Urban viaduct concrete crack width monitoring and early warning method and equipment

Abstract

The invention provides an urban viaduct concrete crack width monitoring and early warning method and equipment, and the method comprises the steps: respectively installing displacement sensors at the concrete cracks of all urban viaducts, carrying out the grouping and collection of the displacement sensors in a Zigbee communication mode, and uploading the monitoring information obtained through sensing to a user side; and based on the received monitoring information, an early warning network model is correspondingly established for each crack at the user side, model parameters of the early warning network models are updated in real time, and large-range cluster type crack width monitoring and early warning are carried out. According to the method, the feature extraction time is shortened and the extraction capability is improved by using data dimension reduction, long and short term memory and a self-attention mechanism network, a corresponding early warning network model is rapidly established for each crack, and the characteristics of the Internet of Things, 5G communication, a one-dimensional convolutional neural network, a recurrent neural network and a self-attention mechanism are fully utilized. The problems of large-range cluster type crack width monitoring and early warning are effectively solved, the crack width monitoring number and early warning precision are improved, and a guarantee is provided for the safety of an urban infrastructure structure.

Description

Technical Field

[0001] The invention belongs to the field of urban infrastructure structure safety monitoring and early warning, and in particular relates to a technical solution for monitoring and early warning of concrete crack width. Background Art

[0002] Urban infrastructure structure safety monitoring and early warning technology refers to the use of various sensors, communication networks, data processing and analysis methods to monitor the operating status of urban infrastructure structures in real time or regularly, as well as to predict and warn of possible dangers or disasters. Urban infrastructure structure safety monitoring and early warning technology is an important means of urban safety management and an important part of urban intelligent construction.

[0003] Urban infrastructure, mainly including viaducts, is mostly made of reinforced concrete. As time goes by, concrete structures are affected by internal and external, human or natural factors, and the material will age and be damaged. The most typical phenomenon is cracks. Concrete cracks not only affect the overall appearance of the structure, but may also lead to a decrease in impermeability, allowing corrosive substances in the environment to enter the structure, thereby affecting the service life of the building. The width of the crack is usually an important indicator to determine whether the crack needs attention. Therefore, monitoring and early warning of the width of concrete cracks is an important part of urban infrastructure structural safety monitoring and early warning.

[0004] The monitoring and early warning method for the width of concrete cracks in urban viaducts currently mainly uses displacement sensors to monitor the width changes in real time, and uses time series analysis algorithms to predict the expected changes in the future. For example, patent document CN 109269457A provides a portable long-term monitoring and early warning system for crack width, including: a sensor for monitoring the changes in the width of concrete cracks; a data acquisition device, which is connected to the nodes of the sensor through lines to form a sensor network, a serial port server connects the sensor network to a 4G router, and the data acquisition device sends the collected data to the Internet through a 4G wireless network; a remote monitoring device, which communicates with the data acquisition device through an Ethernet connection to the Internet, and the remote monitoring device includes a data processing unit and an early warning unit, and the data processing unit parses and calculates the data collected by the data acquisition device, and determines whether to trigger the early warning unit based on the calculated value.

[0005] With the development of smart cities and the aging of urban infrastructure, the number of cracks is gradually increasing. Existing concrete crack width monitoring and early warning technologies are limited by the distance of data transmission and are mostly only suitable for monitoring cracks in a small number and small range. Patent document CN 109269457 A provides a portable crack width long-term monitoring and early warning system, which is only suitable for monitoring in a small range when the user carries it. In addition, due to factors such as the lack of adaptive capabilities of the early warning model, there is a lot of room for improvement in automation, intelligence, high precision and real-time performance, and it is gradually unable to meet the current needs of large-scale and cluster monitoring. The development of modern concrete crack width monitoring and early warning technology is imminent. These shortcomings are mainly reflected in the following points:

[0006] (1) Most of the existing long-term crack width monitoring methods are designed for a small number of cracks in a small range. The data communication sensors used are relatively simple and single, and cannot take into account the characteristics of low power consumption, low cost, self-organizing network and high-speed, large capacity, and long-distance data transmission. They are not suitable for synchronous monitoring of a large number of cracks sparsely distributed in a large area, resulting in the inability to meet the large-scale cluster monitoring needs of urban concrete cracks.

