Bridge health detection method, detection system and internet of things big data system

Abstract

The invention discloses a bridge health detection method, a detection system and an Internet of Things big data system, the bridge health detection Internet of Things big data system comprises LoRa bridge data acquisition terminals 1-n, a LoRa bridge data acquisition gateway, a bridge health monitoring cloud service layer and an application layer, and the application layer comprises a bridge intelligent monitoring terminal and a bridge monitoring APP; each LoRa bridge data acquisition terminal performs data interaction with the server of the bridge health monitoring cloud service layer through the LoRa bridge data acquisition gateway, and the bridge intelligent monitoring terminal and the bridge monitoring APP are used for facilitating management personnel to supervise the operation state of a monitored bridge. According to the invention, the accuracy and reliability of bridge health detection can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of bridge monitoring, and specifically relates to a bridge health detection method, a detection system and an Internet of Things big data system. Background Art

[0002] As the service life of bridges increases and the service environment deteriorates, the safety and health issues of bridges are becoming increasingly prominent. The traditional management and maintenance model is difficult to meet the development needs of modern bridge construction and management and maintenance. Carrying out bridge health monitoring and evaluation is helpful to make scientific assessment and maintenance decisions. At present, bridge health detection is mostly based on the detection data of a single sensor to evaluate the bridge health status, which can only reflect the health status of the bridge from one aspect, and cannot reflect the overall health status of the bridge. Summary of the invention

[0003] Purpose of the invention: The first purpose of the present invention is to provide a bridge health detection method that can accurately evaluate the overall health status of the bridge. The second purpose of the present invention is to provide a bridge health detection system. The third purpose of the present invention is to provide a bridge health detection Internet of Things big data system.

[0004] Technical solution: The bridge health detection method of the present invention comprises:

[0005] Obtain the outputs of parameter sensors 1-5 of the bridge, where the parameter sensors 1-5 are sensors of different types, and the outputs of the parameter sensors 1-5 are all multiple;

[0006] Building a bridge health monitoring model:

[0007] The bridge health detection model includes a bridge parameter module 1-6, a BiGRU neural network model 1-5, a GCN graph convolution network model-fuzzy recursive neural network model 1-2 and a GCN graph convolution network model-BiGRU neural network model, wherein the bridge parameter module is used to obtain triangular fuzzy number prediction values ​​output by multiple parameter sensors;

[0008] The outputs of multiple parameter sensors 1 are respectively used as inputs of the bridge parameter module 1 and the corresponding inputs of the bridge parameter module 6; the outputs of the bridge parameter module 1 are respectively used as inputs of the BiGRU neural network model 1 and the corresponding inputs of the GCN graph convolutional network model-fuzzy recurrent neural network model 2; the outputs of the BiGRU neural network model 1 are used as the corresponding inputs of the GCN graph convolutional network model-fuzzy recurrent neural network model 1;

[0009] The outputs of the multiple parameter sensors 2 are respectively used as the inputs of the bridge parameter module 2 and the corresponding inputs of the bridge parameter module 6; the outputs of the bridge parameter module 2 are respectively used as the inputs of the BiGRU neural network model 2 and the corresponding inputs of the GCN graph convolutional network model-fuzzy recurrent neural network model 2; the outputs of the BiGRU neural network model 2 are respectively used as the corresponding inputs of the GCN graph convolutional network model-fuzzy recurrent neural network model 1;

[0010] The outputs of the multiple parameter sensors 3 are respectively used as the inputs of the bridge parameter module 3 and the corresponding inputs of the bridge parameter module 6; the outputs of the bridge parameter module 3 are respectively used as the inputs of the BiGRU neural network model 3 and the corresponding inputs of the GCN graph convolutional network model-fuzzy recurrent neural network model 2; the outputs of the BiGRU neural network model 3 are respectively used as the corresponding inputs of the GCN graph convolutional network model-fuzzy recurrent neural network model 1;

[0011] The outputs of the multiple parameter sensors 4 are respectively used as the inputs of the bridge parameter module 4 and the corresponding inputs of the bridge parameter module 6; the outputs of the bridge parameter module 4 are respectively used as the inputs of the BiGRU neural network model 4 and the corresponding inputs of the GCN graph convolutional network model-fuzzy recursive neural network model 2; the outputs of the BiGRU neural network model 4 are used as the corresponding inputs of the GCN graph convolutional network model-fuzzy recursive neural network model 1;

[0012] The outputs of the multiple parameter sensors 5 are respectively used as the inputs of the bridge parameter module 5 and the corresponding inputs of the bridge parameter module 6; the outputs of the bridge parameter module 5 are respectively used as the inputs of the BiGRU neural network model 5 and the corresponding inputs of the GCN graph convolutional network model-fuzzy recurrent neural network model 2; the outputs of the BiGRU neural network model 5 are used as the corresponding inputs of the GCN graph convolutional network model-fuzzy recurrent neural network model 1;

[0013] The outputs of the bridge parameter module 6 are respectively used as the corresponding inputs of the GCN graph convolution network model-fuzzy recursive neural network model 1, the GCN graph convolution network model-fuzzy recursive neural network model 2 and the GCN graph convolution network model-BiGRU neural network model. The outputs of the GCN graph convolution network model-fuzzy recursive neural network model 2 are respectively used as the corresponding inputs of the GCN graph convolution network model-BiGRU neural network model. The outputs of the GCN graph convolution network model-fuzzy recursive neural network model 1 are respectively used as the corresponding inputs of the BiGRU neural network model 1, BiGRU neural network model 2, BiGRU neural network model 3, BiGRU neural network model 4, BiGRU neural network model 5 and the GCN graph convolution network model-BiGRU neural network model. The outputs of the GCN graph convolution network model-BiGRU neural network model are respectively used as the corresponding inputs of the GCN graph convolution network model-fuzzy recursive neural network model 1 and the GCN graph convolution network model-fuzzy recursive neural network model 2. The a, b and c output by the GCN graph convolution network model-BiGRU neural network model constitute triangular fuzzy numbers (a, b, c).

[0014] The health status of the bridge is evaluated based on triangular fuzzy numbers (a, b, c).

