Bridge deformation abnormity early warning device and early warning method thereof

Abstract

The invention discloses a bridge deformation abnormity early warning device and an early warning method thereof, the device comprises a data acquisition and analysis device, an acquisition port of the data acquisition and analysis device is connected with a laser tilt angle sensor, an acceleration sensor and an optical fiber strain gauge, and an output port of the data acquisition and analysis device is connected with an audible and visual alarm; the early warning method comprises the following steps: determining a to-be-detected bridge, and obtaining original data of a sensor; performing feature extraction on the third environment parameter signal, and selecting an optimal model; performing abnormal point processing by using an LSTM time sequence prediction model; eliminating the temperature and vehicle load effect based on an auto-encoder in combination with an artificial neural network to obtain a corrected modal frequency after the temperature and vehicle load effect is eliminated; and formulating a scoring detailed rule, and scoring the dynamic characteristics of the bridge. According to the invention, the LSTM neural network is introduced to realize bridge deformation trend prediction, the auto-encoder is used to eliminate temperature and vehicle response, and the early warning perspectiveness is improved.

Description

Technical Field

[0001] This invention belongs to the field of bridge structural health monitoring technology, specifically relating to a bridge deformation anomaly early warning device and its early warning method. Background Technology

[0002] During long-term service, bridges may experience deformation phenomena such as increased deflection, tilting, and strain accumulation due to factors such as vehicle loads, temperature changes, wind, and earthquakes. If these deformations are not monitored and warned of in a timely manner, they may lead to a decrease in the bridge's load-bearing capacity or even structural safety accidents. Therefore, real-time monitoring of bridge structural deformation and timely warnings in case of anomalies are of great significance.

[0003] Most existing bridge monitoring systems rely on single sensors such as strain gauges and inclinometers, and use fixed thresholds as alarm criteria, resulting in problems such as slow response and high false alarm / missed alarm rates. For example, Chinese patent CN107403537A discloses an artificial intelligence early warning system for bridge deformation monitoring, which uses a data acquisition and analysis unit and an alarm module, improving the efficiency of bridge deformation identification to some extent. However, this system does not consider the coupled effect of temperature on deformation and does not introduce deep learning algorithms for trend prediction, resulting in high system energy consumption and making it unsuitable for long-term field deployment. Chinese patent CN219202337U discloses a bridge monitoring data cleaning system based on edge computing. This system mainly collects bridge structure monitoring data and performs data cleaning, which is implemented by an edge computing server, but it does not integrate an analysis and early warning module. Summary of the Invention

[0004] This invention is proposed to solve the problems existing in the prior art, and its purpose is to provide a bridge abnormal deformation early warning device and its early warning method.

[0005] The technical solution of the present invention is: a bridge deformation anomaly early warning device, comprising a data acquisition and analysis device, wherein the acquisition port of the data acquisition and analysis device is connected to a laser tilt sensor, an acceleration sensor, and a fiber optic strain gauge, the output port of the data acquisition and analysis device is connected to an audible and visual alarm, and the communication port of the data acquisition and analysis device is connected to a wireless communication unit.

[0006] Furthermore, the power supply port of the data acquisition and analysis device is connected to a solar power module, which supplies power to the device.

[0007] Furthermore, the laser tilt sensor is fixed to the monitoring point on the web of the main beam of the bridge using stainless steel bolts.

[0008] Furthermore, the accelerometer is installed in a double-layer damping base. The lower layer of the double-layer damping base is connected to the bridge pier by welding, and the upper layer of the double-layer damping base is coupled to the body of the accelerometer through a damping rubber pad.

[0009] Furthermore, the fiber optic strain gauges are arranged at equal intervals along the longitudinal direction of the bridge in the prestressed anchorage zone of the box girder top plate, and the fiber optic strain gauges are bonded to the structure with epoxy resin adhesive.

[0010] A method for early warning of abnormal bridge deformation includes the following steps: A. Identify the bridge to be inspected, acquire raw data from the sensors, and obtain the first deformation signal, the second deformation signal, and the third environmental parameter signal; B. Extract features from the third environmental parameter signal and select the optimal model; C. Use the LSTM time series prediction model to handle outliers; D. Based on the autoencoder and artificial neural network, the temperature and vehicle load effects are removed to obtain the corrected modal frequencies after removing the temperature and vehicle load effects; E. Develop detailed scoring rules and score the dynamic characteristics of the bridge based on the modified modal frequencies.