[0007] (2) Most of the existing crack width change warning methods are to fit the model in advance based on the existing data, obtain the parameters in the model and fix them for subsequent model prediction. However, this method can only obtain the characteristics of a single type of crack and establish an effective prediction model. When facing other types of cracks, the model parameters will not be compatible. It is impossible to automatically analyze the characteristics of the width change of each crack to establish an effective warning model, resulting in low intelligence and adaptability of the prediction algorithm; (3) Most of the existing high-precision crack width change warning methods use recurrent neural network models. Compared with traditional linear fitting or shallow neural network models, they can improve the prediction accuracy by extracting nonlinear features and long-term dependency features. However, the corresponding model complexity is significantly increased, and the time for modeling and prediction is significantly increased. In addition, the prediction of concrete crack width also needs to consider environmental factors such as temperature, humidity and air pressure at the same time. Therefore, the amount of input data of the model will increase significantly compared with only inputting crack width data, further increasing the model modeling and prediction time, and ultimately resulting in the inability to establish an adaptive width prediction model for each crack in real time under high-frequency monitoring conditions. Summary of the invention

[0008] The present invention proposes a technical solution for monitoring and early warning of concrete crack width of urban viaducts to solve the difficulties and problems existing in the above-mentioned background technology. The method can monitor the width changes of concrete cracks in a large-scale cluster manner and improve the prediction accuracy of crack width changes.

[0009] The present invention provides a monitoring and early warning method for the width of concrete cracks of urban viaducts. Displacement sensors are respectively installed at the concrete cracks of each urban viaduct, and the monitoring information obtained by sensing is uploaded to a user end after being grouped and collected through a Zigbee communication method. At the user end, based on the received monitoring information, an early warning network model is respectively established for each crack, and the model parameters of the early warning network model are updated in real time, so as to perform large-scale cluster crack width monitoring and early warning.

[0010] Furthermore, the displacement sensor is fixed on the monitored crack perpendicular to the extension direction of the crack, and a limiter is placed at the rebound pointer of the displacement sensor to monitor the width change data of the crack.

[0011] In addition, a variety of environmental parameter sensors are installed at the concrete cracks of elevated bridges in various cities to monitor the environmental parameter data of the cracks.

[0012] Furthermore, the environmental parameter sensors include a temperature sensor, a humidity sensor and an air pressure sensor.

[0013] Moreover, each displacement sensor is connected to the corresponding sensor communication node, and all sensor communication nodes are divided into multiple groups. A data collection node is set in each group. All sensor communication nodes in the group transmit data with the data collection node through Zigbee communication. All data collection nodes transmit data with the 5G gateway node through Zigbee communication. The 5G gateway node transmits all monitoring information to the user end.

[0014] Moreover, the early warning network model is composed of a one-dimensional convolutional neural network, a single-layer long short-term memory neural network and a self-attention mechanism layer. The one-dimensional convolutional neural network is used to reduce the dimension of the width change data and environmental parameter data obtained by monitoring, and then input into the single-layer long short-term memory neural network to extract the characteristics of the crack width change in the past data. The extracted result integrates the nonlinear characteristics and long-term dependent characteristics of the environmental parameter information; the features extracted from the single-layer long short-term memory neural network at multiple times are input into the self-attention mechanism layer to obtain the predicted value of the crack width change; a fully connected layer is added after the output of the early warning network model to predict the theoretical crack width change value of the next stage.

[0015] Moreover, the model parameters of the early warning network model are updated in real time, including dividing the past multi-source sensor data of each crack into a training set and a verification set according to a preset ratio, and the training set and the verification set are respectively divided into multiple data blocks with a length of an input time step plus an output time step, and then each data block in the training set is sequentially input into the early warning network model for model training and parameter updating, and at the same time, verification is performed on the verification set after each round of training is completed. After multiple rounds of training, the optimal early warning network model of the crack in the current state is obtained according to the optimal result on the verification set, and the model training and selection are re-performed in the next state to achieve real-time updating.

[0016] On the other hand, the present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for monitoring and warning the width of concrete cracks in urban viaducts as described above is implemented.

[0017] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned monitoring and early warning method for the width of concrete cracks in urban viaducts.

[0018] On the other hand, the present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements the above-mentioned monitoring and early warning method for the width of concrete cracks in urban viaducts. Compared with the prior art, the present invention has the following differences and advantages:

[0019] (1) The present invention adopts the form of sensor communication node-aggregation node-gateway node-user, changing the traditional single-point communication method to a multiple sensor relay transmission method, integrating the advantages of each type of sensor. The sensor communication node set at each crack is responsible for collecting all sensor data sparsely distributed over a large range, and is divided into multiple groups according to the maximum communication area for intra-group networking. The aggregation node is responsible for collecting sensor data from each group, while the gateway node is responsible for collecting data from the aggregation node. Data is transmitted between nodes at all levels through the Zigbee protocol, and the gateway node to the user uses a 5G communication method, which combines the advantages of low power consumption, low cost, and self-organizing network of Zigbee communication and the advantages of high speed, large capacity, and long distance of 5G communication, and realizes large-scale, clustered synchronous monitoring and transmission of crack width and environmental parameter information;

[0020] (2) The present invention adopts a single early warning model for a single crack and a method of updating model parameters in real time. Different from the traditional early warning model that uses a large amount of data to establish a universal early warning model with low accuracy, the present invention only uses a certain amount of past data and a recurrent neural network to establish a real-time early warning model. By reducing the modeling difficulty and modeling time of the model, and changing the method of fixed model parameters to a method of updating model parameters in real time for each crack prediction, the modeling intelligence level and adaptive feature extraction capability of the early warning model are improved.

