A method, system, medium and device for detecting safety status of building curtain wall

Through the Kolmogolov-Arnold representation theorem and Transformer neural network, a building curtain wall damage recognition system was built, which solved the problem of difficulty in accurately detecting the overall safety status of the curtain wall in the existing technology, and achieved lossless, fast and accurate detection Effect.

CN118568584BActive Publication Date: 2024-11-26广东省有色工业建筑质量检测站有限公司
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202411025966.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-30
Publication Date
2024-11-26
Estimated Expiration
2044-07-30

AI Technical Summary

Technical Problem

The existing architectural curtain wall detection technology is difficult to accurately reflect the overall safety status of the curtain wall, and the existing methods are effective under specific constraints and cannot truly reflect the deep damage of the curtain wall in actual inspection.

Method used

By pre-preparing standard curtain wall components, obtaining dynamic data and constructing a benchmark dynamic feature model, using the Kolmogolov-Arnold representation theorem and Transformer neural network, a building curtain wall damage recognition system is established to realize the dynamic feature data of curtain wall components Accurate analysis and overall health status assessment.

Benefits of technology

It realizes non-destructive, fast and accurate safety status detection of building curtain walls, which can accurately reveal the connection between each stressed part, reduce the amount of detection parameters, improve the detection efficiency, and infer deep damage.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118568584B_ABST
    Figure CN118568584B_ABST
Patent Text Reader

Abstract

The present invention discloses a method, system, medium and equipment for detecting the safety status of a building curtain wall. The method comprises the following steps: S1, obtaining power data through a pre-prepared standard curtain wall component; S2, constructing a reference power characteristic model based on the power data based on the KA theorem to obtain a reference power characteristic function fd under various working conditions; S3, obtaining a reference power characteristic function fd under various working conditions through J reference power characteristic data d 0,j Train the Transformer neural network to obtain J building curtain wall damage recognition neural networks N j ; S4, obtain the actual building curtain wall component dynamic data W and pre-process, through J building curtain wall damage recognition neural network N j Analyze and obtain the dynamic characteristic data D of J actual building curtain wall components N,j ; S5, calculate dynamic characteristic data D N,j Compared with the benchmark dynamic characteristic data d 0,j The safety status of the actual building curtain wall components can be obtained according to the correlation analysis. The present invention can realize non-destructive, rapid and accurate detection of the safety status of the building curtain wall.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of building curtain wall detection, and in particular, relates to a method, system, medium and equipment for detecting the safety status of a building curtain wall. Background Art

[0002] With economic development, the amount of building curtain walls has increased year by year. As the external protective structure of the building, the curtain wall is exposed to the environment such as sun and rain, atmospheric corrosion and temperature changes for a long time during its service, which leads to the aging of structural adhesives and damage to connectors in various components of the curtain wall, causing them to lose their original bearing capacity and bring the risk of panel falling off. In the existing building curtain walls, curtain walls built earlier than the industry standards may have hidden dangers that do not meet safety requirements. In addition, the existing building glass curtain walls have a large stock and the newly completed glass curtain walls are growing rapidly. In addition, the lack of necessary maintenance during use has led to the gradual emergence of safety issues of glass curtain walls. Therefore, it is of great significance to detect the safety performance of glass curtain walls and ensure that they are in a safe state.

[0003] The existing technical specifications for assessing the safety performance of curtain walls usually adopt a method that combines visual inspection with material performance testing. However, the data obtained by visual inspection for appearance inspection is not accurate enough, and the performance testing of materials requires the destruction of the curtain wall components themselves, which makes later repairs difficult and costly. In most cases, construction units do not accept such destructive testing methods. Therefore, the existing curtain wall inspection results are mostly based on "surface data" obtained by visual inspection, which cannot truly reflect the deep-level safety status of the curtain wall.

[0004] Various researchers have also conducted a lot of research on the safety detection and monitoring technology of curtain walls. Among them, the direction of using structural dynamic data as the core indicator has become a research hotspot. The basic principle is that when the constraints of curtain wall components (i.e., the connection parts or the curtain wall itself) change, its structural dynamic data (such as the first-order natural frequency) will also change linearly accordingly. Therefore, it is only necessary to establish a related model between the constraints and dynamic data of curtain wall components, and then based on the actual measured dynamic data of curtain wall components, the constraints of the curtain wall (i.e., the damage situation) can be quickly inferred. However, this research direction has not been effectively promoted and applied in actual work. After analysis, the main reason is that the dynamic data type is relatively isolated and it is difficult to reflect the overall situation of the curtain wall.

[0005] The composition of curtain wall is mainly divided into two parts: panel system and support system. Curtain wall should be regarded as a system composed of panels and connections, supports and connections. However, the existing curtain wall dynamic data models are mostly studied for a single component of the curtain wall, such as:

[0006] (1) A study established a damage identification model for curtain walls by analyzing the changing trend of the dynamic data response of glass panels under different degrees of debonding of structural adhesives. The model shows that the dynamic data of glass panels decreases as the degree of debonding of structural adhesives increases. The model proposed by this method implies a prerequisite: when the supporting system of the curtain wall is intact, the change in the dynamic data of the curtain wall can reflect the state of the structural adhesive.

[0007] (2) A study established a correlation model between the tightness of the pressure plate connectors connecting the panel components to the supporting components and the dynamic data of the curtain wall. The model shows that the better the tightness of the pressure plate, the larger the measured dynamic data, and vice versa. Similarly, the implicit prerequisite of the model proposed by this method is that when the panel system of the curtain wall is intact, the changes in the dynamic data of the curtain wall can reflect the status of the supporting connection parts.

[0008] The above-mentioned existing methods are only valid under specific and single curtain wall constraint conditions. During actual curtain wall inspection, since it is impossible to know the actual constraint state of the curtain wall, the measured structural dynamic data is input into the existing research model and the measured results are difficult to truly reflect the overall state of the curtain wall. When studying the influencing factors and laws of the dynamic data of the curtain wall, the existing technology does not comprehensively consider the joint action of multiple parts. Summary of the invention

[0009] The first purpose of the invention is to overcome the shortcomings and deficiencies in the prior art and provide a method for detecting the safety status of a building curtain wall, so as to achieve non-destructive, rapid and accurate detection of the safety status of a building curtain wall.

[0010] The second object of the present invention is to provide a building curtain wall safety status detection system.

[0011] The third object of the present invention is to provide a storage medium.

[0012] A fourth object of the present invention is to provide a computing device.

[0013] The purpose of the present invention is achieved through the following technical solution: A method for detecting the safety status of a building curtain wall comprises the steps of:

[0014] S1. Obtain the dynamic data of the preset monitoring position through the pre-prepared standard curtain wall component, wherein the standard curtain wall component includes J working conditions, and the jth working condition is recorded as {i a A.i b B.i c C, D j} j , j = 1, 2, ..., J, A, B, C and D represent the sub-conditions corresponding to different monitoring parts, respectively. Let the sets of power data measured under sub-conditions A, B, C and D be denoted as x A 、xB 、x C and x D ,

[0015] Among them, i a 、i b and i c They represent the different damage levels set for sub-conditions A, B and C respectively;

[0016] S2. Based on the Kolmogorov-Arnold representation theorem, a benchmark dynamic characteristic model is constructed according to the dynamic data to obtain f(x D ) about (x A 、x B 、x C )'s benchmark dynamic characteristic function fd=f(x A , x B , x C ), where f(x D ) represents x D The mapping relationship is as follows: let the reference dynamic characteristic function under the jth working condition be recorded as fd j ;

[0017] S3, data fd j Recorded as the benchmark dynamic characteristic data d 0,j , through J benchmark dynamic characteristic data d 0,j Train the Transformer neural network to obtain J building curtain wall damage recognition neural networks N j ;

[0018] S4, obtain the actual building curtain wall component dynamic data W and pre-process it, through J building curtain wall damage recognition neural network N j The pre-processed data are analyzed respectively to obtain the dynamic characteristic data D of J actual building curtain wall components. N,j ;

[0019] S5. Calculate the dynamic characteristic data D of each actual building curtain wall component N,j The corresponding reference dynamic characteristic data d 0,j The safety status of the actual building curtain wall components is obtained according to the correlation analysis.

