Video AI monitoring system suitable for bridge deck crack / fracture monitoring

Abstract

The invention discloses a video AI monitoring system suitable for bridge deck crack / fracture monitoring, and relates to the technical field of civil engineering structure health monitoring. According to the system, through a multi-source data fusion technology, an intelligent sensing chip, three-dimensional laser scanning, electromagnetic detection and high-definition camera equipment are integrated, and bridge floor structure physical data and crack characteristic data are collected. The core is that the genetic algorithm is adopted, the omission ratio and the monitoring cost are taken as optimization targets, the optimal monitoring frequency and early warning threshold value are dynamically calculated for each partition, and the balance of safety and economical efficiency is realized. The system introduces environmental influence factors and dynamically corrects an early warning threshold value through a deep learning model, so that the false alarm rate is reduced. In addition, the system has real-time feedback and adaptive adjustment capabilities, can identify key areas according to crack development data, and automatically adjusts a monitoring strategy to form closed-loop control. The method improves the accuracy, response speed and decision-making efficiency of crack monitoring, and has remarkable effects on prolonging the service life of the bridge and reducing the operation and maintenance cost.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of civil engineering structure health monitoring, and particularly relates to a video AI monitoring system suitable for bridge deck crack / fracture monitoring. BACKGROUND

[0002] As a main form of bridge structure damage, the early identification and dynamic monitoring of bridge deck cracks are the core topics in the field of civil engineering structure health monitoring. With the increasing heavy traffic load, the increasing service life of bridges and the intensifying influence of extreme environment, the initiation and expansion of cracks may cause major safety accidents such as bridge deck fracture, thus putting forward strict requirements on the accuracy, response speed and cost-effectiveness of the monitoring system. However, the existing bridge deck crack monitoring technology still has many limitations and cannot meet the actual engineering needs. The monitoring parameter setting is rigid, and the safety and cost are unbalanced: the monitoring frequency and early warning threshold of the existing system are mostly based on experience and static setting, and no quantitative optimization mechanism is established. If the monitoring frequency is blindly increased for the purpose of safety, the cost of equipment power consumption, data storage and manual operation and maintenance will increase; if the monitoring frequency or the threshold is reduced to control the cost, the missed detection rate will be significantly increased. In addition, the threshold setting does not consider the regional load difference, further exacerbating the contradiction between excessive monitoring and insufficient monitoring, and it is difficult to balance safety and economy.

[0003] The influence of environment is not considered, and the false alarm and missed alarm rates are high: the monitoring of bridge deck cracks is easily disturbed by environmental factors such as temperature change, humidity fluctuation and rainfall. Due to the lack of dynamic correction mechanism of environmental factors, the early warning threshold of the existing system is fixed for a long time, resulting in high false alarm rate and missed alarm rate, affecting the reliability of operation and maintenance decision.

[0004] The dynamic feedback mechanism is lacking, and the monitoring of key areas is lagging: the existing monitoring mostly adopts a static mode of fixed scheme + regular review, and lacks the ability of real-time feedback of crack development data and adaptive adjustment of strategy. When the local area cracks accelerate expansion, the system cannot identify and strengthen the monitoring of the area in time, resulting in lagging response of the monitoring of key risk areas; at the same time, no regional priority division mechanism based on crack stability is established, and the same monitoring standard is used for high-risk areas and low-risk areas, causing waste of monitoring resources and reducing the overall early warning efficiency. To overcome the above limitations, the present application proposes a bridge deck crack / fracture monitoring system based on video AI and multi-source data fusion. SUMMARY

[0005] In order to overcome the shortcomings and deficiencies of the prior art, the application adopts the following technical solutions: The video AI monitoring system suitable for bridge deck crack / fracture monitoring comprises: The data acquisition module is configured to acquire the physical data of the bridge deck structure, divide the bridge deck into Q monitoring areas, and acquire crack characteristic data corresponding to the Q monitoring areas on the bridge deck. The parameter analysis module is configured to analyze the physical data of the bridge deck structure and the crack characteristic data corresponding to the Q monitoring areas, and obtain the required monitoring frequency and early warning threshold of each area. The dynamic monitoring module is configured to control the monitoring equipment to dynamically monitor the bridge deck according to the required monitoring frequency and early warning threshold of the Q monitoring areas, acquire crack development data of the Q monitoring areas during the monitoring process, and provide real-time feedback. The instruction generation and adjustment module is configured to determine whether to generate a maintenance adjustment instruction according to the real-time feedback of the crack development data of the Q monitoring areas, mark the area for which the adjustment instruction is generated as a key monitoring area, analyze the crack development data of the key monitoring area, calculate the adjusted monitoring parameters, and control the monitoring frequency of the key monitoring area in the monitoring equipment to be adjusted to an optimized frequency.

