Anti-sliding key type gravity anchoring system and state detection method and system
Through anti-slip key gravity anchoring system and GAN technology, the problem of lack of systematicity in foundation monitoring data is solved, and intelligent state detection and abnormal identification of large-span suspension bridge anchor ingot foundation is realized, and structural stability and safety are improved .
Patent Information
- Application Number
- CN202411105190.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-13
- Publication Date
- 2024-11-26
AI Technical Summary
The existing foundation monitoring technology lacks systematicity in the basic data of gravity anchor ingots of large-span suspension bridges, cannot intelligently evaluate the safety and health performance of the structure, and requires digital transformation and upgrading.
The anti-slip-key gravity anchoring system combined with the generative anti-neural network (GAN) technology is used to form a three-bar herringbone structure through anchoring foundation, anchor block and loose cable saddle piers, and the monitoring data is pre-pre-ordered using the GAN model. Processing and training to realize intelligent state detection and abnormal identification of the anchoring system.
The anti-slip and anti-overturn load bearing capacity of the anchor foundation is improved, structural stability is enhanced, intelligent monitoring and real-time alarm are realized, and the safe operation of the anchoring system is ensured.
Smart Images

Figure CN119021206A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of bridge engineering, and in particular relates to an anti-sliding key type gravity anchoring system, a state detection method and a system. Background Art
[0002] Foundation engineering monitoring is an effective means to ensure project quality, safe operation and inspection and acceptance. For current building foundations and gravity anchor foundations of large-span suspension bridges, the original foundation monitoring technology relies on some monitoring sensors to obtain corresponding monitoring parameters. The collected data are independent of each other and lack systematicity. It is impossible to intelligently obtain indicators for evaluating the safety and health performance of the structure from the monitoring data. The existing technology is also in urgent need of completing new digital transformation and upgrades. Summary of the invention
[0003] Purpose of the invention: In order to solve the problems existing in the above-mentioned prior art, the present invention provides an anti-sliding key type gravity anchoring system, a state detection method and system.
[0004] Technical solution: The present invention discloses an anti-slip key type gravity anchoring system, comprising an anchor foundation, an anchor block and a loose cable saddle pier; the loose cable saddle pier and the anchor block are both arranged on the anchor foundation, the anchor block and the loose cable saddle pier together form a complete three-bar herringbone structured spatial force-bearing structure, the top of the anchor block is provided with an anchor chamber, and the bottom is solid, and the bottom of the anchor foundation is provided with a longitudinal serrated groove, which is used as an anti-slip key.
[0005] Furthermore, the anchor foundation adopts a shallow buried enlarged foundation of C30 concrete.
[0006] A state detection method of an anti-slip key type gravity anchoring system specifically comprises the following steps:
[0007] Step 1: Collect historical status data of the anchor system, including: soil stress state around the anchor foundation, lateral displacement of the foundation rock and soil of the anchor system, triaxial acceleration, triaxial angular velocity, magnetic field angle change, vertical settlement and settlement rate of the anchor system, geometric posture of the anchor system, tension and displacement of the connection part between the anchor system and the suspension bridge;
[0008] Step 2: Preprocess the collected historical data;
[0009] Step 3: Establish a GAN model and use historical data to train the established GAN model;
[0010] Step 4: Use the trained GAN model to predict whether the anchoring system is abnormal.
[0011] Furthermore, the generator in the GAN model includes an input layer, first to seventh hidden layers and an output layer connected in sequence; the first hidden layer is a fully connected layer with 256 neurons; the second hidden layer is a LeakyReLU activation function layer; the third hidden layer is a Batch Normalization layer; the fourth hidden layer is a fully connected layer with 512 neurons, the fifth hidden layer is a LeakyReLU activation function layer, the sixth hidden layer is a Batch Normalization layer, and the seventh hidden layer is a fully connected layer for generating target data dimensions; the activation function used in the seventh hidden layer is a hyperbolic tangent activation function.
[0012] Furthermore, the discriminator in the GAN model includes an input layer, eighth to fourteenth hidden layers and an output layer connected in sequence; the eighth hidden layer is a Flatten layer, the input shape of the Flatten layer is (45, 30, 1), and the output shape is (32, 1290), the ninth hidden layer is a fully connected layer with 512 neurons, the tenth hidden layer is a LeakyReLU activation function layer, the eleventh hidden layer is a Dropout layer, the twelfth hidden layer is a fully connected layer with 256 neurons, the thirteenth hidden layer is a LeakyReLU activation function, and the fourteenth hidden layer is a Dropout layer.
[0013] A detection system for a state detection method of an anti-sliding key gravity anchoring system, comprising: a gravity anchoring system detection system, a GAN anomaly detection center and a data cloud platform; the data cloud platform comprises a data storage module, a data analysis module, a remote monitoring module and a user management module, the GAN anomaly detection center comprises a GAN training unit and a GAN model real-time detection unit, and the GAN model real-time detection unit comprises a GAN model;
[0014] The gravity anchor system detection system is used to collect status data of the anchor system and upload the data to the data storage module;
[0015] The data storage module transmits the received data to the data analysis module;
[0016] The data analysis module analyzes the received data, determines the state of the anchoring system at the corresponding moment, and inputs the determined state into the GAN model real-time detection unit;
[0017] The GAN training unit is used to preprocess the received data and train the GAN model;
[0018] The GAN model real-time detection unit uses the trained GAN model to detect the state of the anchoring system according to the state data collected in real time, and when an abnormality is detected, generates an alarm message and sends it to the remote monitoring module;
[0019] The user management is used to manage access rights of different users.
[0020] Furthermore, the gravity anchor foundation detection system includes a foundation deformation monitoring module, a foundation bottom soil pressure monitoring module, a foundation displacement monitoring module, and an anchor system monitoring module;
[0021] The foundation bottom soil pressure monitoring module includes a soil pressure box sensor device for monitoring the soil stress state around the anchor foundation;
[0022] The foundation deformation monitoring module includes a horizontal tilt measurement unit and a vertical settlement measurement unit;
[0023] The horizontal tilt measurement unit includes an intelligent digital inclinometer and an omnidirectional displacement meter. The intelligent digital inclinometer includes a velocity sensor and an electronic compass. The velocity sensor is used to monitor the three-axis acceleration and three-axis angular velocity of the anchoring foundation. The electronic compass is used to monitor the magnetic field angle. The omnidirectional displacement meter is used to monitor the horizontal displacement change of the anchoring system. The vertical settlement measurement unit includes a sliding resistance settlement meter and a settlement tube. The sliding resistance settlement meter is arranged on the settlement tube and is used to monitor the vertical settlement amount and settlement rate.
[0024] The foundation displacement monitoring module includes a total station and a drone equipped with a GPS three-dimensional geometric coordinate monitoring device. The total station is used to automatically monitor the settlement observation mark pasted in the area near the surface of the anchor foundation, measure the elevation value of the observation mark in real time, and obtain the settlement data by comparing it with the initial measured elevation value to complete the verification of the elevation value data; the drone is used to collect the geometric posture of the anchor system foundation;
[0025] The anchor system monitoring module includes an anchor cable tension meter and a displacement sensor. The anchor cable tension meter is used to monitor the tension of the portion where the anchor system is connected to the suspension bridge, and the displacement sensor is used to monitor the displacement of the portion where the anchor system is connected to the suspension bridge.
[0026] Furthermore, the horizontal inclination measurement unit also includes an inclinometer tube for placing an intelligent digital inclinometer. The inclinometer tube and the steel cage are arranged together inside the structure of the anchoring system and poured with concrete. The omnidirectional displacement meter is installed at both ends of the intelligent digital inclinometer.
[0027] Furthermore, the foundation displacement monitoring modules are arranged around the apex of the anchoring system.
[0028] Beneficial effects: The gravity anchor system of the present invention adopts an anti-slip key design. Compared with the previous flat-bottomed gravity anchor foundation with a smooth contact surface with the foundation rock and soil, the anti-slip key design adopted by the present invention can make the gravity anchor foundation use less concrete materials, and can show better anti-slip performance and improve the bearing mechanism of the anchor foundation foundation, while improving the stability and bearing capacity of the structure itself. Due to the clamping effect between the anchor foundation of the anti-slip key gravity anchor system and the rock and soil clamped in front, the bearing capacity of the rock and soil clamped in front of the foundation is more effectively utilized, and the anti-slip and anti-overturning bearing capacity of the anchor foundation is effectively enhanced.
[0029] The present invention combines the generative adversarial neural network (GAN) technology to realize the intelligent monitoring of the anti-sliding key gravity anchor system. Through the mutual adversarial training of the generator G and the discriminator D in the GAN model, abnormal conditions can be automatically identified from the monitoring data, thereby improving the intelligence level of the monitoring system. The generator G in the GAN model can generate pseudo data samples similar to normal data, and the discriminator D is used to identify the authenticity and abnormality of the input data. This method not only improves the processing efficiency of the monitoring data, but also can detect abnormal conditions in the monitoring system in real time, issue alarm information in time, and ensure the safe operation of the anti-sliding key gravity anchor system. Combined with the remote monitoring module of the data cloud platform, maintenance personnel can view the monitoring data and alarm information at any time to achieve remote management and decision-making. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] Figure 1 It is a three-dimensional model diagram of an anti-sliding key type gravity anchor foundation of the present invention;
[0031] Figure 2 It is a two-dimensional side view of a three-dimensional model perspective of an anti-sliding key type gravity anchor foundation of the present invention;
[0032] Figure 3 It is a structural diagram of a condition monitoring system based on an anti-sliding key type gravity anchor foundation of the present invention;
[0033] Figure 4 This is a schematic diagram of the arrangement of monitoring points in the foundation bottom soil pressure monitoring module of the present invention;
[0034] Figure 5 It is a top view of the arrangement of monitoring points in the foundation bottom soil pressure monitoring module of the present invention;
[0035] Figure 6 It is a top view of the arrangement of monitoring points in the basic deformation monitoring module of the present invention;
[0036] Figure 7 It is a side view of the arrangement of monitoring points in the basic displacement monitoring module of the present invention;
[0037] Figure 8 It is a top view of the arrangement of monitoring points in the basic displacement monitoring module of the present invention;
[0038] Fig. 9 It is a monitoring module, monitoring device and corresponding block diagram of quantity and monitoring value of an anti-sliding key type gravity anchor foundation of the present invention;
[0039] Fig.10 It is a schematic diagram of the process of the GAN model training phase of the present invention;
[0040] Fig.11 It is a schematic diagram of the process of the GAN model detection stage of the present invention;
[0041] Fig.12 Schematic diagram of the model structure of the GAN model of the present invention.
[0042] Explanation of the reference numerals: 1. loose cable saddle pier; 2. anchor chamber; 3. anchor block; 4. anchor foundation; 5. anti-slip key. DETAILED DESCRIPTION
[0043] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.
[0044] like Figure 1 and Figure 2 As shown, the present embodiment provides a reference to a gravity anchoring system based on an anti-sliding key, which includes a loose cable saddle pier 1, an anchor chamber 2, an anchor block 3, an anchor foundation 4 and an anti-sliding key 5.
[0045] The loose cable saddle pier 1 mainly bears the main cable pressure transmitted by the loose cable saddle; the anchor chamber 2 is a closed space located in the anchor block, which protects the main cable strand; the anchor block 3 bears the main cable strand tension transmitted by the prestressed anchoring system, and is arranged on the top of the anchor foundation (4). Its structural design needs to meet the tension requirement of the main cable strand to ensure the stability of the bridge; the anchor chamber 2, the anchor block 3 and the loose cable saddle pier 1 together form a complete three-bar herringbone structure of the spatial force-bearing structure, which ensures the overall stability by reasonably distributing the main cable pressure and tension. The stability and safety of the structure; the anchor foundation 4 adopts a shallow buried expanded foundation of C30 concrete, and the bottom surface size of the foundation is designed to be 57.8m×58m×9m (lateral×longitudinal×height). The front part of the anchor foundation 4 adopts a loose cable saddle box pier foundation design, and the rear part adopts a solid anchor block foundation design, which helps to enhance the anti-overturning performance of the structure. Two longitudinal serrated grooves are designed at the bottom of the anchor foundation 4, so that the bedrock under the anchor foundation 4 becomes the anti-slip key 5 of the overall structure; the anti-slip key 5 is arranged at the bottom of the anchor foundation 4.
[0046] The design height of the anti-sliding key 5 groove at the bottom of the anchor foundation 4 is 3m.
[0047] like Figure 3 The figure shows a structural diagram of a state detection system based on an anti-sliding key gravity anchoring system proposed in this embodiment, including a gravity anchoring foundation monitoring system, a GAN abnormal state detection center and a data cloud platform.
[0048] The gravity anchor foundation monitoring system can be divided into four monitoring modules, namely, foundation bottom soil pressure monitoring module, foundation deformation monitoring module, foundation displacement monitoring module and anchor system monitoring module; the GAN abnormal state detection center includes a GAN model training unit and a GAN model real-time detection unit; the data cloud platform includes a data storage module, a data analysis module, a remote monitoring module and a user management module.
[0049] Specifically, the output end of the gravity anchor foundation monitoring system is connected to the input end of the data storage module of the data cloud platform, and the data obtained by the four monitoring modules of the gravity anchor foundation monitoring system will be uploaded to the data storage module. The output end of the data storage module is connected to the input end of the data analysis module; the output end of the data analysis module is connected to the input end of the GAN model real-time detection unit; the input end of the GAN model real-time detection unit is also connected to the output end of the GAN model training unit, and the output end of the GAN model real-time detection unit is connected to the input end of the remote monitoring module; the output end of the remote monitoring module is connected to the user management module.
[0050] The foundation bottom soil pressure monitoring module includes a soil pressure box sensor device and a soil pressure box wire. The soil pressure box sensor device can be used to reflect the stress state of the soil around the bottom of the foundation. The soil pressure box sensor device adopts a range of 2Mpa. The soil pressure box wire is connected to the data storage module of the ground data cloud platform. The foundation bottom soil pressure monitoring module is arranged at the bottom of the anchor foundation. The specific arrangement is as follows: Figure 4 and Figure 5 shown.
[0051] The foundation deformation monitoring module includes a horizontal inclination measurement unit and a vertical settlement measurement unit.
[0052] The horizontal tilt measurement unit structurally includes an intelligent digital inclinometer, an omnidirectional displacement meter, a power signal line and an inclinometer tube, all of which are installed at the measuring point positions set at the front toe angle and the rear toe angle of the anchor foundation; the intelligent inclinometer adopts a serial inclination probe acquisition device, including an MPU6050 sensor, an HMC5883 electronic compass and a data processing circuit; the MPU6050 sensor collects changes in three-axis acceleration and three-axis angular velocity; the HMC5883 electronic compass collects changes in magnetic field angle; the data processing module is used to set configuration information such as acquisition accuracy and rate The data processing module is provided with a real-time clock circuit, which can accurately record the time of data collection; the omnidirectional displacement meter is installed at both ends of the intelligent digital inclinometer to monitor the displacement change of the intelligent digital inclinometer, that is, the anchoring system in three-dimensional space; the power signal line is used to transmit monitoring data to the data storage module of the ground data cloud platform; the inclinometer tube is used to place the intelligent digital inclinometer; the wall of the inclinometer tube is made of PVC pipe with a diameter of 70mm, which is tied together with the steel cage, arranged inside the anchoring system and poured with concrete.
[0053] The vertical settlement measuring unit includes a sliding resistance sedimentation meter, a sedimentation tube and a power signal line, wherein the sedimentation tube adopts a PVC pipe with a diameter of 70mm for placing the sliding resistance sedimentation meter; the sliding resistance sedimentation meter used in the vertical settlement measuring unit automatically records the vertical settlement amount and sedimentation rate of the measured area, and transmits them to the data storage module of the ground data cloud platform through the power signal line.
[0054] The arrangement positions of the inclinometer tube and the settling tube are as follows: Figure 6 As shown in the figure, yellow represents inclinometer tubes and green represents settlement tubes. The inclinometer tubes are numbered from inclinometer tube 1 to inclinometer tube 6 clockwise from the front toe corner of the foundation; the settlement tubes are numbered from settlement tube 1 to settlement tube 4 clockwise from the front toe corner of the foundation.
[0055] The foundation displacement monitoring module specifically includes a total station and an unmanned aerial vehicle (UAV) equipped with a GPS three-dimensional geometric coordinate monitoring device; the total station is used to automatically monitor the settlement observation marks pasted in the area near the surface of the anti-slip key gravity anchor system, measure the elevation value of the observation mark in real time, and obtain the settlement data by comparing it with the elevation value measured initially, and complete the verification of the elevation value data; the unmanned aerial vehicle (UAV) can take high-resolution images, and generate a three-dimensional point cloud and digital surface model (DSM) of the surface structure of the anchor system through image processing technology (Structure from Motion, SfM), so as to obtain real-time geometric posture change information such as the inclination, horizontal displacement and horizontal torsion of the surface part of the anchor foundation, and transmit the monitoring data to the data storage module of the data cloud platform.
[0056] References Figure 7 and Figure 8 As shown in the figure, in the implementation of the foundation displacement monitoring module, the method of arranging four geometric monitoring points around the top surface is adopted to monitor the plane coordinates and elevation of the control points in real time. Through calculation and analysis, the real-time display of the control point position, elevation, torsion angle, inclination and other geometric forms can be realized, and the settlement, horizontal displacement and torsion during the foundation construction and operation can be obtained.
[0057] The anchor system monitoring module specifically monitors the tension and displacement of the portion where the anchor system is connected to the suspension bridge by using an anchor tension meter and a displacement sensor placed on the loose cable saddle buttress.
[0058] Fig. 9 The monitoring modules, monitoring devices, quantities and corresponding monitoring values are displayed, and 45 groups of monitoring data monitored by the present invention in the gravity anchor foundation monitoring system are statistically analyzed.
[0059] The GAN abnormal state detection center includes a GAN model training unit and a GAN model real-time detection unit; Fig.10 As shown in FIG. 1 , the GAN model training unit aggregates 45 groups of monitoring data monitored in the gravity anchor foundation monitoring system into a monitoring data set and performs data preprocessing; the monitoring data that has completed data preprocessing is subjected to adversarial learning of two neural network algorithms, the generator G and the discriminator D, to train the GAN model's ability to recognize abnormal and normal state data, thereby obtaining the original GAN model; as shown in FIG. Fig.11 As shown, the GAN model real-time detection unit inputs the data monitored in real time by the monitoring system into the trained GAN model to detect the abnormal state of the data. When the model detects an abnormality, it generates an alarm message and sends it to the remote monitoring module of the data cloud platform.
[0060] The data cloud platform includes a data storage module, a data analysis module, a remote monitoring module and a user management module; the data storage module provides a large-capacity data storage space to store the data collected by the monitoring system and the training data of the GAN model; the data analysis module uses big data analysis technology to conduct in-depth analysis of historical monitoring data and identify the status of the anchoring system; the remote monitoring module provides a remote monitoring interface based on the cloud platform, and maintenance personnel can view monitoring data and alarm information in real time through the Internet, which is convenient for remote management and decision-making; the user management module manages the access rights of different users to ensure data security and stable operation of the system.
[0061] The present invention does not limit the model of the specific detection device of each module. As a preferred embodiment, the soil pressure box sensor device has a measurement range of 0-2MPa and an accuracy of 0.1%FS; the sensor device included in the intelligent digital inclinometer includes two sensors, MPU6050 sensor, HMC5883 electronic compass and omnidirectional displacement meter. The omnidirectional displacement meter can be selected as JMQJ-7915A model, with a dual-axis inclination of ±90° and an angle of 0.01°. The HMC5883 electronic compass can be selected as a sensor with a built-in ASIC amplifier, a 12-bit A / D converter, and an IIC bus output; the sliding resistance settlement meter sensor device has an accuracy of 0 .3mm / km, resolution 0.1mm, output signal RS485 digital output, voltage 5-24V; total station device distance measurement accuracy 2+2ppm, angle measurement accuracy 2"; anchor tension meter device measurement range 0-500kN, accuracy 0.5% FS (0.25% FS can also be considered); displacement sensor device uses linear displacement sensor, travel range 50-100mm, accuracy ±0.25% FS, resolution 0.01mm, output signal RS485 digital output; data storage module has 64TB storage capacity, supports RAID, uses Gigabit LAN interface, can be used for network attached storage (NAS) or cloud storage services, and supports multi-region backup; data analysis module supports SQL query, automatic expansion and is suitable for real-time data analysis; remote monitoring module can optionally integrate cloud monitoring software, support alarm and automatic response; user management module can be selected as user authority management identity and access management system supporting SSO and AWS IAM.
[0062] As a representative deep learning model, the GAN model is called Generative Adversarial Networks. The GAN model includes a generator G (Generator) and a discriminator D (Discriminator), and both the generator G and the discriminator D are neural network models.
[0063] In the training phase of the GAN model, the generator G is used to learn from the normal data samples that have been preprocessed and generate new data samples similar to normal data, that is, pseudo-structural health and safety data samples, to avoid the recognition of the two types of data by the discriminator D, making it impossible to distinguish between pseudo-structural health and safety data samples and monitored structural health and safety data samples, so as to complete the process of iterative optimization of data samples; when the trained GAN model is applied, the detection phase receives the real-time collected data as input, and after passing through the generator G, generates pseudo-structural health and safety data samples;
[0064] In the GAN model detection stage, the discriminator D is used to discriminate the pseudo-structural health and safety data samples generated by the generator G and the structural health and safety data samples obtained by monitoring. The discriminator D receives the above two types of data samples as input for discrimination, and judges as accurately as possible whether the input data sample represents the original structure to be normal or abnormal based on the loss value of the loss function. If the loss value is less than the preset threshold, it is considered to be a structural health and safety data sample, otherwise it is considered to be an abnormal data sample.
[0065] Before training the GAN model, the data samples need to be preprocessed, that is, normalized, in the data preprocessing process to ensure the quality and consistency of the collected monitoring data samples, that is, the samples are normalized to the range of [0,1]. The formula used to complete the normalization process is:
[0066]
[0067] In the formula, x′ is the data sample in the monitoring data set obtained from the four monitoring modules; x is the healthy and safe data sample obtained after completing data initialization, which exists as a real data sample.
[0068] like Fig.12 As shown in the figure, in the selection of the network architecture parameters of the GAN model, the input dimension of the generator G is designed to be 45×30, the output dimension of the generator is 45×30, the input dimension of the discriminator is 45×30, and the output is a probability value in the range of [0, 1], that is, the output dimension is 1; LeakyReLU and Tanh activation functions and the Adam optimizer with adaptive learning rate are used, and the initial learning rate is set to 2×10 -4 , beta_1=0.5, the number of iterations is 10000 times, and the batch size is 32. The output dimension of 43×30 means that the output dimension includes 45 types of data, and the number of each group of data is 30.
[0069] The generator G includes a 9-layer structure, an input layer, an output layer, and 7 hidden layers, including hidden layer G-1, hidden layer G-2, hidden layer G-3, hidden layer G-4, hidden layer G-5, hidden layer G-6, and hidden layer G-7;
[0070] The hidden layer G-1 is a fully connected layer with 256 neurons;
[0071] The hidden layer G-2 of the GAN model generator G is a LeakyReLU activation function, which is used to introduce nonlinearity. The specific formula of the LeakyReLU activation function used in the hidden layer G-2 is:
[0072]
[0073] Where α is the negative slope coefficient, and its value is α=0.01; G-2 is the output of the hidden layer G-2 of the generator G. It directly outputs when the input is greater than zero, otherwise it becomes a very small negative slope; x G-2 is the input of hidden layer G-2, and is the data sample.
[0074] The hidden layer G-3 of the GAN model generator G is a Batch Normalization layer, which is used to normalize the data and improve the training effect. The Batch Normalization layer used in the hidden layer G-3 divides the data set into smaller subsets (mini-batches), uses only one mini-batch at each parameter update, and performs a batch normalization on the input x of each mini-batch. G-3 The specific formulas for normalization, scaling, and translation are:
[0075]
[0076] In the formula, y G-3 is the output of the hidden layer G-3 of the generator G; x G-3 is the input of hidden layer G-3; μ G-3 and σ G-3 2 They are the input size x of the mini-batch respectively. G-3 The mean and variance of ; ε is a small constant used to prevent division by zero errors, and its value is generally ε = 10 -4 ; γ G-3 is the scaling variable; β G-3 is the translation variable.
[0077] Mini-batch input size x G-3 The mean μ G-3 and variance σ G-3 2 The expressions are:
[0078]
[0079] Where n is the input x in the mini-batch G-3 The number of .
[0080] The hidden layer G-4 is a fully connected layer using 512 neurons;
[0081] The hidden layer G-5 of the GAN model generator G is another LeakyReLU activation function; the LeakyReLU activation function used in the hidden layer G-5 directly outputs when the input is greater than zero, otherwise it becomes a very small negative slope:
[0082]
[0083] In the formula, y G-5 is the output of the hidden layer G-5 of the generator G; y G-5 is the input of hidden layer G-5, and x is the data sample.
[0084] The hidden layer G-6 of the GAN model generator G is another Batch Normalization layer; the hidden layer G-6 is the input x of each mini-batch. G-6 The specific formulas for normalization, scaling, and translation are:
[0085]
[0086] In the formula, y G-6 is the output of the hidden layer G-6 of the generator G; x G-6 is the input of hidden layer G-6; μ G-6 and σ G-6 2 are the mean and variance of the mini-batch respectively; γ G-6 is the scaling variable; β G-6 is the translation variable; the input x of the mini-batch G-6 The mean μ G-6 and variance σ G-6 2 The expressions are:
[0087]
[0088] Where m is the input amount x in the mini-batch G-6 The number of .
[0089] The hidden layer G-7 of the GAN model generator G is a fully connected layer that generates the target data dimension (i.e., 43 groups of 30 dimensions) and uses the Tanh activation function. The Tanh activation function used by the hidden layer G-7 is a hyperbolic tangent activation function, which compresses the input to [-1, 1]. The specific formula is:
[0090]
[0091] In the formula, y G-7 is the output of the hidden layer G-7 of the generator G; x G-7is the input of the hidden layer G-7; e is the base of the natural logarithm and has no other real meaning, and its value is e=2.71828.
[0092] In the selection of network architecture parameters of the GAN model discriminator D, it also includes a 9-layer structure, an input layer, an output layer and 7 hidden layers, including hidden layer D-1, hidden layer D-2, hidden layer D-3, hidden layer D-4, hidden layer D-5, hidden layer D-6, and hidden layer D-7 respectively; the convolution layer selected in the discriminator D adopts a 3×3 convolution kernel, the number of convolution kernels is 32, and the stride is 1; since the pooling layer is used to reduce the spatial dimension of the data, the size of the pooling layer is set to 2×2, and the stride is 2.
[0093] The hidden layer D-1 of the discriminator D is a Flatten layer, which flattens the multi-dimensional vector of the input layer into a one-dimensional vector to facilitate the input of the hidden layer D-2 fully connected layer; the input shape of the hidden layer D-1, Flatten layer is (45, 30, 1). Since each data point in the input shape has only one feature dimension, the number of channels in the input shape is 1, and the output shape after flattening is (32, 1290); the expression is:
[0094] Flatten(X)=Reshape(X,(N,-1))
[0095] Where X is the input tensor with a shape of (45, 30, 1); Reshape is a function that reshapes the input tensor; N is the batch size, which is determined by the training data and is set to N=32; -1 indicates that the dimension size is automatically calculated.
[0096] The hidden layer D-2 is a fully connected layer with 512 neurons;
[0097] The hidden layer D-3 of the GAN model discriminator D is the LeakyReLU activation function, and the specific formula is:
[0098]
[0099] In the formula, y D-3 is the output of the hidden layer D-3 of the discriminator D; x G-3 is the input of hidden layer D-3, and x is the data sample.
[0100] The hidden layer D-4 of the GAN model discriminator D is a Dropout layer. To prevent the GAN model from overfitting, that is, to prevent the model from overfitting the training data, the formula used is:
[0101]
[0102] In the formula, y D-4is the output of the hidden layer D-4 of the discriminator D; x G-4 is the input of hidden layer D-4; z i is a random variable, following a Bernoulli binomial distribution z i ~B(1,p), i=0,1; p is the dropout probability, the value is p=0.2, z i =1, the probability is 1-p=0.8, z i When =0, the probability is p=0.2.
[0103] The hidden layer D-5 is a fully connected layer with 256 neurons;
[0104] The hidden layer D-6 of the GAN model discriminator D is the LeakyReLU activation function, and the specific formula is:
[0105]
[0106] In the formula, y D-6 is the output of the hidden layer D-6 of the discriminator D; x G-6 is the input of hidden layer D-6, and x is the data sample.
[0107] The hidden layer D-7 of the GAN model discriminator D is another Dropout layer; the Dropout layer used in the hidden layer D-7 is
[0108]
[0109] In the formula, y D-7 is the output of the hidden layer D-7 of the discriminator D; x G-7 is the input of hidden layer D-7.
[0110] The training process of the GAN model is achieved through the confrontation between the generator G and the discriminator D. In the process of the game, the optimization target of the generator G and the discriminator D in the training process is defined as the loss function, which is a confrontation loss based on cross entropy. In the data sample training process, the loss function of the GAN abnormal state intelligent monitoring method based on the anti-sliding key gravity anchoring foundation includes the loss function L of the discriminator D. D , the loss function L of the generator G G And the loss function L of the GAN model GAN .
[0111] The loss function L of the discriminator D D It consists of two parts, including the data sample part of the structural health and safety obtained by monitoring and the data sample part of the pseudo structural health and safety generated by the generator G. The loss function L D The specific expression is:
[0112]
[0113] In the formula, z is the data sample of the structure obtained by real-time monitoring, which is the real-time monitoring data and serves as the input of the generator G; G(z) is the data sample of the pseudo structure health and safety generated by the generator, which serves as the output of the generator G; D(G(z)) is the output probability of the discriminator for the pseudo structure health and safety data sample z generated by the generator;
[0114] in is the expectation for the obtained healthy and safe data sample x, that is, it is hoped that the discriminator D can correctly identify the obtained healthy and safe data sample x and output a probability close to 1;
[0115] is the expectation for the generated data G(z), that is, it is expected that the discriminator D can identify the pseudo-structure healthy and safe data samples G(z) generated by the generator and output a probability close to 0; is the z~P from the data distribution z (z) is the expected value of the data sample z of the structure obtained by real-time monitoring, that is, the average value.
[0116] The loss function L of the generator G G It means that the generator G hopes to generate data samples G(z) that can deceive the discriminator D and make the discriminator D output close to 1. The specific expression used is:
[0117]
[0118] The loss function L of the GAN model is GAN It is necessary to maximize the expected value of the discriminator D, that is, it is more hoped that the discriminator D can correctly distinguish the obtained healthy and safe data samples x and the pseudo-structured healthy and safe data samples G(z) generated by the generator; minimize the expected value of the generator G, that is, it is hoped that the pseudo-structured healthy and safe data samples G(z) generated by the generator can deceive the discriminator D and make the discriminator D think that it is real data; the loss function L GAN The specific expression is:
[0119]
[0120] In the formula, the objective function L GAN To minimize the generator G and maximize the discriminator D Among them, max means taking the maximum value, min means taking the minimum value, D means the discriminator, and G means the generator.
[0121] The loss function of the GAN abnormal state intelligent monitoring method based on the anti-sliding key gravity anchor foundation includes the loss function L of the discriminator D D , the loss function L of the generator G G And the loss function L of the GAN model GAN Repeated calculation and iteration until the output result is lower than the preset error threshold or the preset number of iterations is reached to obtain a trained GAN model. At this time, the real-time monitoring data sample can be used as the input of the GAN model, and according to the loss function L GAN The loss value determines whether the input data sample represents the original structure to be normal or abnormal. If the loss value is less than a preset threshold, it is considered to be a data sample with healthy and safe structure, otherwise it is considered to be an abnormal data sample.
[0122] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.
Claims
1. An anti-slip key gravity anchoring system, It is characterized in that It includes an anchor foundation, an anchor block and a loose cable saddle pier; the loose cable saddle pier and the anchor block are both arranged on the anchor foundation, the anchor block and the loose cable saddle pier together form a complete three-bar herringbone structured spatial force-bearing structure, the top of the anchor block is provided with an anchor chamber, and the bottom is solid, and the bottom of the anchor foundation is provided with a longitudinal serrated groove, which is used as an anti-slip key.
2. The anti-slip key type gravity anchoring system according to claim 1, It is characterized in that The anchor foundation adopts a shallow buried enlarged foundation of C30 concrete.
3. The anti-slip key type gravity anchoring system according to claim 1, It is characterized in that The longitudinal sawtooth groove is 3m high.
4. A method for detecting the state of an anti-sliding key type gravity anchoring system according to claim 1, It is characterized in that The specific steps include: Step 1: Collect historical status data of the anchor system, including: soil stress state around the anchor foundation, lateral displacement of the foundation rock and soil of the anchor system, triaxial acceleration, triaxial angular velocity, magnetic field angle change, vertical settlement and settlement rate of the anchor system, geometric posture of the anchor system, tension and displacement of the connection part between the anchor system and the suspension bridge; Step 2: Preprocess the collected historical data; Step 3: Analyze the collected historical data to determine the state of the anchoring system at the corresponding moment; Step 4: Establish a GAN model and use historical data and corresponding states to train the established GAN model; Step 5: Use the trained GAN model to predict whether the anchoring system is abnormal.
