Building quality evaluation method and system based on concrete nondestructive testing and storage medium

Abstract

The invention relates to the technical field of intelligent detection, and discloses a building quality evaluation method and system based on concrete nondestructive testing and a storage medium. The method comprises the steps that a piezoelectric ceramic sensor array is arranged to collect micro-vibration response signals, and an original vibration data set is obtained; extracting an energy distribution coefficient of each frequency band by using a wavelet packet decomposition algorithm, and constructing a damage feature vector matrix; establishing a physical constraint neural network model, and outputting a damage variable time sequence; fusing the damage variable with ultrasonic and rebound data, and calculating comprehensive strength and damage degree indexes; and calculating the remaining service life by using a time sequence prediction algorithm, and generating an evaluation report. According to the method, the technical problem that the existing concrete nondestructive testing technology cannot realize microstructure damage evolution dynamic monitoring and residual life prediction is solved, and the accuracy of building quality evaluation and the scientificity of predictive maintenance decision are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent detection, in particular to a building quality evaluation method and system based on concrete nondestructive testing and a storage medium. BACKGROUND

[0002] In the current booming construction industry, the control of building quality is related to the safety of people's life and property, and the stability and harmony of society. As the core basic material of modern construction, the quality of concrete directly determines the stability and durability of the building structure. Traditional concrete detection methods are mostly destructive detection, such as core drilling method, which drills core samples from the concrete structure for laboratory analysis to evaluate its strength and other performance indicators. Existing nondestructive testing techniques mainly use single or simple combined detection methods such as ultrasonic testing and rebound testing to evaluate the current strength state of concrete by measuring physical parameters such as sound velocity and rebound value.

[0003] However, the existing technology has many drawbacks and limitations. Destructive detection methods can damage the structure, affect the integrity of the structure, and the detection process is tedious, time-consuming, and the number of samples is limited, making it difficult to accurately reflect the actual quality of the entire concrete structure. Although the existing nondestructive detection methods avoid the problem of structural damage, they still have the problems of single detection method, limited data processing capability, and imperfect evaluation standard. These methods mainly focus on the static strength evaluation of concrete, lack dynamic monitoring capability of structural damage evolution process, and cannot identify early microscopic damage signs, nor can they predict future performance degradation trends based on current detection data.

[0004] More importantly, the existing technology lacks a deep understanding of the internal microstructure damage evolution mechanism of concrete and effective monitoring means. Due to the inability to track the dynamic process of micro-crack initiation, expansion and connection in real time, the existing methods are difficult to establish an accurate correlation between damage evolution and macroscopic performance degradation, and thus cannot realize reliable prediction of the remaining service life of the concrete structure. This technical defect not only limits the accuracy and forward-looking of building quality evaluation, but also cannot provide a scientific basis for predictive maintenance decisions, ultimately affecting the effectiveness and economy of building structure safety management. SUMMARY

[0005] The present application provides a building quality evaluation method and system based on concrete nondestructive testing and a storage medium, which solves the technical problem that existing concrete nondestructive testing technology cannot realize dynamic monitoring of microstructure damage evolution and prediction of remaining life, and improves the accuracy of building quality evaluation and the scientificity of predictive maintenance decisions.

[0006] In a first aspect, the application provides a building quality evaluation method based on concrete nondestructive testing, which comprises: S1, a piezoelectric ceramic sensor array is arranged on the surface of the concrete structure, micro-vibration response signals under ultrasonic excitation are collected, and an original vibration data set containing damage sensitive information is obtained; S2, the original vibration data set is decomposed into multiple frequency bands by using a wavelet packet decomposition algorithm, energy distribution coefficients of each frequency band are extracted, and a damage feature vector matrix is constructed; S3, a physically constrained neural network model is established, the damage feature vector matrix is input, and a damage variable time series is output; S4, the damage variable time series, ultrasonic velocity data and rebound value data are fused, a comprehensive strength index and a damage degree index are calculated; S5, based on the historical evolution data of the comprehensive strength index and the damage degree index, a time series prediction algorithm is used to calculate the remaining service life of the structure, and a building quality evaluation report is generated.

[0007] Optionally, S1 further comprises: The piezoelectric ceramic sensors are arranged in an array on the surface of the concrete structure according to a preset matrix arrangement mode, the sensor spacing is set according to the uniform distribution principle, and sensor spatial distribution coordinate information is obtained; An ultrasonic pulse signal based on a specific frequency range is used to excite the internal structure of the concrete structure, the pulse parameters are adjusted according to the density characteristics of the concrete, and a standardized ultrasonic excitation signal is obtained; The structure vibration response received by each sensor node in the sensor spatial distribution coordinate information is processed by high-frequency sampling, the sampling parameters are determined according to the damage detection accuracy requirements, and multi-channel synchronous vibration signal data is obtained; The multi-channel synchronous vibration signal data is amplified and filtered by a signal conditioning circuit, the noise suppression parameters are set according to the signal quality standard, and an original vibration data set containing damage sensitive information is obtained.

[0008] Optionally, S2 further comprises: The Daubechies wavelet function is selected as the mother wavelet, the original vibration data set is processed by multi-layer wavelet packet decomposition, the number of decomposition layers is determined according to the signal frequency band characteristics, and a multi-scale frequency domain decomposition result is obtained; The signal energy of each frequency band in the multi-scale frequency domain decomposition result is calculated and processed, the energy values of each frequency band are obtained by energy integral operation, and frequency band energy distribution data are obtained; The damage sensitive feature parameter set is matrixed and organized according to a time sequence and a spatial position, a mapping relationship between the feature parameters and the sensor positions is established, and a damage feature vector matrix is obtained. The damage sensitive feature parameter set is matrixed and organized according to a time sequence and a spatial position, a mapping relationship between the feature parameters and the sensor positions is established, and a damage feature vector matrix is obtained.

