Steel-concrete composite beam damage early warning and positioning method

Abstract

The invention discloses a steel-concrete composite beam damage early warning and positioning method, and belongs to the technical field of composite beam damage intelligent identification. A reduced-scale steel-concrete composite beam is prepared, a web plate is a corrugated steel web plate, and a bridge deck adopts a laminated structure; the reduced-scale steel-concrete composite beam is horizontally arranged, jacks are loaded at two points from top to bottom in the midspan, and no less than three acoustic emission probes are arranged on each of two opposite surfaces of the bridge deck; in the test loading process, an acoustic emission probe is used for receiving an acoustic signal to determine the damage position of the bridge deck slab, and the concrete cracking damage type is judged based on the RA-AF value of the acoustic emission signal; according to the technical scheme, the steel-concrete composite beam damage early warning and positioning method based on the acoustic emission technology effectively overcomes the defect of inaccurate damage positioning in the prior art; through combination of an acoustic emission technology and a positioning algorithm, the damage position can be accurately positioned, and a reliable basis is provided for subsequent structure damage early warning.

Description

Technical Field

[0001] This invention belongs to the field of intelligent detection technology for composite beam damage, specifically relating to a method for early warning and location of damage in steel-concrete composite beams based on acoustic emission technology. Background Technology

[0002] Steel-concrete composite beams offer advantages such as convenient component connection, high construction efficiency, strong quality control, and low labor costs, making them promising candidates for application in precast component construction. Due to their high strength, high durability, and excellent synergistic properties, steel-concrete composite beams are widely used in modern bridges and long-span structures. However, during service, due to differences in material properties, the complexity of the interface transition zone, and long-term load effects, composite beam structures are prone to damage and hidden cracks at the steel-concrete interface and NC-UHPC interface. This damage not only reduces the overall stiffness and load-bearing capacity of the structure but may also lead to premature failure, seriously threatening engineering safety and service life.

[0003] Damage to steel-concrete composite beams often manifests as a complex interplay of multiple mechanisms, including tensile-controlled cracks, shear cracks, mixed-type failure, and interlaminar delamination. The damage evolution path is highly insidious; microcracks frequently initiate and slowly propagate within the beam structure and at interfaces, making them difficult to detect through traditional visual inspections or surface measurements. Particularly in the multi-interface region of steel-concrete, factors such as localized stress concentration and material discontinuities lead to an extremely complex damage development process, exhibiting characteristics such as early weak signals, interfacial microcracks, and rapid instability.

[0004] Currently, acoustic emission (AE) technology, as an important means in the field of non-destructive testing, has been applied to the health monitoring of concrete and steel structures. AE monitoring can capture the acoustic signals of the initiation and propagation of micro-cracks during the damage process, providing technical possibilities for early damage warning. However, for the actual engineering needs of steel-concrete composite beams, the existing acoustic emission methods still have the following key technical difficulties: (1) Insufficient spatial positioning accuracy, making it difficult to accurately invert interface damage; (2) Insufficient intelligent identification of damage types, with strong subjective dependence on judgment; (3) Complex AE signal patterns, large noise impact, and high difficulty in data interpretation; (4) Existing machine learning methods are singular, and the identification and evaluation lack a closed loop. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for damage early warning and location of steel-concrete composite beams. This method is designed for steel-concrete composite beams and has both high spatial resolution and high intelligent discrimination capability. It can overcome technical bottlenecks such as invisible interface and internal damage, strong subjectivity in type discrimination, and weak spatial tracking and dynamic evaluation capabilities. It can achieve high-precision three-dimensional location and type discrimination of internal structural damage.

[0006] To achieve the above objectives, the present invention is implemented using the following technical solution: This invention provides a method for damage early warning and location of steel-concrete composite beams, including the following steps: A scaled-down steel-concrete composite beam was prepared according to the principles of stiffness equivalence and unchanged relative position of the neutral axis. The web is a corrugated steel web and the bridge deck adopts a laminated structure. The scaled-down steel-concrete composite beam is laid horizontally and jacks are applied from top to bottom at two points in the middle of the span. At least three acoustic emission probes are arranged on each of the two opposite surfaces of the bridge deck. During the test loading process, acoustic emission probes were used to receive acoustic signals to determine the location of bridge deck damage, and the RA-AF value of the acoustic emission signal was used to determine the type of concrete cracking damage. Here, RA is the ratio of rise time to amplitude, and AF is the ratio of signal ring count to duration.