[0021] (3) The present invention is different from the traditional recurrent neural network early warning model, which directly extracts crack width change features from the original massive multi-source data. First, a one-dimensional convolutional neural network is used to remove redundant information in the original data, reduce the dimension of the data, and then input the multi-source data after dimensionality reduction into the recurrent neural network to extract width change features. By reducing the complexity of the original data, the modeling and prediction time of the model is significantly reduced, thereby realizing real-time model updates for each crack each time under high-frequency monitoring. Finally, the self-attention mechanism is used to fuse the features extracted by the LSTM layer at each moment, thereby improving the prediction accuracy of the model.

[0022] In conclusion, the present invention adopts a combination of Zigbee and 5G communication in data wireless communication, making full use of the advantages of low power consumption, low cost, and self-organizing network of Zigbee communication and the advantages of high speed, large capacity, and long distance of 5G communication. This method can effectively solve the problem of synchronous data transmission in large-scale cluster monitoring. The single crack single warning model and real-time update of model parameters fully obtain the unique width change characteristics of each crack, and have the advantages of more intelligence and self-adaptation. The present invention further proposes to use a one-dimensional convolutional neural network for dimensionality reduction of original multi-source data, and then use a single-layer LSTM long short-term memory neural network for the extraction of crack width change characteristics, and finally use a self-attention mechanism for the fusion of crack width change characteristics at different times. It can effectively solve the problem that the modeling and prediction time is significantly increased due to the huge original data and high model complexity, so that it is impossible to realize the real-time update model for each crack in the case of high-frequency monitoring. It can also solve the problem of reduced prediction accuracy due to ignoring the influence of crack change characteristics at past moments on crack width changes at the next moment. The present invention makes full use of the characteristics of the Internet of Things, 5G communication, one-dimensional convolutional neural network, recurrent neural network and self-attention mechanism, effectively solves the problems faced by large-scale cluster crack width monitoring and early warning, improves the number of crack width monitoring and early warning accuracy, and provides protection for the structural safety of urban infrastructure.

[0023] The solution of the present invention is simple and convenient to implement, has strong practicality, solves the problems of low practicality and inconvenience in actual application existing in related technologies, can improve user experience, and has important market value. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 The figure is a schematic diagram of the equipment layout of an embodiment of the present invention.

[0025] Figure 2 This is a diagram of the network model structure of an embodiment of the present invention.

[0026] Figure 3 It is a structural diagram of the LSTM model in the network model of an embodiment of the present invention, wherein part (a) is a schematic diagram of the LSTM layer for extracting crack width change characteristics according to the time dimension, and part (b) is a schematic diagram of the implementation of the LSTM network extracting nonlinear features and long-term dependency features.

[0027] Figure 4 This is a diagram of the model training steps of an embodiment of the present invention.

[0028] Figure 5 The figure is a schematic diagram of the physical structure of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0029] The concept, specific structure and technical effects of the present invention will be further described below in conjunction with the accompanying drawings and embodiments, so as to fully understand the purpose, characteristics and effects of the present invention.

[0030] The present invention proposes a concrete crack width monitoring and early warning method based on the characteristics of the Internet of Things, 5G communication, one-dimensional convolutional neural network, recurrent neural network and self-attention mechanism. The method fully combines the inherent characteristics of Zigbee and 5G communication technologies, overcomes the shortcomings of traditional crack width monitoring methods that are not applicable to large-scale cluster crack width monitoring; and utilizes the characteristics of one-dimensional convolutional neural network, recurrent neural network and self-attention mechanism to overcome the shortcomings of traditional time series processing methods that are unable to establish a high-precision prediction model for each crack. The method has large capacity, high degree of automation and intelligence, and high processing accuracy, and has unique advantages.

[0031] The embodiment of the present invention first installs crack width monitoring equipment, including the installation of high-precision displacement sensors, environmental parameter sensors and sensor communication nodes; installs large-scale cluster wireless data transmission equipment, including the installation of data aggregation nodes and 5G gateway nodes; data acquisition and data preprocessing, including collecting width change information and environmental parameter change information of each crack, and aligning time lines and filling missing data for different types of data; extracting features from crack width change data, including using a recurrent neural network to extract nonlinear features and long-term dependent features to fully capture the characteristics of crack width change; fusing environmental parameter change information, including fusing environmental parameter information such as temperature and air pressure with width change data in a multi-sequence input manner to fully capture the characteristics of crack width change; outputting crack width change prediction values, including adding a fully connected layer at the end of the network model to predict the theoretical crack width change value of the next stage; model training and parameter updating, including preparing multi-source sensor data of each crack for model training, as well as updating model parameters and selecting the optimal model. The present invention makes full use of the characteristics of the Internet of Things, 5G communications, and neural networks, effectively solves the problems faced by large-scale cluster crack width monitoring and early warning, improves the number of crack width monitoring and early warning accuracy, and provides protection for the structural safety of urban infrastructure.

