Method for rapidly evaluating structural safety after bridge fire

Abstract

The invention discloses a method for rapidly evaluating the safety of a structure after a bridge fire, and the method comprises the steps: obtaining an air temperature, flue gas speed distribution, air pressure distribution and a flame flue gas image through simulating a fire scene; the method comprises the following steps: when simulating that a fire enters a fire quasi-steady-state stage, obtaining a fire smoke image for completely describing a dynamic change process of thermal plume and a smoke speed distribution image and an air pressure distribution image above a fire source, and forming a training sample; constructing a long and short term memory network model; training the model; shooting a flame smoke image of a fire scene, inputting the flame smoke image into the trained long-short term memory network model, and outputting a smoke speed distribution image; calculating a final fire source heat release rate according to the flame smoke image and the smoke speed distribution image; the average temperature of the components in the cable at different moments is calculated; according to the average temperature of the components in the cable at different moments, whether the structural components lose efficacy or not is judged. According to the invention, the monitoring accuracy and timeliness are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of bridge safety assessment, and particularly relates to a method for rapid assessment of the structural safety of a bridge after a fire. Background Art

[0002] Fires occurring on the upper part of a bridge often cause serious damage to the load-bearing capacity of the main girder structure, cable components, etc. on the upper part of the bridge, affecting structural safety and hindering the normal operation of local traffic. Traditional bridge structural safety monitoring means, such as manual inspections or fixed monitoring stations, although to a certain extent can monitor the bridge state, are limited by the insufficient monitoring range and monitoring timeliness, and it is difficult to timely and accurately grasp the safety status of the bridge in extreme situations such as fires.

[0003] With the rapid development of unmanned aerial vehicle (UAV) technology, its application in the field of bridge structural safety monitoring has gradually received attention. UAVs have the advantages of strong mobility, wide monitoring range, flexible operation, etc., and can achieve rapid and efficient monitoring of bridge structures. However, how to combine UAV technology with bridge structural safety monitoring and early warning, especially timely monitoring in extreme situations such as fires, is still an urgent problem to be solved. Summary of the Invention

[0004] Object of the Invention: The object of the present invention is to provide a method for rapid assessment of the structural safety of a bridge after a fire, improve the accuracy and timeliness of monitoring, and provide a strong guarantee for the safe operation of the bridge.

[0005] Technical Solution: To achieve the above object, the present invention discloses a method for rapid assessment of the structural safety of a bridge after a fire, including the following steps:

[0006] S1. According to the influencing factors of the fire scenario, construct a number of simulation conditions, set the fire simulation duration, fire simulation space, and the size of the simulation cell, and obtain the air temperature, smoke velocity distribution, air pressure distribution, and flame and smoke images through simulation of the fire scenario.

[0007] S2. When the fire enters the quasi-steady state stage of the fire, set the acquisition period T, save the flame and smoke images of the fire simulation space every time T, and obtain a continuous image of the fire scenario that completely describes the dynamic change process of the thermal plume, as well as the smoke velocity distribution image and air pressure distribution image above the fire source; intercept data every 4 thermal plume oscillation periods to form a set of flame and smoke images, and a number of fire simulation conditions can form a number of non-overlapping training samples.

[0008] S3. Construct a long short-term memory network model. The input of the long short-term memory network model includes a set of flame and smoke images for 4 thermal plume oscillation periods. The output of the long short-term memory network model is the smoke velocity distribution image and the air pressure distribution image above the fire source. The total loss of the loss function in the long short-term memory network model includes a data error term and a physical information error term. The data error term is the mean square error of the corresponding pixel values between the predicted image and the simulated smoke distribution map. The physical information error term is the mean square error of the residual term after substituting the differential form of the Navier-Stokes equation.

[0009] S4. Use the training samples to train the long short-term memory network model.

[0010] S5. Take pictures of the flame and smoke images at the fire scene, convert the flame and smoke images collected by the camera to actual sizes, and input the converted flame and smoke images into the trained long short-term memory network model to output the smoke velocity distribution image.

[0011] S6. Calculate the heat release rate of the fire source according to the flame and smoke images and the smoke velocity distribution image respectively, and take the maximum value of the average of the two heat release rates of the fire source over a period of time as the final heat release rate of the fire source.