[0020] Preferably, step S1 specifically includes the steps of:

[0021] S11. Produce standard curtain wall components according to the supported curtain wall structure style;

[0022] S12, setting several stress-bearing parts of the standard curtain wall component as monitoring parts, and installing vibration sensors respectively, wherein the monitoring parts include structural adhesive, sealant and pressure plate; at the same time, setting the panel of the standard curtain wall component as an excitation part, and installing an acceleration sensor and a pulse excitation sensor;

[0023] S13, set different sub-conditions for structural adhesive, sealant and pressure plate, respectively denoted as sub-condition A={i a A}, sub-condition B={i b B} and sub-case C={i c C},i a 、i b 、i c ∈[1, 0.75, 0.5, 0.25, 0], and [1, 0.75, 0.5, 0.25, 0] correspond to the degree of damage: intact, 1 / 4 damage, 2 / 4 damage, 3 / 4 damage and total loss;

[0024] S14. Under different sub-condition combinations, the panel is stimulated by the pulse excitation sensor, and the response data of all vibration sensors and acceleration sensors are collected to obtain the dynamic data sets x of the structural adhesive, sealant, pressure plate and panel. A 、x B 、x C and x D , that is, x A =x {iaA} 、x B =x {ibB} and x C =x {icC} .

[0025] Preferably, step S2 specifically includes the steps of:

[0026] S21. Since the dynamic performance of the panel is caused by the joint working condition constraints of the structural adhesive, sealant and pressure plate, f(x D ) is considered as about x A 、x B 、x C The ternary function of , according to the Kolmogorov-Arnold representation theorem, namely the KA theorem, obtains the benchmark dynamic characteristic model:

[0027] , formula (1),

[0028] Among them, the number of multidimensional dimensions n=3, x p Take {x1, x2, x3}, where x1, x2, x3 refer to x in turn. A 、x B 、x C , x A 、x B and xC is the input of the model, f(x D ) is the output of the model. The middle of the model is 2 KA layers, which respectively obtain S q,p and P q , p∈[1, 3], q∈[1, 7], S q,p is to convert the variable x p A one-dimensional continuous function mapped to real numbers; P q is a one-dimensional continuous function used to process S q,p The sum after mapping; the basis function form of P and S is: P(x), S(x)=ω(b(x)+spline(x)), where ω is the weight coefficient, b(x)=silu(x)=x / (1+e -x ), spline(x) is the B-spline function;

[0029] S22, let the mapping function group S of x1 q,1 (x1) is denoted as {S1, S2, S3, S4, S5, S6, S7}, and the mapping function set S of x2 q,2 (x2) is denoted as {S8, S9, S 10 , S 11 , S 12 , S 13 , S 14}, the mapping function group S of x3 q,3 (x3) is denoted as {S 15 , S 16 , S 17 , S 18 , S 19 , S 20 , S 21};

[0030] According to formula (1), for the input data x A 、x B 、x C The S basis functions (S1, S2, ..., S 21 ), add the S basis functions at the nodes to get the functions {D1, D2, ..., D7} at the nodes, then fit the functions at each node to get the P basis functions (P1, P2, ..., P7), add all the P basis functions to get the target function, i.e., the dynamic characteristic function fd=f(x A , x B , x C ), where f(x D )=P1+P2+P3+P4+P5+P6+P7, D1=S1+S8+S 15 , D2=S2+S9+S 16 , D3=S3+S 10 +S17 , D4=S4+S 11 +S 18 , D5=S5+S 12 +S 19 , D6=S6+S 13 +S 20 , D7=S7+S 14 +S 21 , f(x A )=S1+S2+S3+S4+S5+S6+S7, f(x B )=S8+S9+S 10 +S 11 +S 12 +S 13 +S 14 , f(x C )=S 15 +S 16 +S 17 +S 18 +S 19 +S 20 +S 21 .

[0031] Preferably, step S3 specifically includes the steps of:

[0032] S31, for the jth group of reference dynamic characteristic data d 0,j Perform Gaussian transformation to obtain data d t,j , the Gaussian diffusion formula is:

[0033] , formula (2),

[0034] Among them, d 0,j Indicates the initial state when data is input, d t,j Indicates d 0,j The data after Gaussian transformation, t is the Gaussian transformation coefficient, t is a positive integer and 1<t≤T, T is a preset positive integer, , , is the weight coefficient term, β t is the preset value and 0.00001≤β t ≤0.01, e is the randomly sampled data that follows the standard normal distribution N (0, 1), that is, e~ N (0, 1);

[0035] S32, group j t,j Denoted as d t , will (d t , t) is input to the Transformer neural network, and the neural network outputs the predicted value e θ (d t , t);

[0036] S33, calculate the loss function Loss=||ee θ (d t , t)|| 2 , where e is the true value, e θ (d t , t) is the predicted value;

[0037] S34, by minimizing the loss function, updating the neural network parameters, repeating steps S31 to S33, when the loss error reaches a preset value, stopping the training, and obtaining the target neural network, i.e., the building curtain wall damage recognition neural network N j .

[0038] Preferably, step S4 specifically includes the steps of:

[0039] S41, preprocessing: The initial state of the dynamic data W of the actual building curtain wall component is recorded as D0, and D0 is Gaussian transformed into D according to formula (3): t ,

[0040] , formula (3),

[0041] Wherein, t is a Gaussian transformation coefficient, t is a positive integer and 1<t≤T, T is a preset positive integer, , , is the weight coefficient term, β t is the preset value and 0.00001≤β t ≤0.01, E is the randomly sampled data that follows the standard normal distribution N (0, 1), that is, E~ N (0, 1);

[0042] S42, D t Input into J building curtain wall damage identification neural networks N j , the data output by the j-th building curtain wall damage identification neural network is recorded as E θ (D t , t), where θ represents the neural network parameters;

[0043] S43, according to formula (4) t Perform an inverse Gaussian transform:

[0044] , formula (4),

[0045] S44, D t-1 Input again into the building curtain wall damage identification neural network N j According to steps S42~S43, the calculation is repeated for t-1 rounds to obtain the final dynamic characteristic data D of the actual building curtain wall component. N,j .

[0046] Preferably, step S5 specifically includes:

[0047] S51. Calculate the dynamic characteristic data D of each actual building curtain wall component according to the relative entropy calculation formula of formula (5) N,j The corresponding reference dynamic characteristic data d 0,j The correlation of , and J correlation coefficient values ​​are obtained:

[0048] , formula (5),

[0049] Among them, the correlation coefficient value M j Indicates the jth dynamic characteristic data D N,j and the jth reference dynamic characteristic data d 0,j The dissimilarity of U(d 0,j ) and V(d N,j ) represent the reference dynamic characteristic data d 0,j and the dynamic characteristic data D of actual building curtain wall components N,j Data distribution;

[0050] S52, select the minimum value among the J correlation coefficient values, namely min{M j}, the jth group of working conditions corresponding to the minimum value is the actual working condition of the actual building curtain wall components, that is, it represents the safety status of the actual building curtain wall components.

