Cement concrete pavement void range identification method based on BP neural network

Abstract

The invention discloses a BP-based cement concrete pavement void range identification method, and relates to the technical field of traffic transportation engineering. According to the method, the finite element fluid-solid coupling simulation model of the cement pavement void damage is constructed, so that the internal mechanism that the development of the void damage is influenced by traffic load and water flow scouring is disclosed; the internal temperature field of the full-scale test pavement is monitored in real time, and a cement pavement dynamic response forward modeling analysis model considering the temperature influence and a void range identification method based on a BP neural network are constructed based on the internal temperature field, so that the void range can be effectively detected, construction site personnel or working personnel are assisted to carry out accurate maintenance, and the working efficiency is improved. Resource waste and possible road damage caused by repair beyond the void range are prevented.

Description

Technical Field

[0001] The invention relates to the technical field of transportation engineering, and in particular to a method for identifying a void range of a cement concrete pavement based on a BP neural network. Background Art

[0002] Cement concrete pavement (hereinafter referred to as cement pavement) is one of the important structural forms of highway pavement in my country due to its advantages of high strength, high water resistance, good stability, long service life and low maintenance cost. Due to the high cost of asphalt concrete pavement and the relative scarcity of corresponding resource reserves in my country, and the characteristics of cement pavement such as high strength, strong durability and long service life, the cement pavement structure occupies an important position in my country's highway construction, which determines that the pavement structure material is mainly cement stabilized material. However, during the use of cement pavement panels, under the repeated action of traffic loads, the cement stabilized base under the board will produce certain plastic deformation. At the same time, under the influence of temperature stress, the cement pavement panels will produce certain warping, resulting in partial discontinuity or peeling of the interface between the pavement panel and the base. Under the combined action of traffic loads, rainfall and structural free water, cement pavement panels will inevitably have problems such as scouring and mud pumping. With the deepening of repeated loads and water scouring and erosion, the bottom of the cement pavement panels will have different degrees of hollowing problems. If the voiding disease is not handled in time, when the traffic load acts on the pavement panel, stress concentration will occur at the edge of the voiding area, which will gradually cause broken panels, misalignment, joint fractures and other diseases. It will seriously affect the traffic quality and bring great safety hazards. In addition, extreme natural disasters have occurred frequently in recent years, the roadbed has been severely damaged, and hidden diseases of cement pavements have occurred frequently. Frequent and long-term maintenance tasks have brought new difficulties to the passage of cement pavements. Therefore, a voiding range identification method is needed to detect cement pavements, thereby effectively maintaining the pavement and traffic quality. Summary of the invention

[0003] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for identifying the void range of cement concrete pavement based on BP neural network, which solves the problem that the prior art lacks an efficient void range identification method, resulting in over-range repair of the void range of the road, reducing the use time of the pavement, causing a certain amount of losses, and being unable to assist staff in performing accurate repairs.

[0004] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0005] A method for identifying a void range of a cement concrete pavement based on BP is provided, which comprises the following steps:

[0006] S1. Acquire the pavement panel data of the cement concrete pavement to be identified;

[0007] S2. Performing deflection detection on the cement concrete pavement to be identified to obtain corresponding deflection data;

[0008] S3, measuring the surface temperature data of the cement concrete pavement to be identified;

[0009] S4, constructing a road surface void recognition model;

[0010] S5. The pavement panel data, deflection data and surface temperature data are normalized and then input into the pavement void identification model to obtain the corresponding void range.

[0011] The beneficial effects of the present invention are as follows: the present invention reveals the inherent mechanism of traffic load and water flow erosion affecting the development of void disease by constructing a finite element fluid-solid coupling simulation model of cement pavement void disease; the internal temperature field of the full-scale test pavement is monitored in real time, and based on this, a cement pavement dynamic response forward analysis model considering the influence of temperature and a void range identification method based on a BP neural network are constructed, which can effectively detect the void range, assist site personnel or workers in performing precise maintenance, and prevent waste of resources and possible road damage caused by repairs beyond the void range. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 is a method flow chart;

[0013] Figure 2 Schematic diagram of the application position of the deflection load;

[0014] Figure 3 This is a schematic diagram of the temperature distribution of the top and bottom of the plate at different times of the day in different seasons;