[0006] Preferably, the system workflow includes the following steps: The physical data of the bridge deck structure is acquired, the bridge deck is divided into Q monitoring areas, and crack characteristic data corresponding to the Q monitoring areas on the bridge deck is acquired. The physical data of the bridge deck structure includes material strength data and structure parameters. The crack characteristic data includes crack length, width, density, and expansion rate. The physical data of the bridge deck structure and the crack characteristic data corresponding to the Q areas are analyzed to obtain the required monitoring frequency and early warning threshold of each area. The monitoring equipment is controlled to dynamically monitor the bridge deck according to the required monitoring frequency and early warning threshold of the Q areas, and crack development data of the Q areas is acquired during the monitoring process and provided with real-time feedback. According to the real-time feedback of the crack development data of the Q areas, it is determined whether to generate a maintenance adjustment instruction. The area for which the adjustment instruction is generated is marked as a key monitoring area. The crack development data of the key monitoring area is analyzed, the adjusted monitoring parameters are calculated, and the monitoring frequency of the key monitoring area in the monitoring equipment is adjusted to an optimized frequency.

[0007] Preferably, the material strength data includes concrete compressive strength, steel tensile strength and interface bonding strength, and the acquisition method is: through the embedded intelligent sensing chip on the bridge deck, the material strength reference data stored in the chip is read by using wireless radio frequency technology, the reference data is corrected by combining with the on-site core sampling detection results through a data fusion algorithm to obtain the actual strength data, the chip ID corresponds to the bridge deck position one by one to form a structured database; preferably, the fine acquisition method of the structure parameter includes: using a ground three-dimensional laser scanning device to perform panoramic scanning on the bridge deck to obtain point cloud data, extracting the bridge deck structure contour through a point cloud segmentation algorithm, and calculating the actual thickness of the bridge deck; adopting electromagnetic induction detection technology to identify the distribution of internal steel bars of the bridge deck, calculating the ratio of the steel bar cross-sectional area in unit area to the bridge deck area to obtain the reinforcement ratio; analyzing the texture characteristics of the bridge deck pavement layer through image recognition technology, combining with the typical structure samples in the database, using a support vector machine classification algorithm to identify the pavement layer structure type, taking thickness, reinforcement ratio and structure type as the core indexes of the structure parameter, and constructing a structured parameter matrix.

[0008] Preferably, the crack feature data acquisition method is: using a combination system of a high-definition camera device and a laser range finder to scan and image the surface of the bridge deck; marking n feature points in each monitoring area, n is an integer greater than 1, measuring the relative displacement and pixel difference between the n feature points in each area through image recognition technology; calculating the crack feature mean value corresponding to each area, and calculating the standard deviation of the feature data of the n feature points in each area, taking the standard deviation as the crack stability coefficient of each area.

[0009] Preferably, the step of obtaining the monitoring frequency and early warning threshold required by each region comprises: randomly selecting a region from the Q regions which is not marked as an analyzed region and marking it as an analyzed region; encoding the monitoring frequency and early warning threshold to obtain a combination of decision variables and constructing an initial solution group; encoding the monitoring frequency as f, the early warning threshold as T, and the combination of decision variables as (f, T), obtaining the frequency range [F1, F2] and the threshold range [T1, T2] that the monitoring system can support, and randomly generating N combinations of decision variables to form the initial solution group; determining the fitness function; the method for obtaining the missed detection rate and the cost coefficient is: inputting the bridge deck structure physical data, the region crack feature data and the combination of decision variables as input parameters into the trained monitoring effectiveness prediction model to predict the corresponding missed detection rate and cost coefficient, the monitoring effectiveness prediction model being a deep neural network model; performing selection operation on the combinations of decision variables in the solution group by combining the tournament selection method with the elite reservation strategy; performing crossover operation on the combinations of decision variables in the solution group, randomly selecting U pairs of combinations of decision variables in the solution group for crossover operation, using arithmetic crossover for the monitoring frequency and single-point crossover for the early warning threshold, generating U new combinations, calculating the fitness of the new combinations, and replacing the U worst individuals in the solution group; performing mutation operation on the combinations of decision variables in the solution group, presetting a mutation probability G, performing Gaussian mutation on the monitoring frequency of the individuals in the solution group, and performing uniform mutation on the early warning threshold; obtaining a new solution group, presetting an iteration number L and a fitness threshold M, and cyclically performing selection, crossover and mutation operations until the iteration number of the new solution group reaches L or there exists a combination of decision variables in the new solution group whose fitness is less than or equal to M, and the cycle ends, the monitoring frequency and early warning threshold of the combination of decision variables with the minimum fitness in the new solution group are obtained as the optimal parameters of the analyzed region; and the above steps are cycled until all Q regions are marked as analyzed regions, the cycle ends, and the monitoring frequency and early warning threshold required by all regions are obtained.