[0051] A building curtain wall safety status detection system, comprising:

[0052] The acquisition module is used to obtain the dynamic data of the preset monitoring position through the pre-prepared standard curtain wall components, and the standard curtain wall components include J working conditions, and the jth working condition is recorded as {i a A.i b B.i c C, D j} j , j = 1, 2, ..., J, A, B, C and D represent the sub-conditions corresponding to different monitoring parts, respectively. Let the sets of power data measured under sub-conditions A, B, C and D be denoted as x A 、x B 、x C and x D ,

[0053] Among them, i a 、i b and i c They represent the different damage levels set for sub-conditions A, B and C respectively;

[0054] The calculation module is used to construct a benchmark dynamic characteristic model based on the Kolmogorov-Arnold representation theorem according to the dynamic data to obtain f(x D ) about (x A 、x B 、x C )'s benchmark dynamic characteristic function fd=f(x A , x B , x C ), where f(x D ) represents x D The mapping relationship is as follows: let the reference dynamic characteristic function under the jth working condition be recorded as fd j ;

[0055] Neural network training module, used to transform data fd j Recorded as the benchmark dynamic characteristic data d 0,j , through J benchmark dynamic characteristic data d 0,j Train the Transformer neural network to obtain J building curtain wall damage recognition neural networks N j ;

[0056] The analysis module is used to obtain the actual dynamic data W of the building curtain wall components and preprocess them, and then use J building curtain wall damage recognition neural networks N j The pre-processed data are analyzed respectively to obtain the dynamic characteristic data D of J actual building curtain wall components. N,j ;

[0057] Analysis module, used to calculate the dynamic characteristic data D of each actual building curtain wall component N,j The corresponding reference dynamic characteristic data d 0,j The safety status of the actual building curtain wall components is obtained according to the correlation analysis.

[0058] A storage medium stores a program, and when the program is executed by a processor, the above-mentioned method for detecting the safety status of a building curtain wall is implemented.

[0059] A computing device comprises a processor and a memory for storing a program executable by the processor. When the processor executes the program stored in the memory, the above-mentioned method for detecting the safety status of a building curtain wall is implemented.

[0060] Compared with the prior art, the present invention has the following advantages and effects:

[0061] (1) The present invention provides a method for detecting the safety status of a building curtain wall. Suitable standard curtain wall components are pre-made, and vibration sensors are installed at key stress-bearing locations as monitoring locations. Various damage conditions are simulated, and the dynamic characteristic data of each monitoring location is measured after applying excitation. Then, according to the KA theorem, the dynamic data of each monitoring location (x A 、x B 、x C ) and the overall data of the building curtain wall components x D The correlation characteristic function is the benchmark dynamic characteristic model, and the dynamic characteristic function fd of the building curtain wall component is obtained through the benchmark dynamic characteristic model. Compared with the dynamic correlation model of the building curtain wall component based on a single constraint condition in the prior art, the dynamic characteristic function obtained by the present invention can accurately reveal the relationship between the overall data change and each stress-bearing part, and evaluate the overall health status of the building curtain wall component with multiple elements or multiple constraints, and the result is more practical. Not only that, the origin of the dynamic characteristic function fd in the present invention is based on the KA representation theorem, while most of the existing research models are obtained by approximate fitting based on the general approximation theorem or finite element analysis. In contrast, the present invention uses the KA theorem to accurately express it, and the result is more accurate.

[0062] (2) The building curtain wall damage recognition neural network provided by the present invention uses the dynamic characteristic function fd obtained by the KA theorem that can express the characteristics of each stress-bearing part to train the Transformer neural network to obtain the building curtain wall damage recognition neural network N j , the neural network N j The neural network learns the characteristics of the benchmark dynamic characteristic function of the standard building curtain wall component, and therefore has the ability to parse the input data into a corresponding dynamic characteristic function. The method of the present invention only needs to measure the dynamic data of one panel of the building curtain wall component, and the function containing the dynamic data of each stress-bearing part can be parsed through the building curtain wall damage identification neural network, and then compared with the test data of the standard building curtain wall component, by calculating the correlation between the two, the deep-level actual working conditions of the actual building curtain wall component measured can be inferred, and its safety status can be analyzed. Compared with the existing curtain wall damage detection technology, the solution of the present invention greatly reduces the number of detection parameters, and can also infer the deep-level damage situation, which greatly improves the efficiency of building curtain wall safety status detection and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 The present invention is a flow chart of a method for detecting the safety status of a building curtain wall.

[0064] Figure 2The schematic diagram of the structure of the benchmark dynamic characteristic model constructed according to the Kolmogorov-Arnold representation theorem of the present invention.

[0065] Figure 3 It is a schematic diagram of the corresponding relationship between various working conditions and corresponding power characteristic functions of the present invention. DETAILED DESCRIPTION

[0066] The present invention will be further described in detail below in conjunction with embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0067] Example 1

[0068] like Figure 1 The figure shows a flow chart of a method for detecting the safety status of a building curtain wall, which includes the following steps:

[0069] S1. Obtain the dynamic data of the preset monitoring position through the pre-prepared standard curtain wall component, wherein the standard curtain wall component includes J working conditions, and the jth working condition is recorded as {i a A.i b B.i c C, D j} j , j = 1, 2, ..., J, A, B, C and D represent the sub-conditions corresponding to different monitoring parts, respectively. Let the sets of power data measured under sub-conditions A, B, C and D be denoted as x A 、x B 、x C and x D ,

[0070] Among them, i a 、i b and i c They represent the different damage levels set for sub-conditions A, B and C respectively;

[0071] S2. Based on the Kolmogorov-Arnold representation theorem, a benchmark dynamic characteristic model is constructed according to the dynamic data to obtain f(x D ) about (x A 、x B 、x C )'s benchmark dynamic characteristic function fd=f(x A , x B , x C ), where f(x D ) represents x D The mapping relationship is as follows: let the reference dynamic characteristic function under the jth working condition be recorded as fd j ;

[0072] S3, data fd j Recorded as the benchmark dynamic characteristic data d0,j , through J benchmark dynamic characteristic data d 0,j Train the Transformer neural network to obtain J building curtain wall damage recognition neural networks N j ;

[0073] S4, obtain the actual building curtain wall component dynamic data W and pre-process it, through J building curtain wall damage recognition neural network N j The pre-processed data are analyzed respectively to obtain the dynamic characteristic data D of J actual building curtain wall components. N,j ;

[0074] S5. Calculate the dynamic characteristic data D of each actual building curtain wall component N,j The corresponding reference dynamic characteristic data d 0,j The safety status of the actual building curtain wall components is obtained according to the correlation analysis.

[0075] Specifically, the present invention provides a method for detecting the safety status of a building curtain wall, wherein suitable standard curtain wall components are pre-made, and vibration sensors are installed at key stress-bearing locations as monitoring locations, various damage conditions are simulated, and the dynamic characteristic data of each monitoring location is measured after applying excitation; then, according to the KA theorem, the dynamic data (x A 、x B 、x C ) and the overall data of the building curtain wall components x D The correlation characteristic function is the benchmark dynamic characteristic model, and the dynamic characteristic function of the building curtain wall component is obtained through the benchmark dynamic characteristic model. Compared with the dynamic correlation model of the building curtain wall component based on a single constraint condition in the prior art, the dynamic characteristic function obtained by the present invention can accurately reveal the relationship between the overall data change and each stress-bearing part, and evaluate the overall health status of the building curtain wall component with multiple elements or multiple constraints, and the result is more practical. Not only that, the origin of the dynamic characteristic function fd in the present invention is based on the KA representation theorem, while most of the existing research models are obtained by approximate fitting based on the general approximation theorem or finite element analysis. In contrast, the present invention adopts the KA theorem to accurately express it, and the result is more accurate.