[0007] Furthermore, during the test loading, a preload force of 30kN was first applied to the beam to check the safety of the test device and whether the sensors were working properly; After the inspection is completed, the formal loading will begin from 0 kN, and the sensor will take an initial reading; thereafter, each loading level will be 30 kN, and the load will be held for 5 minutes, during which the pressure sensor data connected to the jack will be collected. The load is gradually increased by pressure until the load force no longer increases, then the load is switched to displacement control, with each 2mm increment representing one level, until the specimen is completely destroyed.

[0008] Furthermore, determining the location of bridge deck damage using an acoustic emission probe to receive acoustic wave signals includes the following steps: During the test loading process, the raw acoustic emission signals of each channel were collected in real time, including energy, ring count, rise time, amplitude, and duration. The autoregressive Akaike Information Criterion (AR-AIC) algorithm is used to determine the arrival time of the original acoustic emission signal in order to accurately identify the first arrival wave of the event. By utilizing the arrival times of each channel and the known coordinates of the pressure sensor, the actual spatial location of the acoustic emission signal is determined through iterative solution.

[0009] Furthermore, for n Point waveform s ( t The AR-AIC algorithm fits two autoregressive models to windows before and after the change, and minimizes the Akaike information criterion. The time corresponding to the minimum point of the AIC function curve is the arrival time of the first arrival wave. ; in, k It is a model order. LLog-likelihood, minimum value represents pressure sensor i The initial value of the longitudinal wave at that location.

[0010] Furthermore, using the nonlinear least squares method, starting from the assumed initial sound source location, the residual between the theoretical arrival time and the actual arrival time is calculated, and the position coordinates and emission time are compared. x s , y s , z s , t s Perform iterative optimization until the residual converges, as follows: ; In the formula, t i,c For the first i The arrival time recorded by each channel. v p Uniform longitudinal wave velocity ( x i , y i , z i ) represents the location of the pressure sensor. x s , y s , z s (This is an unknown source.) t s The origin time; Linearization is performed and iterative generation of update equations is performed, where the model vector m = ( x s , y s , z s , ts ) and sensitivity matrix G=∂t cal / ∂m; Convergence is typically achieved within five iterations, obtaining the sensor spacing accuracy; ; When the TOA residual reaches its minimum value, the estimated coordinates are obtained. x s , y s , z s This is the actual spatial location of the acoustic emission signal.

[0011] Furthermore, the RA-AF value of the acoustic emission signal is used to determine the type of concrete cracking damage, including the following steps: Key characteristic parameters, including RA value, AF value, energy, and duration, are extracted from acoustic emission monitoring signals, and dimensional differences between different parameters are eliminated through standardization or normalization methods. An unsupervised learning algorithm was used to automatically group the normalized AE parameters, identify potential patterns and feature clusters in the data, and construct a preliminary damage type classification model framework based on the clustering results. By introducing sample data of known damage types as supervision information, and using supervised learning algorithms to train and optimize the preliminary classification model, the accurate identification of unknown damage sample types can be achieved. By associating and mapping damage with the spatial location information of acoustic emission sensors, the distribution patterns of different damage types in the structure can be intuitively displayed through three-dimensional visualization technology, thereby achieving spatial visualization representation of damage state.

[0012] Furthermore, for each AE event, the rise time, amplitude, ring count, and duration are extracted using the AE function; RA Defined as the ratio of rise time to amplitude, expressed as: ; Will AF The ratio of ring count to duration is defined as: ; All AE event data are organized into a two-dimensional feature vector. RA i , AF i ], forming an unlabeled sample set, represented as: ; in, n This represents the total number of AE events. To eliminate the influence of different dimensions of features and improve the stability of the clustering algorithm and the physical meaning of the cluster centers, a normalization method is applied to all features, expressed as: ; in, RA nori and AF nori Characterize the normalized treatment respectively RA Value and AF value.