[0015] Furthermore, the bridge parameter module includes TCN temporal convolutional network model 1-n, GCN graph convolutional network model 1-n, AANN autoassociative neural network model-GCN graph convolutional network model, fuzzy recursive neural network model-metabolism gray GM (1,1) model, BiGRU neural network model-metabolism gray GM (1,1) model 1, BiGRU neural network model-metabolism gray GM (1,1) model 2, fuzzy recursive neural network model-BiGRU neural network model and GCN graph convolutional network model-BiGRU neural network model; wherein n represents the number of parameter sensor outputs of the same type, and the parameter sensor output, TCN temporal convolutional network model and GCN graph convolutional network model correspond one to one;

[0016] The outputs of multiple parameter sensors are respectively used as the inputs of the corresponding TCN temporal convolutional network models and the corresponding inputs of the AANN auto-associative neural network model-GCN graph convolutional network model; the outputs of multiple TCN temporal convolutional network models are respectively used as the inputs of the corresponding GCN graph convolutional network models and the corresponding inputs of the AANN auto-associative neural network model-GCN graph convolutional network model; the outputs of multiple GCN graph convolutional network models are used as the corresponding inputs of the AANN auto-associative neural network model-GCN graph convolutional network model;

[0017] The outputs of the AANN auto-associative neural network model-GCN graph convolutional network model are used as the corresponding inputs of the fuzzy recurrent neural network model-metabolism grey GM (1,1) model, the GCN graph convolutional network model-BiGRU neural network model and the fuzzy recurrent neural network model-BiGRU neural network model respectively; the difference and difference change rate between the output time series of the GCN graph convolutional network model-BiGRU neural network model and the output time series of the fuzzy recurrent neural network model-metabolism grey GM (1,1) model are used as the corresponding inputs of the BiGRU neural network model-metabolism grey GM (1,1) model 1; the outputs of the BiGRU neural network model-metabolism grey GM (1,1) model 1 are used as the corresponding inputs of the fuzzy recurrent neural network model-metabolism grey GM (1,1) model, the GCN graph convolutional network model-BiGRU neural network model and the AANN auto-associative neural network model-GCN graph convolutional network model respectively;

[0018] The difference and difference change rate of the output time series of the GCN graph convolution network model-BiGRU neural network model and the fuzzy recurrent neural network model-BiGRU neural network model are used as the corresponding inputs of the BiGRU neural network model-metabolism gray GM (1,1) model 2, and the output of the BiGRU neural network model-metabolism gray GM (1,1) model 2 are used as the corresponding inputs of the GCN graph convolution network model-BiGRU neural network model, the fuzzy recurrent neural network model-BiGRU neural network model and the AANN autoassociative neural network model-GCN graph convolution network model respectively;

[0019] The outputs of the fuzzy recursive neural network model-metabolic grey GM (1,1) model, the GCN graph convolutional network model-BiGRU neural network model and the fuzzy recursive neural network model-BiGRU neural network model are x, y and z, respectively, forming triangular fuzzy numbers (x, y, z), which are used as the triangular fuzzy number prediction values ​​output by multiple parameter sensors.

[0020] Furthermore, parameter sensor 1 is a settlement sensor, parameter sensor 2 is a crack sensor, parameter sensor 3 is a tilt sensor, parameter sensor 4 is a deformation sensor, and parameter sensor 5 is a vibration sensor.

[0021] Furthermore, the bridge health status includes five levels: very healthy, relatively healthy, healthy, unhealthy and very unhealthy.

[0022] Further,

[0023]

[0024] The bridge health grade is determined based on this table.

[0025] The bridge health detection system of the present invention comprises a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the bridge health detection method is implemented.

[0026] The bridge health detection Internet of Things big data system described in the present invention includes LoRa bridge data acquisition terminals 1-n, a LoRa bridge data acquisition gateway, a bridge health monitoring cloud service layer and an application layer, wherein the application layer includes a bridge intelligent monitoring terminal and a bridge monitoring APP; each LoRa bridge data acquisition terminal exchanges data with a server of the bridge health monitoring cloud service layer through the LoRa bridge data acquisition gateway; the bridge intelligent monitoring terminal and the bridge monitoring APP are used to facilitate management personnel to supervise the operating status of the monitored bridge.

[0027] Furthermore, the LoRa bridge data acquisition terminal includes a camera, a temperature sensor, a settlement sensor, a crack sensor, a tilt sensor, a deformation sensor, a vibration sensor, an STM32 microprocessor, a LoRa communication module and external devices, wherein the external devices include bridge insulation equipment, bridge cleaning equipment and bridge detection equipment, each sensor is connected to the STM32 microprocessor through a corresponding conditioning circuit, and the LoRa communication module, camera and external devices are respectively connected to the STM32 microprocessor.

[0028] Furthermore, the LoRa bridge data acquisition gateway includes an STM32 microprocessor, a 4G communication module, a LoRa communication module, an Ethernet communication interface and an RS485 communication interface. A self-organizing communication network is constructed through the LoRa communication module to realize data interaction between the LoRa bridge data acquisition terminal and the LoRa bridge data acquisition gateway, and two-way data interaction with the bridge health monitoring cloud service layer is realized through the 4G communication module.

[0029] Furthermore, the bridge intelligent monitoring terminal adopts an industrial control computer, and the bridge intelligent monitoring terminal and the bridge monitoring App realize information exchange with the bridge health monitoring cloud service layer through the 5G communication module; the functions of the bridge intelligent monitoring terminal include bridge communication parameter setting, bridge data analysis and management, and bridge health detection model.

[0030] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0031] (1) The TCN temporal convolutional network model receives the output data of the bridge parameter sensor. Through its dilated causal convolution and residual connection modules, it has delay and robustness for the time series values ​​of the bridge parameter sensor output, as well as good bridge parameter sensor output data feature extraction capability and fast convergence level.

[0032] (2) The GCN graph convolutional network model adds filtering to the Fourier transform and acts on vertices and their first-order neighbors to achieve the purpose of extracting the spatial features between vertices of bridge parameters. The GCN graph convolutional network model can obtain the spatial features of bridge parameter information through the hidden relationship between the central node and its surrounding nodes. The essence of the GCN graph convolutional neural network model is to aggregate the features of neighboring nodes of bridge parameter information, and perform graph learning through the relationship between nodes, thereby improving the accuracy, robustness and reliability of extracting the characteristic values ​​of bridge parameter information and the relationship between bridge parameter input information.