[0010] Preferably, the crack development data is the change amount of crack features per unit time, a preset collection interval Δt is set, the development data is collected at intervals, and the method comprises: comparing the position changes of the same feature points in consecutive frames of images through video image difference technology; setting a reference coordinate system in each monitoring region to calculate the displacement change amount of the feature points in the X and Y axis directions; combining laser ranging data to obtain the depth change in the Z axis direction; and synthesizing the three-dimensional change amount to obtain the crack development data.

[0011] Preferably, the method for determining whether to generate a maintenance adjustment instruction comprises: calculating the development threshold of the Q regions, comparing the crack development rate of each region with the corresponding development threshold, and determining whether to generate an adjustment instruction according to the comparison result. If , no adjustment instruction is generated; and if , an adjustment instruction is generated.If yes, a maintenance adjustment instruction is generated; the method for calculating the optimized monitoring frequency comprises: taking the ratio of the crack development rate of the key monitoring area to the development threshold as an adjustment coefficient, and calculating the optimized frequency through a formula.

[0012] Preferably, an environmental impact factor is introduced to dynamically correct the development threshold, the environmental impact factor comprising temperature change amount Delta T, humidity RH and cumulative rainfall R, and the dynamic correction method comprising: dividing the monitoring period into W time periods, each time period lasting for T; standardizing the environmental parameters of each time period to construct an environmental feature vector; inputting the environmental feature vector and the crack development data of the corresponding time period into a trained environmental impact model to predict the environmental correction coefficient of each area, the environmental impact model being a deep neural network model; and the corrected development threshold being calculated through a formula.

[0013] Preferably, the real-time collected crack development data is compared with the corrected development threshold If yes, a reinforcement monitoring instruction is generated to increase the monitoring frequency to 1.5 times of the basic frequency; if no, the original monitoring strategy is maintained.

[0014] In summary, due to the adoption of the above technical solutions, the present application has the following advantages: 1. According to the required monitoring frequency and the early warning threshold of each area, the present application controls the monitoring equipment to dynamically monitor the bridge deck, and real-time collects crack development data and feeds back during the monitoring process. This dynamic monitoring method can flexibly adjust the monitoring intensity according to the actual situation of different areas of the bridge deck, and improves the utilization efficiency of monitoring resources. According to the real-time feedback of the crack development data, the system automatically judges whether to generate a maintenance adjustment instruction, marks the area needing attention as a key monitoring area, analyzes the crack development data thereof, calculates the adjusted monitoring parameters, and optimizes the monitoring frequency of the key monitoring area. This intelligent instruction generation and adjustment mechanism can timely discover abnormal conditions of crack development and quickly respond, ensuring that the monitoring of the key area is more accurate and timely, and effectively preventing the further development of cracks and other serious problems such as rupture.

[0015] ​​2、The application introduces environmental influence factors such as temperature change amount, humidity and cumulative rainfall, and through considering these factors, the system can more truly reflect the actual situation of crack development; by dividing the monitoring period into multiple time periods, standardizing the environmental parameters of each time period, constructing an environmental feature vector, and inputting it and the crack development data of the corresponding time period into the trained environmental influence model, the environmental correction coefficient of each region is predicted, and then the development threshold is dynamically corrected. This dynamic correction mechanism enables the system to timely adjust the judgment standard according to environmental changes, improves the adaptability and reliability of the system under different environmental conditions, and ensures the accuracy and effectiveness of the monitoring results. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, a brief introduction will be given below to the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0017] Figure 1 A module block diagram of the video AI monitoring system for bridge deck crack / fracture monitoring according to the present application is shown; Figure 2 A work flow diagram of the system according to the present application is shown; Figure 3 A flow chart of step two according to the present application is shown. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0019] In addition, the described features, structures or characteristics can be combined in any suitable manner in one or more example embodiments. In the following description, many specific details are provided to give a sufficient understanding of the example embodiments of the present disclosure. However, those skilled in the art will realize that the technical solutions of the present disclosure can be practiced without one or more of the specific details, or other methods, components, steps, etc. can be used. In other cases, well-known structures, methods, implementations or operations are not shown or described in detail to avoid obscuring the aspects of the present disclosure.