[0013] Furthermore, on the RA-AF plane, the K-means++ clustering method is used to minimize the intra-cluster variance, distinguishing between high-frequency "stretching-controlled" and low-frequency "shearing-controlled" events. The horizontal boundary line corresponding to the RA-AF feature space is represented as follows: ; in, J This represents the objective function value for clustering, i.e., the total sum of squared errors within each cluster. C c For the first 𝑐 The set of all sample points contained in a cluster; x j For the first j Each sample point consists of two acoustic emission features. x j = ( RA j , AF j ); μ c For the first c The center point of each cluster, that is, all the points within that cluster. 𝑗 The average value; And, the clustering update center at each step: ; N k For the first k Number of samples in each class; And, the Euclidean distance from each feature point to the center is: ; Categories are assigned based on the principle of proximity, and the center is continuously updated iteratively until the center is stable or the termination condition is met. The GMMSEQ Gaussian mixture model is employed, incorporating event-normalized time information to fit steep vertical boundaries and accurately identify high RA shear signals. It is assumed that each AE event feature vector... x j =[ RA nor,j , AF nor,j , τ j The GMM model assumes that the overall distribution is a mixture of K Gaussian distributions: ; in: K This represents the number of Gaussian distributions, which is typically taken as 2 to 3 in practical applications. π k For the first k Mixed weights of classes, satisfying ; N ( x j | μ k , Σk The distribution is a multivariate Gaussian distribution with parameters being the mean. μ k The covariance matrix Σ k ; The formula for the multivariate Gaussian distribution is expressed as: ; The GMM parameters are achieved by maximizing the log-likelihood, and are solved using the EM algorithm. In E Step / Expected Step, calculate the number of steps each data point belongs to. k Posterior probabilities of a Gaussian distribution , is represented as: ; in, x j For the first j Each sample point consists of two acoustic emission features. x j =( RA j , AF j ); K This represents the total number of mixture components in the Gaussian mixture model; Let be the mixing weight of the nth Gaussian component, representing the proportion of that component in the overall distribution; Let be the Gaussian distribution density function, representing the sample points. x j In μ k For the mean, Σ k The first covariance k The probability density values ​​under a normal distribution; μ k for No. k The mean vector (feature center) of a Gaussian distribution; Σ k For the first k The mean vector (feature center) of a Gaussian distribution; For all Gaussian components, pair the sample points 𝑥 𝑗 The combined effect, normalized term; In the M-step / maximization step, the distribution parameters are updated as follows: ; ; ; ; Then iterate through the E-step and M-step until the parameters converge or the log-likelihood change is below the threshold.

[0014] Furthermore, the consensus events of K-means++ and GMMSEQ clustering results are used as the training set, and a globally optimal hyperplane is trained through a linear support vector machine to achieve the automatic classification of all events into "stretching", "shearing", and "mixed" categories without thresholding. The basic hyperplane equation for SVM classification is defined as follows: ; in, V Let be the normal vector of the hyperplane, representing the classification direction; b This is an offset term used to adjust the position of the hyperplane; H The input features are the characteristics of the sample data. Determine the boundary: ; in, x sv Support vectors are sample points located on the classification boundary; y sv The class label represents the support vector; this formula indicates that the support vector falls exactly on the classification boundary. The constrained optimization problem is: ; ; in, x sv These are support vectors, which are sample points located on the classification boundary; y sv The class label represents the support vector; this formula indicates that the support vector falls exactly on the classification boundary. Solving using the Lagrange multiplier method yields: ; in, δ i These are Lagrange multipliers, corresponding to the "importance weights" of each constraint, used to construct the dual problem; V The optimal normal vector can be represented by a weighted sum of all support vectors; 0 < δ i < C Ensure that the model satisfies the KKT optimality conditions, especially the equality constraints and equilibrium constraints; The minimization problem becomes: ; in, δ i , δ j For Lagrange multipliers; yi , y j For the first i , j The labels of each sample; x i , x j For the first i , j Feature vectors of each sample; The inner product between input features; satisfy: ; in, δ i For Lagrange multipliers; C This is a penalty factor (hyperparameter) that controls the tolerance for misclassification; y i ( x i · V + b The output of the classification function represents the distance between the sample and the boundary hyperplane. The optimal hyperplane segmentation is obtained by iteratively solving the problem using the SMO optimization algorithm. .

[0015] Furthermore, within the RA-AF space, events located above the K-means boundary and to the left of the GMM boundary are classified as stretching events, events located below the K-means boundary and to the right of the GMM boundary are classified as shearing events, and the rest are classified as mixed events. Mapping all event type labels to a three-dimensional structural space visually displays cracks, including main cracks, shear cracks, tensile cracks, and mixed cracks, as well as their dynamic evolution trajectories, obtaining a complete data chain and intuitive diagrams reflecting the health monitoring and quantitative assessment of damage in engineering structures.

[0016] Compared with existing technologies, the beneficial effects achieved by this invention are as follows: The damage early warning and location method for steel-concrete composite beams provided by this invention effectively solves the problems of inaccurate damage location and reliance on manual experience in traditional methods, enabling high-precision three-dimensional location of internal structural damage; it innovatively combines unsupervised clustering and supervised classification algorithms, improving the accuracy and intelligence level of automatic damage type identification, and can distinguish between tensile damage, shear damage, and mixed damage; the process is fully automated, requiring no manual setting of thresholds and discrimination boundaries, adapting to different working conditions and structural types, and possessing good versatility and robustness; it can achieve real-time tracking and dynamic evaluation of the damage development process, providing a scientific and reliable data foundation and decision support for daily monitoring, life prediction, and subsequent maintenance of structures; this method is particularly suitable for damage monitoring of steel-concrete composite structures in practical engineering, ensuring engineering safety, and has broad engineering application prospects and promotional value. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the process of the present invention.