[0012] S7. Match the calculated final heat release rate of the fire source with the simulated working conditions, query the highest air temperature near the bridge structure according to the matching result, and calculate the average temperature of the components in the cable at different times.

[0013] S8. Determine the component resistance R at the average temperature according to the average temperature of the components in the cable at different times. When the component resistance R at the average temperature is less than the internal force S generated by the design load on the component, it is determined that the structural component fails.

[0014] Optionally, when constructing the fire scenario in step (1), it is set that a single vehicle catches fire, and the influencing factors are set as the geometric size of the fire source, the heat release rate of the fire source, the smoke production rate, and the environmental wind speed. The geometric size of the fire source is set as the size of a small passenger car, a large passenger car, a heavy truck, and an oil tanker. The range of the heat release rate of the fire source of the small passenger car is 1.5 - 4.5 MW, and M1 different heat release rates of the fire source are selected from it; the range of the heat release rate of the fire source of the large passenger car is 15 - 45 MW, and M2 different heat release rates of the fire source are selected from it; the range of the heat release rate of the fire source of the heavy truck is 80 - 150 MW, and M3 different heat release rates of the fire source are selected from it; the range of the heat release rate of the fire source of the oil tanker is 180 - 250 MW, and M4 different heat release rates of the fire source are selected from it; the range of the smoke production rate is 0.05 - 0.6 kg / kg, and M5 different sets of smoke production rate values are selected from it; the range of the environmental wind speed change is 0 - 15 m / s, and M6 different sets of environmental wind speed values are selected from it; the orthogonal experiment method is used to construct (M1 + M2 + M3 + M4) × M5 × M6 simulation conditions.

[0015] Optionally, it is assumed in step (1) that the heat release rate of the fire source remains constant from ignition to the end, and the fire simulation duration is set. The length and width directions of the fire simulation space are taken as the vehicle size extended by more than 3 m outward, and the height direction is determined according to the principle that the flame is completely within the simulation space; when performing the simulation, the recommended cell size is set according to the dimensionless grid resolution, where:

[0016]

[0017] The value range of the dimensionless grid resolution is 16 - 24, D * is the characteristic diameter of the fire source, δx is the recommended cell size; the preset recommended cell size is n, and the simulation is performed with n cells, and at the same time, the simulation is performed with 0.5n cells. Compare the two simulation results, compare the temperature-time curves at the same measurement point. If the difference is less than 10%, then take n as the cell setting size; otherwise, further reduce the grid size until the difference between the two simulation results is less than 10%, then output the current cell size as the cell setting size to determine the final grid size.

[0018] Optionally, the method for determining the acquisition period T in step (2) is as follows: According to the vehicle width W, and considering the reduction in the contribution of the heat plume height in the length direction, when calculating the flame length, the effective flame length is 2 times the vehicle width. Therefore, the effective fire source area is A eff = W × 2 × W, then the oscillation frequency f of the heat plume plume = 1.68D -0.5 , where D is the equivalent diameter of the fire source, Calculate to obtain f plume, the oscillation period T of the thermal plume plume is:

[0019]

[0020] The acquisition period T takes an integer multiple of 0.1 times the oscillation period T of the thermal plume plume .

[0021] Optionally, the long short-term memory network model in step (3) is composed of 1 input layer, 1 output layer and 6 hidden layers. Among them, several states of the input layer read the flame and smoke images taken at intervals of the acquisition period T within a period of time, and the three states of the output layer are the horizontal smoke velocity distribution image, the vertical smoke velocity distribution image, and the air pressure distribution image respectively.

[0022] Optionally, the specific calculation method of the total loss in step (3) is:

[0023] Calculate the first-order differential of the smoke velocity and air pressure in space, that is: u x = gradient(u, x), u y = gradient(u, y), v x = gradient(v, x), v y = gradient(v, y), p x = gradient(p, x), p y = gradient(p, y); where u and v respectively refer to the horizontal and vertical smoke velocities, and p is the air pressure;

[0024] Calculate the second-order differential of the smoke velocity in space, that is: u xx = gradient(u x , x), u yy = gradient(u y , y), v xx = gradient(v x , x), v yy = gradient(v y , y);

[0025] Based on the air density rho and the air dynamic annual coefficient mu, calculate the residual terms of the smoke velocity and air pressure, that is residual u = rho × (u × u x + v × u y ) - mu × (u xx + u yy ) + p x , residual v = rho × (u × v x+v×v y ) - mu×(v xx +v yy ) + p y ,residual p =p x +p y ;

