An artificial intelligence-based building material wear quality monitoring method

Abstract

The application discloses a kind of based on artificial intelligence's building material wear quality monitoring method, it is related to building structure health monitoring field;Deployment sensing array acquires initial baseline acoustic signal, uses in periodical monitoring acoustic signal and environmental temperature, humidity, stress data;Extract acoustic feature to generate baseline feature vector, extract monitoring acoustic feature and generate monitoring feature vector after correction;Monitoring feature vector is input based on attention mechanism's bidirectional long short-term memory network, obtains predicted wear and remaining life;Compare prediction result with measured wear, adaptively correct model and feature vector;Based on feature deviation, realize wear early warning according to dynamic threshold, adopt acoustic velocity change tomography algorithm to locate abnormal area.The application significantly improves wear monitoring accuracy and real-time by multi-feature fusion, AI modeling combined with environmental correction and adaptive optimization, realizes early warning and accurate positioning, provides reliable support for building material maintenance.

Description

TECHNICAL FIELD

[0001] The application relates to the field of building structure health monitoring, and particularly discloses a building material wear quality monitoring method based on artificial intelligence. BACKGROUND

[0002] In the field of modern building engineering, the long-term service stability and wear state of building materials (such as concrete pavement, steel structure coating and composite material) are directly related to the safety of building structures, the service life and the operation and maintenance cost control. With the advancement of urbanization, a large number of buildings enter the long-term service stage, and the wear problem of materials under the action of environmental erosion and load gradually becomes prominent. Real-time perception, early hidden danger identification and residual life prediction of the wear state are increasingly demanding, and traditional monitoring methods have been difficult to adapt to the high-precision and continuous requirements of modern building health management.

[0003] Current building material wear monitoring mainly relies on three types of traditional schemes: one is destructive detection, represented by core sampling detection, which determines the wear degree by drilling samples and laboratory analysis, and needs to destroy the integrity of the building structure; the second is semi-non-destructive detection, such as ultrasonic flaw detection, which does not need to destroy the material, but needs manual contact operation of the equipment, which is limited by the range and labor cost, and it is difficult to realize large-scale and long-term continuous monitoring; the third is single parameter monitoring, most systems only focus on macroscopic parameters such as material thickness attenuation, and predict the residual life by combining static empirical formula, without considering the influence of dynamic factors such as temperature, humidity and load frequency in actual service.

[0004] However, the traditional schemes have significant technical bottlenecks: first, the detection is destructive or lagging, core sampling leaves hidden dangers, and manual inspection cannot be continuously monitored, so that the material often has serious damage when the problem is found; second, the evaluation index is single and one-sided, which cannot capture the microstructure changes (such as microcracks and increased porosity) caused by wear, and the monitoring results are difficult to reflect the true health status; third, the model lacks adaptability, the static formula prediction error is large, and it is difficult to guide operation and maintenance; fourth, there is no early warning and positioning capability, which cannot identify micro-damage and cannot lock the local abnormal area, and the operation and maintenance lacks targeting. Therefore, developing a non-destructive, continuous, multi-dimensional sensing and self-adaptive correction wear monitoring method has become a key requirement to break through the bottleneck. SUMMARY

[0005] A building material wear quality monitoring method based on artificial intelligence, comprising the following steps:

[0006] S1, collecting baseline acoustic signals of the building material in the initial state and periodic monitoring acoustic signals in the use process through a sensing array arranged on the building material, and collecting environmental temperature, humidity and stress measurement values;

[0007] S2, extracting a plurality of acoustic features from the baseline acoustic signal collected from S1 to generate a baseline feature vector and store;

[0008] S3, extracting original acoustic feature values from the monitoring acoustic signal collected from S1 each time, and correcting the original acoustic feature values to generate a monitoring feature vector;

[0009] S4, inputting the monitoring feature vector generated by S3 into an AI prediction model to obtain a wear state prediction result of the building material;

[0010] The wear state prediction result includes a predicted wear amount And the remaining life;

[0011] S5, comparing the wear state prediction result obtained by S4 with the measured wear amount obtained by a measurement method, and adaptively correcting the AI prediction model and the feature vector according to the comparison result;

[0012] S6, based on the deviation of the monitoring feature vector generated by S3 relative to the current baseline feature vector, performing wear warning and abnormal area positioning.