[0009] Optionally, the S3 step further includes: A damage evolution physical constraint equation is obtained based on the Mazars damage constitutive model, taking the concrete material parameters and the damage threshold strain as constraint conditions; A long short-term memory neural network architecture is constructed as a data learning layer, a number of hidden units and a learning rate parameter are set, the damage evolution physical constraint equation is embedded in the network structure, and a physical constraint neural network model is obtained; The damage feature vector matrix is input into the physical constraint neural network model according to a time window sequence for training processing, network weight parameters are updated through a back propagation algorithm, and a trained damage prediction model is obtained; Based on the trained damage prediction model, a forward reasoning calculation is performed on the current damage feature vector matrix, a damage evolution prediction value in a future time step range is output, and a damage variable time sequence is obtained.

[0010] Optionally, the long short-term memory neural network architecture is constructed as the data learning layer, the number of hidden units and the learning rate parameter are set, the damage evolution physical constraint equation is embedded in the network structure, and the physical constraint neural network model is obtained, including: Memory cells and gating mechanism parameters of the long short-term memory neural network are initialized according to time sequence characteristics of the damage feature vector matrix, a network input layer dimension is set to match an array number of the damage feature vector matrix, and neural network architecture parameters are obtained; The relationship between the damage variable and the equivalent strain in the damage evolution physical constraint equation is taken as a physical constraint term, and a physical constraint loss function is obtained by weighted combination processing with a network prediction loss function; The number of network hidden layer units is set based on time-dependent characteristics of damage evolution, an activation function type and an output layer damage variable prediction node are configured, and a damage prediction network structure is obtained; The physical constraint loss function is taken as a network training target, network weight parameters are optimized and updated through a time back propagation algorithm, and the physical constraint neural network model is obtained.

[0011] Optionally, the S4 step further includes: The ultrasonic detection device is used for testing the sound wave propagation of the concrete structure, recording the propagation time and path length of the ultrasonic wave in the concrete, calculating the ultrasonic wave speed value, and obtaining the ultrasonic wave speed data; The rebound hammer is used for testing the rebound of the concrete surface, measuring the rebound distance and rebound time parameters after the rebound hammer impacts, and obtaining the rebound value data; The weighted fusion algorithm is used for performing numerical fusion processing on the damage variable time sequence, the ultrasonic wave speed data and the rebound value data, setting the weight coefficients of each data source, calculating the weighted average intensity value, and obtaining the comprehensive intensity index; The damage variable value in the damage variable time sequence is compared with the preset damage threshold, and the damage development rate and the damage distribution range parameters are combined to obtain the damage degree index.

[0012] Optionally, S5 further includes: The comprehensive intensity index and the damage degree index are sequentially processed to form a historical data sequence, a correspondence between a timestamp and an index value is established, and a historical evolution data sequence is obtained; The ARIMA time sequence prediction algorithm is used for performing trend analysis and periodicity identification processing on the historical evolution data sequence, autoregressive parameters and moving average parameters are determined, and a time sequence prediction model is obtained; The historical evolution data sequence is input into the time sequence prediction model for future evolution trend calculation processing, the damage development trajectory and the intensity attenuation law are predicted, and a remaining service life value is obtained; According to the remaining service life value and the current damage degree index, quality grade division processing is performed, the quality grade classification is determined in combination with the preset building quality evaluation standard, and a building quality evaluation report is obtained.

[0013] In a second aspect, the application provides a building quality evaluation system based on concrete nondestructive testing, which comprises: The acquisition module is used for arranging a piezoelectric ceramic sensor array on the surface of the concrete structure, collecting micro-vibration response signals under ultrasonic excitation, and obtaining an original vibration data set containing damage sensitive information; The decomposition module is used for decomposing the original vibration data set into a plurality of frequency bands by using a wavelet packet decomposition algorithm, extracting energy distribution coefficients of each frequency band, and constructing a damage feature vector matrix; The input module is used for establishing a physically constrained neural network model, inputting the damage feature vector matrix, and outputting a damage variable time sequence; The calculation module is used for fusing the damage variable time sequence, the ultrasonic wave speed data and the rebound value data, calculating the comprehensive intensity index and the damage degree index; The generating module is configured to calculate the remaining service life of the structure by using a time series prediction algorithm based on historical evolution data of the comprehensive strength index and the damage degree index, and generate a building quality assessment report.

[0014] In a third aspect, a building quality assessment device based on concrete nondestructive testing is provided, which comprises a memory and at least one processor, and the memory stores instructions; the at least one processor invokes the instructions in the memory to enable the building quality assessment device based on concrete nondestructive testing to perform the building quality assessment method based on concrete nondestructive testing described above.

[0015] In a fourth aspect, a computer readable storage medium is provided, which stores instructions, and when the instructions are run on a computer, the computer is enabled to perform the building quality assessment method based on concrete nondestructive testing described above.

[0016] In the technical scheme provided in the present application, the technical feature of arranging a piezoelectric ceramic sensor array on the surface of the concrete structure to collect micro-vibration response signals breaks through the limitation of traditional detection methods relying on single point or a small number of measuring points, realizes spatial distributed comprehensive perception of the damage state of the concrete structure, and can capture early micro-damage signals that are difficult to detect by traditional methods. The technical feature of using a wavelet packet decomposition algorithm to decompose the original vibration data set into multiple frequency bands, extract energy distribution coefficients of each frequency band, and construct a damage feature vector matrix effectively solves the technical problem of difficulty in extracting damage sensitive information in complex vibration signals, and through multi-scale frequency domain analysis, different types and degrees of internal defects of the concrete can be accurately identified, providing a high-quality feature data basis for subsequent intelligent analysis. The technical feature of establishing a physically constrained neural network model to input the damage feature vector matrix and output a damage variable time series combines the concrete damage mechanics theory and the deep learning algorithm, guarantees that the prediction result conforms to the physical law, fully utilizes the nonlinear mapping ability of the neural network, and realizes accurate modeling and dynamic prediction of the concrete damage evolution process.