[0076] At the same time, the building curtain wall damage recognition neural network provided by the present invention uses the dynamic characteristic function fd obtained by the KA theorem that can express the characteristics of each force-bearing part to train the Transformer neural network to obtain the building curtain wall damage recognition neural network N j , the neural network N jThe characteristics of the benchmark dynamic characteristic function of the standard building curtain wall component are learned, so that the input data has the ability to parse out the corresponding dynamic characteristic function. The method of the present invention only needs to measure the dynamic data of one panel of the building curtain wall component, and the function containing the dynamic data of each stress-bearing part can be parsed through the building curtain wall damage identification neural network, and then compared with the test data of the standard building curtain wall component, by calculating the correlation between the two, the deep-level actual working conditions of the actual building curtain wall component measured can be inferred, and its safety status can be analyzed. Compared with the existing curtain wall damage detection technology, the solution of the present invention greatly reduces the number of detection parameters, and can also infer the deep-level damage situation, which greatly improves the efficiency of building curtain wall safety status detection and has broad application prospects.

[0077] Step S1 specifically includes the following steps:

[0078] S11. Produce standard curtain wall components according to the supported curtain wall structure style;

[0079] S12, setting several stress-bearing parts of the standard curtain wall component as monitoring parts, and installing vibration sensors respectively, wherein the monitoring parts include structural adhesive, sealant and pressure plate; at the same time, setting the panel of the standard curtain wall component as an excitation part, and installing an acceleration sensor and a pulse excitation sensor;

[0080] S13, set different sub-conditions for structural adhesive, sealant and pressure plate, respectively denoted as sub-condition A={i a A}, sub-condition B={i b B} and sub-case C={i c C},i a 、i b 、i c ∈[1, 0.75, 0.5, 0.25, 0], and [1, 0.75, 0.5, 0.25, 0] correspond to the degree of damage: intact, 1 / 4 damage, 2 / 4 damage, 3 / 4 damage and total loss;

[0081] S14. Under different sub-condition combinations, the panel is stimulated by the pulse excitation sensor, and the response data of all vibration sensors and acceleration sensors are collected to obtain the dynamic data sets x of the structural adhesive, sealant, pressure plate and panel. A 、x B 、x C and x D , that is, x A =x {iaA} 、x B =x {ibB} and x C =x {icC} .

[0082] Specifically, when making standard curtain wall components, the present invention can exhaustively list the types of supporting curtain wall structures to make standard curtain wall components, and then set a suitable number of dynamic data monitoring locations according to the force analysis of the corresponding structure. The standard curtain wall components at least include a panel, an auxiliary frame, a fastening kit (including at least a pressure plate, bolts, and nuts) and a sealant, and the four sides of the panel are bonded to the auxiliary frame by structural adhesive. The standard curtain wall components are connected in the following manner: the adjacent side auxiliary frames between the curtain wall components are fastened and connected by a fastening kit with a pressure plate, the pressure plate is pressed on the adjacent auxiliary frame connection part, and the bolts and nuts are fastened and connected perpendicular to the pressure plate; the adjacent panels between the curtain wall components are connected by a sealant. In this embodiment, the standard curtain wall components are made according to the general hidden frame curtain wall structure, and the main dynamic parameter monitoring locations are set at the structural adhesive, sealant, panel, and pressure plate. At the same time, the panel is also used as the location for applying excitation.

[0083] When making standard curtain wall components, vibration sensors are installed at the following locations at the same time. The vibration sensors include optical fiber sensors and acceleration sensors: optical fiber sensors are embedded in structural adhesives and sealants; optical fiber sensors are attached to the flat surface of the pressure plate; acceleration sensors are attached to the center of the outer surface of the panel or at a quarter of the width and height, and pulse excitation sensors are attached next to the acceleration sensors. The wire ends of all sensors are connected to the host equipment (including industrial computers and signal amplifiers, etc.), and the host equipment communicates with computer data processing software.

[0084] After the standard curtain wall components are manufactured, different sub-conditions are set for each monitoring position. In this embodiment, the conditions of the structural adhesive, sealant, and pressure plate are recorded as A, B, and C respectively. Then, the panel is stimulated in turn by the pulse excitation sensor to collect the response data of each sensor. Specifically, the data collected by the optical fiber sensor implanted in the structural adhesive is recorded as x A , where x A The sensor data of the structural adhesive on the four sides of the component panel; the data collected by the optical fiber sensor embedded in the sealant is recorded as x B , where x B The sensor data of the sealant at the adjacent edge between the component and the adjacent component; the data collected by the optical fiber sensor attached to the flat surface of the pressure plate is recorded as x C , where x C The sensor data of the set pressure plate at the four sides where the components are connected to the adjacent components by the fastening set (in this embodiment, one optical fiber sensor is arranged at each of the four sides of the pressure plate near the two ends); the data collected by the acceleration sensor attached to the outer surface of the panel is recorded as x D , the same pulse frequency is used when the pulse excitation sensor is used for excitation.

[0085] Each working condition adopts a linear destruction operation. In this embodiment, the method of destroying 1 / 4, 2 / 4, 3 / 4, and 4 / 4 in sequence is adopted. The intact working condition is represented as 1, that is, the structural adhesive and sealant are evenly and fully applied, there is no leakage, the fastening kit is installed without misalignment and the bolts are tightened; the completely damaged working condition is represented as 0, that is, the structural adhesive and sealant parts are all empty, and the fastening kit is completely loose, that is, [1, 0.75, 0.5, 0.25, 0] correspond to the degree of destruction in sequence: intact, destroyed 1 / 4, destroyed 2 / 4, destroyed 3 / 4, and completely damaged. Among them, the structural adhesive and sealant are destroyed by cutting, and the pressure plate is destroyed by loosening the bolts.

[0086] Then the working conditions of A, B, and C are set as follows: Working condition A includes {1A, 0.75A, 0.5A, 0.25A, 0A}, i a ∈[1, 0.75, 0.5, 0.25, 0], the B condition includes {1B, 0.75B, 0.5B, 0.25B, 0B}, i b ∈[1, 0.75, 0.5, 0.25, 0], C condition includes {1C, 0.75C, 0.5C, 0.25C, 0C}, i c ∈[1, 0.75, 0.5, 0.25, 0], that is, A, B and C each include 5 sub-operating conditions. In this embodiment, as shown in Table 1 and Figure 3 As shown in the figure, by permuting and combining the working condition types {A, B, C} and the damage degree [1, 0.75, 0.5, 0.25, 0], we can get J=5×5×5=125 working conditions. The jth working condition is denoted as {i a A.i b B.i c C} j For example, the first working condition is recorded as {1A, 1B, 1C}1, and so on, marking the combination of A, B and C under all working conditions. When the standard curtain wall components are excited under each working condition, the x A 、x B 、x C and x D For example, under the working condition of {1A, 0.5B, 0C}, the collected data is x 1A 、x 0.5B 、x 0C and {x {1A} , x {0.5B} , x {0C}} 15 That is x D,15 .