5. A method for detecting the state of an anti-sliding key type gravity anchoring system according to claim 4, It is characterized in that The generator in the GAN model includes an input layer, first to seventh hidden layers and an output layer connected in sequence; the first hidden layer is a fully connected layer with 256 neurons; the second hidden layer is a LeakyReLU activation function layer; the third hidden layer is a BatchNormalization layer; the fourth hidden layer is a fully connected layer with 512 neurons, the fifth hidden layer is a LeakyReLU activation function layer, the sixth hidden layer is a Batch Normalization layer, and the seventh hidden layer is a fully connected layer for generating target data dimensions; the activation function used in the seventh hidden layer is a hyperbolic tangent activation function.
6. A method for detecting the state of an anti-sliding key type gravity anchoring system according to claim 4, It is characterized in that The discriminator in the GAN model includes an input layer, eighth to fourteenth hidden layers and an output layer connected in sequence; the eighth hidden layer is a Flatten layer, the input shape of the Flatten layer is (45, 30, 1), and the output shape is (32, 1290), the ninth hidden layer is a fully connected layer with 512 neurons, the tenth hidden layer is a LeakyReLU activation function layer, the eleventh hidden layer is a Dropout layer, the twelfth hidden layer is a fully connected layer with 256 neurons, the thirteenth hidden layer is a LeakyReLU activation function, and the fourteenth hidden layer is a Dropout layer.
7. A detection system for the state detection method of an anti-sliding key type gravity anchoring system according to claim 4, It is characterized in that include: A gravity anchoring system detection system, a GAN anomaly detection center and a data cloud platform; the data cloud platform includes a data storage module, a data analysis module, a remote monitoring module and a user management module, the GAN anomaly detection center includes a GAN training unit and a GAN model real-time detection unit, and the GAN model real-time detection unit includes a GAN model; The gravity anchor system detection system is used to collect status data of the anchor system and upload the data to the data storage module; The data storage module transmits the received data to the data analysis module and the GAN training unit; The data analysis module analyzes the received data, determines the state of the anchoring system at the corresponding moment, and inputs the determined state into the GAN model real-time detection unit; The GAN training unit is used to preprocess the received data and train the GAN model; The GAN model real-time detection unit uses the trained GAN model to detect the state of the anchoring system according to the state data collected in real time, and when an abnormality is detected, generates an alarm message and sends it to the remote monitoring module; The remote monitoring module is used to provide a remote monitoring interface, through which maintenance personnel can view status data, output results of the GAN model and alarm information in real time; The user management is used to manage access rights of different users.
8. The state detection system of the anti-slip key type gravity anchoring system according to claim 7, It is characterized in that The gravity anchor foundation detection system includes a foundation deformation monitoring module, a foundation bottom soil pressure monitoring module, a foundation displacement monitoring module, and an anchor system monitoring module; The foundation bottom soil pressure monitoring module includes a soil pressure box sensor device for monitoring the soil stress state around the anchor foundation; The foundation deformation monitoring module includes a horizontal tilt measurement unit and a vertical settlement measurement unit; The horizontal tilt measurement unit includes an intelligent digital inclinometer and an omnidirectional displacement meter. The intelligent digital inclinometer includes a velocity sensor and an electronic compass. The velocity sensor is used to monitor the three-axis acceleration and three-axis angular velocity of the anchoring foundation. The electronic compass is used to monitor the magnetic field angle. The omnidirectional displacement meter is used to monitor the horizontal displacement change of the anchoring system. The vertical settlement measurement unit includes a sliding resistance settlement meter and a settlement tube. The sliding resistance settlement meter is arranged on the settlement tube and is used to monitor the vertical settlement amount and settlement rate. The foundation displacement monitoring module includes a total station and a drone equipped with a GPS three-dimensional geometric coordinate monitoring device. The total station is used to automatically monitor the settlement observation mark pasted in the area near the surface of the anchor foundation, measure the elevation value of the observation mark in real time, and obtain the settlement data by comparing it with the initial measured elevation value to complete the verification of the elevation value data; the drone is used to collect the geometric posture of the anchor system foundation; The anchor system monitoring module includes an anchor cable tension meter and a displacement sensor. The anchor cable tension meter is used to monitor the tension of the portion where the anchor system is connected to the suspension bridge, and the displacement sensor is used to monitor the displacement of the portion where the anchor system is connected to the suspension bridge.
9. The state detection system of the anti-slip key type gravity anchoring system according to claim 8, It is characterized in that The horizontal tilt measurement unit also includes an inclinometer tube for placing an intelligent digital inclinometer. The inclinometer tube and the steel cage are arranged together inside the structure of the anchoring system and poured with concrete. The omnidirectional displacement meter is installed at both ends of the intelligent digital inclinometer.
10. The state detection system of the anti-slip key type gravity anchoring system according to claim 8, It is characterized in that The foundation displacement monitoring modules are arranged around the apex of the anchoring system.