[0013] Preferably, the sensor array is a piezoelectric ceramic transducer array, which serves as both a sound wave exciter to emit sound waves and a sensor to receive sound wave signals.

[0014] Preferably, the plurality of acoustic features includes sound speed, attenuation coefficient, center frequency, nonlinear acoustic parameter, and scattering intensity parameter.

[0015] Preferably, the nonlinear acoustic parameter is a nonlinear acoustic coefficient, which is calculated by the following formula: ,

[0016] Wherein, is the nonlinear acoustic coefficient, f is the fundamental frequency of the sound wave, c is the sound speed, x is the sound wave propagation distance, is the amplitude of the received signal fundamental wave, is the amplitude of the received signal second harmonic wave.

[0017] Preferably, the original acoustic feature values are corrected, and the original acoustic feature values extracted from the acoustic signal are corrected using the following compensation formula, specifically: ,

[0018] Wherein, is the corrected acoustic feature value, is the original acoustic feature value, are temperature, humidity and stress measurement values respectively, is a reference reference value, and α, β, γ are the corresponding temperature, humidity and stress compensation coefficients.

[0019] Preferably, the AI prediction model is a bidirectional long short-term memory network model based on an attention mechanism, which takes a historical original acoustic feature value sequence as input, calculates feature importance through attention weight, and outputs corresponding historical wear state data.

[0020] Preferably, the AI prediction model outputs a prediction confidence index at the same time, and the calculation formula is: ,

[0021] Wherein, C is the confidence, is the standard deviation of the wear state prediction result, is the mean of the wear state prediction result.

[0022] Preferably, the specific steps of adaptive correction are: calculating the deviation δ between the prediction result and the actual result, and the calculation formula is: ,

[0023] Wherein, predicted wear amount, is the measured wear amount, is the minimum wear amount threshold; when is greater than the set threshold, then the AI prediction model is incrementally learned using the current monitoring feature vector and the measured wear amount data, and the current monitoring feature vector is updated to a new baseline feature vector.

[0024] Preferably, the condition for wear warning is that the change amount of the nonlinear acoustic parameter or the change amount of the scattering intensity parameter contained in the monitoring feature vector exceeds the respective dynamic threshold, and the dynamic threshold is determined according to the following formula: ,

[0025] Wherein, μ is the mean of the historical data of the feature parameter, σ is the standard deviation, and k is an adaptive adjustment coefficient, and the value range is 2-3.

[0026] Preferably, the abnormal area positioning is realized by analyzing the feature difference distribution of the acoustic signals collected by the sensing nodes at different spatial positions in the sensing array, and specifically adopting a tomographic imaging algorithm based on sound velocity change, and the following regularization optimization problem is solved: ,

[0027] Wherein, L is a ray path matrix, m is a sound velocity disturbance distribution to be solved, t is a sound wave travel time change vector, and λ is a regularization parameter.

[0028] Technical effects and advantages of the present application:

[0029] 1、The application can accurately understand the evolution of the microstructure inside the material and improve the comprehensiveness of monitoring by actively exciting sound waves and extracting multi-dimensional acoustic characteristics such as sound velocity and attenuation coefficient, constructing a high-dimensional feature vector, and realizing the transition from single parameter measurement to multi-dimensional fusion analysis.

[0030] 2、The application relies on a piezoelectric ceramic transducer array and a supporting software and hardware system to realize the innovation from offline destructive detection to online non-destructive continuous monitoring, can be monitored for 7x24 hours, and meets the long-term online monitoring demand.

[0031] 3、The application can realize the transition from judging the current state to predicting the remaining life by establishing a dynamic mapping model of "acoustic characteristics-wear amount-remaining life" through a machine learning algorithm, and can also early warning and grasp the best maintenance opportunity.

[0032] 4、The application has dynamic self-calibration capability by periodically calibrating and measuring the deviation degree, online correcting the model and updating the acoustic baseline, can eliminate the error caused by environmental and material batch differences, and ensure the accuracy of long-term monitoring.