[0015] Figure 4 The schematic diagram of the temperature distribution of the top and bottom of the airport runway panel at different times of the day in different seasons;

[0016] Figure 5 It is a schematic diagram of the true curve and fitting curve of the bottom temperature of the plate;

[0017] Figure 6 Schematic diagram of the application position of the deflection load;

[0018] Figure 7 Schematic diagram for comparing the calculated and measured values ​​of the plate edge and plate center deflection; (1) is a schematic diagram for comparing the calculated and measured values ​​of the plate edge deflection, and (2) is a schematic diagram for comparing the calculated and measured values ​​of the plate center deflection;

[0019] Figure 8 This is a curve showing the error changing with the number of iterations. DETAILED DESCRIPTION

[0020] The specific implementation modes of the present invention are described below to facilitate those skilled in the art to understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0021] like Figure 1 As shown, a method for identifying the void range of cement concrete pavement based on BP includes the following steps:

[0022] S1. Acquire the pavement panel data of the cement concrete pavement to be identified;

[0023] S2. Performing deflection detection on the cement concrete pavement to be identified to obtain corresponding deflection data;

[0024] The pavement panel data include pavement panel size, pavement panel modulus, base modulus, and foundation modulus; among them, the pavement panel size includes pavement thickness and pavement plane size; the deflection data is the vertical displacement of the fixed node after the deflection load is applied.

[0025] S3, measuring the surface temperature data of the cement concrete pavement to be identified;

[0026] This method uses the finite element fluid-solid coupling simulation model of void disease composed of FLUNENT fluid analysis module and TRANSIENT STRUCTURE transient structure analysis module to analyze the cement concrete pavement and obtain the corresponding void disease influencing mechanism, that is, the inherent mechanism of traffic load and water flow scouring affecting the development of void disease is analyzed, and a full-scale test section of the pavement is carried out to analyze the influence of temperature on the void disease of the top, middle, bottom and base of the slab. There is a direct connection between deflection data and void damage. Based on the temperature influence and internal mechanism, the relationship between the temperature gradient, pavement panel modulus, base modulus, soil base modulus, pavement panel thickness, pavement panel size, void size, base thickness and other index data and deflection values ​​is analyzed. Based on the analysis results, the deflection value and the temperature gradient, pavement panel size, pavement panel modulus, base modulus, and foundation modulus with a correlation greater than 0.85 with the deflection data are selected as characteristic values, that is, the surface temperature data, pavement panel data, and deflection data are selected as the input of the subsequent BP neural network, and the void size (void range) is selected as the output of the BP neural network. Among them, selecting data with a higher correlation with the deflection value can capture deep-level features, which is conducive to improving the accuracy of identifying the void range.

[0027] S4, constructing a road surface void recognition model;

[0028] The pavement clearance recognition model adopts the BP neural network model, which includes an input layer, a hidden layer and an output layer connected in series; each layer includes multiple neuron nodes, and the neuron nodes are connected through connection weights; the activation function of the output layer adopts the Purelin function; the activation function of the hidden layer adopts the Tasnsig function; the number of neuron nodes J in the hidden layer is based on the formula:

[0029]

[0030] Determine; where a represents the neuron node of the input layer, b represents the neuron node of the output layer, and c represents a constant.

[0031] S5. The pavement panel data, deflection data and surface temperature data are normalized and then input into the pavement void identification model to obtain the corresponding void range.

[0032] The training process of the BP neural network model in step S5 includes the following steps:

[0033] S5-1. Obtaining road panel training data and deflection training data;

[0034] The cement concrete pavement is divided into three parts: cement concrete pavement slab, void disease area and pavement subgrade. The pavement slab is divided into entry slab and exit slab, the fluid part area is the plate joint and void part, and the fluid area is full of water, and the solid stress area is the cement concrete pavement slab part. The dimensions of each part are shown in Table 1.