[0020] Embodiment 1: Reference Figure 1As shown, the video AI monitoring system for bridge deck crack / fracture monitoring of the embodiment comprises: The data acquisition module is used to acquire the structural physical data of the bridge deck, divide the bridge deck into Q monitoring areas, and acquire the crack feature data corresponding to the Q monitoring areas on the bridge deck. The parameter analysis module is used to analyze the structural physical data and the crack feature data corresponding to the Q areas, and obtain the required monitoring frequency and early warning threshold of each area. The dynamic monitoring module is used to control the monitoring equipment to dynamically monitor the bridge deck according to the required monitoring frequency and early warning threshold of the Q areas, and collect the crack development data of the Q areas during the monitoring process and provide real-time feedback. The instruction generation and adjustment module is used to determine whether to generate a maintenance adjustment instruction according to the real-time feedback of the crack development data of the Q areas, mark the area where the adjustment instruction is generated as a key monitoring area, analyze the crack development data of the key monitoring area, calculate the adjusted monitoring parameters, and control the monitoring frequency of the key monitoring area in the monitoring equipment to be adjusted to the optimized frequency.

[0021] The embodiment has the following beneficial effects: through multi-source data fusion (sensing chip, laser scanning, electromagnetic detection), high-precision acquisition of bridge deck structure parameters and crack features is realized; the monitoring frequency and early warning threshold are optimized by genetic algorithm, safety and cost efficiency are considered; the threshold is dynamically corrected by introducing environmental influence factors, which significantly reduces the false alarm rate; the real-time feedback and self-adaptive adjustment mechanism can accurately locate the key area and strengthen the monitoring, improve the crack early warning response speed and maintenance decision efficiency, effectively prolong the service life of the bridge, and reduce the operation and maintenance cost.

[0022] Embodiment 2: Referring to Figure 2 As shown, the video AI monitoring system for bridge deck crack / fracture monitoring of the embodiment has a work flow comprising the following steps: Step 1: Acquire the structural physical data of the bridge deck; divide the bridge deck into Q monitoring areas; and acquire the crack feature data corresponding to the Q monitoring areas on the bridge deck.

[0023] The bridge deck structural physical data includes material strength data and structural parameters; the material strength data includes concrete compressive strength, steel tensile strength, and interface bonding strength; the structural parameters include bridge deck thickness, reinforcement ratio, and pavement layer structure type; the method for obtaining the material strength data includes: through the intelligent sensing chip pre-embedded in the bridge deck, the material strength benchmark data stored in the chip is read by using wireless radio frequency technology; the benchmark data is corrected by a data fusion algorithm in combination with the on-site core sampling detection results to obtain the actual strength data; the chip ID corresponds to the bridge deck position one by one to form a structured database.

[0024] The fine acquisition method of the structure parameter comprises: panoramic scanning of the bridge deck by using a ground three-dimensional laser scanning device to obtain point cloud data; extracting the bridge deck structure contour by a point cloud segmentation algorithm, and calculating the actual thickness of the bridge deck ; wherein, is the total scanning thickness; is the pavement layer thickness; adopting electromagnetic induction detection technology, identifying the internal steel bar distribution of the bridge deck, calculating the ratio of the steel bar cross-sectional area in unit area to the bridge deck area to obtain the reinforcement ratio ; wherein, is the total steel area; is the calculated area of the bridge deck; analyzing the texture characteristics of the bridge deck pavement layer by image recognition technology, combining the typical structure samples in the database, and using a support vector machine classification algorithm to identify the pavement layer structure type (asphalt concrete, cement concrete or composite structure); taking the thickness, reinforcement ratio and structure type as the core indexes of the structure parameter, a structured parameter matrix is constructed.

[0025] Q monitoring areas are divided according to the structural segmentation and traffic load distribution of the bridge deck, wherein each monitoring area corresponds to an independent image acquisition unit; The crack feature data includes crack length, width, density and expansion rate; the acquisition method of the crack feature data comprises: using a combination system of a high-definition camera device and a laser range finder to scan and image the surface of the bridge deck; marking n feature points in each monitoring area, n being an integer greater than 1, measuring the relative displacement and pixel difference between the n feature points in each area by image recognition technology; calculating the crack feature mean value corresponding to each area, and calculating the feature data standard deviation of the n feature points in each area, taking the standard deviation as the crack stability coefficient of each area; the expression of the crack stability coefficient is: ; wherein, is the monitoring area number ( ), corresponding to the independent area divided according to the structural segmentation and load distribution of the bridge deck; is the feature point number in the area ( ), (ensuring statistical reliability and covering the main crack propagation direction); is the crack feature value of the jth feature point in the ith area (physical meaning: measured value of crack width / length / expansion rate, unit: mm / m / mm·d⁻¹); is the crack feature mean value of the ith area (calculation method: ; reflecting the average state of the area crack); crack stability coefficient (engineering significance: crack expansion risk is characterized by the dispersion of the feature value, the larger the value, the worse the stability, which needs to be monitored first).

[0026] Step two, analyze the structure physical data and the crack characteristic data corresponding to the Q regions to obtain the required monitoring frequency and early warning threshold of each region.

[0027] Referring to Figure 3 the step of obtaining the required monitoring frequency and early warning threshold of each region includes: S21: randomly selecting a region from the Q regions that is not marked as an analyzed region and marking it as an analyzed region; S22: encoding the monitoring frequency and early warning threshold to obtain a decision variable combination and constructing an initial solution group; encoding the monitoring frequency as f (unit: times / hour) and the early warning threshold as T (unit: mm), the decision variable combination as (f, T), obtaining the frequency range [F1, F2] and the threshold range [T1, T2] that the monitoring system can support; randomly generating N decision variable combinations to form the initial solution group ; S23: determining the fitness function; the expression of the fitness function is: ; wherein, is the genetic algorithm individual number (N ), N is the population size; is the fitness corresponding to the qth decision variable combination; is the missed detection rate (definition: missed crack sample number / total sample number x 100%, dimensionless, target value ≤5%); is the monitoring cost coefficient (definition: comprehensive quantitative value of equipment power consumption + data storage + labor cost, unit: yuan / hour, which needs to be normalized to the [0, 1] interval); , is the weight coefficient (determination basis: set through analytic hierarchy process or historical data regression, which satisfies , the safety priority scenario takes , ); The method for obtaining the missed detection rate and the cost coefficient includes: taking the bridge deck structure physical data, the region crack characteristic data and the decision variable combination as input parameters; inputting them into the trained monitoring efficiency prediction model to predict the corresponding missed detection rate and cost coefficient; the training process of the monitoring efficiency prediction model includes: collecting multiple groups of input parameters corresponding to the missed detection rate and the cost coefficient in advance, converting the input parameters and the corresponding output into feature vectors; taking each group of feature vectors as model input, taking the predicted missed detection rate and cost coefficient as output, and taking the actual value as target, training by minimizing the sum of prediction errors until convergence; the model is a deep neural network model; S24: Selecting operation is performed on the decision variable combination in the solution group; the selecting operation adopts the tournament selection method combined with the elite reservation strategy; k individuals are randomly selected from the solution group for the tournament, and the winner enters the next generation; at the same time, k1 individuals with the best fitness are directly reserved to enter the next generation, so as to ensure that the total number remains unchanged; S25: The decision variable combination in the solution group is subjected to a crossover operation; U pairs of decision variable combinations are randomly selected from the solution group for the crossover operation, the monitoring frequency adopts arithmetic crossover, and the early warning threshold adopts single-point crossover, thereby generating U new combinations; the fitness of the new combinations is calculated, and the U individuals with the worst fitness in the solution group are replaced; S26: Mutation operation is performed on the decision variable combination in the solution group; the preset mutation probability is G, the monitoring frequency of the individual in the solution group is subjected to Gaussian mutation, and the early warning threshold is subjected to uniform mutation, so as to maintain the diversity of the solution group; S27: A new solution group is obtained, the preset iteration number is L, the fitness threshold is M, L is an integer greater than 0, and M is a real number greater than 0; the cycle of S24-S26 is repeated until the iteration number corresponding to the new solution group reaches L or there exists a decision variable combination in the new solution group whose corresponding fitness is less than or equal to M, the cycle is ended, the monitoring frequency and the early warning threshold of the decision variable combination with the minimum fitness in the new solution group are obtained, and are used as the optimal parameters of the analyzed region in S21; S28: The cycle of S21-S27 is repeated until Q regions are all marked as analyzed regions, the cycle is ended, and the required monitoring frequency and early warning threshold of all regions are obtained.

[0028] Step three: according to the required monitoring frequency and early warning threshold of the Q regions, the monitoring equipment is controlled to perform dynamic monitoring on the bridge deck, and crack development data of the Q regions are collected and fed back in real time during the monitoring process; the crack development data is the change amount of the crack feature per unit time; a preset collection interval Δt is set, and the development data is collected at intervals; the method for obtaining the crack development data includes: comparing the position change of the same feature point in the continuous frame image through video image difference technology; a reference coordinate system is set in each monitoring region, and the displacement change amount of the feature point in the X and Y axis directions is calculated; the depth change in the Z axis direction is obtained in combination with the laser ranging data; the crack development data is obtained by synthesizing the three-dimensional change amount; the expression is: ; wherein, , , respectively, are the displacement change amounts in the three-dimensional directions (source: video difference technology to obtain X / Y axis pixel displacement, laser range finder to obtain Z axis depth change, which needs to be converted into actual physical distance, unit: mm); The collection interval (selection basis: matching video frame rate (≥ 30 fps) and laser range finder sampling period to ensure time resolution ≥ 1 second to meet the needs of micro crack expansion monitoring); The crack development rate (physical meaning: three-dimensional expansion of the crack per unit time, unit: mm / h, reflecting the dynamic expansion trend of the crack).

[0029] Step four, according to the real-time feedback of the crack development data of the Q regions, judge whether to generate maintenance adjustment instruction, mark the region which generates adjustment instruction as key monitoring region, analyze the crack development data of the key monitoring region, calculate the adjusted monitoring parameter, control the monitoring frequency of the key monitoring region in the monitoring equipment to be adjusted to the optimized frequency; The method for judging whether to generate maintenance adjustment instruction includes: calculating the development threshold value of the Q regions The crack development rate of each region Respectively compared with the corresponding If , no adjustment instruction is generated; if , maintenance adjustment instruction is generated; the calculation method of the development threshold value includes: ; Among them, The initial early warning threshold value (definition: based on historical data and specification set crack width / length critical value, unit: mm, such as concrete bridge 2mm); The traffic load level of the i-th region (quantization method: through the traffic flow, axle load spectrum data mapping to a numerical value, such as "low = 1, medium = 2, high = 3"); The crack stability coefficient of the i-th region; 、 、 The preset weight coefficient (determination basis: regression analysis or expert experience, satisfying ; The method for calculating the optimized monitoring frequency includes: taking the ratio of the crack development rate of the key monitoring region to the development threshold value as the adjustment coefficient : ; the optimized frequency ; wherein, The optimized frequency of the h-th key region; The initial monitoring frequency; The actual development rate; The corresponding threshold value, , The number of key regions, .

[0030] Introduce an environmental impact factor to dynamically correct the development threshold; the environmental impact factor includes temperature change amount Delta T, humidity RH and cumulative rainfall R; the dynamic correction method includes: dividing the monitoring period into W time periods, each time period lasting (consistent with the collection interval); standardizing the environmental parameters of each time period to construct an environmental feature vector , ; inputting the environmental feature vector and the crack development data of the corresponding time period into the trained environmental impact model to predict the environmental correction coefficient of each region ; the training process of the environmental impact model includes: collecting historical environmental data and contemporaneous crack development data to establish a sample set; training a deep neural network model with environmental parameters as input and environmental impact proportion of crack development rate as output; minimizing prediction bias as the target until the model converges; the corrected development threshold is: ; wherein, is the corrected threshold of the kth time period of the ith region; is the environmental correction coefficient (source: deep neural network model prediction, input environmental feature vector , output is environmental impact proportion, dimensionless); environmental parameters: temperature change amount (unit: DEG C), humidity (unit: %), cumulative rainfall (unit: mm), which needs to be standardized (such as Z-score standardization); supplementary constraints: the corrected threshold (avoid false positives caused by excessive correction), and needs to be verified by historical data, with a prediction bias of less than or equal to 15%.

[0031] Compare the real-time collected crack development data with to rejudge whether to generate adjustment instructions: if , generate a strengthened monitoring instruction to increase the monitoring frequency to 1.5 times the basic frequency; if , maintain the original monitoring strategy.

[0032] The beneficial effects of this embodiment are: ensuring data comprehensiveness through zoned monitoring and multi-source data fusion; balancing missed detection rate and cost by using genetic algorithm to optimize monitoring frequency and threshold; combining three-dimensional monitoring, environmental correction and dynamic adjustment strategy to accurately capture crack development, strengthen monitoring in key areas and effectively improve the accuracy, adaptability and safety of bridge deck crack monitoring.

[0033] The formulas of the present application are dimensionless and the numerical values are calculated, and the preset parameters in the formulas are set by the person skilled in the art according to the actual situation.

[0034] The weight coefficients of the present application are used to measure the degree of influence of different factors or variables on a certain result or decision. The definition of weight coefficient refers to the numerical value assigned to each factor when comparing and evaluating multiple factors to reflect its importance or priority. These weight coefficients can be determined according to specific circumstances and needs, and are usually formulated and confirmed by professionals or relevant stakeholders. By reasonably setting the weight coefficients, the program or system can help make more accurate decisions or predictions.

[0035] The above is only the preferred embodiment of the present application, but the protection scope of the present application is not limited thereto, any skilled person in the art can make equivalent replacement or change within the technical range disclosed by the present application according to the technical scheme and the inventive concept of the present application, which should be covered within the protection scope of the present application.

[0036] The preferred embodiments of the present application disclosed above are only used to help explain the present application. The preferred embodiments do not describe all the details and limit the present application to the specific embodiments. Obviously, many modifications and changes can be made according to the content of the present application. The present application selects and describes these embodiments in order to better explain the principles and practical applications of the present application, so that those skilled in the art can well understand and utilize the present application. The present application is limited by the claims and their entire scope and equivalents.

Claims

1. A video AI monitoring system applicable to bridge deck crack / fracture monitoring, characterized in that, The system includes: Data acquisition module: used to collect physical data of bridge deck structure, divide the bridge deck into Q monitoring areas, and collect crack feature data corresponding to the Q monitoring areas of the bridge deck; Parameter analysis module: used to analyze the physical data of the bridge deck structure and the crack characteristic data corresponding to Q monitoring areas, and obtain the required monitoring frequency and early warning threshold for each area; Dynamic monitoring module: Used to control the monitoring equipment to dynamically monitor the bridge deck according to the required monitoring frequency and early warning threshold of Q monitoring areas, and to collect crack development data of Q monitoring areas and provide real-time feedback during the monitoring process; Command generation and adjustment module: Based on the crack development data of Q monitoring areas in real time, it determines whether to generate a maintenance adjustment command, marks the area where the adjustment command is generated as a key monitoring area, analyzes the crack development data of the key monitoring area, calculates the adjusted monitoring parameters, and controls the monitoring frequency of the key monitoring area in the monitoring equipment to be adjusted to the optimized frequency.

2. The video AI monitoring system for bridge deck crack / fracture monitoring according to claim 1, characterized in that, The system workflow includes the following steps: Physical data of the bridge deck structure were collected, and the bridge deck was divided into Q monitoring areas. Crack characteristic data corresponding to the Q monitoring areas of the bridge deck were collected. The physical data of the bridge deck structure included material strength data and structural parameters. The crack characteristic data included crack length, width, density and propagation rate. Analyze the physical data of the bridge deck structure and the crack characteristic data corresponding to Q regions to obtain the required monitoring frequency and early warning threshold for each region; based on the required monitoring frequency and early warning threshold for Q regions, control the monitoring equipment to dynamically monitor the bridge deck, and collect crack development data of Q regions during the monitoring process and provide real-time feedback. Based on the real-time feedback of crack development data in Q areas, determine whether to generate a maintenance adjustment command. Mark the areas that generate adjustment commands as key monitoring areas, analyze the crack development data in the key monitoring areas, calculate the adjusted monitoring parameters, and control the monitoring frequency of the key monitoring areas in the monitoring equipment to be adjusted to the optimized frequency.

3. The video AI monitoring system for bridge deck crack / fracture monitoring according to claim 2, characterized in that, The method for acquiring crack feature data is as follows: a combination system of high-definition camera equipment and laser rangefinder is used to scan and image the bridge surface; n feature points are marked in each monitoring area, where n is an integer greater than 1; and the relative displacement and pixel difference between the n feature points in each area are measured by image recognition technology. Calculate the mean crack feature value for each region, and calculate the standard deviation of the feature data of n feature points in each region. Use this standard deviation as the crack stability coefficient for each region.