[0018] Figure 2 This is a schematic diagram illustrating the SVM principle of the present invention.

[0019] Figure 3 This example demonstrates the accuracy of the model testing process based on AE data.

[0020] Figure 4 The image shows the RA-AF clustering results of the steel-concrete composite beam in the example, obtained through SVM clustering analysis.

[0021] Figure 5 This is a schematic diagram of the damage to the steel-concrete composite beam in the embodiment.

[0022] Figure 6 This is a three-dimensional view of the composite beam under load in the embodiment.

[0023] Figure 7 This is a schematic diagram of acoustic emission waveform acquisition in an embodiment.

[0024] Figure 8 This is a schematic diagram of type labels mapped to a three-dimensional structural space in the embodiment. Detailed Implementation

[0025] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0026] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are used only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0027] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0028] like Figure 1 As shown in the figure, this embodiment of the invention provides a method for damage early warning and location of steel-concrete composite beams based on acoustic emission technology. The specific steps are as follows: Step S1) Using a 45m real bridge as a scaled-down prototype, and following the principles of stiffness equivalence and unchanged relative position of the neutral axis, the main dimensions (beam height, slab width, etc.) are scaled down by 1 / 5 to prepare a scaled-down steel-concrete composite beam. The web is a corrugated steel web, and the bridge deck adopts a laminated structure, with the lower layer being C50 concrete and the upper layer being UHPC.

[0029] like Figure 6 As shown, a scaled-down steel-concrete composite beam is used, with jacks applied from top to bottom at two points in the middle of the span. Acoustic emission probes receive acoustic signals, and no fewer than three acoustic emission probes are arranged on each of the two opposite surfaces of the bridge deck.

[0030] In this embodiment, the experimental loading process includes preloading and formal loading.

[0031] During the preloading process, the test procedure and loading control method are as follows: first, a preloading force of 30kN is applied to the beam to check the safety of the test device and whether the sensors are working properly.

[0032] During the formal loading process, the loading started at 0 kN, and the sensor took initial readings. Afterwards, each loading level was 30 kN, held for 5 minutes, and sensor data was collected. Loading was gradually increased according to load control until the load force no longer increased, then displacement control was switched to 2 mm increments until the specimen was completely destroyed.

[0033] Step S2) Acoustic emission (AE) technology, as an economical and efficient non-destructive testing technique, is commonly used to assess the degree of corrosion and damage in civil engineering facilities (bridges, dams, etc.). It can monitor the transient elastic waves generated by materials throughout the entire process from the initial crack formation to final structural failure, thereby reflecting the initiation and propagation of cracks within the material. The main advantage of AE technology is its high sensitivity to minor structural damage, enabling long-term real-time monitoring of structures. The acoustic emission monitoring system is activated to record the acoustic emission signals during the bending process of the steel-concrete composite beam in real time. The AR-AIC (Autoregressive Akaike Information Criterion) algorithm is used to locate the acoustic emission signals. Specific steps include acoustic emission waveform acquisition, acoustic emission signal arrival time determination, and acoustic emission signal localization.

[0034] Step S2.1) Acoustic emission waveform acquisition. During the experimental loading process, the raw acoustic emission signals of each channel were acquired in real time, including basic characteristic parameters such as energy, ring count, rise time, amplitude, and duration. Figure 7 As shown.

[0035] Step S2.2) Determine the arrival time of the acoustic emission signal.

[0036] The AR-AIC (Autoregressive Akaike Information Criterion) algorithm is used to determine the arrival time of the original signal, enabling accurate identification of the first arrival wave and providing high-precision time input for subsequent positioning algorithms. n Point waveform s ( t The algorithm fits two autoregressive models to windows before and after the change, and minimizes the Akaike information criterion. The minimum point of the AIC function curve is found, and the corresponding time is the arrival time of the first arrival wave.

[0037] ; in k It is a model order. L It is the log-likelihood. The minimum value represents the sensor's... i The initial value of the longitudinal wave at that location.

[0038] Step S2.3) Sound emission signal localization.