[0035] Table 1

[0036] structure size Cement concrete road slab 4m×4m×0.3m Board seam 25mm Maximum clearance 25mm Maximum clearance width 800mm

[0037] The process of obtaining the deflection training data in step S5-1 includes the following steps:

[0038] The falling weight deflectometer (FWD) method is used to treat the pavement with a known deflection range, and the deflection basin changes of different pavement structures under the FWD load. The actual falling weight deflectometer with a load plate of 300mm diameter is used, such as Figure 2 As shown, for the plate corner, the deflection load is applied at a position where the center of the loading plate is 25 cm away from the plate edges on both sides; for the plate edge, the deflection load is applied at the middle of the plate edge and the center of the loading plate is 25 cm away from the plate edge, and the corresponding deflection data, i.e., the deflection training data, is obtained.

[0039] S5-2. Conduct full-scale road tests on the road surface to obtain the corresponding temperature field distribution law;

[0040] Step S5-2 includes the following steps:

[0041] S5-2-1. Four layers of temperature sensors are installed on the top, between, bottom and base of the road surface and pavement. Each layer of sensors is installed at four corners and in the middle of the board. One temperature sensor is installed in the road surface to monitor the atmospheric temperature.

[0042] S5-2-2, collect temperature data through various temperature sensors, and classify the temperature data according to the time of spring, summer, autumn and winter to obtain classified temperature data; for example, for road surface, the classified temperature data includes spring road surface temperature data, summer road surface temperature data, autumn road surface temperature data and winter road surface temperature data, and the top of the board, between the boards, the bottom of the board and the base are classified in the same way. Among them, March, April and May are set as spring, June, July and August are set as summer, September, October and November are set as autumn, and December, January and February are set as winter.

[0043] S5-2-3, analyzing the classified temperature data to obtain the corresponding temperature distribution law;

[0044] like Figure 3 and Figure 4 As shown in the figure, the temperature of the top and bottom of the slab fluctuates over time. The temperature of the top of the slab is close to and usually higher than the atmospheric temperature. The temperature separation between the top of the slab and the atmospheric temperature reaches the highest in summer, about 10℃ to 15℃, while it is relatively close in winter. Usually, the peak temperature of the top of the slab occurs at around 14:00 in the afternoon, and the valley value is around 6:00 in the morning. The temperature change of the bottom of the slab is about 4 hours later than that of the top of the slab, because there is a delay in the heat absorption and transfer of cement concrete, and part of the heat is stored in the slab. When the external temperature is lower than the temperature inside the slab, the cement concrete slab will continue to radiate heat to the outside. The temperature change of the bottom of the slab is significantly smaller than that of the top of the slab and varies with the season. In seasons with large day and night temperature differences, such as summer, the temperature change of the bottom of the slab is large; while in seasons with small day and night temperature differences, the change is small. However, from the perspective of the whole year, the daily temperature difference of the bottom of the slab is usually less than 5℃ in each season. The small temperature change of the bottom of the slab makes the temperature difference curve trend between the bottom of the slab and the top of the slab similar to the temperature trend of the top of the slab. As the plate thickness increases, the temperature variation at the bottom of the plate gradually decreases, and its peak appears later because the heat transfer path becomes longer.

[0045] The formula for the temperature distribution law in step S5-2-3 is:

[0046]

[0047]

[0048]

[0049] Among them, T a,i represents the temperature of the ith species, represents the daily average temperature of the ith type, Tm,i represents the daily temperature variation of the ith type, T max,i represents the maximum daily temperature of the ith type, T min,i represents the minimum daily temperature of the ith type, sin(·) represents the sine function, t i represents the phase of the ith type, t 0,i represents the initial phase of the ith species, ω(t i ),ω(t i -t 0,i ) represent the phase t of the i-th type respectively. i , the phase difference is t i -t 0,i The corresponding angular frequency;

[0050] Taking the temperature curve of the bottom of the cement road slab in summer as the benchmark, the corresponding temperature change formula is fitted to verify the feasibility of the formula for temperature distribution law. The following is the specific analysis content:

[0051] The average maximum temperature of the plate bottom varies with the plate thickness:

[0052] T ymax =38.43-2.19×10 -1 H

[0053] Among them, T ymax It represents the maximum average temperature of the plate bottom during the year, and H represents the plate thickness.

[0054] The average minimum temperature of the plate bottom varies with the plate thickness:

[0055] T ymin =2.64+1.34×10 -2 H

[0056] Among them, T ymin Indicates the minimum average temperature of the plate bottom during the year.