[0087] It should be noted that in this embodiment, only three types of working conditions, A, B and C, are set, and each type of working condition only has 5

[0088] It can perform operations with different degrees of destruction. When promoting and applying it, multiple working conditions and destruction strategies can be set.

[0089] Step S2 specifically includes the following steps:

[0090] S21. Since the dynamic performance of the panel is caused by the joint working condition constraints of the structural adhesive, sealant and pressure plate, f(x D ) is considered as about x A 、x B 、x C The ternary function of , according to the Kolmogorov-Arnold representation theorem, namely the KA theorem, obtains the benchmark dynamic characteristic model:

[0091] , formula (1),

[0092] Among them, the number of multidimensional dimensions n=3, x p Take {x1, x2, x3}, where x1, x2, x3 refer to x in turn. A 、x B 、x C , x A 、x B and x C is the input of the model, f(x D ) is the output of the model. The middle of the model is 2 KA layers, which respectively obtain S q,p and P q , p is a positive integer and ∈[1, 3], q is a positive integer and ∈[1, 7], S q,p is to convert the variable x p A one-dimensional continuous function mapped to real numbers; P q is a one-dimensional continuous function used to process S q,p The sum after mapping; the basis function form of P and S is: P(x), S(x)=ω(b(x)+spline(x)), where ω is the weight coefficient, b(x)=silu(x)=x / (1+e -x ), spline(x) is the B-spline function;

[0093] S22, let the mapping function group S of x1 q,1 (x1) is denoted as {S1, S2, S3, S4, S5, S6, S7}, and the mapping function set S of x2 q,2 (x2) is denoted as {S8, S9, S 10 , S 11 , S 12 , S 13 , S 14}, the mapping function group S of x3 q,3 (x3) is denoted as {S 15 , S 16 , S17 , S 18 , S 19 , S 20 , S 21};

[0094] According to formula (1), for the input data x A 、x B 、x C The S basis functions (S1, S2, ..., S 21 ), add the S basis functions at the nodes to get the functions {D1, D2, ..., D7} at the nodes, then fit the functions at each node to get the P basis functions (P1, P2, ..., P7), add all the P basis functions to get the target function, i.e., the dynamic characteristic function fd=f(x A , x B , x C ), where f(x D )=P1+P2+P3+P4+P5+P6+P7, D1=S1+S8+S 15 , D2=S2+S9+S 16 , D3=S3+S 10 +S 17 , D4=S4+S 11 +S 18 , D5=S5+S 12 +S 19 , D6=S6+S 13 +S 20 , D7=S7+S 14 +S 21 , f(x A )=S1+S2+S3+S4+S5+S6+S7, f(x B )=S8+S9+S 10 +S 11 +S 12 +S 13 +S 14 , f(x C )=S 15 +S 16 +S 17 +S 18 +S 19 +S 20 +S 21 .

[0095] Specifically, the mapping function group S of x1 q,1 (x1): {S 1,1 (x1), S 2,1 (x1), S 3,1 (x1), S 4,1 (x1), S5,1 (x1), S 6,1 (x1), S 7,1 (x1)}, x2's mapping function set S q,2 (x2): {S 1,2 (x2), S 2,2 (x2), S 3,2 (x2), S 4,2 (x2), S 5,2 (x2), S 6,2 (x2), S 7,2 (x2)}, the mapping function set S of x3 q,3 (x3): {S 1,3 (x3), S 2,3 (x3), S 3,3 (x3), S 4,3 (x3), S 5,3 (x3), S 6,3 (x3), S 7,3 (x3)}; let x1, x2, x3 refer to x in turn. A 、x B 、x C , let S 1,1 (x1), S 2,1 (x1), S 3,1 (x1), S 4,1 (x1), S 5,1 (x1), S 6,1 (x1), S 7,1 (x1) are denoted as S1, S2, S3, S4, S5, S6, S7 in sequence; S 1,2 (x2), S 2,2 (x2), S 3,2 (x2), S 4,2 (x2), S 5,2 (x2), S 6,2 (x2), S 7,2 (x2) are denoted as S8, S9, S 10 , S 11 , S 12 , S 13 , S 14 ; S 1,3 (x3), S 2,3 (x3), S 3,3 (x3), S 4,3 (x3), S 5,3 (x3), S 6,3 (x3), S 7,3 (x3) is denoted as S 15 , S 16 , S 17 , S18 , S 19 , S 20 , S 21 .

[0096] Specifically, in this example, only three types of working conditions A, B, and C are set, so the measured x D is the final result of the action under the A, B, and C states, that is, the dynamic performance of the outer surface of the panel is caused by the joint constraint of the structural adhesive, sealant, and pressure plate, so f(x D ) and x A 、x B and x C There must be a correlation between f(x D ) is about x A 、x B 、x C Functions of three types of elements.

[0097] According to the Kolmogorov-Arnold Representation Theorem (KART, Kolmogorov-Arnold Representation Theorem, hereinafter referred to as KA Theorem): For any multivariate continuous function, it can be expressed as a combination of a finite number of single variable functions and addition. The KA Theorem formula is:

[0098] , formula (6),

[0099] Among them, n represents the number of multidimensional dimensions, p and q represent the sequence numbers, φ q,p is a single variable x p One-dimensional continuous function mapped to real numbers, Φ q It is also a one-dimensional continuous function, used to process φ q,p After mapping, the sum is D ) is considered as a three-variable function f(x A , x B , x C ), then n=3. It should be noted that when the monitoring location is set to n according to actual needs, f(x D ) is regarded as an n-variable function, which also satisfies the KA theorem.

[0100] According to the KA theorem formula (6), for the ternary function f(x D ) that is, f(x A , x B , x C ), it can be expressed by simply adding and calculating through 2 layers of network functions with a width of (2n+1), i.e. 2*3+1=7. Figure 2The figure shows a schematic diagram of the structure of the benchmark dynamic characteristic model constructed according to the Kolmogorov-Arnold representation theorem of the present invention. The S and P basis functions can be fitted according to the general training method of the neural network in the prior art. First, the basis function parameters are initialized, and the input data x is A 、x B and x C , the output data (i.e. predicted value) is d, by minimizing the loss function (Loss = | predicted value d-true value x D | 2 ) Update the weight parameters of the basis function, and finally fit the ideal P basis function and S basis function to obtain the objective function.

[0101] In this embodiment, it is known that the data pair {x A , x B , x C , x D}, where the input is x A 、x B 、x C , the output is about x D The function f(x D ), according to the KA theorem, the goal is to find a function f that satisfies f(x D ) = f (x A , x B , x C ). In this embodiment, the data directly measured is x A 、x B 、x C and x D , but its corresponding mapping relationship f(x A )、f(x B )、f(x C ) and f(x D ) is unknown, and the data x D The function mapping relationship is related to the data (x A 、x B 、x C ) is related to the function mapping relationship, that is, x D The mapping relationship f(x D ) can be used with data (x A 、x B 、x C ) of the mapping function f(x A 、x B 、x C ) (which can be obtained by KA theorem). Since the present invention requires data (x A 、x B 、x C ) is the mapping function f(x A , xB , x C ) represents x D The mapping function, as a differentiation, is directly obtained by fitting the data x D The obtained mapping function is f(x D ) means that through the data (x A 、x B 、x C ) According to the KA theorem, the mapping function is f(x A , x B , x C ) is represented by fd; to ensure that the obtained fd=f(x D ), by minimizing the loss function (Loss = | predicted value d - true value x D | 2 ) to ensure.