[0033] (3) Bridge parameter information has the characteristics of nonlinearity, large hysteresis and complex dynamic changes. The sensors measuring bridge parameters are easily disturbed, so bridge parameter measurements often contain large noise. On the other hand, the number of bridge parameter measurements is greater than the number of its independent variables, that is, there is redundant information in these measurement parameters. The AANN auto-associative neural network model-GCN graph convolutional network model can use redundant bridge parameter information to suppress its measurement noise through the compression, decompression and graph feature extraction process of bridge parameters. In the process of processing bridge parameter big data, the TCN temporal convolutional network model, the GCN graph convolutional network model and the AANN auto-associative neural network model-GCN graph convolutional network model are applied to the bridge parameter measurement time series features, graph feature extraction and fusion, which can greatly improve the accuracy of bridge parameter measurement.

[0034] (4) BiGRU neural network model - metabolic grey GM (1,1) model is a BiGRU neural network model and a tandem metabolic grey GM (1,1) model; the BiGRU network is constructed by a forward GRU network and a reverse GRU network. The bridge parameter input information sequence is calculated in opposite directions in the two-layer GRU network. This structure enables the BiGRU neural network model to capture the information in the bridge parameter sequence more comprehensively, improving the model's expressive power. The metabolic grey GM (1,1) model gradually removes the earliest BiGRU neural network model output data during modeling, and adds new bridge parameter prediction data, which are supplemented and predicted one by one until all prediction targets are achieved. The metabolic grey GM (1,1) model removes invalid and meaningless data information, and predictively supplements new information, which can effectively avoid the errors of the traditional grey GM (1,1) model.

[0035] In summary, the present invention can improve the accuracy of bridge health detection, especially the accuracy of bridge overall health status detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is a structural schematic diagram of a bridge health detection model in an embodiment of the present application;

[0037] Figure 2 is a structural diagram of a bridge parameter module in an embodiment of the present application;

[0038] Figure 3 It is a structural diagram of the bridge health detection Internet of Things big data system in the embodiment of the present application;

[0039] Figure 4 It is a structural diagram of the LoRa bridge data acquisition terminal in an embodiment of the present application;

[0040] Figure 5 This is a schematic diagram of the structure of the LoRa bridge data acquisition gateway in an embodiment of the present application;

[0041] Figure 6 It is a functional diagram of the bridge intelligent monitoring terminal software in the embodiment of the present application. DETAILED DESCRIPTION

[0042] The present invention will be further described below in conjunction with the accompanying drawings.

[0043] The present application provides a bridge health detection method, comprising the following steps:

[0044] (1) obtaining the outputs of parameter sensors 1-5 of the bridge, where the parameter sensors 1-5 are sensors of different types and the outputs of the parameter sensors 1-5 are multiple;

[0045] In this embodiment, parameter sensor 1 is a settlement sensor, parameter sensor 2 is a crack sensor, parameter sensor 3 is a tilt sensor, parameter sensor 4 is a deformation sensor, and parameter sensor 5 is a vibration sensor.

[0046] (2) Building a bridge health detection model:

[0047] like Figure 1 As shown, the bridge health detection model includes bridge parameter modules 1-6, BiGRU neural network models 1-5, GCN graph convolution network model-fuzzy recursive neural network model 1-2 and GCN graph convolution network model-BiGRU neural network model, wherein the bridge parameter module is used to obtain the triangular fuzzy number prediction value output by multiple parameter sensors; the GCN graph convolution network model-fuzzy recursive neural network model is a GCN graph convolution network model and a fuzzy recursive neural network model connected in series; the GCN graph convolution network model-BiGRU neural network model is a GCN graph convolution network model and a BiGRU neural network model connected in series;

[0048] The outputs of multiple parameter sensors 1 are respectively used as inputs of the bridge parameter module 1 and the corresponding inputs of the bridge parameter module 6; the outputs of the bridge parameter module 1 are respectively used as inputs of the BiGRU neural network model 1 and the corresponding inputs of the GCN graph convolutional network model-fuzzy recurrent neural network model 2; the outputs of the BiGRU neural network model 1 are used as the corresponding inputs of the GCN graph convolutional network model-fuzzy recurrent neural network model 1;

[0049] The outputs of the multiple parameter sensors 2 are respectively used as the inputs of the bridge parameter module 2 and the corresponding inputs of the bridge parameter module 6; the outputs of the bridge parameter module 2 are respectively used as the inputs of the BiGRU neural network model 2 and the corresponding inputs of the GCN graph convolutional network model-fuzzy recurrent neural network model 2; the outputs of the BiGRU neural network model 2 are respectively used as the corresponding inputs of the GCN graph convolutional network model-fuzzy recurrent neural network model 1;

[0050] The outputs of the multiple parameter sensors 3 are respectively used as the inputs of the bridge parameter module 3 and the corresponding inputs of the bridge parameter module 6; the outputs of the bridge parameter module 3 are respectively used as the inputs of the BiGRU neural network model 3 and the corresponding inputs of the GCN graph convolutional network model-fuzzy recurrent neural network model 2; the outputs of the BiGRU neural network model 3 are respectively used as the corresponding inputs of the GCN graph convolutional network model-fuzzy recurrent neural network model 1;

[0051] The outputs of the multiple parameter sensors 4 are respectively used as the inputs of the bridge parameter module 4 and the corresponding inputs of the bridge parameter module 6; the outputs of the bridge parameter module 4 are respectively used as the inputs of the BiGRU neural network model 4 and the corresponding inputs of the GCN graph convolutional network model-fuzzy recursive neural network model 2; the outputs of the BiGRU neural network model 4 are used as the corresponding inputs of the GCN graph convolutional network model-fuzzy recursive neural network model 1;