[0026] Physical information error err phy is the mean square error of three errors residual u 、residual v and residual p ; The total loss err tot is the sum of the data error term and the physical loss term, that is, err tot =err data +err phy 。

[0027] Optionally, the specific method for performing the actual size conversion in step (5) is as follows:

[0028] The actual geometric size L represented by a single pixel in the image can be calculated through the width, height, image format, focal length of the camera sensor, and the distance to the target pxH and L pxV :

[0029]

[0030] Optionally, when calculating the heat release rate of the fire source using the flame and smoke image in step (6), the flame and smoke image collected by the UAV is converted from the RGB space to the HSV space, and the flame area image is extracted through the usual color threshold range of the flame: [0, 100, 100]~[10, 255, 255]; due to the dynamic characteristics of the flame and being blocked by the smoke, the extracted flame image may be discontinuous; the extracted flame area image is binarized, the pixel value of the flame area is 1, and the rest of the pixel values are 0, filtering out the smoke area in the image; the 8-connected method is used to mark all connected areas, and the number N of all non-0 pixel values in each connected area of the flame area image is counted, and the non-0 pixel numbers in each connected area are sorted; according to the area with the most non-0 pixels, according to the actual geometric size L pxV represented by a single pixel, the actual height L of the flame area is determined, and the heat release rate of the fire source is inversely calculated according to the flame height

[0031]

[0032] Among them, ρ ∞ ,c p ,T ∞, g are the density, specific heat capacity, temperature and gravitational acceleration of air at room temperature, and D is the equivalent diameter of the fire source.

[0033] Optionally, when the smoke velocity distribution image is used to calculate the heat release rate in step (6), the maximum value u0 of the average smoke velocity at different vertical heights is calculated based on the identified vertical smoke velocity u distribution, and then the heat release rate of the fire source is inversely calculated.

[0034]

[0035] Among them, z0 is the height of the virtual fire source, which is defined as:

[0036]

[0037]

[0038] The heat release rate of the fire source can be obtained by combining the two equations

[0039] Optionally, the average temperature T in step (7) s The calculation formula is:

[0040]

[0041] Where t is the time instant, ρ s and c s are the density and specific heat capacity of steel, α=α c +α r , α c is the heat convection coefficient, ε r and σ are the thermal radiation coefficient and Boltzmann constant respectively; the shape coefficient 4 / d is determined according to the specific size of the cable, where d is the cable diameter.

[0042] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: the present invention simulates the fire by constructing several simulated working conditions to obtain sufficient training samples to train the long short-term memory network model. The error function in the long short-term memory network model takes into account the data error term and the physical information error term, thereby enhancing the reliability and generalization ability of the predicted data; the smoke velocity distribution image in the fire scene image taken by the drone is extracted through the long short-term memory network model, and the heat release rate of the fire source is inverted. Finally, combined with the high-temperature performance of the material, whether the bridge structure components have failed is evaluated, thereby providing a scientific basis for post-disaster emergency decision-making; the present invention can timely monitor the fire scene on the upper part of the bridge to quickly evaluate the safety of the bridge structure under fire, improve the accuracy and timeliness of monitoring, and provide a strong guarantee for the safe operation of the bridge. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a schematic diagram of the framework of the long short-term memory network model in the present invention. Specific implementation manners

[0044] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0045] As Figure 1 shown, the present invention discloses a method for rapid assessment of the structural safety of a bridge after a fire, including the following steps:

[0046] S1. According to the influencing factors of the fire scenario, several simulation conditions are constructed, the fire simulation duration, the fire simulation space and the size of the simulation cell are set, and the fire scenario is simulated through numerical simulation to obtain the air temperature, the distribution image of the smoke velocity, the air pressure distribution and the flame and smoke image.