[0033] 5、The application can locate the local abnormal wear area by analyzing the acoustic characteristic field, greatly reduce the artificial investigation range, reduce the maintenance labor cost, and improve the problem handling efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the application, and for the basic field technical personnel, other drawings can also be obtained without creative labor on the premise of these drawings.

[0035] Figure 1 It is a method step flow chart.

[0036] Figure 2 It is an adaptive correction flow chart. DETAILED DESCRIPTION

[0037] The technical solutions in the embodiments of the application will be described clearly and completely in the following with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only some embodiments of the application, not all embodiments. Based on the embodiments in the application, all other embodiments obtained by the ordinary skilled in the art without creative labor are within the protection scope of the application.

[0038] Referring to Figure 1 The figure shows a building material wear quality monitoring method based on artificial intelligence, which comprises:

[0039] S1, collecting baseline acoustic signals in initial state and periodic monitoring acoustic signals in use process through the sensor array deployed on the building material, and collecting environmental temperature, humidity and stress measurement values at the same time;

[0040] The embodiment needs to be specifically explained that the deployed sensor array is a piezoelectric ceramic transducer array, which has a dual function of emitting acoustic waves to the building material as an acoustic wave exciter and receiving acoustic wave signals fed back by the building material as a sensor.

[0041] When collecting baseline acoustic signals in initial state, it is necessary to ensure that the building material is in the state of non-use, no wear and tear and complete internal and surface structure, and to ensure that the collection environment is stable, to start the piezoelectric ceramic transducer array to continuously collect acoustic signals until continuous and stable signal data is obtained, which is used as baseline acoustic signals and temporarily stored in the back-end data storage unit.

[0042] When collecting periodic monitoring acoustic signals in use process, the collection period needs to be set according to the use scene and wear characteristics of the building material. For example, the ceramic tile material used for building outer wall has less influence from external environment and slower wear rate, so the monitoring period can be set to once every 72 hours. For the concrete material used for building ground, the monitoring period can be set to once every 24 hours due to frequent daily stepping and faster wear rate. The acoustic wave emission frequency and signal receiving time length of the piezoelectric ceramic transducer array need to be consistent with the collection of baseline acoustic signals to avoid signal comparison deviation caused by equipment parameter difference. The collected periodic monitoring acoustic signals are transmitted to the back-end data storage unit in real time.

[0043] At the same time of collecting the above two types of acoustic signals, environmental temperature, humidity and stress measurement values need to be collected synchronously. The environmental temperature is measured by a sensor with an accuracy of ±0.5℃, the measurement range covers the temperature range of the actual use environment of the building material, which is set to -30℃-90℃. The environmental humidity is measured by a sensor with an accuracy of ±3%RH, the measurement range is set to 0-100%RH. The stress measurement is measured by a sensor with an accuracy of ±0.1MPa, the stress range is determined according to the design bearing capacity of the building material to ensure accurate capture of stress changes in the use process of the material.

[0044] All collected environmental temperature values, humidity values and stress measurement values need to be stored in association with the acoustic signals (baseline acoustic signals or periodic monitoring acoustic signals) at the corresponding time, transmitted to the back-end data storage unit through shielded data transmission cable, the transmission rate is set to be more than 1Mbps to avoid data transmission delay or loss, and to ensure that each set of acoustic signals can correspond to complete environmental and stress data, providing reliable original data support for subsequent feature extraction and wear monitoring.

[0045] S2: Extracting a plurality of acoustic features from the baseline acoustic signal collected in S1 to generate a baseline feature vector and storing;

[0046] S3: Extracting original acoustic feature values from the monitoring acoustic signal collected in S1 each time, and correcting the original acoustic feature values to generate a monitoring feature vector;

[0047] It is particularly pointed out in this embodiment that the plurality of acoustic features include but are not limited to sound velocity, attenuation coefficient, center frequency, nonlinear acoustic parameter and scattering intensity parameter. The above five types of acoustic features are extracted from the baseline acoustic signal collected in S1, and the specific extraction method and correlation are as follows:

[0048] When extracting the sound velocity, the fixed distance between the transmitting node and the receiving node of the piezoelectric ceramic transducer array in S1 is taken as the sound wave propagation distance x (this distance has been measured and recorded in advance when the array is deployed, for example, for a rectangular building panel, the distance between adjacent sensing nodes is set to 0.6 meters), and the signal acquisition system captures the propagation time difference of the sound wave from the transmitting node to the receiving node, and the sound velocity value is calculated according to:

[0049] Wherein, c is the sound velocity value, x is the propagation distance, and t is the propagation time. The sound velocity value will be one of the basic parameters for subsequent calculation of nonlinear acoustic parameters.