[0057] The maximum daily temperature change of the plate bottom varies with the plate thickness:

[0058] T Amax =5.81-6.63×10 -2 H

[0059] Among them, T Amax Indicates the maximum daily temperature change at the bottom of the plate.

[0060] The minimum amplitude of daily temperature change at the bottom of the plate changes with the plate thickness:

[0061] T Amin =3.79-3.32×10 -2 H

[0062] Among them, TAmin Indicates the minimum daily temperature change at the bottom of the plate.

[0063] The initial phase of daily temperature variation at the bottom of the plate changes with plate thickness:

[0064] T d0 =10.75+4.53×10 -2 H

[0065] Among them, T d0 It indicates the change value of the initial phase of daily temperature change at the bottom of the plate along with the plate thickness.

[0066] The formula for the average temperature of the bottom of the plate changing with the month is:

[0067] T mon =A+A m [0.96sinω(2t m -9)+0.14sin2ω(2t m -9)]

[0068]

[0069]

[0070] Among them, T mon represents the monthly change of the average temperature of the plate bottom, A represents the average temperature of the plate bottom during the year, and A m Indicates the temperature variation of the bottom of the plate, t m Indicates phase.

[0071] The formula for the change of the daily temperature variation of the bottom of the plate with the month is:

[0072] T day =B+B m [0.96sinω(2t m -T d0 )+0.14sin2ω(2t m -T d0 )]

[0073]

[0074]

[0075] Among them, T day represents the change of the daily temperature of the plate bottom with the month, B represents the daily average temperature of the plate bottom, and B m Indicates the daily temperature variation amplitude of the plate bottom.

[0076] Then, calculate the average daily plate bottom temperature variation curve T for different plate thicknesses in different months. d The calculation formula is:

[0077] T d =T mon +T day [0.96sinω(2t m -9)+0.14sin2ω(2t m -9)]

[0078] like Figure 4 As shown, the summer temperature curve of the plate bottom obtained by the above method is close to the real curve, which can effectively reflect the temperature change trend of the plate bottom.

[0079] S5-2-4. Construct a non-steady-state temperature field based on the temperature distribution law, heat transfer mode, energy conservation principle and Fourier's law;

[0080] The present invention regards the temperature field in the road panel as non-steady state, and uses the calculation method of non-steady state temperature field to solve its temperature distribution at any time. It is assumed that the thermophysical parameters related to any point in the road panel do not change with the change of spatial position, that is, the physical parameters such as density, thermal conductivity, specific heat capacity, etc. are unchanged. Select a micro unit at any position of the road panel, and based on the principle of conservation of energy and Fourier's law, the following equation is obtained:

[0081] Heat input and heat output in the x direction:

[0082]

[0083]

[0084] Among them, q x represents the incoming heat in the x direction, q x+dx represents the heat transfer in the x direction, λ represents the thermal conductivity, d(·) represents the derivative, and y and z represent the ordinate and vertical coordinate directions, respectively.

[0085] Therefore, the heat increment Δq in the x direction is x for:

[0086]

[0087] By analogy, the heat increment Δq in the y and z directions can be obtained y , Δq z for:

[0088]

[0089]

[0090] Then the heat increment Δq of the micro-unit is:

[0091]

[0092] The heat increment in the unit body changes with time and temperature, so:

[0093]

[0094] Among them, Δ=dxdydz.

[0095] Therefore, the formula of the unsteady temperature field in step S5-2-4 is:

[0096]

[0097] Among them, T i represents the temperature of the i-th micro-unit, t represents the time, represents the partial derivative of the temperature of the i-th micro-unit, ρ represents the density of the micro-unit material, C represents the specific heat capacity of the micro-unit material, represents the partial derivative, x i ,y i 、z i They respectively represent the heat increment of the i-th micro-unit in the x, y, and z directions; wherein the micro-unit is a unit set at any position of the cement concrete pavement.

[0098] S5-2-5. Based on the first initial condition and the first boundary condition, the finite element method is used to solve the non-steady-state temperature field to obtain the temperature field distribution of the pavement at the beginning of heat transfer and the heat transfer state at its boundary, that is, the temperature field distribution law.

[0099] The first initial condition is the temperature distribution state in the track plate when time is 0; the first boundary condition includes the first temperature boundary condition, the heat flux boundary condition and the surrounding environment heat exchange boundary condition.