[0017] The technical features of fusing the time series of damage variables with the ultrasonic wave speed data and the rebound value data to calculate the comprehensive strength index and the damage degree index effectively improve the reliability and accuracy of the evaluation results through the multi-source data fusion strategy, and overcome the limitations and uncertainties of single detection means. The technical features of calculating the remaining service life of the structure and generating the building quality evaluation report based on the historical evolution data of the comprehensive strength index and the damage degree index using the time series prediction algorithm realize the important change from static quality evaluation to dynamic life prediction, and provide a scientific basis for predictive maintenance and risk control of building structures. In the specific application field of concrete nondestructive testing, the multiscale analysis characteristics of the wavelet packet decomposition algorithm enable the system to simultaneously capture local damage information with high frequency and overall structural response with low frequency, and the introduction of the physically constrained neural network model ensures the rationality of the prediction results in the physical sense. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor based on these drawings.

[0019] Figure 1 An embodiment of the building quality evaluation method based on concrete nondestructive testing in the embodiments of the present application is shown. Figure 2 A flowchart of the concrete structure damage prediction based on the physically constrained LSTM in the embodiments of the present application is shown. Figure 3 A comparison result graph of the office building damage prediction based on the physically constrained LSTM in the embodiments of the present application is shown. Figure 4 An embodiment of the building quality evaluation system based on concrete nondestructive testing in the embodiments of the present application is shown. Figure 5 A structural schematic block diagram of the building quality evaluation equipment based on concrete nondestructive testing in the embodiments of the present application is shown. DETAILED DESCRIPTION

[0020] The embodiment of the application provides a building quality evaluation method and system based on concrete nondestructive testing and a storage medium. The terms "first", "second", "third", "fourth" and the like (if any) in the specification and claims of the application and the above drawings are used to distinguish similar objects and do not have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" or "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0021] For ease of understanding, the specific process of the embodiment of the application is described below. Please refer to Figure 1 One embodiment of the building quality evaluation method based on concrete nondestructive testing in the embodiment of the application comprises the following steps: S1, a piezoelectric ceramic sensor array is arranged on the surface of a concrete structure, micro-vibration response signals under ultrasonic excitation are collected, and an original vibration data set containing damage sensitive information is obtained; S2, a wavelet packet decomposition algorithm is used to decompose the original vibration data set into a plurality of frequency bands, energy distribution coefficients of each frequency band are extracted, and a damage feature vector matrix is constructed; S3, a physically constrained neural network model is established, the damage feature vector matrix is input, and a damage variable time series is output; S4, the damage variable time series, ultrasonic velocity data and rebound value data are fused, a comprehensive strength index and a damage degree index are calculated; S5, based on the historical evolution data of the comprehensive strength index and the damage degree index, a time series prediction algorithm is used to calculate the remaining service life of the structure, and a building quality evaluation report is generated.

[0022] It can be understood that the execution subject of the application can be a building quality evaluation system based on concrete nondestructive testing, and can also be a terminal or a server, which is not limited here. The embodiment of the application takes the server as the execution subject for example.

[0023] In a specific embodiment, S1 further comprises: The piezoelectric ceramic sensors are arranged in an array on the surface of the concrete structure according to a preset matrix arrangement mode, the spacing between the sensors is set according to the uniform distribution principle, and sensor spatial distribution coordinate information is obtained. The ultrasonic pulse signal based on a specific frequency range is used to excite the internal acoustic waves of the concrete structure, and the pulse parameters are adjusted according to the density characteristics of the concrete to obtain a standardized ultrasonic excitation signal. The structural vibration responses received by each sensor node in the spatial distribution coordinate information of the sensor are processed by high-frequency sampling, and the sampling parameters are determined according to the damage detection accuracy requirements to obtain multi-channel synchronous vibration signal data. The multi-channel synchronous vibration signal data is amplified and filtered by a signal conditioning circuit, and the noise suppression parameters are set according to the signal quality standard to obtain an original vibration data set containing damage sensitive information.

[0024] Specifically, piezoelectric ceramic sensors are arranged in a preset matrix pattern on the surface of the concrete structure, with uniform spacing to ensure coverage of the entire detection area and acquisition of spatial distribution coordinate information. Based on the characteristics of the concrete structure, ultrasonic pulse signals with a specific frequency range are used to excite acoustic waves, and the parameters of the pulse signal are adjusted according to the density characteristics of the concrete to ensure signal standardization. For the vibration response signals received by the sensors, high-frequency sampling is performed according to accurate damage detection requirements to ensure that small changes in the structure can be captured, thereby obtaining multi-channel synchronous vibration signal data. To enhance signal quality, these synchronous vibration signals are amplified and filtered by a signal conditioning circuit, and the noise suppression parameters are set according to the signal quality standard to effectively remove background noise, and finally an original vibration data set containing damage sensitive information is obtained. This series of processing steps provides high-quality input signals for subsequent data analysis, ensuring that damage features can be accurately extracted. The sensors can be uniformly arranged in a rectangular array on the surface of the concrete structure, with a distance of 10 cm between each sensor to ensure that there are no gaps in the coverage area. To optimize the ultrasonic pulse excitation signal, the frequency range used is 20 kHz to 50 kHz, which can effectively excite micro-cracks and defects in the concrete without causing excessive impact on the concrete. According to the density characteristics of the concrete, the pulse width of the ultrasonic pulse signal is set to 200 microseconds, and the power of the signal is set to 5 W to ensure that the excitation signal can penetrate the surface of the concrete and be transmitted to the deep layer of the structure.