[0032] The high-precision displacement sensor model used in the embodiment is LVDT-G125-25MM-A2, with a measurement accuracy of 1 micron and a measurement range of 25 mm. It adopts the principle of rebound displacement measurement, measures the displacement of the object by monitoring the displacement of the rebound pointer, and uses RS485 for data communication. The sensor communication node used reserves an RS485 interface for sensor data reception, and an environmental parameter sensor is integrated on the main board of the sensor communication node to detect environmental parameters such as temperature, humidity and air pressure. Zigbee communication sensors are integrated and set in the data collection node, sensor communication node and 5G gateway node respectively for wireless data transmission, with a transmission distance of less than 100 meters. The data collection node used forwards data between the sensor communication node and the 5G gateway node by Zigbee communication, with a coverage radius of less than 100 meters. The 5G gateway node used must be used within the coverage range of the mobile signal, without distance restrictions. The user end used is preferably a personal computer or server that has been connected to the Internet and has AI computing capabilities, such as an NVIDIA graphics card.

[0033] In the embodiment, a high-precision prediction model is established for each crack on the user side, the hardware device for algorithm model training is NVIDIA RTX 2060Super GPU, and the algorithm model adopts Keras framework in deep learning.

[0034] The concrete crack width monitoring and early warning method provided in the embodiment is specifically implemented as follows:

[0035] First, equipment is deployed to establish a sensor relay transmission channel of sensor communication node-aggregation node-gateway node-user, so as to achieve large-scale, cluster-type synchronous monitoring and transmission of crack width and environmental parameter information. Figure 1 shown.

[0036] Equipment deployment includes two aspects. One is the installation of crack width monitoring equipment, including the installation of high-precision displacement sensors, environmental parameter sensors and sensor communication nodes. First, fix the high-precision displacement sensor on the monitored crack perpendicular to the crack extension direction, and then place a limiter at the rebound pointer of the displacement sensor to monitor the change in the crack width. Finally, use the RS485 transmission interface to connect the high-precision displacement sensor to the sensor communication node and establish communication, adjust the sending format and frequency of the displacement data and environmental parameter data, as well as information such as the sensor number, to prepare for large-scale cluster networking. The environmental parameter sensors are preferably recommended to include temperature sensors, humidity sensors and air pressure sensors.

[0037] In the implementation method preferably adopted in the embodiment, in the transmission data format setting, the sending frequency is set to once every 30 minutes, and the sensor number is set according to the group number-sequence number, such as 01-01, which respectively represent the grouping and sequence of the sensor.

[0038] On the other hand, there is the installation of large-scale clustered wireless data transmission equipment, including the installation of data collection nodes and 5G gateway nodes. First, according to the distance and method of Zigbee stable transmission, all sensor communication nodes are divided into multiple groups based on the principle of networking as much as possible in one area. The sensor communication nodes in each group uniformly transmit data with a data collection node, and it is recommended to use clustering criteria to optimize the layout of data collection nodes so that they can transmit data with all sensor communication nodes in the group with as few as possible. Then choose a location with a higher geographical location to deploy 5G gateway nodes, so that the mobile signal is stable and all data collection nodes can transmit data with the node through Zigbee, and finally transmit all monitoring information to the user end through 5G data transmission.

[0039] In specific implementation, it is recommended to divide the sensor communication nodes into multiple blocks according to the distance, place a data collection node in the central position of each block, and ensure that the data collection node and all sensor communication nodes in the block can use the Zigbee protocol for data transmission. In addition, it is necessary to ensure that the data collection nodes in all blocks can transmit data with a 5G gateway node through the Zigbee protocol, and the 5G gateway node then transmits data with the user end through 5G communication.

[0040] Then, at the user end, according to the received monitoring information, an early warning model is established for each crack, and the model parameters are updated in real time. Different from the traditional early warning model, which uses a large amount of data to establish a universal early warning model with low accuracy, the present invention only uses a certain amount of past data and a one-dimensional convolutional neural network, a recurrent neural network and a self-attention mechanism to establish a real-time early warning model. By reducing the complexity of the input data, the modeling and prediction time of the model is reduced, and the data change characteristics at each moment in the past are integrated through the self-attention mechanism to enhance the feature extraction ability of the model. In addition, the fixed model parameter method is changed to a real-time update of the model parameters for each crack prediction, which improves the modeling intelligence, adaptive feature extraction ability and prediction accuracy of the early warning model.