[0100] Table 2 shows the specific parameters of the second boundary condition.

[0101] Table 2

[0102]

[0103]

[0104] The weighted residual method is used to calculate the boundary effect of the unsteady temperature field. Assuming that the temperature function of the heat transfer object is T1, the residuals generated by each boundary condition are weighted and integrated to obtain

[0105]

[0106] Among them, ζ2, ζ3, ζ represent boundaries, n x 、n y 、n zrepresents the components of the normal in the horizontal direction x, vertical direction y and vertical direction z, q represents the heat flux density, T α represents the ambient environment parameters, w represents the heat convection exchange coefficient, represents the average temperature equation, k1, k2, k3 represent weighting parameters, and V represents the volume of the heat transfer object;

[0107] The object to be solved is discretized into N units, each unit has m nodes, and the temperature interpolation function of the unit body is:

[0108]

[0109] Among them, N i (x, y, z) represents the i unit, T i Represents the temperature data corresponding to the i-th node, [N1, N2, ..., N m ] represents each unit, represents the temperature data corresponding to each node, ∑(·) represents the summation function;

[0110] The temperature interpolation function is derived twice and substituted into the weighted integral function, and the weighted parameters are set to k1 = -k2 = -k3, the margin on the boundary is 0, and k1 = N1 = N2 = ... = N m ,get:

[0111]

[0112] Among them, [C] m represents the specific heat capacity matrix of the unit, represents the partial differential of the unit temperature with respect to time, [K] m represents the heat transfer matrix of the unit, {T} m represents the temperature matrix of the unit, {P} m represents the element temperature load vector;

[0113] Solving the overall heat conduction equation requires parameter transformation. The finite element unit body uses a hexahedral eight-node unit, and the shape function represents a continuous function of the internal coordinates of the unit; the shape function is partially differentiated in the local coordinates and sorted out to obtain:

[0114]

[0115] Where [J] represents the transposed matrix of the Jacobian matrix.

[0116] The unit strain matrix and stiffness matrix are:

[0117] {ε}=[B i ]{δ} e

[0118]

[0119] [K] = ∫∫∫[B i ] T [D][B i ]dxdydz

[0120] Among them, [B i ] represents the unit geometry matrix, {ε} represents the unit strain matrix, δ represents the stress matrix, e represents the unit node, [K] represents the stiffness matrix, [·] T represents the transposed matrix, [D] represents the elastic matrix, and ∫∫∫(·)dxdydz represents the triple integral. The rearranged shape function is taken as the derivative in the global coordinates, and then it is fed into the temperature interpolation function of the unit cell to be solved to obtain the temperature field.

[0121] For the discretization in the time domain, take the time increment as dt, and substitute the corresponding time domain shape function into the temperature equation and the heat conduction equation in turn, and after sorting and integrating, we get:

[0122]

[0123]

[0124] in, represents the time matrix, T P represents the temperature load, {T P} e represents the element temperature load vector, [C] e represents the unit heat capacity matrix, represents the intermediate parameter, and k(Δ) represents the integral function with respect to time.

[0125] For temperature stress, the stress in the three directions of xyz is calculated based on Hooke's law, and the total potential energy of the unit is obtained by calculating the unit strain energy increment and according to the law of conservation of energy. The equilibrium equation and the unit displacement and strain expressions are further obtained as follows:

[0126] {u} e =[N]{u i}{ε} c =[B]{u} e

[0127] Among them, {ε} c represents the strain matrix, {u i} represents the unit displacement matrix, {u} e represents the nodal displacement matrix, and [B] represents the element geometry matrix.

[0128] The corresponding program steps are:

[0129] The geometric parameters, material parameters and boundary conditions are used as input, and the cement concrete road slab is meshed; the heat conduction equation of each unit is calculated; the overall heat conduction equation is calculated based on the heat conduction of each unit; the temperature of each unit node and the unit stress, that is, the temperature field distribution law, are calculated.

[0130] The structural layers of the road slab are divided into surface layer, base layer and subbase layer, as shown in Table 3, which shows the specific material parameters of each structural layer.

