A method, system, and computer storage medium for bridge durability testing

Abstract

This invention relates to a bridge durability testing method, system, and computer storage medium, comprising: preparing standard concrete test specimens and collecting chloride ion concentration data; constructing and training a diffusion flow model based on conditional-flow matching; acquiring chloride ion concentration detection data at multiple discrete depths for representative engineering parts of the bridge concrete under test, and collecting corresponding physical parameters; generating a chloride ion concentration diffusion flow matching field for the engineering parts under test within a continuous depth range; extracting parameters required for bridge durability assessment and substituting them into a preset standard calculation formula to obtain the bridge durability assessment result. The beneficial effects of this invention are: the invention generates a chloride ion concentration diffusion flow matching field without requiring a preset data distribution form, naturally adapting to the unsteady diffusion distribution of chloride ions in bridge concrete structures, characterized by an increase followed by a decrease from the surface to the deeper layers, thus achieving a leap from "point prediction" to "field reconstruction."

Description

Technical Field

[0001] This invention relates to the field of bridge engineering durability assessment technology, and more specifically, to a bridge durability testing method, system, and computer storage medium. Background Technology

[0002] As a core component of transportation infrastructure, the structural durability of bridges directly determines their service safety and life-cycle benefits. Chloride media can penetrate and diffuse into the concrete through its pores. When the chloride ion concentration around the reinforcing steel exceeds a critical value, it can damage the passivation film on the steel, causing corrosion. The volume expansion of corrosion products leads to concrete cracking and peeling of the protective layer, further accelerating the vicious cycle of chloride penetration and steel corrosion. Ultimately, this reduces the load-bearing capacity of key components such as bridge beams and piers, increasing maintenance costs and potentially triggering structural safety accidents, significantly shortening the bridge's designed service life.

[0003] To assess the durability of bridge concrete structures, a series of technical standards related to concrete durability have been promulgated in China. Commonly used standards include GB / T 51355 "Standard for Durability Assessment of Existing Concrete Structures". Among them, the following provisions (1) to (3) are mainly made for the durability assessment of concrete structures in chloride-eroded environments: (1) The chloride ion concentration in concrete shall be determined according to the method of GB / T 50344 "Technical Standard for Testing of Building Structures". The sample shall be prepared by on-site core drilling and grinding; the core sample diameter should be 100 mm.

[0004] (2) Powder samples for chloride ion concentration on concrete surface are taken from within 5 mm of the component surface; when detecting chloride ion concentration distribution, samples are taken from the component surface every 2-3 mm along the depth.

[0005] (3) The chloride ion deposition process on the concrete surface and the chloride ion diffusion process in the concrete protective layer shall be determined according to the formula for evaluating the durability of concrete structures in chloride salt erosion environment in GB / T 51355.

[0006] In simple terms, the most important thing in assessing the durability of concrete structures in a chloride environment is to determine the parameter ti (the number of years at which the steel reinforcement begins to corrode). ti is ultimately deduced from several important parameters. The parameters that actually need to be measured are: the measured chloride ion concentration Cse on the concrete surface and the chloride ion concentration C(x,t0) at depth x during the test.

[0007] Inside concrete, the chloride ion concentration distribution trend is as follows: it gradually increases in concentration along the concrete surface, reaches a peak at a certain location, and then gradually decreases in concentration along the depth direction, eventually stabilizing. The situation at the point of maximum chloride ion concentration is of most noteworthy interest.

[0008] Numerous scholars have conducted research on the diffusion distribution of chloride ions within concrete, proposing various evaluation models and measurement methods for chloride ion diffusion behavior. The diffusion models for chloride ions can be broadly categorized as follows: diffusion models based on Fick's law, diffusion models based on the Nernst-Einstein equation, ion diffusion models based on multiphase composite media theory, and diffusion behavior prediction models based on machine learning. Methods for measuring chloride ion diffusion can be broadly divided into laboratory and field testing methods.

[0009] Overall, when assessing the durability of bridge concrete structures under chloride salt environments, and measuring the chloride ion concentration Cse on the bridge concrete surface and the chloride ion concentration C(x,t0) at a depth of x according to relevant standards and studies, the following issues arise: Question 1: According to the specifications, the chloride ion concentration value is measured by sampling at a position about 5mm away from the surface of the concrete component. However, it is impossible to determine whether this value represents the maximum chloride ion concentration inside the concrete, which may lead to inaccurate durability parameters obtained later.