[0039] Using the arrival times of each channel and the known sensor coordinates, the three-dimensional spatial location of the acoustic emission event is determined through iterative solution. Let (x i , y i , z i ) represents the sensor position, ( x s , y s , z s (This is an unknown source.) t s The time interval is the origin. For uniform longitudinal wave velocity... v p , t i,c For the first i The arrival times are recorded for each channel. Using the nonlinear least squares method, starting from the assumed initial sound source location, the residual between the theoretical and actual arrival times is calculated, and this residual is then compared with the position coordinates and transmission time. x s , y s , z s , t s Iterative optimization is performed until the residual converges.

[0040] ; Linearization and iteration produce the following update equation, where the model vector m = ( x s , y s , z s , ts ) and sensitivity matrix G=∂t cal / ∂m. Convergence is typically achieved within five iterations, thus providing accuracy in sub-sensor spacing.

[0041] ; When the TOA residual reaches its minimum value, the estimated coordinates are ( x s , y s , z s This is the actual spatial location of the acoustic emission signal.

[0042] Step S3) Acoustic emission characteristic parameters are highly sensitive to the fracture mode of concrete. The distribution characteristics of RA-AF are often used to interpret the cracking modes at different stages of concrete fracture. The RA value represents the ratio of rise time to amplitude, measured in s / V, and the AF value represents the ratio of ring count to duration, measured in kHz. It is a key factor in crack classification. Acoustic emission signals with low AF values ​​and high RA values ​​indicate shear cracks, while signals associated with tensile cracks show the opposite.

[0043] Step S3.1) Acoustic emission parameter extraction and normalization.

[0044] Acoustic emission signals were collected during the loading process of a steel-concrete composite beam structure using an acoustic emission system. For each acoustic emission event, basic parameters such as rise time, amplitude, ring count, and duration were extracted using the acoustic emission system. RA Defined as the ratio of rise time to amplitude, i.e.: ; AF Define the ratio of the ring count to the duration, i.e.: ; All acoustic emission event data are organized into a two-dimensional feature vector. RA i , AF i This constitutes an unlabeled sample set; ; in, n This represents the total number of acoustic emission events.

[0045] To eliminate the influence of different features' dimensions and improve the stability of the clustering algorithm and the physical meaning of the cluster centers, a normalization method is applied to all features: ; in, RA nori and AF nori Characterize the normalized treatment respectively RA Value and AF value.

[0046] Step S3.2) Construction of unsupervised clustering and classification models.

[0047] On the RA-AF plane, the K-means++ clustering method is used to minimize the intra-cluster variance, which can usually distinguish between high-frequency "stretching master" events and low-frequency "shearing master" events, corresponding to the horizontal boundary of the RA-AF feature space.

[0048] ; in, J This represents the objective function value for clustering, i.e., the total sum of squared errors within each cluster. C c For the first 𝑐 The set of all sample points contained in a cluster; x j For the first j Each sample point consists of two acoustic emission features. ; μ c For the first c The center point of each cluster (i.e., all nodes in that cluster) 𝑗 (average).

[0049] Clustering update center at each step: ; N k For the first k Number of samples in each class.

[0050] The Euclidean distance from each feature point to the center is: ; Categories are assigned based on the principle of proximity, and the center is continuously updated iteratively until the center is stable or the termination condition is met.

[0051] A Gaussian mixture model (GMMSEQ) is employed, incorporating event-normalized time information to fit steep vertical boundaries, accurately identifying high-RA shear signals. This is achieved by assuming a feature vector for each acoustic emission event. x j =[ RA nor,j , AF nor,j , τ j The GMM model assumes that the overall distribution is a mixture of K Gaussian distributions: ; in: K The number of Gaussian distributions (usually taken as 2 to 3 in practical applications); π k For the first k Mixed weights of classes, satisfying ; N ( x j | μ k , Σ k The distribution is a multivariate Gaussian distribution with parameters being the mean. μk The covariance matrix Σ k .

[0052] The formula for the multivariate Gaussian distribution is: ; The GMM parameters are achieved by maximizing the log-likelihood, which is solved using the EM (Expectation-Maximization) algorithm. exist E Step (Expected Step): Calculate which data point belongs to the first step. k Posterior probabilities ("soft classification") from a Gaussian distribution: ; in, x j For the first j Each sample point consists of two acoustic emission features. ; K This represents the total number of mixture components in the Gaussian mixture model; Let be the mixing weight of the nth Gaussian component, representing the proportion of that component in the overall distribution; Let be the Gaussian distribution density function, representing the sample points. x j In μ k For the mean, Σ k The first covariance k The probability density values ​​under a normal distribution; μ k for No. k The mean vector (feature center) of a Gaussian distribution; Σ k For the first k The mean vector (feature center) of a Gaussian distribution; For all Gaussian components, pair the sample points 𝑥 𝑗 The combined effects (normalized term).