[0102] According to the steps of S21-S22, calculate f(x D ) about (x A 、x B 、x C )'s dynamic characteristic function fd=f(x A , x B , x C ), specifically, the working conditions and damage degrees are arranged and combined according to the calculation method in Table 1, and the corresponding relationship between various working conditions and corresponding dynamic characteristic functions is obtained as follows: Figure 3 As shown:

[0103] Table 1 Calculation method for the combination of various working conditions and damage degree

[0104]

[0105] It can be seen from the above table that in this embodiment, J=125, that is, 125 fds are obtained accordingly.

[0106] Step S3 specifically includes the following steps:

[0107] S31, for the jth group of reference dynamic characteristic data d 0,j Perform Gaussian transformation to obtain data d t,j , the Gaussian diffusion formula is:

[0108] , formula (2),

[0109] Among them, d 0,j Indicates the initial state when data is input, d t,j Indicates d 0,j The data after Gaussian transformation, t is the Gaussian transformation coefficient, t is a positive integer and 1<t≤T, T is a preset positive integer, , , is the weight coefficient term, β t is the preset value and 0.00001≤β t ≤0.01, e is the randomly sampled data that follows the standard normal distribution N (0, 1), that is, e~ N (0, 1);

[0110] S32, group j t,j Denoted as d t , will (d t , t) is input to the Transformer neural network, and the neural network outputs the predicted value e θ (d t , t);

[0111] S33, calculate the loss function Loss=||ee θ (d t , t)|| 2 , where e is the true value, e θ (d t , t) is the predicted value;

[0112] S34, by minimizing the loss function, updating the neural network parameters, repeating steps S31 to S33, when the loss error reaches a preset value, stopping the training, and obtaining the target neural network, i.e., the building curtain wall damage recognition neural network N j .

[0113] Specifically, the building curtain wall damage identification neural network N in this embodiment j That is, a dynamic characteristic function parsing neural network, which has the ability to parse out the dynamic characteristic functions of structural adhesives, sealants and pressure plates by measuring the dynamic data W on the outer surface of the panel. It is trained based on the data of hidden frame building curtain wall components. When it comes to other types of building curtain wall structures, such as exposed frame, semi-hidden frame, stone and aluminum plate curtain walls, etc., the corresponding parsing neural network is trained according to the above steps S31~S34, which is used to parse the dynamic characteristic data of the corresponding force-bearing parts such as structural adhesives, sealants and pressure plates according to the dynamic data at the panel.

[0114] Step S4 specifically includes the following steps:

[0115] S41, preprocessing: The initial state of the dynamic data W of the actual building curtain wall component is recorded as D0, and D0 is Gaussian transformed into D according to formula (3): t ,

[0116] , formula (3),

[0117] Wherein, t is a Gaussian transformation coefficient, t is a positive integer and 1<t≤T, T is a preset positive integer, , , is the weight coefficient term, β t is the preset value and 0.00001≤β t ≤0.01, E is the randomly sampled data that follows the standard normal distribution N (0, 1), that is, E~ N (0, 1);

[0118] S42, D t Input into J building curtain wall damage identification neural networks N j , the data output by the j-th building curtain wall damage identification neural network is recorded as E θ (D t , t), where θ represents the neural network parameters;

[0119] S43, according to formula (4) t Perform an inverse Gaussian transform:

[0120] , formula (4),

[0121] S44, D t-1 Input again into the building curtain wall damage identification neural network N j According to steps S42~S43, the calculation is repeated for t-1 rounds to obtain the final dynamic characteristic data D of the actual building curtain wall component. N,j .

[0122] Step S5 specifically includes:

[0123] S51. Calculate the dynamic characteristic data D of each actual building curtain wall component according to the relative entropy calculation formula of formula (5) N,j The corresponding reference dynamic characteristic data d 0,j The correlation of , and J correlation coefficient values ​​are obtained:

[0124] , formula (5),

[0125] Among them, the correlation coefficient value M j Indicates the jth dynamic characteristic data D N,j and the jth reference dynamic characteristic data d 0,j The dissimilarity of U(d 0,j ) and V(d N,j ) represent the reference dynamic characteristic data d 0,j and the dynamic characteristic data D of actual building curtain wall components N,j Data distribution;

[0126] S52, select the minimum value among the J correlation coefficient values, namely min{M j}, the jth group of working conditions corresponding to the minimum value is the actual working condition of the actual building curtain wall components, that is, it represents the safety status of the actual building curtain wall components.

[0127] Specifically, for example, when min{M j} corresponds to Figure 3 The figure in the figure is the working condition of group J. At this time, there are (0A, 0B, 0C), that is, the corresponding damage degree of the structural adhesive, sealant and pressure plate is total loss. The actual working condition or safety status of the building curtain wall actually tested is analyzed and judged as follows: the structural adhesive and sealant parts are all empty, and the fastening kit is completely loose.

[0128] Example 2

[0129] A building curtain wall safety status detection system, comprising:

[0130] The acquisition module is used to obtain the dynamic data of the preset monitoring position through the pre-prepared standard curtain wall components, and the standard curtain wall components include J working conditions, and the jth working condition is recorded as {i a A.i b B.i c C, D j} j , j = 1, 2, ..., J, A, B, C and D represent the sub-conditions corresponding to different monitoring parts, respectively. Let the sets of power data measured under sub-conditions A, B, C and D be denoted as x A 、x B 、x C and x D ,

[0131] Among them, i a 、i b and i c They represent the different damage levels set for sub-conditions A, B and C respectively;

[0132] The calculation module is used to construct a benchmark dynamic characteristic model based on the Kolmogorov-Arnold representation theorem according to the dynamic data to obtain f(x D ) about (x A 、x B 、x C )'s benchmark dynamic characteristic function fd=f(x A , x B , x C ), where f(x D ) represents x D The mapping relationship is as follows: let the reference dynamic characteristic function under the jth working condition be recorded as fd j ;

[0133] Neural network training module, used to transform data fd jRecorded as the benchmark dynamic characteristic data d 0,j , through J benchmark dynamic characteristic data d 0,j Train the Transformer neural network to obtain J building curtain wall damage recognition neural networks N j ;

[0134] The analysis module is used to obtain the actual dynamic data W of the building curtain wall components and preprocess them, and then use J building curtain wall damage recognition neural networks N j The pre-processed data are analyzed respectively to obtain the dynamic characteristic data D of J actual building curtain wall components. N,j ;

[0135] Analysis module, used to calculate the dynamic characteristic data D of each actual building curtain wall component N,j The corresponding reference dynamic characteristic data d 0,j The safety status of the actual building curtain wall components is obtained according to the correlation analysis.