[0041] During specific implementation, it is preferred to perform the following steps on each crack:

[0042] Step 1: data collection and data preprocessing, including collecting the width change information and environmental parameter change information of each crack, and aligning the time lines and filling in missing data for different types of data.

[0043] Furthermore, data collection and data preprocessing are implemented by uniformly adjusting the collected data of various types to the same frequency, and aligning the time lines of various types of data in such a way that the data with similar time are corresponding data. Then, linear interpolation is used to fill in a small amount of missing data, and least squares fitting is used to fill in a large amount of missing data. If there are a large number of missing data, the data should be discarded, and various types of equipment should be checked as soon as possible to see if there are any hardware defects.

[0044] In the implementation method preferably adopted in the embodiment, the frequency of timeline alignment is set to once every 30 minutes, and the rules for filling in missing data are set as follows: when 1 to 6 periods of missing data are missing, linear interpolation is used to fill in the missing data; when 7 to 24 periods of missing data are missing, least squares fitting is used to fill in the missing data; when 25 periods of data or more are missing, various types of equipment should be checked as soon as possible for hardware defects.

[0045] Step 2: Extract the features from the data of each crack width change, including using a one-dimensional convolutional neural network to reduce the dimensionality of the original data, using a recurrent neural network to extract nonlinear features and long-term dependent features from the fused data, and using a self-attention mechanism to fuse the crack width change features at different times, so as to quickly and accurately capture the characteristics of each crack width change.

[0046] In the specific implementation, firstly, a one-dimensional convolutional neural network is used to reduce the dimension of the width change data and the environmental parameter data, remove the redundant information in the original data, retain the feature information, and thus reduce the complexity of the data;

[0047] Then, the environmental parameter change information and crack width change information are fused, and a recurrent neural network is used to extract the nonlinear characteristics and long-term dependent characteristics of the crack width change in the fused information. Then, the self-attention mechanism is used to fuse the features of the recurrent neural network output in each time period to obtain the final crack width change prediction value.

[0048] The preferred implementation method is that the original data is first reduced to half its length by passing through a one-dimensional convolutional neural network with a convolution kernel size of 2 and a step size of 2. Then it is input into a single-layer long short-term memory neural network (LSTM) to capture the characteristics of crack width changes in past data, including nonlinear features and long-term dependent features that integrate environmental parameter information. Finally, the features extracted by the LSTM model at each moment are input into the self-attention mechanism layer to obtain the final prediction value.

[0049] The early warning model proposed in the present invention is composed of a one-dimensional convolutional neural network, a single-layer long short-term memory neural network and a fully connected layer. It has a simple structure and can support the simultaneous modeling of each crack. It is suitable for synchronous monitoring of massive cracks sparsely distributed in a large area, and meets the needs of large-scale cluster monitoring of urban concrete cracks.

[0050] Furthermore, the embodiment of the present invention proposes a method for quickly and accurately extracting crack width change features in data by inputting the crack width change and environmental parameter change time series into the one-dimensional convolution layer of the model, then inputting the output into the LSTM layer of the model, and finally inputting the output result of the LSTM layer at each moment into the self-attention mechanism layer. The network model structure of the embodiment is as follows: Figure 2 As shown in the figure, let t represent the starting time of the data and n represent the amount of input data. The input data of the model is the mixed time series of crack width change and environmental parameter change after time line alignment. The environmental parameter change time series also includes: temperature, humidity and air pressure change time series. Therefore, the input data dimension of the model is 4, and the input time step is 2n, that is, [4,X t+1 ~X t+2n]. Since the total amount of input data is large, directly performing subsequent feature extraction will significantly increase the model building and prediction time of the model. Therefore, the model designed by the present invention first uses a one-dimensional convolution layer to reduce the dimension of the original data. The dimension of each convolution kernel in the one-dimensional convolution layer is [4,2], the moving step is 2, and the number of convolution kernels is preferably recommended to be determined according to the search results of the grid search method. And no padding is used in the convolution process, so the length of the vector will be reduced to half of the original after the one-dimensional convolution, that is, n. Since the convolution operation has the functions of feature extraction, knowledge distillation and nonlinearization, the model first uses a one-dimensional convolution layer to process the original data, which helps to improve the nonlinear fitting ability of the model. In addition, it can also filter the redundant information in the original data and reduce the length of the input data, thereby improving the fitting speed of the model and reducing the time required for model training once. After being processed by the one-dimensional convolution layer, the output data is input into the LSTM layer, and the dimension of the implicit state vector in the LSTM will also be determined by the search results of the grid search method. After the LSTM layer is operated, the model will generate a total of n implicit state vectors based on the context information in the input data. The implicit state vector at each moment contains the state information of all previous moments, that is, the implicit state vector h at the last moment. t+n Contains the state information of all previous moments, and so on. After obtaining the state vectors of n moments, all vectors are input into a self-attention mechanism layer. The network of the self-attention mechanism layer first calculates h t+n And each state vector [h t+1 ~h t+n ] t+1 ~α t+n ], the calculation method is matrix dot product, and then each state vector [h t+1 ~h t+n ] respectively with their respective correlation [α t+1 ~α t+n ] and finally sum these quantities to get the final output feature vector.