[0025] In a specific embodiment, in the signal acquisition stage, the sampling frequency is set to 100 kHz to ensure that the high-frequency vibration response of the structure can be captured. The vibration signals received by each sensor node are recorded in multiple channels and synchronized through the data obtained by high-frequency sampling. For example, in one sampling, the data collected by the 6 sensors at the same time are recorded synchronously to ensure that there is no time delay error. The collected data are amplified and filtered through a signal conditioning circuit, a high-pass filter is used to remove low-frequency noise, and the bandwidth is set to 1 Hz to 10 kHz, which can effectively remove noise interference from the external environment. After signal processing, the obtained raw vibration data set contains key information of the micro-damage of the structure, providing high-quality input signals for subsequent analysis.

[0026] In a specific embodiment, the S2 step further comprises: selecting a Daubechies wavelet function as a mother wavelet, performing multi-layer wavelet packet decomposition processing on the raw vibration data set, the number of decomposition layers being determined according to the signal frequency band characteristics, and obtaining a multi-scale frequency domain decomposition result; calculating and processing the signal energy of each frequency band in the multi-scale frequency domain decomposition result, obtaining the energy value of each frequency band through energy integral operation, and obtaining frequency band energy distribution data; filtering the frequency band energy distribution data based on a concrete damage sensitive frequency band identification criterion, extracting specific frequency band energy coefficients sensitive to microstructure changes, and obtaining a damage sensitive feature parameter set; matrixing and organizing the damage sensitive feature parameter set according to time sequence and spatial position, establishing a mapping relationship between the feature parameters and the sensor positions, and obtaining a damage feature vector matrix.

[0027] Specifically, a Daubechies wavelet function is selected as a mother wavelet for wavelet packet decomposition of the raw vibration data to capture the micro changes in the concrete structure in multiple frequency bands. The number of decomposition layers is adjusted according to the frequency band characteristics of the signal to ensure that the frequency domain information at different scales can be effectively decomposed. Through multi-scale frequency domain decomposition, different frequency bands of the vibration signal can be distinguished in detail. Next, the signal energy of each frequency band is calculated, the energy value of each frequency band is obtained through energy integral operation, and frequency band energy distribution data is generated to reflect the energy distribution characteristics of each frequency band in the entire signal. On this basis, the frequency band energy distribution data is filtered according to the concrete damage sensitive frequency band identification criterion, and the frequency band energy coefficients particularly sensitive to microstructure changes are extracted. These specific frequency band energy coefficients constitute a damage sensitive feature parameter set, which provides effective feature information for subsequent damage detection. The damage sensitive feature parameters are matrixed and organized according to time sequence and spatial position, a mapping relationship between the feature parameters and the sensor positions is established, and a damage feature vector matrix is obtained.

[0028] For example, when the Daubechies wavelet function is selected as the mother wavelet, the Daubechies-4 wavelet can be used to process the original vibration data. In an embodiment, it is assumed that the original vibration data set contains 5000 sampling points. Through wavelet packet decomposition, this signal is decomposed into 8 layers, where the first layer corresponds to the low frequency band, capturing the macro response of the concrete structure, and the eighth layer extracts the high frequency band information, which can capture the details of micro cracks or local damage. According to the signal frequency band characteristics, the number of decomposition layers will be adjusted according to the signal variation range and damage identification requirements. When calculating the energy of the second layer frequency band, it is assumed that the energy calculation result of this frequency band is 0.035, and the energy proportion of this frequency band in the whole signal is obtained through integral operation. For the fourth layer frequency band, the energy is 0.041, indicating that this frequency band has a greater contribution to the signal in the damage state. After processing, a frequency band energy distribution data is obtained, which describes the energy values of each frequency band.

[0029] In a specific embodiment, the S3 step further comprises: Based on the Mazars damage constitutive model, a physical constraint layer is established, the concrete material parameters and damage threshold strain are taken as constraint conditions, and a damage evolution physical constraint equation is obtained; A long short-term memory neural network architecture is constructed as a data learning layer, the number of hidden units and the learning rate parameters are set, the damage evolution physical constraint equation is embedded in the network structure, and a physical constraint neural network model is obtained; The damage feature vector matrix is input into the physical constraint neural network model according to the time window sequence for training processing, the network weight parameters are updated through the back propagation algorithm, and a trained damage prediction model is obtained; Based on the trained damage prediction model, the current damage feature vector matrix is calculated and processed by forward reasoning, the damage evolution prediction value in the future time step range is output, and a damage variable time series is obtained.

[0030] Specifically, based on the Mazars damage constitutive model, the material parameters of concrete (such as elastic modulus, Poisson's ratio) and damage threshold strain are taken as constraint conditions to establish the physical constraint equation of damage evolution. This equation can describe the damage evolution law of concrete under different stress conditions, providing a physical background for the subsequent neural network model. In the construction process of the neural network, the long short-term memory neural network (LSTM) architecture is selected as the data learning layer, and the number of hidden units is set to 128 and the learning rate is set to 0.001. These parameters are optimized through experiments to ensure that the network can capture the long and short-term dependencies in the time series. The physical constraint equation of damage evolution is embedded in the LSTM network to ensure that the neural network considers the actual physical limitations during training. The damage feature vector matrix is input into this physically constrained neural network according to the time window sequence for training. During training, the network weights are continuously adjusted through the backpropagation algorithm to ensure that the network accurately learns the damage evolution law after each iteration. After training, the trained model is used to input the current damage feature vector matrix for forward reasoning to output the damage evolution prediction value for future time steps, obtaining the time series of damage variables, thereby achieving effective prediction of the damage evolution process of building structures.