[0047] Since the heat release rate generated during vehicle combustion is related to the vehicle, the two factors are not completely independent; although in an actual bridge fire, multiple vehicles may catch fire simultaneously due to a traffic accident, most scenarios involve a single vehicle catching fire. Therefore, it is set as a single vehicle catching fire when constructing the fire scenario, and the influencing factors are set as the geometric size of the fire source, the heat release rate of the fire source, the smoke production rate and the environmental wind speed. The geometric size of the fire source is set as the size of a small passenger car, a large passenger car, a heavy truck and an oil tanker. The size of the small passenger car is 4m and 2.5m, and the range of the heat release rate of the fire source of the small passenger car is 1.5 - 4.5MW. The heat release rate values are 1.5MW, 3MW, 4.5MW, and 3 different heat release rates are selected from them; the size of the large passenger car is 8m and 2.5m, and the range of the heat release rate of the fire source of the large passenger car is 15 - 45MW. The heat release rate values are 15MW, 30MW, 45MW, and 3 different heat release rates are selected from them; the size of the heavy truck is 12m and 2.5m, and the range of the heat release rate of the fire source of the heavy truck is 80 - 150MW. The heat release rate values are 80MW, 100MW, 150MW, and 3 different heat release rates are selected from them; the size of the oil tanker is 12m and 2.5m, and the range of the heat release rate of the fire source of the oil tanker is 180 - 250MW. The heat release rate values are 180MW, 200MW, 250MW, and 3 different heat release rates are selected from them.

[0048] The range of the smoke production rate is 0.05 - 0.6 kg / kg. The fuel type of ordinary vehicles is gasoline, and its smoke production rate is 0.05 - 0.15 kg / kg; the main fuel types such as the seat interior in the vehicle are plastics, and its smoke production rate is 0.3 - 0.6 kg / kg; considering that the actual fire is a mixed fuel mainly composed of gasoline, five different values are formed: 0.05 kg / kg, 0.10 kg / kg, 0.15 kg / kg, 0.30 kg / kg, 0.45 kg / kg, and five different smoke production rate values are selected from them.

[0049] The present invention considers that the range of the lateral environmental wind speed of the suspension bridge is 0 - 15 m / s, and five different environmental wind speed values are selected from it, 0 m / s (no wind), 3 m / s, 6 m / s, 10 m / s, 15 m / s; since when using a drone for shooting, the shooting point and angle of the drone can be controlled, and shooting is carried out in the direction perpendicular to the environmental wind direction, therefore, in order to reduce the number of simulation conditions and improve the numerical simulation efficiency, the influence of the environmental wind direction is not considered.

[0050] The orthogonal experiment method is used to construct (3 + 3 + 3 + 3)×5×5 = 300 simulation conditions; since the initial development stage after the vehicle catches fire is short, its development stage is not considered in the simulation, and it is assumed that the heat release rate of the fire source remains constant from ignition to the end. The vehicle fire can last up to 2 hours. Considering that the smoke in the fully developed stage of the fire is periodically oscillating, in order to capture the fire characteristics of a sufficient number of cycles while reducing the simulation time, the fire simulation duration is set to 30 min. The length and width directions of the fire simulation space are taken as the vehicle size extended by more than 3 m outward. In the present invention, 5 m is taken, and the height direction is determined according to the principle that the flame is completely within the simulation space; when performing the simulation, the recommended cell size is set according to the dimensionless grid resolution, where:

[0051]

[0052] The value range of the dimensionless grid resolution is 16 - 24, D * is the characteristic diameter of the fire source, and δx is the recommended cell size; the preset recommended cell size is n = 0.1 m, and the simulation is carried out with cells of n = 0.1 m. At the same time, the simulation is carried out with cells of 0.5n = 0.05 m. The temperature-time curves of the same measurement point are compared for the two simulation results. If the difference is less than 10%, then n = 0.1 m is taken as the cell setting size; otherwise, the grid size is further reduced until the difference between the two simulation results before and after is less than 10%, and then the current cell size is output as the cell setting size to determine the final grid size of 0.1 m; while ensuring the accuracy of the simulation results, the simulation efficiency is improved.

[0053] S2. Extract the fire simulation result data under each simulated condition, that is, the smoke velocity distribution image, the air pressure distribution image, and the flame and smoke image to form training samples;

[0054] Use the open-source fire simulation result visualization software SmokeView, turn on the GPU display switch, and load the smoke density and the heat release rate per unit volume at the same time to display the superimposed image of the smoke and the flame. The period from 5 minutes to 30 minutes after the start of the fire is the quasi-steady state stage of the fire. At this time, the thermal plume movement shows periodic oscillation characteristics. Set the acquisition period T to 0.1 s, and save the RGB flame and smoke image of the simulation space every time T to obtain a continuous image of the fire scene that completely describes the dynamic change process of the thermal plume;