[0136] Example 3

[0137] A storage medium stores a program, and when the program is executed by a processor, a building curtain wall safety status detection method described in Example 1 is implemented as follows:

[0138] S1. Obtain the dynamic data of the preset monitoring position through the pre-prepared standard curtain wall component, wherein the standard curtain wall component includes J working conditions, and the jth working condition is recorded as {i a A.i b B.i c C, D j} j , j = 1, 2, ..., J, A, B, C and D represent the sub-conditions corresponding to different monitoring parts, respectively. Let the sets of power data measured under sub-conditions A, B, C and D be denoted as x A 、x B 、x C and x D ,

[0139] Among them, i a 、i b and i c They represent the different damage levels set for sub-conditions A, B and C respectively;

[0140] S2. Based on the Kolmogorov-Arnold representation theorem, a benchmark dynamic characteristic model is constructed according to the dynamic data to obtain f(x D ) about (x A 、x B 、x C )'s benchmark dynamic characteristic function fd=f(x A, x B , x C ), where f(x D ) represents x D The mapping relationship is as follows: let the reference dynamic characteristic function under the jth working condition be recorded as fd j ;

[0141] S3, data fd j Recorded as the benchmark dynamic characteristic data d 0,j , through J benchmark dynamic characteristic data d 0,j Train the Transformer neural network to obtain J building curtain wall damage recognition neural networks N j ;

[0142] S4, obtain the actual building curtain wall component dynamic data W and pre-process it, through J building curtain wall damage recognition neural network N j The pre-processed data are analyzed respectively to obtain the dynamic characteristic data D of J actual building curtain wall components. N,j ;

[0143] S5. Calculate the dynamic characteristic data D of each actual building curtain wall component N,j The corresponding reference dynamic characteristic data d 0,j The safety status of the actual building curtain wall components is obtained according to the correlation analysis.

[0144] In the above process, the specific processing process is as described in Example 1 and will not be repeated here.

[0145] In this embodiment, the storage medium may be a magnetic disk, an optical disk, a computer memory, a read-only memory, a random access memory, a USB flash drive, a mobile hard disk, or the like.

[0146] Example 4

[0147] A computing device includes a processor and a memory for storing a program executable by the processor. When the processor executes the program stored in the memory, a building curtain wall safety status detection method described in Example 1 is implemented as follows:

[0148] S1. Obtain the dynamic data of the preset monitoring position through the pre-prepared standard curtain wall component, wherein the standard curtain wall component includes J working conditions, and the jth working condition is recorded as {i a A.i b B.i c C, D j} j , j = 1, 2, ..., J, A, B, C and D represent the sub-conditions corresponding to different monitoring parts, respectively. Let the sets of power data measured under sub-conditions A, B, C and D be denoted as x A 、xB 、x C and x D ,

[0149] Among them, i a 、i b and i c They represent the different damage levels set for sub-conditions A, B and C respectively;

[0150] S2. Based on the Kolmogorov-Arnold representation theorem, a benchmark dynamic characteristic model is constructed according to the dynamic data to obtain f(x D ) about (x A 、x B 、x C )'s benchmark dynamic characteristic function fd=f(x A , x B , x C ), where f(x D ) represents x D The mapping relationship is as follows: let the reference dynamic characteristic function under the jth working condition be recorded as fd j ;

[0151] S3, data fd j Recorded as the benchmark dynamic characteristic data d 0,j , through J benchmark dynamic characteristic data d 0,j Train the Transformer neural network to obtain J building curtain wall damage recognition neural networks N j ;

[0152] S4, obtain the actual building curtain wall component dynamic data W and pre-process it, through J building curtain wall damage recognition neural network N j The pre-processed data are analyzed respectively to obtain the dynamic characteristic data D of J actual building curtain wall components. N,j ;

[0153] S5. Calculate the dynamic characteristic data D of each actual building curtain wall component N,j The corresponding reference dynamic characteristic data d 0,j The safety status of the actual building curtain wall components is obtained according to the correlation analysis.

[0154] In the above process, the specific processing process is as described in Example 1 and will not be repeated here.

[0155] In this embodiment, the computing device may be a terminal device such as a desktop computer, a laptop computer, a PDA handheld terminal, a tablet computer, etc.

[0156] The above embodiments are preferred implementations of the present invention and are not intended to limit the present invention. Any other changes or other equivalent replacement methods that do not deviate from the technical solutions of the present invention are included in the protection scope of the present invention.

Claims

1. A method for detecting the safety status of a building curtain wall, It is characterized in that Includes steps: S1. Obtain the dynamic data of the preset monitoring position through the pre-prepared standard curtain wall component, wherein the standard curtain wall component includes J working conditions, and the jth working condition is recorded as {i a A.i b B.i c C, D j } j , j = 1, 2, ..., J, A, B, C and D represent the sub-conditions corresponding to different monitoring parts, respectively. Let the sets of power data measured under sub-conditions A, B, C and D be denoted as x A 、x B 、x C and x D , Among them, i a 、i b and i c They represent the different damage levels set for sub-conditions A, B and C respectively; S2. Based on the Kolmogorov-Arnold representation theorem, a benchmark dynamic characteristic model is constructed according to the dynamic data to obtain f(x D ) about (x A 、x B 、x C ) of the benchmark dynamic characteristic function fd=f(x A , x B , x C ), where f(x D ) represents x D The mapping relationship is as follows: let the reference dynamic characteristic function under the jth working condition be recorded as fd j ; S3, data fd j Recorded as the reference dynamic characteristic data d 0,j , through J benchmark dynamic characteristic data d 0,j Train the Transformer neural network to obtain J building curtain wall damage recognition neural networks N j ; S4, obtain the actual building curtain wall component dynamic data W and pre-process it, through J building curtain wall damage recognition neural network N j The pre-processed data are analyzed respectively to obtain the dynamic characteristic data D of J actual building curtain wall components. N,j ; S5. Calculate the dynamic characteristic data D of each actual building curtain wall component N,j The corresponding reference dynamic characteristic data d 0,j The safety status of the actual building curtain wall components is obtained according to the correlation analysis.

2. A method for detecting the safety status of a building curtain wall according to claim 1, It is characterized in that Step S1 specifically includes the following steps: S11. Produce standard curtain wall components according to the supported curtain wall structure style; S12, setting several stress-bearing parts of the standard curtain wall component as monitoring parts, and installing vibration sensors respectively, wherein the monitoring parts include structural adhesive, sealant and pressure plate; at the same time, setting the panel of the standard curtain wall component as an excitation part, and installing an acceleration sensor and a pulse excitation sensor; S13, set different sub-conditions for structural adhesive, sealant and pressure plate, respectively denoted as sub-condition A={i a A}, sub-condition B={i b B} and sub-case C={i c C},i a 、i b 、i c ∈[1, 0.75, 0.5, 0.25, 0], and [1, 0.75, 0.5, 0.25, 0] correspond to the degree of damage: intact, 1 / 4 damage, 2 / 4 damage, 3 / 4 damage and total loss; S14. Under different sub-condition combinations, the panel is stimulated by the pulse excitation sensor, and the response data of all vibration sensors and acceleration sensors are collected to obtain the dynamic data sets x of the structural adhesive, sealant, pressure plate and panel. A 、x B 、x C and x D , that is, x A =x {iaA} 、x B =x {ibB} and x C =x {icC} .