[0031] Taking the concrete structure of a certain bridge as an example, according to the Mazars damage constitutive model, the elastic modulus of concrete is selected as 30 GPa, the Poisson's ratio is 0.2, and the damage threshold strain is set to 0.002. On this basis, the physical constraint equation of damage evolution is established. The long short-term memory neural network (LSTM) architecture is used as the data learning layer, with 128 hidden units and a learning rate of 0.005 to capture the damage evolution law of concrete structures under different environmental conditions. The above physical constraint equation of damage evolution is embedded in the LSTM network to ensure that the training process of the neural network considers the physical limitations. The damage feature vector matrix collected by the piezoelectric ceramic sensor array is input into the physically constrained neural network according to the time window sequence for training. During training, the backpropagation algorithm is used to gradually optimize the network weights, allowing the neural network to accurately fit the time series of damage evolution. After training, the trained model is used to input the current damage feature matrix, and the damage evolution value for a future period of time is calculated through forward reasoning to obtain the predicted damage variable time series. This prediction value can help structural engineers evaluate the remaining service life of the bridge and guide maintenance decisions. Referring to Figure 2 The figure shows the process of concrete structure damage prediction based on physically constrained LSTM.

[0032] In a specific embodiment, the process of performing steps of constructing a long short-term memory neural network architecture as a data learning layer, setting the number of hidden units and learning rate parameters can specifically include the following steps: Initialize the memory cell and gating mechanism parameters of the long short-term memory neural network based on the time sequence characteristics of the damage feature vector matrix, set the network input layer dimension to match the number of columns of the damage feature vector matrix, and obtain the neural network architecture parameters; Take the relationship between the damage variable in the damage evolution physical constraint equation and the equivalent strain as a physical constraint term, and perform weighted combination processing with the network prediction loss function to obtain a physical constraint loss function; Set the number of hidden layer units of the network based on the time-dependent characteristics of damage evolution, configure the type of activation function and the output layer damage variable prediction node, and obtain the damage prediction network structure; Take the physical constraint loss function as the network training target, and optimize and update the network weight parameters through the time backpropagation algorithm to obtain a physical constraint neural network model.

[0033] Specifically, the initialization step of the long short-term memory neural network (LSTM) first sets the memory cell and gating mechanism parameters based on the time sequence characteristics of the damage feature vector matrix, so that the neural network can effectively process time series data. The dimension of the network input layer matches the number of columns of the damage feature vector matrix, which ensures that the input of each time step is consistent with the feature data processed by the network. In terms of physical constraints, the relationship between the damage variable in the damage evolution physical constraint equation and the equivalent strain is taken as a physical constraint term, and it is combined with the network prediction loss function to form a physical constraint loss function. This loss function helps to ensure that the prediction results of the neural network are consistent with the actual physical laws. Then, according to the time-dependent characteristics of damage evolution, the number of hidden layer units of the network is set to 256, and the ReLU activation function is configured to ensure that the network can capture complex nonlinear relationships. The output layer is set to predict the damage variable node to accurately output the predicted value of structural damage. Take the physical constraint loss function as the training target, and optimize and update the network weight through the time backpropagation algorithm, so as to obtain the final trained physical constraint neural network model. This model can accurately predict the damage evolution process of the concrete structure according to the input damage feature vector matrix.

[0034] For example, when initializing the long short-term memory neural network, suppose we are analyzing the damage of a bridge's concrete structure, and the damage feature vector matrix contains vibration data of the structure at multiple time points. Each time step data contains the response signal of 5 sensors. According to these data, the input layer dimension of the neural network is set to 5, that is, 5 feature values are input at each time step. These feature values represent the vibration response data of each sensor, ensuring that the data at each time point can match the processing capacity of the network.

[0035] For example, in the setting process of physical constraint term, it is assumed that the relationship between damage variable and equivalent strain is included in the damage evolution physical constraint equation. According to the specific material and damage model of the bridge, it is assumed that the relationship between damage variable and equivalent strain is linear. This relationship is used as a physical constraint term, which is combined with the network prediction loss function by weighting. For example, the weight of damage variable is 0.7, and the weight of equivalent strain is 0.3. Through this weighted combination, the network can optimize damage prediction and follow the actual physical law during training.

[0036] For example, in the setting process of network hidden layer unit number, it is assumed that the damage evolution process has a long time dependence, so we set the number of hidden layer units to 256, which can ensure that the network can capture the information in the long time sequence, and use the ReLU activation function to handle the nonlinear relationship, so as to enhance the learning ability and prediction accuracy of the network.

[0037] For example, in the training process, it is assumed that the damage feature matrix of the training data contains vibration response data from the beginning of construction to the current time. The network optimizes these data through the time back propagation algorithm. At each training time, the network updates its weights according to the back propagation result, so as to minimize the error of damage prediction, and finally obtains a trained physical constraint neural network model. This model can accurately predict the future damage evolution process based on the input feature data, and provide a scientific basis for the maintenance of the structure.

