing distributed power distribution network based on deep learning model | |
CN110470259A (en) | Landslide displacement dynamic prediction method based on LSTM | |
CN115758212B (en) | Mechanical equipment fault diagnosis method based on parallel network and transfer learning | |
CN109034191A (en) | One-dimensional telemetry exception interpretation method based on ELM | |
CN112507479B (en) | Oil drilling machine health state assessment method based on manifold learning and softmax | |
CN110516907A (en) | A kind of rock burst grade evaluation method based on AHP- entropy weight cloud model | |
CN113973403B (en) | Temperature-induced strain field redistribution intelligent sensing method based on structure discrete measurement point topology | |
CN107832789A (en) | Characteristic weighing k nearest neighbor method for diagnosing faults based on the conversion of average influence Value Data | |
CN110555235A (en) | Structure local defect detection method based on vector autoregressive model | |
CN114548375A (en) | Cable-stayed bridge main beam dynamic deflection monitoring method based on bidirectional long-short term memory neural network | |
Lei et al. | Fault diagnosis of rotating machinery based on a new hybrid clustering algorithm | |
CN110082106A (en) | A kind of Method for Bearing Fault Diagnosis of the depth measure study based on Yu norm | |
CN116757078A (en) | Method and system for measuring flow velocity of pulverized coal based on acting force | |
CN114046816B (en) | Sensor signal fault diagnosis method based on lightweight gradient lifting decision tree | |
CN112990601B (en) | Worm wheel machining precision self-healing system and method based on data mining | |
CN113642822B (en) | VAE-based sample set directivity extension method for evaluating building group structure safety | |
Bi et al. | A fault diagnosis algorithm for wind turbine blades based on bp neural network | |
CN108053093A (en) | A kind of k- neighbour's method for diagnosing faults based on the conversion of average influence Value Data | |
CN105699043A (en) | Method for improving measuring stability and precision of wind tunnel sensor | |
CN112784462A (en) | Hydraulic structure stress deformation prediction system based on finite element method | |
CN111291490B (en) | Nonlinear mapping intelligent modeling method for structure multi-scale heterogeneous response | |
CN112528849B (en) | Structure health monitoring method based on inner product matrix and deep learning | |
CN115688544B (en) | Microwave attenuation snowfield chromatography reconstruction method based on MLP neural network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |