Concrete structure damage intelligent identification and evaluation method and system based on acoustic emission waveform

Abstract

The invention is suitable for the technical field of concrete structure health monitoring, and provides an acoustic emission waveform-based concrete structure damage intelligent identification and evaluation method and system. The method comprises the steps of acoustic emission signal collection and microscomic damage synchronous monitoring, waveform feature extraction based on wavelet packet decomposition, intelligent clustering recognition of microscomic damage types and multi-parameter fusion damage evaluation model construction. According to the method, through multi-scale sensing arrangement, wavelet packet decomposition, a variational Bayesian Gaussian mixture model clustering algorithm and a multi-parameter fusion model, high-precision identification and quantitative evaluation of concrete structure damage are realized; the method has high recognition accuracy in complex structures such as steel fiber reinforced concrete and prestressed concrete, generalization ability is improved, and efficient and reliable technical support is provided for safe operation and maintenance of concrete structures.

Description

Technical Field

[0001] This invention relates to the field of concrete structure health monitoring technology, specifically to a method and system for intelligent identification and assessment of concrete structure damage based on acoustic emission waveforms. Background Technology

[0002] Under long-term loads, environmental erosion (such as freeze-thaw cycles and chemical corrosion) and material deterioration, concrete structures are prone to microscopic damage such as interfacial transition zone (ITZ) cracking, aggregate breakage, fiber pull-out, and steel reinforcement corrosion. The accumulation of these damages will gradually lead to the degradation of macroscopic mechanical properties and even structural failure.

[0003] Acoustic emission technology, as a dynamic non-destructive testing method, can reflect the internal damage state of a structure in real time by capturing the elastic wave signals released during the damage evolution process. However, it has the following limitations in practical applications: (1) Insufficient accuracy of microscopic damage identification: Traditional methods rely on single parameter analysis such as the counting and energy of acoustic emission events, which makes it difficult to distinguish between ITZ damage cracking and damage coupled with fiber pull-out, resulting in low identification sensitivity in complex structures such as steel fiber reinforced concrete. (2) Single signal processing method: Time-frequency domain feature extraction mostly uses Fourier transform or single wavelet decomposition with fixed resolution, which cannot adapt to the local changes of signals of different damage types (such as the mixture of high-frequency friction signals and low-frequency cracking signals). Traditional noise reduction algorithms (such as threshold noise reduction) retain less than 50% of the effective high-frequency signals, resulting in feature distortion. (3) The generalization ability of intelligent algorithms is weak: existing deep learning models rely on a large amount of labeled data, and the recognition accuracy drops by 20%-30% when crossing structural types or working conditions, making it difficult to adapt to diverse monitoring scenarios in engineering. (4) The assessment system lacks quantitative correlation: the assessment of damage degree mostly relies on a single parameter, and a quantitative mapping relationship between micro-damage mechanism and macro-bearing capacity has not been established, which cannot directly support engineering maintenance decisions.

[0004] Therefore, in view of the above situation, there is an urgent need to provide a method and system for intelligent identification and assessment of concrete structure damage based on acoustic emission waveforms, so as to overcome the shortcomings in current practical applications. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for intelligent identification and assessment of concrete structure damage based on acoustic emission waveforms, effectively solving the problems mentioned in the background art.

[0006] This invention is implemented as follows: a method for intelligent identification and assessment of concrete structure damage based on acoustic emission waveforms, the method comprising the following steps: Step 1: Acoustic emission signal acquisition and microscopic damage monitoring simultaneously: Multi-scale acoustic emission sensor arrays are deployed at key parts of the concrete structure to acquire full waveform data of acoustic emission, and a non-contact digital image correlation (DIC) system is deployed to simultaneously acquire images of structural damage and cracking. Step 2: Waveform feature extraction based on wavelet packet decomposition: The acquired acoustic emission signal is decomposed into four layers of wavelet packets using the DB3 wavelet basis, and then dimensionality reduction is performed by combining kernel principal component analysis (KPCA) to extract multidimensional feature vectors. Step 3: Intelligent clustering identification of micro-damage types: Using the variational Bayesian Gaussian mixture model (VBGMM) clustering algorithm, the micro-damage types of concrete structures are classified, and the mapping relationship between clustering results and damage mechanisms is constructed. Step 4: Construction of a multi-parameter fusion damage assessment model: By fusing acoustic emission HI index, Sr value and b value, a quantitative index system that can correlate microscopic damage and macroscopic performance is established to assess the degree of damage to concrete structures.

[0007] As a further aspect of the present invention: In step 1, the deployment of the multi-scale acoustic emission sensor follows the JT / T1037-2022 standard, wherein a high-frequency sensor of 100-400kHz is used on the concrete surface, and a low-frequency sensor of 20-100kHz is used on the embedded prestressed steel bar.

[0008] As a further aspect of the present invention: Step 1 also includes a graded loading test, in which a four-point shear test is used for reinforced concrete structures and a splitting tensile test is used for fiber-reinforced concrete structures. The load is applied in stages according to 20%, 40%, 60%, 80% and 90% of the ultimate load F, with each loading stage lasting for 10 minutes. During the loading process, acoustic emission signals, load-displacement curves and DIC images are recorded simultaneously to establish the correspondence between "load level - damage characteristics - acoustic emission signal".

[0009] As a further aspect of the present invention: In step 2, the frequency band range of wavelet packet decomposition is 0-500kHz, wherein the damage signal of the interface transition zone (ITZ) between concrete aggregate and mortar is concentrated in 62.5-125kHz, the fiber pull-out signal is concentrated in 0-62.5kHz, the aggregate crushing signal is concentrated in 125-250kHz, and the steel bar damage signal is concentrated in 250-500kHz.

[0010] As a further aspect of the present invention: In step 3, the parameters of the VBGMM clustering algorithm are optimized by the Markov Chain Monte Carlo (MCMC) method. The optimized parameters include the number of iterations, the convergence threshold, and the prior distribution of the mixing coefficients, wherein the number of iterations is set to 5000 and the convergence threshold is set to 1e-6.

[0011] As a further aspect of the present invention: In step 4, based on the acoustic emission cumulative energy versus load curve, the concrete structure damage process is divided into three stages: Elastic phase: Cumulative energy less than 10 -3 J, This stage is mainly characterized by initial pore closure, with no obvious damage; Stable damage phase: Cumulative energy is at 10 -3 -10 -1 Between J, the damage is mainly dominated by ITZ and mortar damage, with an energy slope of less than 0.01 J / kN; Accelerated damage phase: Cumulative energy greater than 10 -1 J, dominated by aggregate crushing and steel / fiber damage, with high-energy signals accounting for over 80%.

[0012] As a further aspect of the present invention: In step 4, the formula for calculating the HI index is: ; in, For damage weight, This represents the percentage of such injury events; The Sr value represents the frequency of high-energy events occurring per unit time. The value of b is the logarithmic linear slope of the acoustic emission signal amplitude distribution, and its calculation formula is: ; in, The number of events with an amplitude greater than or equal to A. When the value of b is less than 1.0, it indicates that the damage has entered the accelerated phase.

[0013] As a further aspect of the present invention: In step 4, the damage assessment model sets a three-level early warning threshold based on the Sr value and the HI index: Mild damage: Log10(Sr) < 5.0 and HI < 0.77, corresponding to the early stage of stable damage, requiring increased monitoring frequency; Moderate damage: 5.0≤Log10(Sr)<6.0 and 0.77≤HI<3.40, corresponding to the late stage of stable damage, it is recommended to carry out local reinforcement work; Severe damage: Log10(Sr)≥6.0 and HI≥3.40, corresponding to the accelerated damage stage, requiring immediate reinforcement and repair.

[0014] This invention also provides an intelligent identification and assessment system for concrete structure damage based on acoustic emission waveforms, for implementing the above-mentioned method. The system includes: A multi-scale acoustic emission sensor array is used to acquire full waveform data of acoustic emission. Full waveform data acquisition equipment is used to store and perform preliminary processing on the acquired data; An intelligent identification model, which integrates the VBGMM algorithm, is used to identify the types of damage to concrete structures.

[0015] As a further aspect of the present invention: the intelligent recognition model uses ResNet18 as the basic network and combines the MAML algorithm to optimize the initial weight parameters, so as to reduce the model's requirement for the number of samples and improve the model's generalization ability under different structural types and working conditions.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) Significantly improved recognition accuracy: By optimizing the probability density estimation and parameter adaptation capability of VBGMM, it has a high recognition accuracy for ITZ, mortar, aggregate, and steel / fiber damage. In particular, the recognition sensitivity of "matrix cracking-fiber pull-out" coupled damage is improved compared with traditional algorithms.

[0017] (2) Strong generalization ability: VBGMM does not require a preset number of clusters. It handles data uncertainty through variational inference. The accuracy of recognition decreases less in cross-structure type tests. Combined with the ResNet18 model optimized by meta-learning (MAML algorithm), the sample requirement is reduced.

[0018] (3) The assessment system is quantitative and controllable: a mapping relationship between "microscopic damage ratio and macroscopic bearing capacity" is established, with a quantitative error of <5%. The three-level early warning threshold can directly guide engineering decisions and provide a scientific basis for the operation and maintenance of the structure throughout its entire life cycle.

[0019] (4) Wide applicability to engineering: The sensor deployment is compatible with the JT / T1037-2022 standard and is applicable to various structures such as steel fiber concrete and prestressed concrete. Attached Figure Description

[0020] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0022] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] The present invention will be further explained below with reference to specific embodiments.

[0024] Please see Figure 1 The present invention provides a method for intelligent identification and assessment of concrete structure damage based on acoustic emission waveforms, the method comprising: 1. Simultaneous acquisition of acoustic emission signals and monitoring of microscopic damage: (1) Multi-scale sensing arrangement: An 8-channel acoustic emission sensor array is arranged in key parts of the structure (such as tension zone and node zone). The sensor type is selected according to the JT / 1037-2022 standard: 100-400kHz high-frequency sensors are used for concrete surface, and 20-100kHz low-frequency sensors are used for embedded prestressed steel bars to ensure coverage of signal frequency bands for different damage types. The sensors are installed using vacuum grease coupling and metal cable ties. The protective layer is encapsulated with epoxy resin (thickness 3-5mm) to reduce environmental noise interference. The trigger threshold is set to 40dB and the sampling rate is 1MHz. A digital image correlation (DIC) system is deployed simultaneously to collect images of the displacement field and crack evolution of the structural surface, realizing the spatiotemporal correlation between acoustic emission signals and macroscopic cracks.

[0025] (2) Graded loading test design: Four-point shear test was used for reinforced concrete structure and splitting tensile test was used for fiber concrete. The load was applied in stages of 20%F, 40%F, 60%F, 80%F and 90%F (F is the ultimate load). Each stage was held for 10 minutes. Acoustic emission signal, load-displacement curve and DIC image were recorded simultaneously to establish the correspondence between "load level - damage characteristics - acoustic emission signal".

[0026] 2. Waveform feature extraction based on wavelet packet decomposition: (1) Optimal basis function and decomposition parameters: The acoustic emission waveform was decomposed into 4-layer wavelet packets using the DB3 wavelet basis (based on the minimum Shannon entropy criterion, with Shannon entropy value ≤ 0.3) to obtain the energy distribution of 16 frequency bands (0-500kHz). Among them, the ITZ damage signal is concentrated in 62.5-125kHz, the fiber pull-out signal is concentrated in 0-62.5kHz, the aggregate crushing signal is concentrated in 125-250kHz, and the steel bar damage signal is concentrated in 250-500kHz.

[0027] (2) Feature dimensionality reduction and enhancement: Kernel principal component analysis (KPCA, Gaussian kernel function γ=10) was used to reduce the 16-dimensional frequency band energy features to 4-5 dimensions, retaining more than 95% of the feature information. Mix-up data augmentation technology was used to linearly interpolate the acoustic emission waveforms of different damage types with weights of 0.3-0.7 to generate mixed samples (such as mixed waveforms of ITZ damage and fiber pull-out), thereby improving the model's ability to identify coupled damage.

[0028] 3. Intelligent clustering identification of detailed damage types: End-to-end damage clustering is achieved using a variational Bayesian Gaussian mixture model (VBGMM): (1) Model parameter optimization: The core parameters of VBGMM are optimized by Markov chain Monte Carlo (MCMC) method, including: 5000 iterations, convergence threshold 1e-6, and prior distribution of mixing coefficients, so that the model can adapt to the data distribution characteristics and automatically classify 4 types of damage signals (corresponding to ITZ, mortar, aggregate, and steel / fiber damage).

[0029] (2) Coupling damage processing: For coupling damage signals such as “matrix cracking - fiber pull-out”, the probability density estimation characteristics of VBGMM are used to calculate the posterior probability of the signal belonging to different damage categories. Combined with the frequency band energy ratio (e.g., when the energy ratio of 0-62.5kHz is >40%, it is judged as coupling damage), the accurate identification of coupling damage is achieved.

[0030] (3) Mechanism mapping verification: Combined with actual experiments, the VBGMM clustering results are spatially matched with the experimental damage area to establish a one-to-one mapping relationship between the clustering categories and the actual damage mechanisms, ensuring that the physical meaning of the classification results is clear.

[0031] 4. Multi-parameter fusion damage assessment model: (1) Damage stage division: Based on the acoustic emission cumulative energy and load curve, the damage process is divided into three stages: I. Elastic Stage: Cumulative Energy < 10 -3 J, mainly characterized by initial pore closure, with no obvious damage; II. Stable Damage Phase: Cumulative Energy 10 -3 -10 -1 J, ITZ and mortar damage are dominant, with an energy slope <0.01J / kN; III. Accelerated Damage Phase: Cumulative Energy > 10 -1 J, aggregate fragmentation and steel / fiber damage are dominant, high-energy signals (>10) -3 J) accounts for >80%.

[0032] (2) Quantitative index system: Microscopic damage index (HI): calculated based on the proportion of different cluster signals, the formula is: ; in Damage weight (reinforcement / fiber damage) oh =0.4, aggregate oh =0.3, mortar oh =0.2, ITZ oh =0.1), This represents the percentage of this type of damage event. Severity Rate (Sr value): defined as the percentage of high-energy events (>10) per unit time. - 4 The frequency of J's occurrence reflects the rate of damage development; b Value: The logarithmic linear slope of the acoustic emission signal amplitude distribution, expressed by the formula: ; Where N is the number of events with amplitude ≥ A, and a b value < 1.0 indicates that the damage has entered the accelerated phase.

[0033] (3) Three-level early warning mechanism: Mild injury: Log 10 (Sr)<5.0, HI<0.77, corresponding to the early stage of stable damage, the monitoring frequency needs to be increased; Moderate injury: 5.0 ≤ Log 10 (Sr)<6.0, 0.77≤HI<3.40, corresponding to the late stage of stable damage, local reinforcement is recommended; Severe injury: Log 10 (Sr)≥6.0, HI≥3.40, corresponding to the accelerated damage stage, requiring immediate reinforcement and repair.

[0034] The practical applications of the above technical solutions are as follows: Sensor Deployment and Data Acquisition: A prestressed concrete beam (5000×500×300mm) was selected, and 8-channel sensors (4 G150 high-frequency sensors and 4 low-frequency sensors) were deployed in the tension zone at mid-span, with a spacing of 100mm. The DIC system covered the entire beam, with a sampling frequency of 5Hz. Loading was applied in stages of 20%F (F=500kN), with each stage held for 10 minutes. Acoustic emission signals (sampling rate 1MHz), load-displacement curves, and DIC images were acquired simultaneously.

[0035] Data acquisition parameters: Sampling rate set to 1MHz, trigger threshold 40dB. Auxiliary information such as structural strain and load is recorded synchronously.

[0036] Signal preprocessing and feature extraction: A 5-layer soft thresholding denoising method using the sym8 wavelet basis (threshold 0.02V) was employed. The denoised waveform was then decomposed using a 4-layer DB3 wavelet basis to calculate the energy proportion of each of the 16 frequency bands. Dimensionality was reduced to 5 dimensions using KPCA (γ=10), and 2000 mixed samples were generated using Mix-up.

[0037] VBGMM Clustering and Damage Recognition: VBGMM parameters were set (5000 iterations, convergence threshold 1e-6), and four types of damage signals were obtained by clustering the dimensionality-reduced features. Verification using an actual experimental model showed that the matching rate between ITZ damage and simulated cracked areas was 92%, and the matching rate for aggregate breakage was 95%. Test set recognition results: ITZ damage 96%, fiber pull-out 92%, aggregate breakage 95%, and rebar damage 93%, with an average of 94%.

[0038] Damage assessment and early warning: When the load reaches 60% of the ultimate load, Log is detected. 10 (Sr)=5.2, HI=0.89, triggering a moderate damage warning.

[0039] In this embodiment, the present invention aims to address the problems of insufficient accuracy in identifying microscopic damage, limited signal processing methods, weak generalization ability of intelligent algorithms, and lack of quantitative correlation in the existing damage monitoring of concrete structures. It provides a method for intelligent damage identification and assessment based on acoustic emission waveforms. By optimizing the clustering performance of the variational Bayesian Gaussian mixture model (VBGMM), the accuracy of microscopic damage identification is improved, and a quantitative assessment system for the correlation between microscopic and macroscopic damage is established, providing technical support for the safe operation and maintenance of complex concrete structures.

[0040] This invention also provides an intelligent identification and assessment system for concrete structure damage based on acoustic emission waveforms, for implementing the above-described method. The system includes: A multi-scale acoustic emission sensor array is used to acquire full waveform data of acoustic emission. Full waveform data acquisition equipment is used to store and perform preliminary processing on the acquired data; An intelligent identification model, which integrates the VBGMM algorithm, is used to identify the types of damage to concrete structures.

[0041] The intelligent recognition model uses ResNet18 as the base network and combines the MAML algorithm to optimize the initial weight parameters, thereby reducing the model's requirement for a large number of samples and improving the model's generalization ability under different structural types and working conditions.

[0042] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent identification and assessment of concrete structure damage based on acoustic emission waveforms, characterized in that, The method includes the following steps: Step 1: Acoustic emission signal acquisition and microscopic damage monitoring simultaneously: Multi-scale acoustic emission sensor arrays are deployed at key parts of the concrete structure to acquire full waveform data of acoustic emission, and a non-contact digital image correlation (DIC) system is deployed to simultaneously acquire images of structural damage and cracking. Step 2: Waveform feature extraction based on wavelet packet decomposition: The acquired acoustic emission signal is decomposed into four layers of wavelet packets using the DB3 wavelet basis, and then dimensionality reduction is performed by combining kernel principal component analysis (KPCA) to extract multidimensional feature vectors. Step 3: Intelligent clustering identification of micro-damage types: Using the variational Bayesian Gaussian mixture model (VBGMM) clustering algorithm, the micro-damage types of concrete structures are classified, and the mapping relationship between clustering results and damage mechanisms is constructed. Step 4: Construction of a multi-parameter fusion damage assessment model: By fusing acoustic emission HI index, Sr value and b value, a quantitative index system that can correlate microscopic damage and macroscopic performance is established to assess the degree of damage to concrete structures.

2. The method according to claim 1, characterized in that, In step 1, the deployment of multi-scale acoustic emission sensors follows the JT / T1037-2022 standard, in which high-frequency sensors of 100-400kHz are used on concrete surfaces and low-frequency sensors of 20-100kHz are used on embedded prestressed steel bars.

3. The method according to claim 1, characterized in that, Step 1 also includes a graded loading test. For reinforced concrete structures, a four-point shear test is used, and for fiber-reinforced concrete, a splitting tensile test is used. The load is applied in stages at 20%, 40%, 60%, 80%, and 90% of the ultimate load F, with each loading stage lasting 10 minutes. Acoustic emission signals, load-displacement curves, and DIC images are recorded simultaneously during the loading process to establish the correspondence between "load level - damage characteristics - acoustic emission signal".

4. The method according to claim 1, characterized in that, In step 2, the frequency band of wavelet packet decomposition is 0-500kHz, where the damage signal of the interface transition zone (ITZ) between concrete aggregate and mortar is concentrated in 62.5-125kHz, the fiber pull-out signal is concentrated in 0-62.5kHz, the aggregate crushing signal is concentrated in 125-250kHz, and the steel bar damage signal is concentrated in 250-500kHz.

5. The method according to claim 1, characterized in that, In step 3, the parameters of the VBGMM clustering algorithm are optimized using the Markov Chain Monte Carlo (MCMC) method. The optimized parameters include the number of iterations, the convergence threshold, and the prior distribution of the mixing coefficients. The number of iterations is set to 5000, and the convergence threshold is set to 1e-6.

6. The method according to claim 1, characterized in that, In step 4, based on the acoustic emission cumulative energy versus load curve, the damage process of the concrete structure is divided into three stages: Elastic phase: Cumulative energy less than 10 -3 J, This stage is mainly characterized by initial pore closure, with no obvious damage; Stable damage phase: Cumulative energy is at 10 -3 -10 -1 Between J, the damage is mainly dominated by ITZ and mortar damage, with an energy slope of less than 0.01 J / kN; Accelerated damage phase: Cumulative energy greater than 10 -1 J, dominated by aggregate crushing and steel / fiber damage, with high-energy signals accounting for over 80%.

7. The method according to claim 1, characterized in that, In step 4, the formula for calculating the HI index is: ； in, For damage weight, This represents the percentage of such injury events; The Sr value represents the frequency of high-energy events occurring per unit time. The value of b is the logarithmic linear slope of the acoustic emission signal amplitude distribution, and its calculation formula is: ； in, The number of events with an amplitude greater than or equal to A. When the value of b is less than 1.0, it indicates that the damage has entered the accelerated phase.

8. The method according to claim 1, characterized in that, In step 4, the damage assessment model sets three warning thresholds based on Sr values ​​and the HI index: Mild damage: Log10(Sr) < 5.0 and HI < 0.77, corresponding to the early stage of stable damage, requiring increased monitoring frequency; Moderate damage: 5.0≤Log10(Sr)<6.0 and 0.77≤HI<3.40, corresponding to the late stage of stable damage, it is recommended to carry out local reinforcement work; Severe damage: Log10(Sr)≥6.0 and HI≥3.40, corresponding to the accelerated damage stage, requiring immediate reinforcement and repair.

9. A smart identification and assessment system for concrete structure damage based on acoustic emission waveforms, used to implement the method described in any one of claims 1-8, characterized in that, The system includes: A multi-scale acoustic emission sensor array is used to acquire full waveform data of acoustic emission. Full waveform data acquisition equipment is used to store and perform preliminary processing on the acquired data; An intelligent identification model, which integrates the VBGMM algorithm, is used to identify the types of damage to concrete structures.

10. The system according to claim 9, characterized in that, The intelligent recognition model uses ResNet18 as the base network and combines the MAML algorithm to optimize the initial weight parameters, thereby reducing the model's requirement for a large number of samples and improving the model's generalization ability under different structural types and working conditions.

Images