[0131] Table 3

[0132] Structural Layer <![CDATA[Density / kg·m -3 > <![CDATA[Thermal conductivity / W·m -1 ·k -1 > <![CDATA[Specific heat capacity / kJ·kg -1 ·k -1 > Surface 2350 0.35 0.97 Grassroots 1600 1.20 0.91 Subbase 1400 1.10 0.90

[0133] S5-3. Construct a dynamic response forward analysis model;

[0134] The dynamic response forward analysis model adopts a multi-layer finite element analysis model based on the Winkler foundation model theory; the Winkler foundation model is a theoretical model widely used in civil engineering to describe the interaction between soil foundation and structure. The model proposes three assumptions: ① The foundation is continuous, uniform and elastic; ② The stiffness is mainly determined by the elastic modulus and shear modulus; ③ It has infinite stiffness in the horizontal direction and finite stiffness in the vertical direction.

[0135] The application background of the dynamic response analysis model in this method is to analyze the changes in the deflection basin of different pavement structures under the action of FWD loads. The vertical displacement caused by the FWD load is small for the pavement structure, and when unloaded, the pavement deformation recovers instantly, indicating that it is reasonable to use the Winkler foundation model theory in the dynamic response forward analysis model. S5-4. Based on the temperature field distribution law, the pavement panel training data and deflection training data are analyzed through the dynamic response forward analysis model to obtain the corresponding forward analysis sample library and its labels;

[0136] The analysis process in step S5-4 includes the following steps:

[0137] S5-4-1. Obtain the geometric parameters, material parameters and second temperature boundary conditions of the pavement;

[0138] S5-4-2, inputting all the data in step S4-1 into the dynamic response forward analysis model;

[0139] S5-4-3. Unit grid division of cement concrete pavement to be identified;

[0140] S5-4-4. Calculate the temperature stress distribution of each unit grid;

[0141] S5-4-5. Calculate the stiffness matrix of each unit grid;

[0142] S5-4-6, judging whether the unit grid is empty based on the stiffness matrix of each unit grid; if so, modifying the stiffness matrix of the unit grid; otherwise, retaining the stiffness matrix of the unit grid; and proceeding to step S4-7;

[0143] S5-4-7, calculating the overall stiffness matrix based on the data in step S4-6;

[0144] S5-4-8. Based on the overall stiffness matrix and temperature stress distribution, the dynamic response analysis of the cement concrete pavement to be identified is carried out to obtain the node displacement value of each unit grid and the temperature value of the node at each time, that is, to obtain the forward analysis sample library.

[0145] The effectiveness of the dynamic response analysis program embedded with temperature stress calculation is verified by comparing the measured deflection values ​​with the deflection values ​​calculated by the program. The deflection in the plate (without voids) and the deflection at the edge of the plate (with voids) are used as examples for testing. The deflection detection time is 8:00, 12:00, 16:00 and 20:00 to represent the different temperature stress distributions of different pavement structural layers. The deflection detection position at the edge of the plate is as follows: Figure 6 The model parameters input into the program are shown in Table 3 and Table 4.

[0146] Table 4

[0147]

[0148] like Figure 7 As shown in the figure, the results show that the errors between the calculated and measured values ​​are both less than 9%, and the accuracy meets the requirements of engineering applications.

[0149] In order to verify the effectiveness of the quantitative identification method of voids based on the BP neural network algorithm, all the data in the sample library were trained, and the training samples: inspection samples: test samples were adapted according to the 70%:15%:15% method. The sample data was input into the neural network and the learning training was completed, with 12 hidden layer nodes.

[0150] Taking the plate edge void as an example, 10 sets of bending sink basin data were recalculated using the forward model for prediction verification, as shown in Table 5. These 10 sets of data are different from any data in the sample library.

[0151] Table 5

[0152] Top temperature 38.0℃ Top temperature 28.3℃ Pavement panel modulus 32000MPa Road panel size 5m×5m×26m Base modulus 1500MPa Base thickness 40cm Soil base modulus 140MPa

[0153] The mean square error method is used to count the errors in the training process, such as Figure 8As shown in the figure, the error gradually decreases with the increase of the number of iterations, and finally approaches 0.013. Based on the trained neural network program, the prediction group data is simulated and predicted. The prediction results and error analysis are shown in Table 6. The results show that the absolute error between the predicted value and the expected value is less than 10%, indicating that this method is feasible for quantitative identification of the range of cement concrete to panel degassing.