[0052] The outputs of the multiple parameter sensors 5 are respectively used as the inputs of the bridge parameter module 5 and the corresponding inputs of the bridge parameter module 6; the outputs of the bridge parameter module 5 are respectively used as the inputs of the BiGRU neural network model 5 and the corresponding inputs of the GCN graph convolutional network model-fuzzy recurrent neural network model 2; the outputs of the BiGRU neural network model 5 are used as the corresponding inputs of the GCN graph convolutional network model-fuzzy recurrent neural network model 1;

[0053] The outputs of the bridge parameter module 6 are respectively used as the corresponding inputs of the GCN graph convolution network model-fuzzy recursive neural network model 1, the GCN graph convolution network model-fuzzy recursive neural network model 2 and the GCN graph convolution network model-BiGRU neural network model. The outputs of the GCN graph convolution network model-fuzzy recursive neural network model 2 are respectively used as the corresponding inputs of the GCN graph convolution network model-BiGRU neural network model. The outputs of the GCN graph convolution network model-fuzzy recursive neural network model 1 are respectively used as the corresponding inputs of the BiGRU neural network model 1, BiGRU neural network model 2, BiGRU neural network model 3, BiGRU neural network model 4, BiGRU neural network model 5 and the GCN graph convolution network model-BiGRU neural network model. The outputs of the GCN graph convolution network model-BiGRU neural network model are respectively used as the corresponding inputs of the GCN graph convolution network model-fuzzy recursive neural network model 1 and the GCN graph convolution network model-fuzzy recursive neural network model 2. The a, b and c output by the GCN graph convolution network model-BiGRU neural network model constitute triangular fuzzy numbers (a, b, c).

[0054] The bridge health status is evaluated according to the triangular fuzzy numbers (a, b, c). Specifically, the bridge health status includes five levels: very healthy, relatively healthy, healthy, unhealthy and very unhealthy.

[0055] Table 1 Correspondence between the five levels of bridge health and triangular fuzzy numbers

[0056] Serial number Health level Triangular fuzzy numbers 1 Very healthy (0.0，0.1，0.2) 2 Healthier (0.2，0.3，0.4) 3 healthy (0.4，0.5，0.6) 4 Unhealthy (0.6，0.7，0.8) 5 Very unhealthy (0.8，0.9，1.0)

[0057] The bridge health level is determined according to Table 1.

[0058] Combination Figure 2 , the bridge parameter modules include TCN temporal convolutional network models 1-n, GCN graph convolutional network models 1-n, AANN autoassociative neural network model-GCN graph convolutional network model, fuzzy recursive neural network model-metabolism gray GM (1,1) model, BiGRU neural network model-metabolism gray GM (1,1) model 1, BiGRU neural network model-metabolism gray GM (1,1) model 2, fuzzy recursive neural network model-BiGRU neural network model and GCN graph convolutional network model-BiGRU neural network model; where n represents the number of parameter sensor outputs of the same type, and the parameter sensor output, TCN temporal convolutional network model and GCN graph convolutional network model correspond one to one;

[0059] The outputs of multiple parameter sensors are respectively used as the inputs of the corresponding TCN temporal convolutional network models and the corresponding inputs of the AANN auto-associative neural network model-GCN graph convolutional network model; the outputs of multiple TCN temporal convolutional network models are respectively used as the inputs of the corresponding GCN graph convolutional network models and the corresponding inputs of the AANN auto-associative neural network model-GCN graph convolutional network model; the outputs of multiple GCN graph convolutional network models are used as the corresponding inputs of the AANN auto-associative neural network model-GCN graph convolutional network model;

[0060] The outputs of the AANN auto-associative neural network model-GCN graph convolutional network model are used as the corresponding inputs of the fuzzy recurrent neural network model-metabolism grey GM (1,1) model, the GCN graph convolutional network model-BiGRU neural network model and the fuzzy recurrent neural network model-BiGRU neural network model respectively; the difference and difference change rate between the output time series of the GCN graph convolutional network model-BiGRU neural network model and the output time series of the fuzzy recurrent neural network model-metabolism grey GM (1,1) model are used as the corresponding inputs of the BiGRU neural network model-metabolism grey GM (1,1) model 1; the outputs of the BiGRU neural network model-metabolism grey GM (1,1) model 1 are used as the corresponding inputs of the fuzzy recurrent neural network model-metabolism grey GM (1,1) model, the GCN graph convolutional network model-BiGRU neural network model and the AANN auto-associative neural network model-GCN graph convolutional network model respectively;

[0061] The difference and difference change rate of the output time series of the GCN graph convolution network model-BiGRU neural network model and the fuzzy recurrent neural network model-BiGRU neural network model are used as the corresponding inputs of the BiGRU neural network model-metabolism gray GM (1,1) model 2, and the output of the BiGRU neural network model-metabolism gray GM (1,1) model 2 are used as the corresponding inputs of the GCN graph convolution network model-BiGRU neural network model, the fuzzy recurrent neural network model-BiGRU neural network model and the AANN autoassociative neural network model-GCN graph convolution network model respectively;

[0062] The outputs of the fuzzy recursive neural network model-metabolic grey GM (1,1) model, the GCN graph convolutional network model-BiGRU neural network model and the fuzzy recursive neural network model-BiGRU neural network model are x, y and z, respectively, forming triangular fuzzy numbers (x, y, z), which are used as the triangular fuzzy number prediction values ​​output by multiple parameter sensors.

[0063] 1) TCN temporal convolutional network model

[0064] The TCN temporal convolutional network model uses a convolutional neural network to process the time series data output by the bridge parameter sensor. The TCN temporal convolutional network model mainly includes two parts: dilated causal convolution and residual connection. The output of the dilated causal convolution at each moment depends on the output data of the bridge parameter sensor at the previous moment. The dilation coefficient z is exponentially related to the number of network layers n, that is, z = 2 n As the expansion coefficient increases, the range of bridge parameter sensor output information covered becomes wider. Compared with ordinary convolutional networks, the TCN temporal convolutional network model only needs fewer layers to achieve a wider range of perceived bridge parameter sensor output values, thereby avoiding repeated extraction of bridge parameter sensor output value information.