[0050] When extracting the attenuation coefficient, the amplitude of the initial transmitted sound wave is obtained through the transmitting node of the piezoelectric ceramic transducer array, and the amplitude of the received signal fundamental wave is obtained through the receiving node, and the propagation distance x is combined to calculate the attenuation coefficient according to:

[0051] Wherein is the initial transmitted sound wave amplitude, the amplitude of the received signal fundamental wave, which reflects the energy loss of the sound wave in the building material during propagation.

[0052] When extracting the center frequency, the baseline acoustic signal is subjected to fast Fourier transform (FFT) to convert the time domain signal to the frequency domain signal. In the generated frequency domain graph, the frequency value corresponding to the maximum signal amplitude is located, which is the center frequency. The value should be consistent with the preset transmitting frequency of the piezoelectric ceramic transducer array (such as 100 kHz for concrete materials) to verify the effectiveness of the signal acquisition.

[0053] The nonlinear acoustic parameter is the nonlinear acoustic coefficient, which is calculated by the following formula:

[0054] Wherein, is the nonlinear acoustic coefficient, f is the fundamental frequency of the sound wave, c is the sound velocity, and x is the sound wave propagation distance,​​​ is the amplitude of the fundamental wave of the received signal, is the amplitude of the second harmonic wave of the received signal.

[0055] Substitute the obtained parameters for calculation: where is the nonlinear acoustic coefficient to be solved, f is the fundamental frequency of the sound wave, that is, the preset transmission frequency of the piezoelectric ceramic transducer array in S1, such as 80 kHz-120 kHz, the specific value is determined according to the density of the building material, and the higher value is taken for the higher density, c is the sound speed value, and x is the sound wave propagation distance, that is, the fixed interval between the sensing nodes, is the amplitude of the fundamental wave of the received signal, which is measured after the fundamental wave component is separated by filtering the baseline acoustic signal, is the amplitude of the second harmonic wave of the received signal, which is measured after the second harmonic wave component is separated by filtering, and the nonlinear acoustic coefficient can be calculated by substituting the above parameters into the formula .

[0056] When extracting the scattering intensity parameter, the scattering wave component in the baseline acoustic signal except the main wave is analyzed, where the main wave refers to the sound wave directly propagating from the transmission node to the receiving node; the amplitude data of all scattering waves are extracted by signal separation algorithm, the arithmetic mean of these amplitude data is calculated, and the average value is then compared with the amplitude of the main wave to obtain the scattering intensity parameter, which is used to represent the influence of the uniformity of the internal structure of the building material on the sound wave propagation.

[0057] After the extraction of the above five types of acoustic characteristics and the acquisition of specific values, the values of the five types of characteristics are arranged in the fixed order of "sound speed-attenuation coefficient-central frequency-nonlinear acoustic parameter-scattering intensity parameter" to form a one-dimensional baseline feature vector. Then, the baseline feature vector is stored in the back-end data storage unit, and the corresponding baseline acoustic signal collected in S1, the synchronous collected environmental temperature value, humidity value and stress measurement value are associated and marked (for example, named by combining time stamp and material number) during storage, so that the corresponding original collected data can be quickly matched when the baseline feature vector is called in the subsequent steps, providing an accurate benchmark for the generation and deviation comparison of the monitoring feature vector.

[0058] Correct the original acoustic characteristic value, and correct the original acoustic characteristic value extracted from the acoustic signal by using the following compensation formula, which is: ,

[0059] wherein, is the corrected acoustic characteristic value, is the original acoustic characteristic value, are the temperature, humidity and stress measurement values respectively, is the reference benchmark value, and α, β and γ are the corresponding temperature, humidity and stress compensation coefficients.

[0060] Take the correction of attenuation coefficient as an example: if the original attenuation coefficient in a certain monitoring is 5dB / m, the current temperature T = 25℃, the humidity H = 55%RH, the stress = 0.4MPa, and the formula is calculated as follows: First, calculate the deviation of environmental parameters from the reference value:

[0061] = 25-20 = 5℃, = 55-45 = 10%RH, = 0.4-0.1 = 0.3MPa; Second, calculate the sum of compensation terms:

[0062] = 0.0209; Third, calculate the corrected attenuation coefficient:

[0063] = 5.1045dB / m.

[0064] The correction process of other acoustic features (such as sound speed, nonlinear acoustic parameters, etc.) is consistent with this, only the corresponding compensation coefficient needs to be adjusted according to the feature type (such as the temperature compensation coefficient of sound speed α = 0.001 / ℃), and finally all the corrected feature values are obtained to generate the monitoring feature vector.

[0065] S4: input the monitoring feature vector generated by S3 into the AI prediction model to obtain the wear state prediction result of the building material;

[0066] The wear state prediction result includes the predicted wear amount and the remaining life;

[0067] It should be noted that the AI prediction model is a bidirectional long short-term memory network model based on attention mechanism, which takes the historical sequence of original acoustic feature values as input, calculates the feature importance through attention weight, and outputs the corresponding historical wear state data.

[0068] At the same time, the AI prediction model outputs the prediction confidence index, and its calculation formula is: ,

[0069] Where C is the confidence, is the standard deviation of the wear state prediction result, is the mean of the wear state prediction result.

[0070] The specific implementation details of the AI prediction model are as follows:

[0071] ​​Firstly, the composition of the model input data is determined. The "historical original acoustic characteristic value sequence" refers to the time series data sequence composed of the original acoustic characteristic values of the building material from the start of use, collected by S1 and extracted by S3 in chronological order. The sequence length needs to be determined in combination with the wear rate of the building material and the monitoring period. For example, for building floor concrete materials that are frequently stepped on and have a fast wear rate, the monitoring period is 24 hours, and the sequence length can be set to 30 groups, i.e., containing nearly 30 times of collected original acoustic characteristic values. For building external wall ceramic tile materials with a slow wear rate, the monitoring period is 72 hours, and the sequence length can be set to 20 groups, i.e., containing nearly 20 times of collected original acoustic characteristic values, to ensure that the sequence can cover the typical stage characteristics of material wear.

[0072] Further, the model is composed of an attention weight calculation module and a bidirectional long short-term memory network module. The attention weight calculation module assigns different weights to different time and different types of characteristic values in the input historical original acoustic characteristic value sequence. For example, when the material wear enters the middle stage, the change amplitude of nonlinear acoustic parameters and scattering intensity parameters will be significantly larger than other features. The module will automatically assign higher attention weights to these two types of parameters by calculating the correlation coefficient of the features and the wear state, such as weight values of 0.3-0.4, while assigning lower weights to features with relatively flat changes, such as weight values of 0.1-0.2, to highlight the contribution of key features to wear state prediction and improve prediction accuracy. The bidirectional long short-term memory network module includes a forward LSTM layer and a backward LSTM layer. The forward LSTM layer transmits information from the start time of the sequence to the back, capturing the forward time dependence of historical data, such as the cumulative influence of early wear rate on current wear amount. The backward LSTM layer transmits information from the end time of the sequence to the front, capturing the reverse time dependence of historical data (such as the feedback correlation of current wear trend to early wear regularity). The outputs of the two layers are fused by concatenation to form a feature representation containing complete time sequence information.

[0073] Finally, before the model is formally put into use, it needs to be trained with historical monitoring data of the same type of building materials. The training input is the historical original acoustic feature value sequence of the material, and the training output is the historical wear state data, which is composed of the measured wear amount in S5 and the remaining life reference value labeled by artificial according to the material scrap wear threshold. Through the iterative adjustment of the model parameters by the back propagation algorithm, the error between the historical wear state data output by the model and the actual data is reduced to below the preset threshold (such as less than 5%). When the monitoring feature vector generated by S3 is input into the trained model, the model will first supplement the monitoring feature vector to the end of the historical original acoustic feature value sequence to form an updated complete input sequence, and then through the collaborative processing of the attention weight calculation module and the bidirectional long short-term memory network module, finally output the wear state prediction result of the building material. The predicted wear amount is in millimeters (mm) and reflects the current cumulative wear degree of the material. The remaining life is in days and reflects the time the material can still be used normally from the current state to reach the preset scrap wear threshold, such as the concrete material scrap threshold set to 5 mm.

[0074] At the same time, the standard deviation of the prediction result obtained by calculating the prediction confidence index is the mean of the wear state prediction result. For example, the predicted wear amounts of 5 times are 2.1 mm, 2.3 mm, 2.2 mm, 2.4 mm, and 2.0 mm, and the calculation is = 2.2 mm, ≈ 0.141 mm, and the formula gives C = 1 - (0.141 / 2.2) ≈ 1 - 0.064 = 0.936, i.e. the confidence is 93.6%.

[0075] The confidence index, together with the predicted wear amount and the remaining life, is the wear state prediction result. When the confidence is less than 80%, the system will prompt that manual detection needs to be combined for further confirmation.

[0076] Referring to Figure 2 , the predicted wear amount output by the AI mapping model is compared with the actual wear amount obtained by the actual measurement method, and the deviation degree δ is calculated. It is judged whether the deviation degree δ is greater than the set threshold. If the judgment is yes, the model updating mechanism is triggered, the current monitoring feature vector and the actual wear amount data are used for incremental learning of the AI mapping model, and at the same time the current acoustic feature vector is updated to a new theoretical acoustic baseline, so as to realize the collaborative self-adaptive correction of the model and the baseline. If the judgment is no, the AI mapping model and the theoretical acoustic baseline remain unchanged. This closed-loop feedback mechanism ensures that the monitoring system can optimize itself as the material state changes, significantly improving the accuracy and reliability of long-term monitoring.

[0077] S5: comparing the wear state prediction result obtained in S4 with the measured wear amount obtained by measurement, and adaptively correcting the AI prediction model and the feature vector according to the comparison result;

[0078] It is particularly pointed out in this embodiment that first, the measurement method of the measured wear amount is determined: a high-precision laser profiler is used to measure the wear area of the building material, and the measurement position needs to completely correspond to the monitoring area of the sensor array in S1, so as to ensure that the measured data and the predicted data are for the same monitoring object; the measurement accuracy of the laser profiler is set to ±0.01 mm, and 5-8 measurement points need to be uniformly selected in the monitoring area each time, and the arithmetic mean of the wear amounts of the measurement points is taken as the final measured wear amount , in millimeters.

[0079] The specific steps of adaptive correction are: calculating the deviation δ between the prediction result and the actual result, and the calculation formula is: ,

[0080] wherein, is the predicted wear amount, is the measured wear amount, is the minimum wear threshold; when is greater than the set threshold, then the AI prediction model is incrementally learned using the current monitoring feature vector and the measured wear amount data, and the current monitoring feature vector is updated to a new baseline feature vector.

[0081] The specific implementation process of adaptive correction is as follows:

[0082] First, the deviation δ is calculated . Wherein, is the predicted wear amount in the wear state prediction result output by S4, and the unit is consistent with (mm); is the minimum wear threshold, which needs to be determined according to the type of building material and the design use standard, for example, for building floor concrete material, considering its daily wear demand, 0.5 mm is set; for building outer wall ceramic tile material, because the wear resistance is low, 0.2 mm is set. The specific numerical value of , , is substituted into the formula to calculate the specific numerical value of the deviation .

[0083] Second step, set the deviation judgment threshold. Combined with the precision requirement of building material wear monitoring, the deviation judgment threshold is set to 0.1 (i.e. allowing deviation within 10%); if the calculated δ≤0.1, it indicates that the prediction accuracy of the AI prediction model meets the requirements, and there is no need to modify the model and feature vector, only the prediction data, measured data and monitoring feature vector of this time are associated and stored; if δ>0.1, the adaptive correction process needs to be started.

[0084] Third step, perform incremental learning of the AI prediction model. The current monitoring feature vector generated in S3 and the measured wear amount of this time form a new training sample, which is added to the incremental training data set of the AI prediction model; the small batch gradient descent method is used to update the model parameters, the training batch size is set to 16, the learning rate is set to 0.001, and the iteration number is set to 50 times, so that the model can quickly adapt to the change of the wear characteristics of the current material, and there is no need to retrain the entire model, which improves the correction efficiency.

[0085] Fourth step, update the baseline feature vector. The current monitoring feature vector used for incremental learning is directly updated as the new baseline feature vector, replacing the original stored baseline feature vector; after updating, the effective time of the new baseline, the corresponding measured wear amount and the deviation degree need to be marked in the back-end data storage unit, so that the latest baseline feature vector can be used as the reference for comparison and analysis when generating the monitoring feature vector in S3 in the future, and the monitoring deviation caused by the lag of the baseline is avoided.

[0086] S6: Based on the deviation of the monitoring feature vector generated in S3 relative to the current baseline feature vector, wear warning and abnormal area positioning are performed.

[0087] The condition for wear warning is that the change amount of the nonlinear acoustic parameter or the change amount of the scattering intensity parameter contained in the monitoring feature vector exceeds the respective dynamic threshold value, and the dynamic threshold value is determined according to the following formula:

[0088] wherein μ is the mean value of the historical data of the feature parameter, σ is the standard deviation, and k is an adaptive adjustment coefficient, with a value range of 2-3.

[0089] It needs to be specifically explained in this embodiment that the specific process of wear warning is as follows: first, for the nonlinear acoustic parameter and the scattering intensity parameter, the feature values corrected by S3 in the past 20 times (corresponding to 20 monitoring periods) are collected respectively, the historical mean value μ and the standard deviation σ of the two types of parameters are calculated, and then k value is adjusted according to environmental conditions, k=2 in indoor dry environment and k=3 in outdoor rainy environment. Taking indoor environment as an example, the dynamic threshold value of the nonlinear acoustic parameter is calculated to be 1.5×10 -6 ​, the dynamic threshold of the scattering intensity parameter is 0.36; finally, the deviation amount of the nonlinear acoustic parameter in the monitoring feature vector and the deviation amount of the scattering intensity parameter are calculated, if any deviation amount exceeds the corresponding dynamic threshold, wear early warning is triggered, and the early warning information includes the deviation value, the dynamic threshold and the monitoring time.

[0090] The abnormal area positioning is realized by analyzing the feature difference distribution of the acoustic signals collected by the sensing nodes at different spatial positions in the sensing array, specifically, a tomography algorithm based on the change of the sound velocity is adopted, and the following regularization optimization problem is solved: ,

[0091] Wherein, L is a ray path matrix, m is a sound velocity disturbance distribution to be solved, t is a sound wave travel time change vector, and lambda is a regularization parameter.

[0092] The embodiment is explained as follows: first, the ray path matrix L, the sound wave travel time change vector t and the regularization parameter lambda are constructed - L is constructed based on the node coordinates of the sensing array, such as 8 nodes, the coordinates are (x1, y1) to (x8, y8), including 28 ray paths between all nodes, which is a 28 row x 100 column matrix (the material surface is divided into 100 grids), if the path passes through the i-th grid, the corresponding element in L is 1, otherwise it is 0; t is a 28x1 vector, where each element is the difference between the current travel time of the i-th path and the baseline travel time; lambda is set to 0.02 to balance the fitting accuracy and the stability of the solution; then the conjugate gradient method is used to solve the above minimization problem, and the sound velocity disturbance distribution m (100x1 vector, each element corresponds to the sound velocity disturbance value of the grid) is iteratively updated until the objective function converges (the convergence threshold is 10 -4 ); finally, the grids with absolute value exceeding 0.6m / s in m (such as the m values of grids 12-15 are-0.7m / s to-0.8m / s) are screened, and the continuous area composed of these grids is the abnormal area, and its coordinate range (such as (x 12 ,y 12 ) to (x 15 ,y 15 )) is output to complete positioning. Finally, the wear early warning and the abnormal area positioning result are associated and stored, and a visual report is generated to provide a basis for maintenance decision.

[0093] Secondly: the present application discloses the structure involved in the embodiment of the present application, other structures can refer to the usual design, and in the case of no conflict, the same embodiment and different embodiments of the present application can be combined with each other;

[0094] Finally: the above only for the preferred embodiments of the present application, and not for limiting the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application, should be included in the scope of protection of the present application.

Claims

1. An artificial intelligence-based building material wear quality monitoring method, characterized by, The method comprises the following steps: S1, collecting baseline acoustic signals in an initial state and periodic monitoring acoustic signals in use through a sensing array deployed on the building material, while collecting environmental temperature, humidity and stress measurement values; S2, extracting a plurality of acoustic features from the baseline acoustic signals collected in S1 to generate a baseline feature vector and store it; S3, extracting original acoustic feature values from the monitoring acoustic signals collected each time in S1, and correcting the original acoustic feature values to generate a monitoring feature vector; S4, inputting the monitoring feature vector generated in S3 into an AI prediction model to obtain a wear state prediction result of the building material; The AI prediction model in S4 is a bidirectional long short-term memory network model based on an attention mechanism, which takes a historical sequence of original acoustic feature values as input, calculates feature importance through attention weight, and outputs corresponding historical wear state data; The wear state prediction result includes a predicted wear amount and a remaining life. The AI prediction model outputs a prediction confidence index at the same time, and its calculation formula is: ， wherein C is a confidence level, is a standard deviation of the wear state prediction result, is a mean value of the wear state prediction result; S5, comparing the wear state prediction result obtained in S4 with the actual wear amount obtained by actual measurement at regular intervals, and adaptively correcting the AI prediction model and the baseline feature vector according to the comparison result; The specific steps of adaptive correction in S5 are: calculating the deviation δ between the prediction result and the actual result of the wear state, and the calculation formula is: ， wherein, is a predicted wear amount, is a measured wear amount, is a minimum wear threshold; when is greater than a set threshold, then the AI prediction model is incrementally learned with the current monitoring feature vector and the measured wear amount data, and simultaneously the current monitoring feature vector is updated as a new baseline feature vector; S6, based on the deviation of the monitoring feature vector generated in S3 relative to the current baseline feature vector, wear warning and abnormal area positioning are performed; The condition for performing wear warning in S6 is that the change amount of the nonlinear acoustic parameter or the change amount of the scattering intensity parameter contained in the monitoring feature vector exceeds the respective dynamic threshold value, and the dynamic threshold value is calculated according to the following formula: Sure, Where μ is the mean of the historical data of the nonlinear acoustic parameter or the scattering intensity parameter, σ is the standard deviation, and k is an adaptive adjustment coefficient with a value range of 2-3.

2. The method of claim 1, wherein the method is based on artificial intelligence. The sensing array is a piezoelectric ceramic transducer array, which not only serves as a sound wave exciter to emit sound waves, but also serves as a sensor to receive sound wave signals.

3. The method of claim 1, wherein the method is based on artificial intelligence. The plurality of acoustic features include sound velocity, attenuation coefficient, center frequency, nonlinear acoustic parameter and scattering intensity parameter.

4. The method of claim 3, wherein the method is based on artificial intelligence. The nonlinear acoustic parameter is a nonlinear acoustic coefficient, which is calculated by the following formula: ， wherein, is a nonlinear acoustic coefficient, f is the fundamental frequency of the acoustic wave, c is the sound speed, x is the acoustic wave propagation distance, is the received signal fundamental amplitude, is the received signal second harmonic amplitude.

5. The method of claim 1, wherein the method is based on artificial intelligence. The correction of the original acoustic feature value uses the following compensation formula to correct the original acoustic feature value extracted from the acoustic signal, specifically: ， wherein, is the corrected acoustic characteristic value, is the original acoustic characteristic value, are temperature, humidity and stress measurement values, respectively, is a reference reference value, and a, β, γ are the corresponding temperature, humidity and stress compensation coefficients.

6. The method of claim 1, wherein the method is based on artificial intelligence. The abnormal area positioning in S6 is realized by analyzing the feature difference distribution of the acoustic signals collected by the sensing nodes at different spatial positions in the sensing array, and a tomographic imaging algorithm based on sound velocity change is adopted, and the following regularization optimization problem is solved: ， Where L is a ray path matrix, m is the sound velocity disturbance distribution to be solved, t is the sound wave travel time change vector, and λ is the regularization parameter.

Images