[0011] Furthermore, in step A, the raw data from the sensor is acquired to obtain the first deformation signal, the second deformation signal, and the third environmental parameter signal. The specific process is as follows: First, the three-axis tilt angle data of the main beam are collected using a laser tilt sensor array and marked as the first deformation signal; Then, the micro-strain changes inside the structure are obtained using fiber optic strain gauges and recorded as the second deformation signal; Finally, acceleration data of bridge vibration is collected using accelerometers to form a third environmental parameter signal.

[0012] Furthermore, step A also requires the transmission and processing of the first deformation signal, the second deformation signal, and the third environmental parameter signal. The specific process is as follows: First, the third environmental parameter signal is transmitted via bus to the edge computing unit in the data acquisition and analysis device; Then, the first deformation signal and the second deformation signal are subjected to moving average filtering.

[0013] Furthermore, step B involves feature extraction of the third environmental parameter signal and selection of the optimal model. The specific process is as follows: First, the third environmental parameter signal is input into a pre-trained LSTM network for feature extraction; Then, the prediction performance of the univariate prediction model is obtained; Then, the prediction results of the multivariate prediction model were obtained; Finally, the optimal model is selected by comparing the prediction performance of univariate and multivariate prediction models.

[0014] Furthermore, step C utilizes the LSTM time series prediction model to handle outliers, as detailed below: First, an outlier caused by sensor failure or human interference is removed using an LSTM time series prediction model. Then, the drifted data is corrected; Finally, missing data were supplemented, providing a reliable data foundation for subsequent health assessments.

[0015] The beneficial effects of this invention are as follows: This invention integrates a laser tilt sensor, a fiber optic sensor, and an accelerometer to achieve multi-source data fusion and improve deformation monitoring accuracy.

[0016] This invention introduces an LSTM neural network to predict bridge deformation trends and uses an autoencoder to eliminate temperature and vehicle response, improving early warning foresight. The invention employs a low-power design combined with solar power supply to enhance system adaptability and deployment flexibility. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the device of the present invention; Figure 2 This is a flowchart of the data analysis module algorithm of the present invention; The components include: 1. Laser tilt sensor; 2. Accelerometer; 3. Fiber optic strain gauge; 4. Data acquisition and analysis device; 5. Audible and visual alarm; 6. Wireless communication unit; and 7. Solar power module. Detailed Implementation

[0018] The present invention will now be described in detail with reference to the accompanying drawings and embodiments: like Figures 1 to 2 As shown, a bridge deformation anomaly early warning device includes a data acquisition and analysis device 4. The acquisition port of the data acquisition and analysis device 4 is connected to a laser tilt sensor 1, an acceleration sensor 2, and a fiber optic strain gauge 3. The output port of the data acquisition and analysis device 4 is connected to an audible and visual alarm 5. The communication port of the data acquisition and analysis device 4 is connected to a wireless communication unit 6.

[0019] The power supply port of the data acquisition and analysis device 4 is connected to the solar power module 7, which supplies power to the device.

[0020] The laser tilt sensor 1 is fixed to the monitoring point on the web of the main beam of the bridge by stainless steel bolts.

[0021] The accelerometer 2 is installed in a double-layer damping base. The lower layer of the double-layer damping base is connected to the bridge pier by welding, and the upper layer of the double-layer damping base is coupled to the body of the accelerometer 2 through a damping rubber pad.

[0022] The fiber optic strain gauges 3 are arranged at equal intervals along the longitudinal direction of the bridge in the prestressed anchorage area of ​​the top plate of the box girder, and the fiber optic strain gauges 3 are bonded to the structure with epoxy resin adhesive.

[0023] Specifically, the laser emission direction of the laser tilt sensor 1 forms a certain angle with the bridge axis, which is used to monitor the angle changes of bridge components in real time.