[0065] The TCN temporal convolutional network model also has excellent parallel performance and can efficiently train and process long time series of bridge parameter sensor output values, thereby reducing the running time of the TCN temporal convolutional network model. The TCN temporal convolutional network model uses a deeply separable convolutional layer, which not only reduces the number of bridge parameters in the model, but also improves the generalization ability of the TCN temporal convolutional network model. The TCN temporal convolutional network model uses a residual connection module to solve the problems of gradient explosion and network degradation in deep traditional neural networks. The use of this residual structure can reduce a large amount of information loss in the process of extracting features from the output values ​​of bridge parameter sensors, thereby improving the accuracy of the TCN temporal convolutional network model.

[0066] 2) GCN graph convolutional network model

[0067] The essence of the GCN graph convolutional network model is to extract the spatial features of the input bridge parameter information. At present, the graph convolution methods include spectrum-based graph convolution and spatial domain-based graph convolution. The topological graph of the input bridge parameter information consists of nodes and undirected edges, which contains the node characteristics of the nodes themselves and the structural characteristics between nodes, and can constitute the adjacency matrix of the topological graph. The spectrum-based graph convolution method is used to study the spatial features of the input bridge parameter information; the spectrum-based graph convolution mainly performs Fourier transform on the graph, converts the bridge parameter graph signal from the spatial domain to the spectral domain, and then performs convolution operation on it; the GCN graph convolutional network model adds filtering to the Fourier transform, acting on the vertices and their first-order neighbors. In the domain, the purpose of extracting the spatial features between the vertices of the bridge parameters is achieved. The GCN graph convolutional network model can obtain the spatial features of the input bridge parameter information through the hidden relationship between the central node and its surrounding nodes; the essence of the GCN graph convolutional neural network model is to aggregate the features of the neighboring nodes of the input bridge parameter information, and perform graph learning through the relationship between nodes. The GCN graph convolutional neural network model extracts the features of the bridge parameter nodes by using two graph convolutional layers and uses two fully connected layers to classify the features; the GCN graph convolutional neural network model is trained using the bridge parameter training set, and the learnable bridge parameters of the GCN graph convolutional neural network model are updated through the error back propagation algorithm.

[0068] 3) AANN auto-associative neural network model-GCN graph convolutional network model

[0069] The AANN auto-associative neural network model-GCN graph convolutional network model is a series connection of the AANN auto-associative neural network model and the GCN graph convolutional network model. The output of the AANN auto-associative neural network model is used as the input of the GCN graph convolutional network model. The AANN auto-associative neural network model realizes the compression of the input bridge parameter information through the bridge parameter input layer, mapping layer and bottleneck layer, and extracts the most representative low-dimensional subspace reflecting the system structure of the input bridge parameter information from the high-dimensional parameter space of the input bridge parameter information. At the same time, it effectively filters out the noise and measurement errors in the input bridge parameters, and then realizes the decompression of the input bridge information through the bottleneck layer, demapping layer and output layer, and restores the previously compressed information to each bridge parameter value, thereby realizing the reconstruction of each input bridge parameter.

[0070] In order to achieve the purpose of compressing the input bridge parameter information, the number of nodes in the bottleneck layer of the AANN self-associative neural network model is significantly smaller than that in the input layer. In order to prevent the formation of a simple single mapping between the input layer and the output layer, except for the linear function of the output layer excitation function, all other layers use nonlinear excitation functions. In essence, the first hidden layer of the AANN self-associative neural network model is called the mapping layer. The node transfer function of the mapping layer may be an S-type function or other similar nonlinear functions; the second hidden layer is called the bottleneck layer. The dimension of the bottleneck layer is the smallest in the network. Its transfer function may be linear or nonlinear. The bottleneck layer avoids the one-to-one mapping relationship between output and input that is easy to achieve. It enables the AANN self-associative neural network model to encode and compress the input bridge parameter signal to obtain the relevant model of the input bridge parameter sensor data, and decode and decompress the input bridge parameter information after the bottleneck layer to generate the estimated value of the input bridge parameter signal; the third or last hidden layer is called the demapping layer. The node transfer function of the demapping layer is usually a nonlinear S-type function. The AANN self-associative neural network model uses the error back propagation algorithm to train the bridge parameters.

[0071] The design method of the GCN graph convolutional network model refers to 2).

[0072] 4) Fuzzy recursive neural network model-metabolism grey GM (1,1) model

[0073] The fuzzy recursive neural network model-metabolism grey GM (1,1) model is a fuzzy recursive neural network model and a metabolic grey GM (1,1) model in series; the fuzzy recursive neural network model consists of 4 input layers, a membership function layer, a rule layer and an output layer. The first layer is the input layer. Each input bridge parameter node of this layer is directly connected to the input bridge parameter variable; the second layer is the membership function layer. The nodes of this layer fuzzify the input bridge parameter variable. Each node represents a membership function, and the Gaussian basis function is used as the membership function input; the third layer is the fuzzy reasoning layer, i.e., the rule layer. Dynamic feedback is added to make the network have better learning efficiency and memory of the historical information of the network bridge parameters. The feedback link introduces internal variables, and the sigmoid function is selected as the activation function of the internal variables of the feedback link. The fourth layer is the defuzzification layer, i.e., the output layer. The nodes of this layer perform summation operations on the input bridge parameters.

[0074] The grey GM (1,1) model prediction is a dynamic prediction model. The data used is derived from the accumulated historical data of bridge parameters, and the differential method is used to fit it to obtain a dynamic data prediction model. The residual identification method is used to further process the bridge parameters to ensure the accuracy of the model's predicted bridge parameters. However, the grey GM (1,1) model only considers all the past bridge parameter data. Over time, some potential disturbance factors will continue to interfere with the stability of the system. When the model is used to predict bridge parameters, the accuracy of the predicted bridge parameters will drop rapidly over time. In order to make up for the above-mentioned defects of the model, the metabolic grey GM (1,1) model is introduced, that is, the earliest bridge data is gradually eliminated during modeling, and the newly predicted bridge data is added at the same time, and the bridge parameters are predicted one by one in turn until all the prediction goals are achieved. The metabolic grey GM (1,1) model is based on the basic principle of grey prediction. It removes invalid bridge data information or information of little significance, and predictively supplements new bridge parameter information. This can effectively avoid the bridge parameter errors of the traditional grey GM (1,1) model, and also effectively overcome the defect that the overall operation procedure of the model is cumbersome due to excessive accumulation of bridge parameter information.