[0154] Table 6

[0155]

[0156]

[0157] S5-5, inputting a set of sample data and labels of the forward analysis sample library into the BP neural network model, and using the back propagation algorithm to adjust the weights and thresholds of the BP neural network model to obtain the BP neural network model after initial training;

[0158] BP neural network uses back propagation algorithm for training, and selects Trainbfg function as training function and Learndm function as learning function. The back propagation algorithm propagates the error from the output layer backward through the chain rule, and adjusts the weight and threshold of each neuron according to the error gradient, thereby continuously optimizing the performance of the network. It includes five steps: forward propagation, error calculation, back propagation, parameter update and repeated iteration.

[0159] S5-6. Repeat step S5-5 until each set of sample data is input into the BP neural network model to obtain a trained BP neural network model.

[0160] In the process of debugging the BP neural network program through the sample library, first, compile the forward model, indicating that in a specific void range and a specific road panel data, the deflection data obtained by the deflection detection is different, which means that a set of deflection data and road panel data corresponds to a void data one by one. This is actually the basis for the subsequent BP neural network recognition, so thousands of data sets are required, that is, the forward analysis sample library, to meet the common needs in the project. In the subsequent steps, the void data is obtained in reverse through the neural network recognition method. In addition, due to the large influence of temperature, the influence of temperature is considered. Temperature will change the mechanical properties of the material, but the temperature data at different positions of the road panel are different, so the significance of the full-scale test section is to obtain all the road surface temperature data. The temperature in the road panel can be calculated theoretically. The road surface temperature and the bottom temperature of the plate can be calculated. The road surface temperature can be directly measured in actual engineering, and the bottom temperature of the plate changes less. The empirical formula summarized by the experimental results of the test section, that is, the formula of step S5-2-3 and step S5-2-4, is embedded in the forward program, and the temperature change of the entire road panel can be calculated by the road surface temperature. Finally, through countless forward calculations, the sample library is obtained and imported into the BP neural network for learning. The learned BP neural network can be directly applied to actual engineering.

[0161] In summary, the present invention reveals the inherent mechanism of how traffic load and water flow scouring affect the development of voiding diseases by constructing a finite element fluid-solid coupling simulation model of cement pavement voiding diseases; it monitors the internal temperature field of the full-scale test pavement in real time, and based on this, constructs a forward analysis model of the dynamic response of a cement pavement considering the influence of temperature and a voiding range identification method based on a BP neural network, which can effectively detect the voiding range, assist construction site personnel or staff in performing precise maintenance, and prevent waste of resources and possible road damage caused by repairs beyond the voiding range.

Claims

1. A method for identifying the void range of a cement concrete pavement, characterized in that: The following steps are involved: S1. Acquire the pavement panel data of the cement concrete pavement to be identified; S2. Performing deflection detection on the cement concrete pavement to be identified to obtain corresponding deflection data; S3, measuring the surface temperature data of the cement concrete pavement to be identified; S4, constructing a road surface void recognition model; S5. The pavement panel data, deflection data and surface temperature data are normalized and then input into the pavement void identification model to obtain the corresponding void range.

2. The method for identifying the void range of cement concrete pavement according to claim 1 is characterized in that: The pavement panel data include pavement panel size, pavement panel modulus, base modulus, and foundation modulus; wherein the pavement panel size includes pavement thickness and pavement plane size; the deflection data is the vertical displacement of the fixed node after the deflection load is applied.

3. The method for identifying the void range of cement concrete pavement according to claim 2, characterized in that: The pavement clearance recognition model adopts a BP neural network model, including an input layer, a hidden layer and an output layer connected in series; each layer includes a plurality of neuron nodes, and the neuron nodes are connected through connection weights; wherein the activation function of the output layer adopts the Purelin function; the activation function of the hidden layer adopts the Tasnsig function; the number of neuron nodes J in the hidden layer is calculated according to the formula: Determine; where a represents the neuron node of the input layer, b represents the neuron node of the output layer, and c represents a constant.