[0024] Specifically, the microprocessor SPI interface of the data acquisition and analysis device 4 is connected to the laser tilt sensor 1, the ADC channel is connected to the output of the accelerometer 2, and the GPI port communicates with the fiber optic strain gauge 3 through an optical coupler isolation circuit.

[0025] Specifically, the LSTM neural network is stored in external memory and interacts with the microprocessor through a DMA channel to process multi-source sensor data and identify abnormal deformation trends.

[0026] Specifically, the sound and light alarm 5 includes a piezoelectric buzzer and a high-brightness LED array. A solid-state relay is connected in series in the power supply circuit of the buzzer, and the LED driving circuit adopts a constant current source design.

[0027] Specifically, the wireless communication unit 6 includes a 4G module and a LoRa module. The 4G module antenna is installed on the non-metallic area at the top of the protective housing, and the LoRa antenna is connected to the PCB board through a connector.

[0028] Specifically, the solar power module 7 includes a flexible solar panel, a lithium iron phosphate battery pack, and an intelligent charge and discharge controller. A diode is connected in parallel to the output terminal of the solar panel and then connected to the charging circuit. A self-resetting fuse is connected in series in the positive terminal line of the battery pack.

[0029] Specifically, each sensor is connected to the data acquisition and analysis device 4 via wires or wireless means. The LSTM neural network in the data acquisition and analysis device 4 can perform trend modeling based on historical monitoring data and compare it with real-time monitoring data to determine whether there is an abnormal trend. If it exceeds the set threshold, the early warning module will be automatically triggered.

[0030] Specifically, after the data acquisition device 4 is powered on, it reads historical initialization data for preliminary training, or it can upload the data to the platform for model initialization and updating by the cloud server; all sensors complete zero-point calibration, and the system enters a monitoring standby state. Various sensors collect data according to a set cycle, such as every 5 minutes, and send it to the data analysis module to learn and predict parameters such as tilt angle changes, strain development trends, and bridge modal frequencies, automatically determining whether there are any abnormal trends.

[0031] Specifically, if continuous irregular trend fluctuations are detected in the data or if the future trend is predicted to exceed the limit, the audible and visual alarm 5 will be automatically activated and the alarm information will be sent to the manager's mobile phone via the wireless communication unit 6.

[0032] A method for early warning of abnormal bridge deformation includes the following steps: A. Identify the bridge to be inspected, acquire raw data from the sensors, and obtain the first deformation signal, the second deformation signal, and the third environmental parameter signal; B. Extract features from the third environmental parameter signal and select the optimal model; C. Use the LSTM time series prediction model to handle outliers; D. Based on the autoencoder and artificial neural network, the temperature and vehicle load effects are removed to obtain the corrected modal frequencies after removing the temperature and vehicle load effects; E. Develop detailed scoring rules and score the dynamic characteristics of the bridge based on the modified modal frequencies.

[0033] Step A involves acquiring raw sensor data to obtain the first deformation signal, the second deformation signal, and the third environmental parameter signal. The specific process is as follows: First, a set of laser tilt sensors was used to collect the three-axis tilt angle data of the main beam and marked it as the first deformation signal. Then, the micro-strain changes inside the structure are obtained using fiber optic strain gauge 3 and recorded as the second deformation signal; Finally, acceleration data of bridge vibration is collected using accelerometer 2 to form a third environmental parameter signal.

[0034] Step A also requires the transmission and processing of the first deformation signal, the second deformation signal, and the third environmental parameter signal. The specific process is as follows: First, the third environmental parameter signal is transmitted via bus to the edge computing unit in the data acquisition and analysis device 4; Then, the first deformation signal and the second deformation signal are subjected to moving average filtering.

[0035] Step B involves feature extraction from the third environmental parameter signal and selection of the optimal model. The specific process is as follows: First, the third environmental parameter signal is input into a pre-trained LSTM network for feature extraction; Then, the prediction performance of the univariate prediction model is obtained; Then, the prediction results of the multivariate prediction model were obtained; Finally, the optimal model is selected by comparing the prediction performance of univariate and multivariate prediction models.

[0036] Step C utilizes the LSTM time series prediction model to handle outliers. The specific process is as follows: First, an outlier caused by sensor failure or human interference is removed using an LSTM time series prediction model. Then, the drifted data is corrected; Finally, missing data were supplemented, providing a reliable data foundation for subsequent health assessments.