[0010] Question 2: When measuring the chloride ion concentration distribution C(x,t0) at depth x according to the specifications, samples need to be taken from the surface of the core sample every 2-3 mm along the depth. From the perspective of on-site sampling, it is very difficult to cut the core sample into at least 5 2-3 mm specimens. Moreover, during the cutting process, the external working environment (such as long-term water flow) will cause environmental pollution to the small 2-3 mm specimens, which will interfere with the distribution of chloride ions inside the specimens to a certain extent, thus affecting the chloride ion concentration measurement value.

[0011] Question 3: Existing research on evaluation models for chloride ion concentration distribution has certain limitations: 1) When using diffusion models based on Fick's law or the Nernst-Einstein equation, the diffusion characteristics within the concrete structure are typically considered to be singular (steady-state or unsteady-state). This does not conform to the actual diffusion situation in the complex environment of concrete, and many sub-models have been derived based on this law, with no specific evaluation model clearly defined in current research or regulations; 2) When using ion diffusion models based on multiphase composite media theory, the diffusion behavior of the entire internal structure is inferred from the diffusion behavior of each phase. The problem is that the required parameters need to be determined... The number of experimental parameters is very large and the testing process is very complex. Moreover, each test is biased towards the microscopic level and is not suitable for the determination of chloride ion diffusion in existing concrete structures on site. 3) Existing machine learning methods for predicting concrete durability mostly use classic deep learning network models directly, taking relevant experimental parameters such as water-cement ratio, humidity, and time as inputs, and then fitting / predicting. However, they do not take into account the actual physical constraints of ion diffusion, which can lead to deviations or even overfitting in the prediction results. In addition, deep neural networks often require a large amount of data to train to achieve good results, while the amount of historical detection data related to chloride ions in concrete is insufficient to train deeper neural networks.

[0012] Question 4: Regarding the methods for determining chloride ion concentration distribution, laboratory methods are not suitable for determining the diffusion behavior of chloride ions inside bridge concrete in the field. Existing field testing methods, such as the Permit method, suffer from the problem of limited availability of testing equipment. The AC impedance technique has considerable potential for field testing, but there is limited research on it and the interpretable theory for its applicability is not yet fully mature.

[0013] In summary, accurately determining the surface concentration and internal distribution of chloride ions in concrete is the most critical step in assessing the durability of bridge concrete under chloride salt environments. However, a rapid and accurate method for measuring chloride ion diffusion in existing bridge concrete is currently lacking. Considering the advantages and disadvantages of the above research results and the problems encountered in practical testing, this paper proposes a method for determining the chloride ion concentration in existing bridge concrete. This method eliminates the need for complex sample preparation and, under conditions that satisfy the internal chloride ion diffusion characteristics of concrete, can accurately measure the maximum concentration and distribution of chloride ions inside bridge concrete, thereby assessing bridge durability. Summary of the Invention

[0014] The purpose of this invention is to overcome the shortcomings of the prior art and provide a bridge durability testing method, system, and computer storage medium.

[0015] Firstly, a bridge durability testing method is provided, including: S1. Prepare standard concrete test specimens and collect chloride ion concentration data; S2. Based on the chloride ion concentration data, construct and train a diffusion flow model based on condition-flow matching; the model is used to learn the transformation law of chloride erosion flow from the initial distribution to the target concentration distribution field; S3. Obtain chloride ion concentration detection data at multiple discrete depths for representative engineering parts of the concrete of the bridge under test, and collect the corresponding physical parameters; S4. Input the detection data and physical parameters obtained in S3 into the trained diffusion flow model. The model generates a chloride ion concentration diffusion flow matching field for the engineering part under test within a continuous depth range. S5. Extract the parameters required for bridge durability assessment from the chloride ion concentration diffusion flow matching field, and substitute them into the preset standard calculation formula to obtain the bridge durability assessment result.

[0016] As a preferred embodiment, in S1, chloride ion concentration data of concrete specimens with different mix proportions, different chloride salt erosion concentrations, and different service ages are collected at multiple discrete depths as discrete observation data, and chloride ion concentration data at continuous depths within the corresponding depth range are collected as complete label data.

[0017] Preferably, in S1, the plurality of discrete depths include a plurality of different depth points within the depth range of the diffusion zone; the continuous depth range is a complete diffusion interval from the concrete surface to the interior where the chloride ion concentration tends to stabilize.