[0055] The method for determining the acquisition period T is as follows: According to the vehicle width W = 2.5 m, and considering the reduction in the contribution of the thermal plume height in the length direction, when calculating the flame length, the effective length of the flame is 2 times the vehicle width. Therefore, the effective fire source area is A eff = W × 2 × W = 2.5 m × 2 × 2.5 m = 12.5 m 2 , then the oscillation frequency f of the thermal plume plume = 1.68D -0.5 , where D is the equivalent diameter of the fire source, Calculate to get f plume to be 0.67, and the oscillation period of the thermal plume The acquisition period T takes an integer multiple of 0.1 times the oscillation period T of the thermal plume plume , so the acquisition period T is 0.1 s;

[0056] In the SmokeView simulation software, reload the smoke velocity distribution and air pressure distribution slices that pass through the fire source center and are perpendicular to the fire source width direction. The smoke velocity distribution includes the vertical V pluH and the horizontal V pluV , and use the fire simulation post-processing program fds2ascii to convert the data recorded in the image into txt format data at an interval of the acquisition period T = 0.1 s to obtain the smoke velocity distribution image and air pressure distribution image above the fire source;

[0057] For the flame and smoke images, smoke velocity distribution images, and air pressure distribution images under each fire simulation condition, intercept the data every 4 thermal plume oscillation periods (every 6 s) to form a set of continuous images. In the 5 - 30 minute quasi-steady state stage, several groups of 250 sets of continuous images can be formed. Then, 300 fire simulation conditions can form 75,000 non-overlapping training samples; Respecify the resolution of all the images in the training samples to 128 pixels × 128 pixels, shuffle the order of the training samples, and divide them into a training set, a validation set, and a test set according to 5:3:2;

[0058] S3. Construct a long short - term memory network model embedded with the mechanism of flue gas flow

[0059] The input of the long short - term memory network model includes a set of flame flue gas images for 4 thermal plume oscillation periods. The output of the long short - term memory network model is the flue gas velocity distribution image and the air pressure distribution image above the fire source. The sizes of the input and output images are both 128 pixels × 128 pixels.

[0060] As Figure 1 shown, the long short - term memory network model consists of 1 input layer, 1 output layer and 6 hidden layers. Among them, 60 states of the input layer read the flame flue gas images captured at intervals of the acquisition period T = 0.1 s within a time period of 6 s. The three states of the output layer are the horizontal flue gas velocity distribution image, the vertical flue gas velocity distribution image, and the air pressure distribution image. The 3rd and 6th layers are defined as Dropout layers to reduce the risk of model overfitting.

[0061] To enhance the reliability and generalization ability of the predicted data, the loss function of the long short - term memory network model includes a data error term and a physical information error term; the data error term err data is the mean square error of the corresponding pixel values between the predicted image and the flue gas distribution map obtained by simulation. Considering that the flue gas movement conforms to the laws of fluid motion and follows the Navier - Stokes equation, the physical information error term err phy is the mean square error of the residual term after substituting the differential form of the Navier - Stokes equation.

[0062] The specific calculation method of the total loss in the long short - term memory network model is as follows:

[0063] Calculate the first - order differentials of the flue gas velocity and air pressure in space, that is: u x = gradient(u, x), u y = gradient(u, y), v x = gradient(v, x), v y = gradient(v, y), p x = gradient(p, x), p y = gradient(p, y); where u and v respectively refer to the horizontal and vertical flue gas velocities, and p is the air pressure.

[0064] Calculate the second - order differentials of the flue gas velocity in space, that is: u xx = gradient(u x , x), u yy = gradient(u y , y), v xx = gradient(vx , x), v yy = gradient(v y , y);

[0065] Based on the air density rho = 1.225 kg / m 3 and the aerodynamic annual coefficient mu = 2×10 -5 , calculate the flue gas velocity and air pressure residual term, i.e., residual i = rho×(u×u x + v×u y ), -mu×(u xx + u yy ) + p x , residual v = rho×(u×v x + v×v y ), -mu×(v xx + v yy ) + p y , residual p = p x + p y ;

[0066] The physical information error err phy is the mean square error of the three errors residual u , residual v and residual p ; The total loss err tot is the sum of the data error term and the physical loss term, i.e., err tot = err data + err phy ;