[0053] M-step (maximization step): Updates the parameters of each distribution, expressed as follows: ; ; ; ; Iterate through the E-step and M-step until the parameters converge or the log-likelihood change is below the threshold.

[0054] Step S3.3) Supervise classification and type identification.

[0055] The consensus events from both clustering results are used as the training set. A globally optimal hyperplane is trained using a linear support vector machine (SVM), enabling the automatic, threshold-free classification of all events into "stretched," "sheared," and "mixed" classes. A schematic diagram of the SVM principle is shown below. Figure 2 The hyperplane for classification is defined as: ; in V Let be the normal vector of the hyperplane, representing the classification direction; b This is an offset term used to adjust the position of the hyperplane; H The input features are the features of the sample data.

[0056] Determine the boundary: ; in x sv Support vectors are sample points located on the classification boundary; y sv The expression represents the class label of the support vector; it indicates that the support vector falls exactly on the classification boundary.

[0057] The constrained optimization problem is: ; ; in x sv These are support vectors, which are sample points located on the classification boundary; y sv The expression represents the class label of the support vector; it indicates that the support vector falls exactly on the classification boundary.

[0058] Solving using the Lagrange multiplier method yields: ; in These are Lagrange multipliers, corresponding to the "importance weights" of each constraint, used to construct the dual problem; V The optimal normal vector can be represented by a weighted sum of all support vectors; Ensure that the model satisfies the KKT optimality conditions, especially the equality constraints and equilibrium constraints.

[0059] The minimization problem becomes: ; in , For Lagrange multipliers; y i , y j For the first i ,j The labels of each sample; x i , x j For the first i , j Feature vectors of each sample; This is the inner product between the input features.

[0060] satisfy: ; in For Lagrange multipliers; C This is a penalty factor (hyperparameter) that controls the tolerance for misclassification; The output of the classification function represents the distance between the sample and the boundary hyperplane.

[0061] The optimal hyperplane segmentation is obtained by iteratively solving the problem using optimization algorithms such as SMO. AF = w 1 RA + b。

[0062] Step S3.4) Spatial distribution of types and visualization of damage.

[0063] Within the RA-AF space, events located above the K-means boundary and to the left of the GMM boundary are classified as stretching events, events located below the K-means boundary and to the right of the GMM boundary are classified as shearing events, and the rest are classified as mixed events.

[0064] By mapping all event type labels to a three-dimensional structural space, cracks (main cracks, shear cracks, tension cracks, and mixed cracks) and their dynamic evolution trajectories are displayed intuitively, providing a complete data chain and intuitive diagrams for the health monitoring and quantitative assessment of damage in engineering structures.

[0065] Step S4) In this step, taking the example of the embodiment, two clustering algorithms, K-means++ and GMMSEQ, are used to extract horizontal and vertical dividing lines in the RA-AF feature plane, respectively. K-means++ can effectively identify frequency-dominated stretching events, while GMMSEQ introduces the event occurrence order (normalized time) to accurately distinguish shearing events with high RA. Next, the "consensus events" jointly identified by the two clustering methods are used as the training set to construct a linear support vector machine (SVM) classification model.

[0066] During the SVM training phase, 70% of the samples were randomly selected from the constructed labeled AE database as training data, and the remainder was used as the test set. To further verify the model's stability and generalization ability, 20 test subsets (each containing 20% ​​of the database's samples) were randomly selected from the database for independent testing. The results show that the SVM model achieves a classification accuracy exceeding 99.9% for AE events. Figure 3 As shown, the three types of events (stretching, shearing, and mixing) are clearly distributed in the RA-AF feature space, with well-defined classification boundaries, as follows. Figure 4 As shown. Finally, the identification results are mapped onto a 3D structural model to visualize the crack type and location, as shown. Figure 5 As shown, this provides an efficient and reliable intelligent method for damage mode recognition and structural safety assessment of steel-concrete composite beams.

[0067] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for damage early warning and positioning of a steel-concrete composite beam, characterized in that, The method comprises the steps of: A scaled steel-concrete composite beam is prepared according to the principles of stiffness equivalence and invariable relative position of neutral axis, the web is a corrugated steel web, and the deck slab adopts a laminated structure; The scaled steel-concrete composite beam is vertically placed and loaded by two upward-pointing jacks in the middle of the span, and each surface of the two opposite surfaces of the deck slab is arranged with no less than three acoustic emission probes; In the process of test loading, the damage position of the deck slab is determined by using the acoustic emission probes to receive acoustic wave signals, and the RA-AF value of the acoustic emission signals is used to identify the concrete cracking damage type, wherein RA is the ratio of rise time to amplitude, and AF is the ratio of signal ring count to duration.