4. The video AI monitoring system for bridge deck crack / fracture monitoring according to claim 2, characterized in that, The steps for obtaining the required monitoring frequency and early warning threshold for each region include: randomly selecting a region from Q regions that was not marked as an analyzed region and marking it as an analyzed region; encoding the monitoring frequency and early warning threshold to obtain a combination of decision variables and constructing an initial solution group; encoding the monitoring frequency as f, the early warning threshold as T, and the combination of decision variables as (f, T), obtaining the frequency range [F1, F2] and threshold range [T1, T2] that the monitoring system can support, and randomly generating N combinations of decision variables to form an initial solution group; determining the fitness function; performing a selection operation on the combination of decision variables in the solution group, using a combination of tournament selection and elite retention strategy; performing cross operations on the combination of decision variables in the solution group, randomly selecting U in the solution group to perform cross operations on the combination of decision variables, using arithmetic cross operations on the monitoring frequency, and performing cross operations on the early warning... The threshold is determined by single-point crossover, generating U new combinations. The fitness of each new combination is calculated, and the U individuals with the worst fitness in the solution group are replaced. Mutation is performed on the decision variable combinations in the solution group, with a preset mutation probability of G. Gaussian mutation is applied to the monitoring frequency of individuals in the solution group, and uniform mutation is applied to the warning threshold. A new solution group is obtained, with a preset iteration count of L and a fitness threshold of M. The selection, crossover, and mutation operations are repeated until the iteration count for the new solution group reaches L or the fitness of a decision variable combination in the new solution group is less than or equal to M. The loop ends when the iteration count reaches L or when the fitness of a decision variable combination in the new solution group is less than or equal to M. The monitoring frequency and warning threshold of the decision variable combination with the lowest fitness in the new solution group are obtained and used as the optimal parameters for the analyzed region. The above steps are repeated until all Q regions are marked as analyzed regions. The loop ends when the required monitoring frequency and warning threshold for all regions are obtained.

5. The video AI monitoring system for bridge deck crack / fracture monitoring according to claim 4, characterized in that, It also includes methods for obtaining the missed detection rate and cost coefficient: the physical data of the bridge deck structure, the regional crack feature data and the combination of decision variables are used as input parameters and input into the trained monitoring effectiveness prediction model to predict the corresponding missed detection rate and cost coefficient. The monitoring effectiveness prediction model is a deep neural network model.

6. The video AI monitoring system for bridge deck crack / fracture monitoring according to claim 2, characterized in that, The crack development data is the change in crack characteristics per unit time. The preset acquisition interval Δt is used to collect development data at intervals. The acquisition method includes: comparing the positional changes of the same feature point in consecutive frame images using video image differential technology; setting a reference coordinate system in each monitoring area to calculate the displacement changes of the feature point in the X and Y axis directions; obtaining the depth change in the Z axis direction by combining laser ranging data; and synthesizing crack development data through three-dimensional change data.

7. The video AI monitoring system for bridge deck crack / fracture monitoring according to claim 2, characterized in that, The method for determining whether a maintenance adjustment command has been generated includes: calculating the development thresholds for Q regions and the crack development rate of each region. Each with its corresponding development threshold Comparison; if If so, no adjustment command will be generated; if If so, a maintenance adjustment command is generated; the method for calculating the optimized monitoring frequency includes: using the ratio of the crack development rate to the development threshold in the key monitoring area as the adjustment coefficient, and calculating the optimized frequency through a formula.

8. The video AI monitoring system for bridge deck crack / fracture monitoring according to claim 6, characterized in that, Environmental impact factors are introduced to dynamically adjust the development threshold. These factors include temperature change ΔT, humidity RH, and cumulative rainfall R. The dynamic adjustment method includes: dividing the monitoring period into W time periods, each with a duration of ; standardizing the environmental parameters for each time period to construct an environmental feature vector; inputting the environmental feature vector and the corresponding crack development data into a trained environmental impact model to predict the environmental correction coefficient for each region. The environmental impact model is a deep neural network model; and calculating the adjusted development threshold using a formula.

9. The video AI monitoring system for bridge deck crack / fracture monitoring according to claim 7, characterized in that, Real-time collected crack development data With the revised development threshold Compare and reassess whether to generate an adjustment command: If If so, an enhanced monitoring command will be generated, increasing the monitoring frequency to 1.5 times the base frequency; if If so, the original monitoring strategy will be maintained.

Images