[0075] 5) BiGRU neural network model - metabolic gray GM (1,1) model

[0076] The BiGRU neural network model-metabolic grey GM (1,1) model is a series connection of the BiGRU neural network model and the metabolic grey GM (1,1) model. The BiGRU network is constructed by the forward GRU network and the reverse GRU network. The input bridge parameter information sequence is calculated in the two-layer GRU network in opposite directions. The two layers of GRU networks are responsible for capturing the historical information and future information of the bridge parameters respectively. Finally, the outputs of the two layers of networks are integrated according to their corresponding positions to obtain the final output of the BiGRU neural network model. This structure enables the BiGRU neural network model to capture the information in the bridge parameter sequence more comprehensively and improves the expression ability of the model. The GRU neural network model is an optimized gate structure designed on the basis of the recurrent neural network, so that the bridge parameter information can be selectively transmitted in the hidden layer, and the important information of the bridge parameter can be memorized, and the gradient vanishing and gradient exploding problems that occur during the long-term sequence training process can be solved. Similar to the LSTM neural network model structure, the GRU neural network model is designed to solve the long-term memory of bridge parameters and the gradient problem in back propagation. Compared with the LSTM neural network model, each unit of the GRU neural network model is simpler, containing only update gates and reset gates. This simplified structure helps to reduce the complexity and computational cost of the model while maintaining the performance when processing input bridge parameter sequence data. The GRU neural network model parameters are simpler, which can effectively prevent overfitting of bridge parameters while speeding up training.

[0077] The design method of metabolic grey GM (1,1) model refers to 4).

[0078] 6) Fuzzy Recurrent Neural Network Model - BiGRU Neural Network Model

[0079] The fuzzy recursive neural network model-BiGRU neural network model is a fuzzy recursive neural network model and a BiGRU neural network model connected in series, wherein the design method of the fuzzy recursive neural network model refers to 4), and the design method of the BiGRU neural network model refers to 5).

[0080] 7) GCN graph convolutional network model-BiGRU neural network model

[0081] The GCN graph convolutional network model-BiGRU neural network model is a series connection of the GCN graph convolutional network model and the BiGRU neural network model, wherein the design method of the GCN graph convolutional network model refers to 2), and the design method of the BiGRU neural network model refers to 5).

[0082] The embodiment of the present application also provides a bridge health detection system, including a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the bridge health detection method described in the embodiment of the present application is implemented. The bridge health detection system has the advantages of the bridge health detection method.

[0083] like Figure 3 The embodiment of the present application also provides a bridge health detection Internet of Things big data system, including LoRa bridge data acquisition terminals 1-n, LoRa bridge data acquisition gateways, bridge health monitoring cloud service layers and application layers, the application layer including bridge intelligent monitoring terminals and bridge monitoring APP; LoRa bridge data acquisition terminals are responsible for collecting sensor data related to the health status of the bridge, as well as controlling external devices, receiving and executing control commands issued by the LoRa bridge data acquisition gateway; the LoRa bridge data acquisition gateway integrates the sensor data of each LoRa bridge data acquisition terminal and sends it to the server of the bridge health monitoring cloud service layer, providing a data basis for bridge health status analysis and evaluation; the bridge intelligent monitoring terminal and the bridge monitoring APP are used to facilitate management personnel to supervise the operating status of the monitored bridge.

[0084] The bridge health monitoring cloud service layer uses the powerful data storage and computing capabilities of cloud services to provide massive storage, supercomputing, dynamic expansion and other services for bridge health monitoring. The bridge health monitoring cloud service layer is used to store, process, analyze, display and query bridge monitoring data. It can dynamically call relevant data and models for data integration and analysis according to the needs of the application layer, and evaluate the health status of the bridge. Through Web services, a unified application data access interface is provided. Users access the monitoring data cloud service layer through computers or mobile smart devices to achieve real-time sharing of bridge health status data, breaking the phenomenon of "information islands" under the traditional monitoring model. The bridge health monitoring cloud server uses Alibaba Cloud servers. The health monitoring cloud platform is mainly composed of five functional modules: bridge project management, configuration management, bridge data analysis, bridge status assessment and bridge system management, which jointly realize the functions of real-time sharing of bridge health monitoring data, bridge data storage, bridge data analysis and bridge data display. System users can log in to the bridge health monitoring cloud platform remotely through a web page or a bridge monitoring APP to intuitively view the real-time data of different bridge health monitoring projects.

[0085] Specifically, Figure 4As shown, the LoRa bridge data acquisition terminal includes a camera, a temperature sensor, a settlement sensor, a crack sensor, a tilt sensor, a deformation sensor, a vibration sensor, an STM32 microprocessor, a LoRa communication module and external devices, wherein the external devices include bridge insulation equipment, bridge cleaning equipment and bridge detection equipment. Each sensor is connected to the STM32 microprocessor through a corresponding conditioning circuit, and the LoRa communication module, the camera and the external device are connected to the STM32 microprocessor respectively. The LoRa bridge data acquisition terminal and the LoRa bridge data acquisition gateway transmit data and command interactions through the self-organizing network composed of the LoRa communication module. When the LoRa bridge data acquisition terminal receives the command issued by the LoRa bridge data acquisition gateway node, it first checks the CRC of the command, and executes the corresponding operation command according to the operation requirements of the frame command in the command under the condition of no error, that is, the bridge data is collected and the collected bridge data is transmitted to the gateway node.

[0086] like Figure 5 As shown, the LoRa bridge data acquisition gateway includes an STM32 microprocessor, a 4G communication module, a LoRa communication module, an Ethernet communication interface and an RS485 communication interface. The LoRa communication module is used to build a self-organizing communication network to realize data interaction between the LoRa bridge data acquisition terminal and the LoRa bridge data acquisition gateway. The LoRa bridge data acquisition gateway realizes two-way data interaction with the bridge health monitoring cloud service layer through the 4G communication module. The detection parameters of the LoRa bridge data acquisition terminal are uploaded to the server of the bridge health monitoring cloud service layer through the LoRa bridge data acquisition gateway to realize the monitoring of the detected bridge. The LoRa bridge data acquisition gateway is responsible for collecting and integrating the information of the LoRa bridge data acquisition terminal. After the LoRa bridge data acquisition gateway is integrated, the data is packaged and sent to the data storage server of the bridge health monitoring cloud service layer to realize bridge data collection, storage and management.