3. A method for detecting the safety status of a building curtain wall according to claim 2, It is characterized in that Step S2 specifically includes the following steps: S21. Since the dynamic performance of the panel is caused by the joint working condition constraints of the structural adhesive, sealant and pressure plate, f(x D ) is considered as about x A 、x B 、x C The ternary function of , according to the Kolmogorov-Arnold representation theorem, namely the KA theorem, obtains the benchmark dynamic characteristic model: , formula (1), Among them, the number of multidimensional dimensions n=3, x p Take {x 1 、x 2 、x 3 }, x 1 、x 2 、x 3 Refers to x in sequence A 、x B 、x C , x A 、x B and x C is the input of the model, f(x D ) is the output of the model. The middle of the model is 2 KA layers, which respectively obtain S q,p and P q , p∈[1, 3], q∈[1, 7], S q,p is to convert the variable x p A one-dimensional continuous function mapped to real numbers; P q is a one-dimensional continuous function used to process S q,p The sum after mapping; the basis function form of P is: P(x) = ω (b(x) + spline(x)), the basis function form of S is: S(x) = ω (b(x) + spline(x)), where ω is the weight coefficient, b(x) = silu(x) = x / (1 + e -x ), spline(x) is the B-spline function; S22, let x 1 The mapping function group S q,1 (x 1 ) is denoted as {S 1 , S 2 , S 3 , S 4 , S 5 , S 6 , S 7 }, x 2 The mapping function group S q,2 (x 2 ) is denoted as {S 8 , S 9 , S 10 , S 11 , S 12 , S 13 , S 14 }, x 3 The mapping function group S q,3 (x 3 ) is denoted as {S 15 , S 16 , S 17 , S 18 , S 19 , S 20 , S 21 }; According to formula (1), for the input data x A 、x B 、x C The S basis function (S 1 , S 2 ,…,S 21 ), add the S basis functions at the node and get the function {D 1 , D 2 ,…,D 7 }, and then fit the function at each node to obtain the P basis function (P 1 , P 2 ,…,P 7 ), add all P basis functions to get the objective function, i.e. the dynamic characteristic function fd=f(x A , x B , x C ), where f(x D )=P 1 +P 2 +P 3 +P 4 +P 5 +P 6 +P 7 , D 1 =S 1 +S 8 +S 15 , D 2 =S 2 +S 9 +S 16 , D 3 =S 3 +S 10 +S 17 , D 4 =S 4 +S 11 +S 18 , D 5 =S 5 +S 12 +S 19 , D 6 =S 6 +S 13 +S 20 , D 7 =S 7 +S 14 +S 21 , f(x A )=S 1 +S 2 +S 3 +S 4 +S 5 +S 6 +S 7 , f(x B )=S 8 +S 9 +S 10 +S 11 +S 12 +S 13 +S 14 ,f(x C )=S 15 +S 16 +S 17 +S 18 +S 19 +S 20 +S 21 。 4. A method for detecting the safety status of a building curtain wall according to claim 1, It is characterized in that Step S3 specifically includes the following steps: S31, for the jth group of reference dynamic characteristic data d 0,j Perform Gaussian transformation to obtain data d t,j , the Gaussian diffusion formula is: , formula (2), Among them, d 0,j Indicates the initial state when data is input, d t,j Indicates d 0,j The data after Gaussian transformation, t is the Gaussian transformation coefficient, t is a positive integer and 1<t≤T, T is a preset positive integer, , , is the weight coefficient term, β t is the preset value and 0.00001≤β t ≤0.01, e is the randomly sampled data that follows the standard normal distribution N (0, 1), that is, e~ N (0, 1); S32, group j t,j Denoted as d t , will (d t , t) is input to the Transformer neural network, and the neural network outputs the predicted value e θ (d t , t); S33, calculate the loss function Loss=||ee θ (d t , t)|| 2 , where e is the true value, e θ (d t , t) is the predicted value; S34, by minimizing the loss function, updating the neural network parameters, repeating steps S31 to S33, when the loss error reaches a preset value, stopping the training, and obtaining the target neural network, i.e., the building curtain wall damage recognition neural network N j .

5. A method for detecting the safety status of a building curtain wall according to claim 1, It is characterized in that Step S4 specifically includes the following steps: S41, preprocessing: the initial state of the actual building curtain wall component power data W is recorded as D 0 According to formula (3), D 0 Gaussian transform to D t , , formula (3), Wherein, t is the Gaussian transformation coefficient, t is a positive integer and 1<t≤T, T is a preset positive integer, , , is the weight coefficient term, β t is the preset value and 0.00001≤β t ≤0.01, E is the randomly sampled data that follows the standard normal distribution N (0, 1), that is, E~ N (0, 1); S42, D t Input into J building curtain wall damage identification neural networks N j , the data output by the j-th building curtain wall damage identification neural network is recorded as E θ (D t , t), where θ represents the neural network parameters; S43, according to formula (4) t Perform an inverse Gaussian transform: , formula (4), S44, D t-1 Input again into the building curtain wall damage identification neural network N j According to steps S42~S43, the calculation is repeated for t-1 rounds to obtain the final dynamic characteristic data D of the actual building curtain wall component. N,j .

6. A method for detecting the safety status of a building curtain wall according to claim 1, It is characterized in that Step S5 specifically includes: S51. Calculate the dynamic characteristic data D of each actual building curtain wall component according to the relative entropy calculation formula of formula (5) N,j The corresponding reference dynamic characteristic data d 0,j The correlation of , and J correlation coefficient values ​​are obtained: , formula (5), Among them, the correlation coefficient value M j Indicates the jth dynamic characteristic data D N,j and the jth reference dynamic characteristic data d 0,j The dissimilarity of U(d 0,j ) and V(d N,j ) represent the reference dynamic characteristic data d 0,j and the dynamic characteristic data D of actual building curtain wall components N,j Data distribution; S52, select the minimum value among the J correlation coefficient values, namely min{M j }, the jth group of working conditions corresponding to the minimum value is the actual working condition of the actual building curtain wall components, that is, it represents the safety status of the actual building curtain wall components.

7. A building curtain wall safety status detection system, It is characterized in that include: The acquisition module is used to obtain the dynamic data of the preset monitoring position through the pre-prepared standard curtain wall components, and the standard curtain wall components include J working conditions, and the jth working condition is recorded as {i a A.i b B.i c C, D j } j , j = 1, 2, ..., J, A, B, C and D represent the sub-conditions corresponding to different monitoring parts, respectively. Let the sets of power data measured under sub-conditions A, B, C and D be denoted as x A 、x B 、x C and x D , Among them, i a 、i b and i c They represent the different damage levels set for sub-conditions A, B and C respectively; The calculation module is used to construct a benchmark dynamic characteristic model based on the Kolmogorov-Arnold representation theorem according to the dynamic data to obtain f(x D ) about (x A 、x B 、x C ) of the benchmark dynamic characteristic function fd=f(x A , x B , x C ), where f(x D ) represents x D The mapping relationship is as follows: let the reference dynamic characteristic function under the jth working condition be recorded as fd j ; Neural network training module, used to transform data fd j Recorded as the reference dynamic characteristic data d 0,j , through J benchmark dynamic characteristic data d 0,j Train the Transformer neural network to obtain J building curtain wall damage recognition neural networks N j ; The analysis module is used to obtain the actual dynamic data W of the building curtain wall components and preprocess them, and then use J building curtain wall damage recognition neural networks N j The pre-processed data are analyzed respectively to obtain the dynamic characteristic data D of J actual building curtain wall components. N,j ; Analysis module, used to calculate the dynamic characteristic data D of each actual building curtain wall component N,j The corresponding reference dynamic characteristic data d 0,j The safety status of the actual building curtain wall components is obtained according to the correlation analysis.

8. A storage medium, It is characterized in that A program is stored, and when the program is executed by a processor, a building curtain wall safety status detection method as described in any one of claims 1 to 6 is implemented.

9. A computing device, It is characterized in that It comprises a processor and a memory for storing a program executable by the processor. When the processor executes the program stored in the memory, a building curtain wall safety status detection method as described in any one of claims 1 to 6 is implemented.

Citation Information

Patent Citations

  • Hidden frame glass curtain wall structural adhesive damage identification method based on inherent frequency signal

    CN116008391A

  • Steel structure and curtain wall intelligent monitoring system based on data analysis

    CN116838114A

  • Chloride ion diffusion concentration detection method, system, medium and equipment

    CN117686442A

  • Building structure monitoring method and device and storage medium

    CN117824756A