[0067] S4. Train the long short - term memory network model using the training samples and deploy the trained long short - term memory network model to the server;

[0068] S5. Use a drone to take images of the flame and flue gas at the fire scene. Assuming that the width, height, focal length, image format of the camera sensor and the distance to the target are known, convert the actual size of the flame and flue gas images collected by the camera, and input the converted flame and flue gas images into the trained long short - term memory network model to output the flue gas velocity distribution image;

[0069] Among them, the drone remains stationary when shooting videos or continuous images, and takes upward backlit shots with the sky as the background. The distance between the drone and the target can be measured by a laser sensor installed on the drone. The actual geometric size L represented by a single pixel in the image can be calculated based on the width, height, image format, focal length of the camera sensor, and the distance to the target. pxH and L pxV :

[0070]

[0071] Input the converted flame and smoke image into the trained long short-term memory network model to output the smoke velocity distribution image.

[0072] S6. Calculate the heat release rate of the fire source based on the flame and smoke image and the smoke velocity distribution image respectively, and take the maximum value of the average of the two heat release rates of the fire source over a period of time as the final heat release rate of the fire source.

[0073] When calculating the heat release rate using the flame and smoke image, convert the flame and smoke image collected by the drone from the RGB space to the HSV space, and extract the flame area image through the normal color threshold range of the flame: [0, 100, 100]~[10, 255, 255]. Due to the dynamic characteristics of the flame and being blocked by the smoke, the extracted flame image may be discontinuous. Binarize the extracted flame area image, with the pixel value of the flame area being 1 and the rest being 0, to filter out the smoke area in the image. Use the 8-connected method to label all connected areas, count the number N of all non-zero pixel values in each connected area in the flame area image, and sort the number of non-zero pixels in each connected area. According to the area with the most non-zero pixels, and according to the actual geometric size L pxV , determine the actual height L of the flame area, and calculate the heat release rate of the fire source by back-calculating based on the flame height.

[0074]

[0075] Among them, ρ ∞ , c p , T ∞ , g are the density, specific heat capacity, temperature of normal temperature air, and gravitational acceleration respectively, and D is the equivalent diameter of the fire source.

[0076] When calculating the heat release rate using the smoke velocity distribution image, first calculate the maximum value u0 of the average smoke velocity at different heights along the vertical direction based on the identified vertical smoke velocity u distribution, and then back-calculate the heat release rate of the fire source.

[0077]

[0078] Among them, z0 is the height of the virtual fire source, which is defined as:

[0079]

[0080] Combining the two equations, we can get The heat release rate of the fire source is calculated based on the smoke velocity distribution and the flame area. and The maximum value among the average values ​​is taken as the final fire source heat release rate;

[0081] S7. Match the calculated final fire source heat release rate with the simulated working condition, and query the maximum air temperature T near the bridge structure based on the matching result. g , conservatively consider that there is no fire protection outside the cable, determine its shape factor 4 / d according to the specific size of the cable, d is the cable diameter, and calculate the average temperature T of the components inside the cable at different times s :

[0082]

[0083] Where t is the time instant, ρ s and c s are the density and specific heat capacity of steel, α=α c +α r , α c is the heat convection coefficient, ε r and σ are the thermal radiation coefficient and Boltzmann constant, respectively;

[0084] S8. According to the average temperature T of the components in the cable at different times s , in order to assess the damage extent of bridge components.

[0085] Determine whether the structure has failed based on the calculated average temperature of the components and in combination with the code provisions; when the component resistance R at the average temperature is less than the internal force S generated by the design load on the component, the structural component is deemed to have failed;

[0086] For example, for ordinary steel components, when the temperature reaches 581°C, the bearing capacity drops to 50% of that at room temperature. At this time, the component resistance is R. By comparing the internal force S generated by the design load on the ordinary steel component, it is judged whether the structural component has failed. When the component resistance R