[0087] The bridge intelligent monitoring terminal uses an industrial control computer. The bridge intelligent monitoring terminal and the bridge monitoring app are connected to the bridge health monitoring cloud service layer through the 5G communication module to realize the information interaction between the LoRa bridge data acquisition terminal and the LoRa bridge data acquisition gateway of the detected bridge, and realize the parameter monitoring and bridge health detection of the detected bridge; the functions of the bridge intelligent monitoring terminal include bridge communication parameter setting, bridge data analysis and management, and bridge health detection model, such as Figure 6As shown. The present invention makes full use of the characteristics of LoRa technology such as low power consumption, fast self-organizing network and long communication distance, connects the bridge data acquisition terminal and the bridge data acquisition gateway, realizes real-time monitoring of the bridge structure from the whole to the part, uploads the bridge monitoring data to the bridge health monitoring cloud platform through remote wireless communication technology, and uses the output of multiple sensor groups to accurately judge the health status of the bridge through the bridge health detection model, thereby improving the accuracy and reliability of bridge health status detection. By utilizing the powerful data storage and computing capabilities of the cloud server, a complete bridge cluster health monitoring data platform is established. It has the advantages of low power consumption, strong expansion capability, convenient system networking, and low operating cost, which can meet the remote collection of real-time bridge data and realize the deep sharing of bridge data resources.

[0088] The bridge health monitoring system proposed by the present invention, which combines LoRa communication module, cloud service, mobile communication and other technologies, consists of three parts: bridge data perception layer, bridge health monitoring cloud service layer and application layer. The LoRa communication module topology networking and cloud service platform technology can provide a complete set of solutions for bridge users in terms of bridge operation status data collection and integration, data storage, status assessment, management display and query functions.

Claims

1. A bridge health detection method, characterized in that: include: Obtain the outputs of parameter sensors 1-5 of the bridge, where the parameter sensors 1-5 are sensors of different types, and the outputs of the parameter sensors 1-5 are all multiple; Building a bridge health monitoring model: The bridge health detection model includes a bridge parameter module 1-6, a BiGRU neural network model 1-5, a GCN graph convolution network model-fuzzy recursive neural network model 1-2 and a GCN graph convolution network model-BiGRU neural network model, wherein the bridge parameter module is used to obtain triangular fuzzy number prediction values ​​output by multiple parameter sensors; The outputs of multiple parameter sensors 1 are respectively used as inputs of the bridge parameter module 1 and the corresponding inputs of the bridge parameter module 6; the outputs of the bridge parameter module 1 are respectively used as inputs of the BiGRU neural network model 1 and the corresponding inputs of the GCN graph convolutional network model-fuzzy recurrent neural network model 2; the outputs of the BiGRU neural network model 1 are used as the corresponding inputs of the GCN graph convolutional network model-fuzzy recurrent neural network model 1; The outputs of the multiple parameter sensors 2 are respectively used as the inputs of the bridge parameter module 2 and the corresponding inputs of the bridge parameter module 6; the outputs of the bridge parameter module 2 are respectively used as the inputs of the BiGRU neural network model 2 and the corresponding inputs of the GCN graph convolutional network model-fuzzy recurrent neural network model 2; the outputs of the BiGRU neural network model 2 are respectively used as the corresponding inputs of the GCN graph convolutional network model-fuzzy recurrent neural network model 1; The outputs of the multiple parameter sensors 3 are respectively used as the inputs of the bridge parameter module 3 and the corresponding inputs of the bridge parameter module 6; the outputs of the bridge parameter module 3 are respectively used as the inputs of the BiGRU neural network model 3 and the corresponding inputs of the GCN graph convolutional network model-fuzzy recurrent neural network model 2; the outputs of the BiGRU neural network model 3 are respectively used as the corresponding inputs of the GCN graph convolutional network model-fuzzy recurrent neural network model 1; The outputs of the multiple parameter sensors 4 are respectively used as the inputs of the bridge parameter module 4 and the corresponding inputs of the bridge parameter module 6; the outputs of the bridge parameter module 4 are respectively used as the inputs of the BiGRU neural network model 4 and the corresponding inputs of the GCN graph convolutional network model-fuzzy recursive neural network model 2; the outputs of the BiGRU neural network model 4 are used as the corresponding inputs of the GCN graph convolutional network model-fuzzy recursive neural network model 1; The outputs of the multiple parameter sensors 5 are respectively used as the inputs of the bridge parameter module 5 and the corresponding inputs of the bridge parameter module 6; the outputs of the bridge parameter module 5 are respectively used as the inputs of the BiGRU neural network model 5 and the corresponding inputs of the GCN graph convolutional network model-fuzzy recurrent neural network model 2; the outputs of the BiGRU neural network model 5 are used as the corresponding inputs of the GCN graph convolutional network model-fuzzy recurrent neural network model 1; The outputs of the bridge parameter module 6 are respectively used as the corresponding inputs of the GCN graph convolution network model-fuzzy recursive neural network model 1, the GCN graph convolution network model-fuzzy recursive neural network model 2 and the GCN graph convolution network model-BiGRU neural network model. The outputs of the GCN graph convolution network model-fuzzy recursive neural network model 2 are respectively used as the corresponding inputs of the GCN graph convolution network model-BiGRU neural network model. The outputs of the GCN graph convolution network model-fuzzy recursive neural network model 1 are respectively used as the corresponding inputs of the BiGRU neural network model 1, BiGRU neural network model 2, BiGRU neural network model 3, BiGRU neural network model 4, BiGRU neural network model 5 and the GCN graph convolution network model-BiGRU neural network model. The outputs of the GCN graph convolution network model-BiGRU neural network model are respectively used as the corresponding inputs of the GCN graph convolution network model-fuzzy recursive neural network model 1 and the GCN graph convolution network model-fuzzy recursive neural network model 2. The a, b and c output by the GCN graph convolution network model-BiGRU neural network model constitute triangular fuzzy numbers (a, b, c). The health status of the bridge is evaluated based on triangular fuzzy numbers (a, b, c).