[0051] Furthermore, the LSTM layer in the model, which is mainly used to extract crack width variation characteristics, is expanded according to the time dimension as follows: Figure 3 As shown in part (a) of the figure. r represents the parameter input at time t, h t and c t They represent the implicit state vector and the unit state vector calculated using the nonlinear formula, and the state vector is consistent with the input x at the next moment. t+1 Will generate the implicit state vector h at the next moment together t+1 and the cell state vector c t+1, and always carries the characteristics of each period of data and passes it on, so as to extract the nonlinear characteristics and long-term dependence characteristics in the crack width change data.

[0052] Furthermore, the LSTM network extracts nonlinear features and long-term dependent features in the following way: Figure 3 As shown in part (b) of t Represents the input gate, which is used to control the amount of input to the current LSTM unit. It and the candidate value It will affect the value of the unit state vector in the current state. The calculation formula is:

[0053] i t =f(w i ·[h t-1 ,x t ]+b i )

[0054]

[0055] where h t-1 represents the implicit state vector at the previous moment, x t Represents the input at the current moment, w i , and b i , They represent the input gate, the corresponding weight matrix of the candidate value, and the bias vector, which are the training parameters of the model. f() represents the Sigmoid function, which is a nonlinear function, thus extracting nonlinear features from the data.

[0056] f t Represents the forget gate, which is used to control how much information in the unit state vector of the previous moment needs to be deleted. The calculation formula is:

[0057] f t =f(w f ·[h t-1 ,x t ]+b f )

[0058] Among them, w f and b f They represent the corresponding weight matrix and bias vector of the forget gate, that is, the training parameters of the model.

[0059] Through the input gate and the forget gate, we can use x at time t t ,h t-1 and c t-1 Three quantities update the cell state vector at this time: c t , and store useful information all the time and delete useless information, thereby extracting long-term dependent features in the data. The calculation formula is:

[0060]

[0061] In the formula Represents a matrix dot product operation.

[0062] o t Represents the output gate, and c calculated above t The implicit state vector of the output is determined by the following formula:

[0063] o t =f(w o ·[h t-1 ,x t ]+b0)

[0064]

[0065] Among them, w o and b0 represent the corresponding weight matrix and bias vector of the output gate, i.e., the training parameters of the model.

[0066] Finally, through the calculation of three gates, c t and h t are updated and passed to the next moment with features, and the h obtained at the final moment is t+n That is, it is a feature vector that carries the nonlinear characteristics and long-term dependence characteristics of the crack width change data within this period of time.

[0067] In the embodiment, the time step of the network model input data is 20, the dimension is 4, and the output time step is 1, the dimension is 1. The dimension of the implicit state vector of the LSTM model is 512, and the model structure is a single layer; the input dimension of the self-attention mechanism is 10, and the output dimension is 1. That is, the trained model cyclically calculates the past 20 periods of width change data, obtains the features in the past data, and then outputs the width change feature vector of the next period in the future.

[0068] Step 3: output the predicted value of crack width change, including adding a fully connected layer at the end of the network model to predict the theoretical crack width change value in the next stage.

[0069] The present invention notes that the existing recurrent neural network early warning model only uses the output of the model at the last moment as the extracted feature and is used for the prediction of the next stage. However, in time series data, the predicted value at the next moment is not only strongly correlated with the previous moment, but also with the data in the previous period of time, and the correlation gradually weakens as time goes by. Therefore, the present invention proposes that it is not enough to use only the output of the model at the last moment to extract the change characteristics of the crack width. It is also necessary to establish a correlation with the features extracted by the model in the previous period of time to jointly determine the predicted value at the next moment, thereby improving the accuracy of the prediction.

[0070] Furthermore, the embodiment proposes that the method for outputting the predicted value of crack width change is to add a fully connected layer after the output feature vector finally obtained by the early warning network model adopted in step 2, so that the dimension of the output vector is reduced to 1, and the calculation result is nonlinearly transformed through the Sigmoid activation function to calculate the output value of the model at the next moment, which is the theoretical crack width change value of the next stage predicted by the model.

[0071] Step 4, model training and parameter updating, includes preparing multi-source sensor data of each crack for model training, as well as updating model parameters and selecting the optimal model.