2. The method of claim 1, wherein, During the test loading, a pre-load of 30kN is first applied to the beam to check the safety of the test device and whether the sensors are working normally; After the completion of the check, the formal loading starts from 0kN, and the sensors perform initial reading; then, every 30kN is a loading level, and the pressure sensor data connected to the jacks are collected for 5 minutes; After the pressure load control is used for step-by-step loading until the loading force no longer rises, displacement control is used, every 2mm is a level, and the test piece is completely destroyed.

3. The method of claim 2, wherein the method further comprises: The method for determining the damage position of the deck slab by using the acoustic emission probes to receive acoustic wave signals comprises the steps of: In the process of test loading, the original acoustic emission signals of each channel are collected in real time, including energy, ring count, rise time, amplitude and duration; The autoregressive Akaike information criterion (AR-AIC) algorithm is used to determine the arrival time of the original acoustic emission signals to accurately identify the event first arrival wave; The actual spatial position of the acoustic emission signals is determined by using the arrival time of each channel and the known pressure sensor coordinates through iterative solution.

4. The method of claim 3, wherein the method further comprises: For n point waveform s ( t ), the AR-AIC algorithm fits two autoregressive models to the pre- and post-change windows and minimizes the Akaike information criterion, obtaining the time corresponding to the minimum point of the AIC function curve as the arrival time of the first arrival wave: ； wherein, k is the model order, L is the log-likelihood, the minimum value indicating the P-wave onset value at the pressure sensor i .

5. The method of claim 4, wherein the method further comprises: Using the nonlinear least squares method, starting from the assumed initial sound source location, the residual between the theoretical arrival time and the actual arrival time is calculated, and the position coordinates and emission time are compared. x s , y s , z s , t s Perform iterative optimization until the residual converges, as follows: ； wherein t i,c is the arrival time recorded for the i p v p is the uniform P-wave velocity x i , y i , z i is the pressure sensor location, x s , y s , z s is the unknown source, t s is the origin time; Linearization and iteration are performed to produce an update equation where the model vector m=( x s , y s , z s , ts ) and the sensitivity matrix G = ∂t cal / ∂m; convergence is typically achieved within five iterations, obtaining sensor spacing accuracy; ； When the TOA residual takes the minimum value, the estimated coordinates (x, y, z) are obtained x s , y s , z s ) are the actual spatial position of the acoustic emission signal.

6. The method of claim 1 or 5, wherein, The method for identifying the concrete cracking damage type based on the RA-AF value of the acoustic emission signals comprises the steps of: Key feature parameters are extracted from the acoustic emission monitoring signals, including RA value, AF value, energy, duration, and the dimension differences between different parameters are eliminated through standardization or normalization methods; An unsupervised learning algorithm is used to automatically group the normalized AE parameters, identify the potential patterns and feature clusters in the data, and construct a preliminary damage type classification model framework based on the clustering results; Sample data of known damage types are introduced as supervision information, and a supervised learning algorithm is used to train and optimize the preliminary classification model to accurately identify unknown damage sample types; The spatial position information of the damage and the acoustic emission sensors is associated and mapped, and the distribution law of different damage types in the structure is intuitively displayed through three-dimensional visualization technology to realize the spatial visualization representation of the damage state.

7. The method of claim 6, wherein the method further comprises: For each AE event, the AE extraction rise time, amplitude, ring count, and duration are utilized; the ratio of rise time to amplitude is defined as the AE figure of merit and is expressed as: RA defined as the ratio of rise time to amplitude, is expressed as: ； The AF The ratio of ring count to duration is defined as: ； Data of all AE events are collated into two-dimensional feature vectors [ RA i , AF i ] to form an unlabeled sample set, denoted as: ； wherein, n is the total number of AE events; To eliminate the dimension influence of different features and improve the stability of the clustering algorithm and the physical meaning of the clustering center, all features are processed by a normalization method and represented as: ； wherein, RA nori and AF nori respectively represent the normalized values of the RA values and AF values.