[0038] Taking a high-rise office building concrete structure as an example, first, according to the time sequence characteristics of the damage feature vector matrix, the memory unit and gating mechanism parameters of the long short-term memory neural network (LSTM) are initialized. It is assumed that sensors are arranged at the four corners of each floor of the building, and each sensor records the vertical vibration response data. The input layer dimension of the network is set to 4, ensuring that the data of four sensors at each time step can be input into the network for processing at the same time. The relationship between the elastic modulus of concrete material and the damage variable is defined in the damage evolution physical constraint equation. These physical constraint terms are added to the loss function of damage prediction, and are processed by weighted combination, in which the weight of elastic modulus is set to 0.7 and the weight of damage variable is 0.3, to ensure that the prediction result of the network follows the physical law. In order to capture the damage evolution in a long time span, the number of hidden layer units of the network is set to 150, and the ReLU activation function is used to handle nonlinear problems. In the training process, the network weight is updated and optimized according to the damage feature vector matrix by using the back propagation algorithm. After multiple rounds of training, an optimized physical constraint neural network model is obtained, which can predict the possible damage of the building structure in the next few months according to the sensor data, provide early warning signals, and provide data support for the maintenance and risk management of the building, reference Figure 3The figure shows the office building damage prediction comparison results based on physical constraint LSTM.

[0039] In a specific embodiment, the S4 step further comprises: The ultrasonic detection device is used to test the sound wave propagation in the concrete structure, record the propagation time and path length of the ultrasonic wave in the concrete, calculate the ultrasonic wave speed value, and obtain the ultrasonic wave speed data. The rebound hammer is used to test the rebound on the surface of the concrete, measure the rebound distance and rebound time parameters after the impact of the rebound hammer, and obtain the rebound value data. The weighted fusion algorithm is used to process the numerical fusion of the damage variable time series, ultrasonic wave speed data and rebound value data, set the weight coefficients of each data source, calculate the weighted average intensity value, and obtain the comprehensive intensity index. According to the comparison and analysis of the damage variable value in the damage variable time series and the preset damage threshold, combined with the damage development rate and damage distribution range parameters, the damage degree index is obtained.

[0040] Specifically, the ultrasonic detection device is used to test the sound wave propagation in the concrete structure, record the propagation time and path length of the ultrasonic wave in the concrete, calculate the ultrasonic wave speed value according to these data, and obtain the ultrasonic wave speed data, reflecting the compactness and possible crack situation inside the concrete. The rebound hammer is used to test the rebound on the surface of the concrete, measure the rebound distance and rebound time parameters after the impact of the rebound hammer, and obtain the rebound value data, which can reflect the surface strength and hardness of the concrete. Based on the weighted fusion algorithm, the damage variable time series, ultrasonic wave speed data and rebound value data are processed by numerical fusion. In order to ensure that the influence degree of each data source matches its importance, different weight coefficients are set, such as assigning a higher weight to the ultrasonic wave speed data, calculating the weighted average intensity value, and forming the comprehensive intensity index. According to the comparison and analysis of the damage variable value in the damage variable time series and the preset damage threshold, combined with the damage development rate and damage distribution range parameters, the damage degree index is obtained, which can comprehensively evaluate the health status and damage degree of the concrete structure, and provide data support for subsequent structure maintenance and safety evaluation.

[0041] For example, when the ultrasonic detection device tests the sound wave propagation in a certain section of the concrete structure of the bridge, it measures the time required for the sound wave to propagate from the transmitting end to the receiving end by transmitting an ultrasonic wave pulse with a known frequency. In this example, the path length of the ultrasonic wave propagation is 2 meters, and the measurement result shows that the propagation time of the ultrasonic wave is 0.06 seconds, and the speed of the ultrasonic wave in the concrete is calculated to be 33.33 m / s. This data can reflect the compactness and possible cracks of the concrete.

[0042] For example, the data in the damage variable time series is used to compare with the preset damage threshold. In an actual embodiment, the set damage threshold is 50, and when the damage variable value reaches or exceeds this threshold, it indicates that the structure has suffered a large damage. According to the parameters of the damage development rate and the distribution range, the damage degree index is 0.7, indicating that the damage of the concrete structure has reached a high degree, and further repair or reinforcement treatment is needed.

[0043] In a specific embodiment, the S5 step further comprises: The comprehensive strength index and the damage degree index are processed in time sequence to construct a historical data sequence, and a corresponding relationship between a time stamp and an index value is established to obtain a historical evolution data sequence; Based on the ARIMA time series prediction algorithm, the historical evolution data sequence is analyzed for trend and periodicity to determine autoregressive parameters and moving average parameters, and a time series prediction model is obtained; The historical evolution data sequence is input into the time series prediction model for future evolution trend calculation, to predict the damage development trajectory and the strength decay law, and to obtain a residual service life value; According to the residual service life value and the current damage degree index, quality grade division processing is performed, and the quality grade classification is determined in combination with the preset building quality evaluation standard to obtain a building quality evaluation report.

[0044] Specifically, the comprehensive strength index and the damage degree index are processed in time sequence to construct a historical data sequence. The index value at each time point is corresponded to a time stamp to form a continuous historical evolution data sequence. Based on the ARIMA time series prediction algorithm, the historical evolution data sequence is analyzed to identify the trend and periodicity changes in the data. By calculating the autoregressive parameters and the moving average parameters, a time series prediction model is obtained, which can capture the law of damage index change over time. After the model is established, the historical evolution data sequence is input to predict the future trend, calculate the damage development trajectory and the strength decay law, and obtain the residual service life value. According to the residual service life value and the current damage degree index, in combination with the preset building quality evaluation standard, the building structure is divided into quality grades. By comparing the actual data with the standard grades, a building quality evaluation report is finally generated, providing a scientific maintenance decision basis for engineers.