[0037] Specifically, in step B, to avoid overfitting of the LSTM prediction model, the input features were standardized, and the input variables were adjusted to zero mean and unit variance to ensure that the input features were at the same scale during model training, thereby improving the stability and speed of training.

[0038] Specifically, step D uses an autoencoder combined with an artificial neural network to remove temperature and vehicle load effects, obtaining the corrected modal frequencies after removing these effects. The specific process is as follows: First, an autoencoder (AE) is used to preprocess the environmental variables, with acceleration, deflection, and temperature data used as the input and output of the autoencoder to decouple the environmental variables. Then, the PSO-BP neural network was selected as the network structure of the artificial neural network (ANN), and the decoupled environmental variables were used as inputs and the measured structure frequencies were used as outputs for network training. Finally, based on the formula for calculating the modified modal frequency, the modified modal frequency after eliminating the effects of temperature and vehicle load is calculated.

[0039] Specifically, step E involves developing detailed scoring rules and scoring the bridge's dynamic characteristics based on the modified modal frequencies. The specific process is as follows: These scoring rules are based on the provisions for bridge frequency assessment in the Technical Specifications for Highway Bridge Structure Monitoring.

[0040] Specifically, the loss function selected for training the LSTM network in this invention is obtained through the following formula: ; in, Y n and y nThese represent the actual value and the predicted value, respectively. N Represents the length of the sequence; The loss function mentioned above is used in model training to measure the difference between the model's predicted values ​​and the true values, helping to guide model optimization.

[0041] Mean squared error (MSE) is chosen as the loss function because it is easy to calculate and widely used. By penalizing large errors, it helps the model fit the data more accurately. In optimization problems, the squared term of MSE guarantees that it is differentiable, which is very important for optimization algorithms such as gradient descent.

[0042] Specifically, in this invention, the Adam (Adaptive Moment Estimation) optimizer is selected to train the LSTM network. It considers both the first moment (mean) and the second moment (uncentered variance) of the gradient, and adaptively adjusts the learning rate according to the gradient of each parameter. In this embodiment, the update parameter is obtained by the following formula: ; in, m t It is an exponential moving average of the gradient. v t It is the exponential moving average of the squared gradient. In time step t gradient, In time step t The parameters, It's the learning rate. and These are the decay rates of the gradient and its squared moving average, respectively. It is a very small number to avoid division by zero.

[0043] Specifically, the modified modal frequency used in this invention utilizes the encoder part of the autoencoder to decouple and reduce the environmental variables, resulting in a new output variable. The measured modal frequency is used as the output to train the network parameters.

[0044] The network structure has 2 input layers, 1 output layer, and 20 hidden layers. The dataset is divided into training and test sets using a random partitioning method.

[0045] In practical applications, based on the acceleration in each span, the random subspace method is used for modal identification. The measured modal frequencies of the first four vibration modes are identified in a 5-minute time interval and then input into the neural network to obtain the first four reference frequencies.

[0046] Specifically, according to the technical specifications for monitoring the structure of highway bridges (JT_T 1037-2022CN), the main frequency changes of bridges are defined as follows: frequency changes exceeding 3% after excluding environmental influences are classified as Level II over-limit, and 5% are classified as Level III over-limit; based on this, scoring rules are formulated for the assessment of bridge dynamic characteristics.

[0047] Specifically, the bridge dynamic characteristic score is based on the percentage difference between the corrected frequency and the theoretical frequency to determine the bridge condition; If the frequency change is between 0% and 3%, the bridge's service dynamic characteristics are considered good, with a score between 100 and 80. If the frequency change is between 3% and 5%, the bridge's service dynamic characteristics are considered normal, with a score between 80 and 60. If the frequency change exceeds 5%, the bridge's service dynamic characteristics are considered unqualified, with a score below 60.

[0048] This invention integrates a laser tilt sensor, a fiber optic sensor, and an accelerometer to achieve multi-source data fusion and improve deformation monitoring accuracy.