8. The method of claim 7, wherein the method further comprises: On the RA-AF plane, the K-means++ clustering method is used to minimize the intra-cluster variance to distinguish high-frequency "stretching main control" and low-frequency "shearing main control" events, and the horizontal dividing line corresponding to the RA-AF feature space is represented as: ； wherein, J is the objective function value of the cluster, i.e. the total within-cluster sum of squared errors; C c is the set of all sample points contained in the 𝑐 th cluster; x j is the j th sample point, consisting of two acoustic emission features, x j = RA j , AF j ; μ c is the center point of the c th cluster, i.e. the mean of all 𝑗 x​ In each step, the center is updated as: ； N k For the first k Class sample number; In each step, the Euclidean distance of each feature point to the center is: ； The centers are iteratively updated until convergence or a termination condition is met; GMMSEQ Gaussian Mixture Model, the event normalized time information is introduced to fit the steep vertical boundary and accurately identify the high RA shear signal; assuming that each AE event feature vector x j [ RA nor,j , AF nor,j , τ j ]GMM model assumes that the overall distribution is a mixture of K Gaussian distributions: ； wherein: K is the number of Gaussian distributions, usually taken as 2~3 in practical applications; π k is the weight of the k th class, satisfying ; N x j is the number of classes, and μ k is the number of samples in the k th class; and μ k is the multivariate Gaussian distribution, with parameters being the mean k and the covariance matrix; The multivariate Gaussian distribution is expressed as: ； The GMM parameters are obtained by maximizing the log-likelihood using the EM algorithm: In the E Step / Expected Step, compute the posterior probability that each data point belongs to the k ith Gaussian distribution , denoted as: ； in, x j For the first j Each sample point consists of two acoustic emission features. x j =( RA j , AF j ); K This represents the total number of mixture components in the Gaussian mixture model; Let be the mixing weight of the nth Gaussian component, representing the proportion of that component in the overall distribution; Let be the Gaussian distribution density function, representing the sample points. x j In μ k For the mean, Σ k The first covariance k The probability density values ​​under a normal distribution; μ k In the M step, the parameters of each distribution are updated as: No. k The mean vector (feature center) of a Gaussian distribution; Σ k For the first k The mean vector (feature center) of a Gaussian distribution; For all Gaussian components, pair the sample points 𝑥 𝑗 The combined effect, normalized term; Then the E step and M step are iterated until the parameters converge or the change in log-likelihood is below a threshold. ； ； ； ； The consensus events of K-means++ and GMMSEQ clustering results are taken as the training set, and the global optimal hyperplane is trained by linear support vector machine to automatically classify all events into "stretching class", "shearing class", and "mixed class" without threshold.

9. The method of claim 8, wherein, The basic hyperplane equation of SVM classification is defined as: The discriminant boundary is: ； wherein, V is a normal vector of the hyperplane, indicating the classification direction; b is a bias term, used to adjust the position of the hyperplane; H is an input feature, i.e., a feature of the sample data; The constrained optimization problem is: ； wherein, x sv support vectors, i.e. sample points that lie on the classification boundary; y sv a class label representing the support vector; this equation indicates that the support vector falls exactly on the classification boundary; The Lagrange multiplier method is used to solve the problem: ； ； wherein, x sv to support vectors, i.e. sample points lying on the classification boundary; y sv a class label representing the support vector; this equation indicates that the support vector falls exactly on the classification boundary; δ ； where, δ i are Lagrange multipliers, "importance weights" corresponding to each constraint, used to construct the dual problem; V denotes the optimal normal vector, which can be expressed by the weighted sum of all support vectors; 0 < a < 1 The minimization problem becomes: i < C ensure that the model satisfies the KKT optimality conditions, in particular the equality constraints and the balanced limit condition; δ ； wherein, δ i , Satisfies: j is a Lagrange multiplier; y i , y j is the label of the i , j th sample; x i , x j is the feature vector of the i , j th sample; is the inner product between input features; δ ； where, In the RA-AF space, the events above the K-means boundary and left of the GMM boundary are determined as stretching type, the events below the K-means boundary and right of the GMM boundary are determined as shearing type, and the rest are classified as mixed type. i is the Lagrange multiplier; C is the penalty factor (hyperparameter) that controls the tolerance to misclassification; y i ( x i · V + b ) is the classification function output, representing the distance of the sample to the separating hyperplane; The SMO optimization algorithm is used for iterative solution to obtain the optimal hyperplane partition: .

10. The method of claim 9, wherein the method further comprises: The type labels of all events are mapped to the three-dimensional structure space to visually display the cracks, including the main crack, shear, stretch, and mixed, as well as their dynamic evolution trajectories, to obtain a complete data chain and intuitive graphical solution reflecting the health monitoring and quantitative damage assessment of engineering structures. ​

Images