4. The method for identifying the void range of cement concrete pavement according to claim 2, characterized in that: The training process of the BP neural network model in step S5 comprises the following steps: S5-1. Obtaining road panel training data and deflection training data; S5-2. Conduct full-scale road tests on the road surface to obtain the corresponding temperature field distribution law; S5-3. Construct a dynamic response forward analysis model; S5-4. Based on the temperature field distribution law, the pavement panel training data and deflection training data are analyzed through the dynamic response forward analysis model to obtain the corresponding forward analysis sample library and its labels; S5-5, inputting a set of sample data and labels of the forward analysis sample library into the BP neural network model, and using the back propagation algorithm to adjust the weights and thresholds of the BP neural network model to obtain the BP neural network model after initial training; S5-6. Repeat step S5-5 until each set of sample data is input into the BP neural network model to obtain a trained BP neural network model.

5. The method for identifying the void range of cement concrete pavement according to claim 4 is characterized in that: The step S5-2 comprises the following steps: S5-2-1. Install temperature sensors on the top, between, bottom and base of the road surface and pavement. S5-2-2, collecting temperature data through various temperature sensors, and classifying the temperature data according to the time of spring, summer, autumn and winter to obtain classified temperature data; S5-2-3, analyzing the classified temperature data to obtain the corresponding temperature distribution law; S5-2-4. Construct a non-steady-state temperature field based on the temperature distribution law, heat transfer mode, energy conservation principle and Fourier's law; S5-2-5. Based on the first initial condition and the first boundary condition, the finite element method is used to solve the non-steady-state temperature field to obtain the temperature field distribution of the pavement at the beginning of heat transfer and the heat transfer state at its boundary, that is, the temperature field distribution law.

6. The method for identifying the void range of cement concrete pavement according to claim 5, characterized in that: The temperature distribution law of step S5-2-3 is: Where Ta,i represents the temperature of the i-th type, represents the daily average temperature of the ith type, Tm,i represents the daily temperature variation of the ith type, Tmax,i represents the daily maximum temperature of the ith type, Tmin,i represents the daily minimum temperature of the ith type, sin(v) represents the sine function, ti represents the phase of the ith type, t0,i represents the initial phase of the ith type, ω(ti) and ω(ti-t0,i) represent the angular frequency corresponding to the phase ti of the ith type and the phase difference ti-t0,i, respectively; The formula of the unsteady temperature field in step S5-2-4 is: Where Ti represents the temperature of the i-th micro-unit, t represents the time, represents the partial derivative of the temperature of the i-th micro-unit, ρ represents the density of the micro-unit material, C represents the specific heat capacity of the micro-unit material, represents the partial derivative, xi, yi, zi represent the heat increment of the ith micro-unit in the x, y, and z directions respectively; wherein the micro-unit is a unit set at any position of the cement concrete pavement.

7. The method for identifying the void range of cement concrete pavement according to claim 5, characterized in that: The first initial condition is the temperature distribution state in the track plate when time is 0; the first boundary condition includes a first temperature boundary condition, a heat flux density boundary condition and an ambient heat exchange boundary condition.

8. The method for identifying the void range of cement concrete pavement according to claim 4, characterized in that: The dynamic response forward analysis model adopts a multi-layer finite element analysis model based on the Winkler foundation model theory; The analysis process in step S5-4 includes the following steps: S5-4-1. Obtain the geometric parameters, material parameters and second temperature boundary conditions of the pavement; S5-4-2, inputting all the data in step S4-1 into the dynamic response forward analysis model; S5-4-3. Unit grid division of cement concrete pavement to be identified; S5-4-4. Calculate the temperature stress distribution of each unit grid; S5-4-5. Calculate the stiffness matrix of each unit grid; S5-4-6, judging whether the unit grid is empty based on the stiffness matrix of each unit grid; if so, modifying the stiffness matrix of the unit grid; otherwise, retaining the stiffness matrix of the unit grid; and proceeding to step S4-7; S5-4-7, calculating the overall stiffness matrix based on the data in step S4-6; S5-4-8. Based on the overall stiffness matrix and temperature stress distribution, the dynamic response analysis of the cement concrete pavement to be identified is carried out to obtain the node displacement value of each unit grid and the temperature value of the node at each time, that is, to obtain the forward analysis sample library.

Images