Claims

1. A method for rapid assessment of bridge structural safety after fire, characterized in that: The steps include: S1. According to the factors affecting the fire scene, several simulation conditions are constructed, the fire simulation duration, fire simulation space and simulation cell size are set, and the air temperature, smoke velocity distribution, air pressure distribution and flame and smoke images are obtained by simulating the fire scene; S2. When the simulated fire enters the quasi-steady-state stage, set the acquisition period T, save the flame and smoke images of the fire simulation space every time T, obtain the continuous images of the fire scene that fully describe the dynamic change process of the heat plume, and the smoke velocity distribution image and air pressure distribution image above the fire source; intercept the data every 4 heat plume oscillation periods to form a group of flame and smoke images, and several fire simulation conditions can form several groups of non-overlapping training samples; S3. Construct a long short-term memory network model. The input of the long short-term memory network model includes a set of flame smoke images of 4 thermal plume oscillation cycles. The output of the long short-term memory network model is the smoke velocity distribution image and air pressure distribution image above the fire source. The total loss of the loss function in the long short-term memory network model includes a data error term and a physical information error term. The data error term is the mean square error of the pixel values ​​corresponding to the predicted image and the simulated smoke distribution map, and the physical information error term is the mean square error of the residual term after substituting into the differential form of the Navier-Stokes equation. S4. Use training samples to train the long short-term memory network model; S5. Take a flame and smoke image of the fire scene, convert the flame and smoke image collected by the camera to the actual size, input the converted flame and smoke image into the trained long short-term memory network model, and output the smoke velocity distribution image; S6. Calculate the heat release rate of the fire source according to the flame smoke image and the smoke velocity distribution image, and take the maximum value of the average values ​​of the heat release rates of the two fire sources over a period of time as the final heat release rate of the fire source; S7. Match the calculated final fire source heat release rate with the simulated working condition, query the maximum air temperature near the bridge structure according to the matching result, and calculate the average temperature of the components in the cable at different times; S8. According to the average temperature of the components in the cable at different times, determine the component resistance R at the average temperature. When the component resistance R at the average temperature is less than the internal force S generated by the design load on the component, the structural component is deemed to have failed.

2. A method for rapid assessment of bridge structure safety after fire according to claim 1, characterized in that: When constructing the fire scene in the step (1), it is set as a single vehicle on fire, and the influencing factors are set as the geometric size of the fire source, the heat release rate of the fire source, the smoke production rate and the ambient wind speed. The geometric size of the fire source is set as the size of a small passenger car, the size of a large passenger car, the size of a heavy truck and the size of a tanker truck. The heat release rate of the fire source of a small passenger car ranges from 1.5 to 4.5 MW, from which M1 different heat release rates of the fire source are selected; the heat release rate of the fire source of a large passenger car ranges from 15 to 45 MW, from which M2 different heat release rates of the fire source are selected; the heat release rate of the fire source of a heavy truck ranges from 15 to 45 MW, from which M2 different heat release rates of the fire source are selected; The range of fire source heat release rate is 80~150MW, from which M3 different fire source heat release rates are selected; the range of fire source heat release rate of tank truck is 180~250MW, from which M4 different fire source heat release rates are selected; the range of smoke production rate is 0.05~0.6kg / kg, from which M5 groups of different smoke production rate values ​​are selected; the range of ambient wind speed is 0~15m / s, from which M6 groups of different ambient wind speed values ​​are selected; the orthogonal experimental method is used to construct (M1+M2+M3+M4)×M5×M6 simulation working conditions.

3. A method for rapid assessment of bridge structure safety after fire according to claim 1, characterized in that: In the step (1), it is assumed that the heat release rate of the fire source remains constant from the beginning to the end of ignition, and the fire simulation duration is set. The length and width of the fire simulation space are taken as the distance extending outward by more than 3m from the vehicle size, and the height direction is determined based on the principle that the flame is completely within the simulation space; during the simulation, the recommended cell size is set according to the dimensionless grid resolution, where: The dimensionless grid resolution ranges from 16 to 24. * is the characteristic diameter of the fire source, δx is the recommended cell size; the preset recommended cell size is n, and the simulation is performed with a cell of n, and a cell of 0.5n is used for simulation at the same time, and the two simulation results are compared, and the temperature-time curves of the same measuring point are compared. If the difference is less than 10%, n is taken as the cell setting size; otherwise, the grid size is further reduced until the difference between the two simulation results is less than 10%, then the current cell size is output as the cell setting size, and the final grid size is determined.