[0045] Taking the structure of an old residential building as an example, first, the comprehensive strength index and the damage degree index are processed according to the detection data in chronological order. The strength and damage degree values recorded at each detection time are correspondingly associated with the time stamp to obtain a historical evolution data sequence. Assuming that regular detection has been carried out for two years, 30 detection data have been accumulated, which are used to construct a time series. Then, the ARIMA time series prediction algorithm is used to analyze the trend of the historical data sequence to determine the autoregressive parameters and the moving average parameters. By this method, the long-term development trend and periodic changes of the structural damage of the residential building can be identified, and the future damage trend can be predicted. Next, based on the obtained time series prediction model, the historical evolution data are input into the model for processing to calculate the damage development trajectory and the strength attenuation law in the next five years, and it is concluded that the remaining service life is 7 years. According to this remaining service life value, combined with the current damage degree index and the preset building quality evaluation standard, it is concluded that the quality grade of the residential building is “good”, and a building quality evaluation report is generated, suggesting that the monitoring should be continued, and timely maintenance and reinforcement should be carried out according to the prediction of future damage. This process provides a scientific basis for structural maintenance and ensures the safe use of the building.

[0046] The building quality evaluation method based on nondestructive testing of concrete in the embodiments of the present application is described above, and the building quality evaluation system based on nondestructive testing of concrete in the embodiments of the present application is described below. Please refer to Figure 4 The building quality evaluation system based on nondestructive testing of concrete in the embodiments of the present application includes one embodiment: The acquisition module is used to arrange a piezoelectric ceramic sensor array on the surface of a concrete structure, collect micro-vibration response signals under ultrasonic excitation, and obtain a raw vibration data set containing damage sensitive information. The decomposition module is used to decompose the raw vibration data set into multiple frequency bands by using a wavelet packet decomposition algorithm, extract energy distribution coefficients of each frequency band, and construct a damage feature vector matrix. The input module is used to establish a physically constrained neural network model, input the damage feature vector matrix, and output a damage variable time series. The calculation module is used to fuse the damage variable time series with ultrasonic sound velocity data and rebound value data to calculate a comprehensive strength index and a damage degree index. The generation module is used to calculate the remaining service life of the structure based on the historical evolution data of the comprehensive strength index and the damage degree index by using a time series prediction algorithm, and generate a building quality evaluation report.

[0047] The above Figure 4The concrete nondestructive testing based building quality evaluation system in the embodiment of the application is described in detail from the perspective of the modular functional entity, and the concrete nondestructive testing based building quality evaluation device in the embodiment of the application is described in detail from the perspective of hardware processing.

[0048] With reference to Figure 5 The embodiment of the application also provides a concrete nondestructive testing based building quality evaluation device, which can be a server, and the internal structure of the concrete nondestructive testing based building quality evaluation device can be as shown in Figure 5 The concrete nondestructive testing based building quality evaluation device comprises a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. The processor of the computer is used to provide computing and control capabilities. The memory of the concrete nondestructive testing based building quality evaluation device comprises a nonvolatile storage medium and an internal memory. The nonvolatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operating system and the computer program in the nonvolatile storage medium. The database of the concrete nondestructive testing based building quality evaluation device is used to store corresponding data in the embodiment. The network interface of the concrete nondestructive testing based building quality evaluation device is used to communicate with an external terminal through a network connection. The computer program is executed by the processor to implement the above method.

[0049] Those skilled in the art can understand that Figure 5 The structure shown in the embodiment of the application is only a block diagram of part of the structure related to the scheme of the application, and does not constitute a limitation on the concrete nondestructive testing based building quality evaluation device to which the scheme of the application is applied.

[0050] The application also provides a computer readable storage medium, which can be a nonvolatile computer readable storage medium or a volatile computer readable storage medium. The computer readable storage medium stores instructions, and when the instructions are run on a computer, the computer is caused to perform the steps of the concrete nondestructive testing based building quality evaluation method.

[0051] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, system and unit can refer to the corresponding processes in the foregoing method embodiments, which will not be described herein.

[0052] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application or the entire or part of the technical solutions that essentially contribute to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a building quality evaluation device based on concrete nondestructive testing (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0053] The above embodiments are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A building quality assessment method based on non-destructive testing of concrete, characterized in that, The method includes: Step S1: Arrange a piezoelectric ceramic sensor array on the surface of the concrete structure to collect micro-vibration response signals under ultrasonic excitation and obtain the original vibration dataset containing damage-sensitive information. Step S2: Decompose the original vibration dataset into multiple frequency bands using the wavelet packet decomposition algorithm, extract the energy distribution coefficients of each frequency band, and construct the damage feature vector matrix. Step S3: Establish a physical constraint neural network model, input the damage feature vector matrix, and output the damage variable time series. Step S4: Integrate the time series of the damage variables with the ultrasonic velocity data and rebound value data to calculate the comprehensive strength index and damage degree index. Step S5: Based on the historical evolution data of the comprehensive strength index and damage degree index, the remaining service life of the structure is calculated using a time series prediction algorithm, and a building quality assessment report is generated.

2. The building quality assessment method based on non-destructive testing of concrete according to claim 1, characterized in that, Step S1 further includes: Piezoelectric ceramic sensors are arrayed on the surface of a concrete structure according to a preset matrix arrangement pattern, and the sensor spacing is set according to the principle of uniform distribution to obtain the spatial distribution coordinate information of the sensors. The ultrasonic pulse signal within a specific frequency range is used to excite the interior of the concrete structure. The pulse parameters are adjusted according to the density characteristics of the concrete to obtain a standardized ultrasonic excitation signal. The structural vibration response received by each sensor node in the spatial distribution coordinate information of the sensor is subjected to high-frequency sampling processing. The sampling parameters are determined according to the damage detection accuracy requirements to obtain multi-channel synchronous vibration signal data. The multi-channel synchronous vibration signal data is amplified and filtered through a signal conditioning circuit, and the noise suppression parameters are set according to the signal quality standards to obtain the original vibration dataset containing damage-sensitive information.