[0072] In specific implementation, the multi-source sensor data of each crack can be divided into training set, validation set and test set, and the loss function and gradient descent method are used to update the model parameters on the training set. The model with the smallest loss on the validation set is selected as the optimal warning model for the crack in the current state, and the model is retrained and selected in the next state.

[0073] In the embodiment, the past multi-source sensor data of each crack is divided into a training set and a verification set in a ratio of 3:1, and each set is divided into multiple data blocks with a length of the input time step (Input Timestep) plus the output time step (Output Timestep). Then, each data block in the training set is sequentially input into the model for model training and parameter update. At the same time, each data in the training set is verified on the verification set after completing a round of training. After multiple rounds of training, the model with the best warning result on the verification set is selected as the optimal warning model for the crack in the current state, and the model training and selection are re-performed in the next state to achieve the effect of real-time updating of the model.

[0074] Furthermore, the implementation of model training and parameter updating in the embodiment is as follows: Figure 4As shown in the figure. Let the number of training data be TrainDataNum. First, use the multi-source sensor data of the previous TrainDataNum period to train the model. Before training, divide the data of this TrainDataNum period into multiple data blocks of length Input Timestep+1 period, and divide these data blocks into training set and validation set in a ratio of 3:1. The training set is used to update the model parameters, and the validation set is used to select the optimal model. The specific size of TrainDataNum and Input Timestep depends on the complexity of the data. Usually, the empirical value obtained after multiple trial and error is used. Therefore, different values ​​can be used for different tasks in specific implementation. In the training process, the first Input Timestep digits in each data block in the training set are the input data of the model, and the last digit is the label value. After the input data is calculated by the initial model, a predicted value will be generated. Then, the predicted value and the label value are compared using the loss function to obtain the loss value of the model. Finally, the model optimizer is used to update the parameters in the model so that the loss is minimized, that is, the predicted value is closest to the label value. The data blocks in the validation set do not participate in the update of the model parameters, but are only used to calculate the loss value of the updated model in the data set. Therefore, after multiple rounds of training, the model with the smallest loss on the validation set is the optimal model. After obtaining the optimal model, the last Input Timestep period data of the multi-source sensor data of TrainDataNum period is input into the optimal model as input data, so as to obtain the predicted value of TrainDataNum+1 period, that is, the theoretical crack width change value of the next stage.

[0075] Since the above process takes a short time, a prediction model can be generated for each crack data through this method, so that the model can be established for the characteristics of the crack, and the accuracy of crack width change prediction can be improved. And the prediction model can be re-trained with recent past data over time to achieve the effect of real-time model update, so that the model can achieve the best fitting effect for the current crack width change trend, and further improve the accuracy of crack width change prediction.

[0076] In the embodiment, the preferred TrainDataNum and Input Timestep are 300 and 10 respectively; the preferred model optimizer is Adam (setting parameters β1=0.9, β2=0.99), and the learning rate is 1×10 -3; The loss function is the mean square error loss function; the batch size of the training data is 32, that is, the model requires 32 data blocks for each iteration; the training rounds of the training data are 500, that is, the model loops the training data 500 times in total, and a model is obtained after each loop, for a total of 500 models, and the model with the smallest loss on the validation set is selected as the optimal model.

[0077] In order to demonstrate the technical effect of the present invention, a model test was conducted based on the embodiment method, including testing the proposed method on real multi-source sensor data to verify the prediction and forecasting performance of the method.

[0078] In the embodiment, when testing the model, the predicted value of the crack width change of the model is first calculated by the proposed method, and then the sensor is waited to return the true width change value. After multiple tests, the root mean square error RMSE is used to calculate the gap between all the predicted values ​​and the true value to obtain the actual prediction accuracy of the method. The proposed method is compared with several traditional methods for accuracy to verify the superiority of the method. The traditional methods used include the moving average autoregressive model (AutoRegressive Integrated Moving Average, ARIMA), support vector regression (Support VectorRegression, SVR) and BP neural network (Back Propagation Neural Network, BPNN) and LSTM model. The data used are the width change data of 50 cracks in different regions within two weeks, that is, a total of 50×672 data. The first 300 data of each crack are used as model training data, and the last 372 data are used as test data. The traditional method uses all the test data to establish a universal warning model and verifies all the test data; the method proposed in the present invention establishes a model for each crack in real time each time it is predicted, and verifies the effectiveness of improving the prediction accuracy by establishing a model for each crack in real time each time it is predicted through ablation experiments using the LSTM model and the real-time modeling method. The comparison of the prediction accuracy of several methods is shown in Table 1 below:

[0079] Table 1

[0080]

[0081] As shown in Table 1, the method of the present invention has the highest prediction accuracy compared with the traditional method under the same data, indicating that the method can effectively improve the prediction accuracy of large-scale clustered concrete crack width. And the ablation experiment proves the effectiveness of the method of real-time modeling for each crack prediction and the self-attention mechanism in improving the prediction accuracy of the model. At the same time, since the complexity of the input parameters is reduced by half, the modeling and prediction speed is greatly reduced compared to using only the LSTM model, which can effectively improve the model's real-time model update capability for each crack prediction under high-frequency monitoring conditions.