2. The bridge health detection method according to claim 1, characterized in that: The bridge parameter module includes TCN temporal convolutional network model 1-n, GCN graph convolutional network model 1-n, AANN autoassociative neural network model-GCN graph convolutional network model, fuzzy recursive neural network model-metabolism gray GM (1,1) model, BiGRU neural network model-metabolism gray GM (1,1) model 1, BiGRU neural network model-metabolism gray GM (1,1) model 2, fuzzy recursive neural network model-BiGRU neural network model and GCN graph convolutional network model-BiGRU neural network model; where n represents the number of parameter sensor outputs of the same type, and the parameter sensor output, TCN temporal convolutional network model and GCN graph convolutional network model correspond one to one; The outputs of multiple parameter sensors are respectively used as the inputs of the corresponding TCN temporal convolutional network models and the corresponding inputs of the AANN auto-associative neural network model-GCN graph convolutional network model; the outputs of multiple TCN temporal convolutional network models are respectively used as the inputs of the corresponding GCN graph convolutional network models and the corresponding inputs of the AANN auto-associative neural network model-GCN graph convolutional network model; the outputs of multiple GCN graph convolutional network models are used as the corresponding inputs of the AANN auto-associative neural network model-GCN graph convolutional network model; The outputs of the AANN auto-associative neural network model-GCN graph convolutional network model are used as the corresponding inputs of the fuzzy recurrent neural network model-metabolism grey GM (1,1) model, the GCN graph convolutional network model-BiGRU neural network model and the fuzzy recurrent neural network model-BiGRU neural network model respectively; the difference and difference change rate between the output time series of the GCN graph convolutional network model-BiGRU neural network model and the output time series of the fuzzy recurrent neural network model-metabolism grey GM (1,1) model are used as the corresponding inputs of the BiGRU neural network model-metabolism grey GM (1,1) model 1; the outputs of the BiGRU neural network model-metabolism grey GM (1,1) model 1 are used as the corresponding inputs of the fuzzy recurrent neural network model-metabolism grey GM (1,1) model, the GCN graph convolutional network model-BiGRU neural network model and the AANN auto-associative neural network model-GCN graph convolutional network model respectively; The difference and difference change rate of the output time series of the GCN graph convolution network model-BiGRU neural network model and the fuzzy recurrent neural network model-BiGRU neural network model are used as the corresponding inputs of the BiGRU neural network model-metabolism gray GM (1,1) model 2, and the output of the BiGRU neural network model-metabolism gray GM (1,1) model 2 are used as the corresponding inputs of the GCN graph convolution network model-BiGRU neural network model, the fuzzy recurrent neural network model-BiGRU neural network model and the AANN autoassociative neural network model-GCN graph convolution network model respectively; The outputs of the fuzzy recursive neural network model-metabolic grey GM (1,1) model, the GCN graph convolutional network model-BiGRU neural network model and the fuzzy recursive neural network model-BiGRU neural network model are x, y and z, respectively, forming triangular fuzzy numbers (x, y, z), which are used as the triangular fuzzy number prediction values ​​output by multiple parameter sensors.

3. The bridge health detection method according to claim 1, characterized in that: Parameter sensor 1 is a settlement sensor, parameter sensor 2 is a crack sensor, parameter sensor 3 is a tilt sensor, parameter sensor 4 is a deformation sensor, and parameter sensor 5 is a vibration sensor.

4. The bridge health detection method according to claim 1, characterized in that: The bridge health status includes five levels: very healthy, relatively healthy, healthy, unhealthy and very unhealthy.

5. The bridge health detection method according to claim 4, characterized in that: The bridge health grade is determined based on this table.

6. A bridge health detection system, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the computer program is executed by a processor, the bridge health detection method according to any one of claims 1 to 5 is implemented.

7. A bridge health detection Internet of Things big data system, characterized in that: It includes LoRa bridge data acquisition terminals 1-n, LoRa bridge data acquisition gateways, bridge health monitoring cloud service layers and application layers, wherein the application layer includes bridge intelligent monitoring terminals and bridge monitoring APPs; Each LoRa bridge data collection terminal exchanges data with the server of the bridge health monitoring cloud service layer through the LoRa bridge data collection gateway; the bridge intelligent monitoring terminal and the bridge monitoring APP are used to facilitate management personnel to supervise the operating status of the monitored bridge.

8. The bridge health detection Internet of Things big data system according to claim 7 is characterized in that: The LoRa bridge data acquisition terminal includes a camera, a temperature sensor, a settlement sensor, a crack sensor, a tilt sensor, a deformation sensor, a vibration sensor, an STM32 microprocessor, a LoRa communication module and external devices, wherein the external devices include bridge insulation equipment, bridge cleaning equipment and bridge detection equipment. Each sensor is connected to the STM32 microprocessor through a corresponding conditioning circuit, and the LoRa communication module, camera and external devices are respectively connected to the STM32 microprocessor.

9. The bridge health detection Internet of Things big data system according to claim 8 is characterized in that: The LoRa bridge data acquisition gateway includes an STM32 microprocessor, a 4G communication module, a LoRa communication module, an Ethernet communication interface and an RS485 communication interface. A self-organizing communication network is constructed through the LoRa communication module to realize data interaction between the LoRa bridge data acquisition terminal and the LoRa bridge data acquisition gateway, and two-way data interaction with the bridge health monitoring cloud service layer is realized through the 4G communication module.

10. The bridge health detection Internet of Things big data system according to claim 7 is characterized in that: The bridge intelligent monitoring terminal uses an industrial control computer. The bridge intelligent monitoring terminal and the bridge monitoring App realize information exchange with the bridge health monitoring cloud service layer through the 5G communication module; the functions of the bridge intelligent monitoring terminal include bridge communication parameter setting, bridge data analysis and management, and bridge health detection model.

Images