[0049] This invention introduces an LSTM neural network to predict bridge deformation trends and uses an autoencoder to eliminate temperature and vehicle response, improving early warning foresight. The invention employs a low-power design combined with solar power supply to enhance system adaptability and deployment flexibility.

Claims

1. A bridge deformation anomaly early warning device, comprising a data acquisition and analysis device (4), characterized in that: The acquisition port of the data acquisition and analysis device (4) is connected to the laser tilt sensor (1), the acceleration sensor (2), and the fiber optic strain gauge (3). The output port of the data acquisition and analysis device (4) is connected to the audible and visual alarm (5). The communication port of the data acquisition and analysis device (4) is connected to the wireless communication unit (6).

2. The bridge deformation anomaly early warning device according to claim 1, characterized in that: The power supply port of the data acquisition and analysis device (4) is connected to the solar power module (7), and the solar power module (7) supplies power to the device.

3. The bridge deformation anomaly early warning device according to claim 1, characterized in that: The laser tilt sensor (1) is fixed to the monitoring point of the web of the main beam of the bridge by stainless steel bolts.

4. The bridge deformation anomaly early warning device according to claim 1, characterized in that: The acceleration sensor (2) is installed in a double-layer damping base. The lower layer of the double-layer damping base is connected to the bridge pier by welding, and the upper layer of the double-layer damping base is coupled to the body of the acceleration sensor (2) through a damping rubber pad.

5. The bridge deformation anomaly early warning device according to claim 1, characterized in that: The fiber optic strain gauges (3) are arranged at equal intervals along the longitudinal direction of the bridge in the prestressed anchorage area of ​​the top plate of the box girder. The fiber optic strain gauges (3) are bonded to the structure with epoxy resin.

6. The early warning method for a bridge deformation anomaly early warning device according to claim 1, characterized in that: Includes the following steps: A. Identify the bridge to be inspected, acquire raw data from the sensors, and obtain the first deformation signal, the second deformation signal, and the third environmental parameter signal; B. Extract features from the third environmental parameter signal and select the optimal model; C. Use the LSTM time series prediction model to handle outliers; D. Based on the autoencoder and artificial neural network, the temperature and vehicle load effects are removed to obtain the corrected modal frequencies after removing the temperature and vehicle load effects; E. Develop detailed scoring rules and score the dynamic characteristics of the bridge based on the modified modal frequencies.

7. The early warning method for a bridge deformation anomaly early warning device according to claim 1, characterized in that: Step A involves acquiring raw sensor data to obtain the first deformation signal, the second deformation signal, and the third environmental parameter signal. The specific process is as follows: First, the three-axis tilt angle data of the main beam are collected using a laser tilt sensor (1) group and marked as the first deformation signal; Then, the micro-strain changes inside the structure are obtained using fiber optic strain gauges (3), and recorded as the second deformation signal; Finally, the acceleration data of the bridge vibration is collected using the accelerometer (2) to form the third environmental parameter signal.

8. The early warning method for a bridge deformation anomaly early warning device according to claim 7, characterized in that: Step A also requires the transmission and processing of the first deformation signal, the second deformation signal, and the third environmental parameter signal. The specific process is as follows: First, the third environmental parameter signal is transmitted via bus to the edge computing unit in the data acquisition and analysis device (4); Then, the first deformation signal and the second deformation signal are subjected to moving average filtering.

9. The early warning method for a bridge deformation anomaly early warning device according to claim 1, characterized in that: Step B involves feature extraction from the third environmental parameter signal and selection of the optimal model. The specific process is as follows: First, the third environmental parameter signal is input into a pre-trained LSTM network for feature extraction; Then, the prediction performance of the univariate prediction model is obtained; Then, the prediction results of the multivariate prediction model were obtained; Finally, the optimal model is selected by comparing the prediction performance of univariate and multivariate prediction models.

10. The early warning method of the bridge deformation anomaly early warning device according to claim 1, characterized in that: Step C utilizes the LSTM time series prediction model to handle outliers. The specific process is as follows: First, an outlier caused by sensor failure or human interference is removed using an LSTM time series prediction model. Then, the drifted data is corrected; Finally, missing data were supplemented, providing a reliable data foundation for subsequent health assessments.

Images