[0082] In specific implementation, the method proposed in the technical solution of the present invention can be implemented by technical personnel in this field using computer software technology to realize automatic operation process. System devices for implementing the method, such as computer-readable storage media storing the corresponding computer program of the technical solution of the present invention and computer equipment running the corresponding computer program, should also be within the scope of protection of the present invention.

[0083] The following is a description of the urban viaduct concrete crack width monitoring and early warning device provided by the present invention. The urban viaduct concrete crack width monitoring and early warning device described below and the urban viaduct concrete crack width monitoring and early warning method described above can correspond to each other.

[0084] Figure 5 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5 As shown, the electronic device may include: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus. The processor may call the logic instructions in the memory to execute the concrete crack width monitoring and early warning method, which mainly includes the above steps 1 to 4.

[0085] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0086] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the urban viaduct concrete crack width monitoring and early warning methods provided by the above methods.

[0087] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the urban viaduct concrete crack width monitoring and early warning method provided by the above-mentioned methods.

[0088] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0089] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A monitoring and early warning method for the width of concrete cracks in urban viaducts, characterized by: Displacement sensors are installed at the concrete cracks of elevated bridges in each city. The monitoring information obtained by the sensors is grouped and uploaded to the user end through Zigbee communication. At the user end, based on the received monitoring information, an early warning network model is established for each crack, and the model parameters of the early warning network model are updated in real time to carry out large-scale cluster crack width monitoring and early warning.

2. The monitoring and early warning method for the width of concrete cracks of urban viaducts according to claim 1 is characterized by: The displacement sensor is fixed on the monitored crack perpendicular to the extension direction of the crack, and a limiter is placed at the rebound pointer of the displacement sensor to monitor the width change data of the crack.

3. The monitoring and early warning method for the width of concrete cracks of urban viaducts according to claim 1 is characterized by: A variety of environmental parameter sensors are also installed at the concrete cracks of elevated bridges in various cities to monitor the environmental parameter data of the cracks.

4. The monitoring and early warning method for the width of concrete cracks in urban viaducts according to claim 3 is characterized by: The environmental parameter sensors include a temperature sensor, a humidity sensor and an air pressure sensor.

5. The monitoring and early warning method for the width of concrete cracks of urban viaducts according to claim 1 is characterized by: Connect each displacement sensor to the corresponding sensor communication node, divide all sensor communication nodes into multiple groups, set a data collection node in each group, all sensor communication nodes in the group and the data collection node transmit data through Zigbee communication, all data collection nodes transmit data with the 5G gateway node through Zigbee communication, and the 5G gateway node transmits all monitoring information to the user end.

6. The monitoring and early warning method for the width of concrete cracks of urban viaducts according to claim 3 is characterized by: The early warning network model is composed of a one-dimensional convolutional neural network, a single-layer long short-term memory neural network and a self-attention mechanism layer. The one-dimensional convolutional neural network is used to reduce the dimension of the width change data and environmental parameter data obtained by monitoring, and then input into the single-layer long short-term memory neural network to extract the characteristics of crack width changes in past data. The extracted results integrate the nonlinear characteristics and long-term dependent characteristics of environmental parameter information; The features extracted from the single-layer long short-term memory neural network at multiple times are input into the self-attention mechanism layer to obtain the predicted value of the crack width change; a fully connected layer is added after the output of the early warning network model to predict the theoretical crack width change value in the next stage.

7. The monitoring and early warning method for the width of concrete cracks of urban viaducts according to claim 1 is characterized by: The model parameters of the early warning network model are updated in real time, including dividing the past multi-source sensor data of each crack into a training set and a verification set according to a preset ratio, the training set and the verification set are respectively divided into multiple data blocks with a length of an input time step plus an output time step, and then each data block in the training set is sequentially input into the early warning network model for model training and parameter updating, and at the same time, verification is performed on the verification set after each round of training is completed. After multiple rounds of training, the optimal early warning network model of the crack in the current state is obtained according to the optimal result on the verification set, and the model training and selection are re-performed in the next state to achieve real-time updating.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the monitoring and early warning method for the width of concrete cracks in an urban viaduct as described in any one of claims 1 to 7 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the monitoring and early warning method for the width of concrete cracks in an urban viaduct as claimed in any one of claims 1 to 7 is implemented.

10. A computer program product, comprising a computer program, characterized in that: When the computer program is executed by a processor, the monitoring and early warning method for the width of concrete cracks in an urban viaduct as claimed in any one of claims 1 to 7 is implemented.

Images