4. A method for rapid assessment of bridge structure safety after fire according to claim 1, characterized in that: The method for determining the acquisition period T in step (2) is as follows: based on the width W of the vehicle and taking into account the reduction in the contribution of the heat plume height in the length direction, when calculating the flame length, the effective flame length is twice the width of the vehicle, so the effective fire source area is A eff =W×2×W, then the oscillation frequency of the thermal plume is f plume =1.68D -0.5 , where D is the equivalent diameter of the fire source, Calculate f plume , the oscillation period of the thermal plume is T mlume for: The acquisition period T is the oscillation period T of the heat plume. plume An integer that is 0.1 times of .

5. The method for rapid assessment of bridge structure safety after fire according to claim 1 is characterized by: In the step (3), the long short-term memory network model consists of 1 input layer, 1 output layer and 6 hidden layers, wherein several states of the input layer read the flame smoke images taken at intervals of acquisition period T within a time period, and the three states of the output layer are the horizontal smoke velocity distribution image, the vertical smoke velocity distribution image, and the air pressure distribution image.

6. The method for rapid assessment of bridge structure safety after fire according to claim 1 is characterized by: The specific calculation method of the total loss in step (3) is: Calculate the first-order differential of smoke velocity and air pressure in space, that is: u x = gradient(u,x), u y = gradient(u,y),v x = gradient(v,x), v y = gradient(v,y), p x = gradient(p,x), p y = gradient (p, y); where u and v refer to the horizontal and vertical smoke velocities respectively, and p is the air pressure; Calculate the second-order differential of the smoke velocity in space, that is: u xx = gradient(u x ,x),u yy = gradient(u y ,y),v xx = gradient(v x ,x),v yy = gradient9v y ,y); Based on the air density rho and the annual aerodynamic coefficient mu, the residual term of smoke velocity and air pressure is calculated. u =rho×(u×u x +v×u y )-mu×(u xx +u yy )+p x , residual v =rho×(u×v x +v×v y )-mu×(v xx +v yy )+p y , residual p =p x +p y ; Physical information error err phy is the residual error of three items u 、residual v and residual p The mean square error of the total loss err tot is the sum of the data error term and the physical loss term, that is, err tot =err data +err phy .

7. The method for rapid assessment of bridge structure safety after fire according to claim 1 is characterized by: The specific method for converting the actual size in step (5) is as follows: Assuming that the camera sensor width, height, focal length, image frame, and distance to the target are known, the actual geometric size L represented by a single pixel in the image can be calculated through the camera sensor width, height, image frame, focal length, and distance to the target pxH and L pxV :

8. The method for rapid assessment of bridge structure safety after fire according to claim 1 is characterized by: When the flame and smoke image is used to calculate the heat release rate of the fire source in step (6), the flame and smoke image collected by the drone is converted from the RGB space to the HSV space, and the flame area image is extracted through the common color threshold range of the flame: [0, 100, 100] to [10, 255, 255]. Due to the dynamic characteristics of the flame and the obstruction of the smoke, the extracted flame image may be discontinuous; the extracted flame area image is binarized, the pixel value of the flame area is 1, and the other pixel values ​​are 0, and the smoke area in the image is filtered out; The 8-connectivity method is used to mark all connected areas, and the number of non-zero pixel values ​​N in each connected area in the flame area image is counted, and the number of non-zero pixels in each connected area is sorted; according to the area with the most non-zero pixels, the actual geometric size L represented by a single pixel is sorted. pxV , determine the actual height L of the flame area, and calculate the heat release rate of the fire source based on the flame height Among them, ρ ∞ , c p , T ∞ , g are the density, specific heat capacity, temperature and gravitational acceleration of air at room temperature, and D is the equivalent diameter of the fire source.

9. The method for rapid assessment of bridge structure safety after fire according to claim 1 is characterized by: When the smoke velocity distribution image is used to calculate the heat release rate in step (6), first, according to the identified vertical smoke velocity u distribution, the maximum value u0 of the average smoke velocity at different vertical heights is calculated, and then the heat release rate of the fire source is inversely calculated. Among them, z0 is the height of the virtual fire source, which is defined as: The heat release rate of the fire source can be obtained by combining the two equations 10. The method for rapid assessment of bridge structure safety after fire according to claim 1 is characterized by: The average temperature T in step (7) s The calculation formula is: Where t is the time instant, ρ s and c s are the density and specific heat capacity of steel, α=α c +α r , α c is the heat convection coefficient, ε r and σ are the thermal radiation coefficient and Boltzmann constant respectively; the shape coefficient 4 / d is determined according to the specific size of the cable, where d is the cable diameter.

Images