3. The building quality assessment method based on non-destructive testing of concrete according to claim 1, characterized in that, Step S2 further includes: The Daubechies wavelet function is selected as the mother wavelet, and the original vibration dataset is subjected to multi-level wavelet packet decomposition. The number of decomposition levels is determined according to the signal frequency band characteristics to obtain multi-scale frequency domain decomposition results. The signal energy of each frequency band in the multi-scale frequency domain decomposition result is calculated and processed. The energy value of each frequency band is obtained by energy integration operation, and the frequency band energy distribution data is obtained. Based on the identification criteria for concrete damage-sensitive frequency bands, the energy distribution data of the frequency bands are screened and processed, and the energy coefficients of specific frequency bands that are sensitive to microstructure changes are extracted to obtain a set of damage-sensitive characteristic parameters. The set of damage-sensitive feature parameters is organized into a matrix according to time series and spatial location, and a mapping relationship between feature parameters and sensor location is established to obtain a damage feature vector matrix.

4. The building quality assessment method based on non-destructive testing of concrete according to claim 1, characterized in that, Step S3 further includes: A physical constraint layer is established based on the Mazars damage constitutive model, and the concrete material parameters and damage threshold strain are used as constraints to obtain the physical constraint equation for damage evolution. A long short-term memory neural network architecture is constructed as the data learning layer. The number of hidden units and the learning rate parameters are set, and the physical constraint equation of damage evolution is embedded into the network structure to obtain a physical constraint neural network model. The damage feature vector matrix is ​​input into the physical constraint neural network model according to the time window sequence for training. The network weight parameters are updated through the backpropagation algorithm to obtain the trained damage prediction model. Based on the trained damage prediction model, forward inference calculation is performed on the current damage feature vector matrix to output the damage evolution prediction value within the future time step range, thus obtaining the damage variable time series.

5. The building quality assessment method based on non-destructive testing of concrete according to claim 4, characterized in that, The construction of a long short-term memory neural network architecture as the data learning layer, setting the number of hidden units and learning rate parameters, and embedding the physical constraint equation of damage evolution into the network structure, yields a physically constrained neural network model, including: The memory unit and gating mechanism parameters of the long short-term memory neural network are initialized based on the temporal characteristics of the damage feature vector matrix. The dimension of the network input layer is set to match the number of columns of the damage feature vector matrix to obtain the neural network architecture parameters. The damage variable and equivalent effect change relationship in the physical constraint equation of damage evolution are used as physical constraint terms and weighted combination with the network prediction loss function to obtain the physical constraint loss function. Based on the time-dependent characteristics of damage evolution, the number of hidden layer units in the network is set, the activation function type and the damage variable prediction node in the output layer are configured to obtain the damage prediction network structure. Using the physical constraint loss function as the network training objective, the network weight parameters are optimized and updated through the time backpropagation algorithm to obtain a physical constraint neural network model.

6. The building quality assessment method based on non-destructive testing of concrete according to claim 1, characterized in that, Step S4 further includes: The ultrasonic wave propagation test is performed on the concrete structure by using an ultrasonic testing device. The propagation time and path length of the ultrasonic wave in the concrete are recorded, and the ultrasonic speed value is calculated to obtain the ultrasonic speed data. A rebound hammer was used to conduct an impact rebound test on the concrete surface. The rebound distance and rebound time parameters after the rebound hammer impact were measured to obtain the rebound value data. The time series of damage variables, ultrasonic velocity data and rebound value data are numerically fused based on a weighted fusion algorithm. The weight coefficients of each data source are set, and the weighted average intensity value is calculated to obtain a comprehensive intensity index. The damage degree index is obtained by comparing and analyzing the damage variable values ​​in the damage variable time series with the preset damage threshold, and combining the damage development rate and damage distribution range parameters.

7. The building quality assessment method based on non-destructive testing of concrete according to claim 1, characterized in that, Step S5 further includes: The comprehensive strength index and damage degree index are processed into a historical data sequence according to time order, and the correspondence between timestamps and index values ​​is established to obtain the historical evolution data sequence. The historical evolution data sequence is analyzed for trend and periodicity based on the ARIMA time series prediction algorithm to determine the autoregressive parameters and moving average parameters, and a time series prediction model is obtained. The historical evolution data sequence is input into the time series prediction model to calculate the future evolution trend, predict the damage development trajectory and intensity decay law, and obtain the remaining service life value. The quality grade is determined based on the remaining service life value and the current damage level index. The quality grade classification is then determined in conjunction with the preset building quality assessment standards to obtain a building quality assessment report.

8. A building quality assessment system based on non-destructive testing of concrete, characterized in that, For implementing the building quality assessment method based on nondestructive testing of concrete as described in any one of claims 1-7, the building quality assessment system based on nondestructive testing of concrete comprises: The acquisition module is used to deploy a piezoelectric ceramic sensor array on the surface of a concrete structure to acquire micro-vibration response signals under ultrasonic excitation and obtain raw vibration datasets containing damage-sensitive information. The decomposition module is used to decompose the original vibration dataset into multiple frequency bands using the wavelet packet decomposition algorithm, extract the energy distribution coefficients of each frequency band, and construct a damage feature vector matrix. The input module is used to establish a physical constraint neural network model, input the damage feature vector matrix, and output the time series of damage variables. The calculation module is used to fuse the time series of the damage variables with ultrasonic velocity data and rebound value data to calculate the comprehensive strength index and damage degree index. The generation module is used to calculate the remaining service life of the structure based on the historical evolution data of the comprehensive strength index and damage degree index, and generate a building quality assessment report.

9. A building quality assessment device based on non-destructive testing of concrete, characterized in that, The method includes a memory and a processor, the memory storing a computer program that can run on the processor, and the processor executing the computer program to implement the building quality assessment method based on non-destructive testing of concrete as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by the processor, it causes the processor to perform the building quality assessment method based on non-destructive testing of concrete as described in any one of claims 1 to 7.

Images