[0018] Preferably, in S2, the physical parameters include the proportion of fly ash, the proportion of slag, the service life, and the chloride erosion concentration; the diffusion flow model includes a flow transformation module, which is used to learn the velocity field under the constraints of physical parameters and discrete observations, driving the initial Gaussian distribution to gradually evolve into the target concentration diffusion distribution field.

[0019] Preferably, in S2, the training process of the model incorporates physical constraints, including: hard constraints, soft constraints, and peak enhancement constraints. The hard constraints include: ensuring non-negative concentration through the activation function of the model output layer, and introducing the influence coefficient of admixtures into the flow transformation velocity field to make the chloride ion diffusion coefficient negatively correlated with the proportion of fly ash and slag. The soft constraints include: guiding the gradient of the concentration distribution field output by the model through the loss function to approach zero in the depth range of the stable region, and showing a trend of first increasing and then decreasing in the depth range of the diffusion region; The peak enhancement constraint includes: assigning higher weight to the prediction error of the peak region depth range in the reconstruction loss, thereby improving the fitting accuracy of peak concentration and peak depth.

[0020] Preferably, in S2, the total loss function of the model is a weighted sum of the conditional flow matching basic loss, peak weighted reconstruction loss, gradient constraint loss, and trend constraint loss.

[0021] In a second aspect, a bridge durability testing system is provided for performing any of the methods described in the first aspect, including: The preparation module is used to prepare standard concrete test specimens and collect chloride ion concentration data. A construction module is used to construct and train a diffusion flow model based on conditional-flow matching based on the chloride ion concentration data; the model is used to learn the transformation law of chloride erosion flow from the initial distribution to the target concentration distribution field; The acquisition module is used to acquire chloride ion concentration detection data at multiple discrete depths of representative engineering parts of the concrete of the bridge under test, and to collect the corresponding physical parameters. The input module is used to input the detection data and physical parameters acquired by the acquisition module into the trained diffusion flow model. The model generates a chloride ion concentration diffusion flow matching field for the engineering part under test within a continuous depth range. The extraction module is used to extract the parameters required for bridge durability assessment from the chloride ion concentration diffusion flow matching field, and substitute them into the preset standard calculation formula to obtain the bridge durability assessment result.

[0022] Thirdly, a computer storage medium is provided, wherein a computer program is stored therein; when the computer program is run on a computer, the computer causes the computer to perform any of the methods described in the first aspect.

[0023] Fourthly, an electronic device is provided, comprising: Memory, used to store computer programs; A processor for executing the computer program to implement the method as described in any of the first aspects.

[0024] The beneficial effects of this invention are: 1. This invention introduces a flow matching model based on flow matching, with the flow transformation from "distribution field to distribution field" as the core, to generate a chloride ion concentration diffusion flow matching field. It does not require a preset data distribution form and is naturally adapted to the unsteady diffusion distribution of chloride ions in bridge concrete structures, which "increases first and then decreases from the surface to the deep layer". It achieves a leap from "point prediction" to "field reconstruction", which the existing Fick's law model cannot accurately describe.

[0025] 2. The model established in this invention incorporates reversible physical constraints, transforming the physical laws of chloride ion diffusion (non-negative concentration, correlation between admixture and diffusion coefficient, trend of initial increase followed by decrease, and gradient constraint in the stable region) into a triple constraint of "velocity field weighting + output layer forcing + loss function guidance" in the flow matching field, thus solving the problem of "lack of physical constraints" in existing machine learning models.

[0026] 3. This invention employs a discrete sampling data acquisition method, utilizing discrete sampling at four core depths. The concentration distribution across the entire depth range of 0-29mm is reconstructed through flow matching, eliminating the need to damage the sample and effectively avoiding contamination and operational complexity issues caused by continuous physical cutting of core samples. Simultaneously, based on the model-reconstructed data distribution field, peak concentration, peak depth, and other global parameters, as well as all parameters required for standardized calculations, can be easily extracted, solving the core technical challenge of "inaccurate peak location" and replacing the traditional operation and mindset of "using fixed 5mm sampling for detection and evaluation." Attached Figure Description

[0027] Figure 1 A flowchart of a bridge durability testing method provided in this application; Figure 2 This is a structural schematic diagram of a bridge durability testing system provided in this application. Detailed Implementation

[0028] The present invention will be further described below with reference to embodiments. The description of the embodiments below is only for the purpose of helping to understand the present invention. It should be noted that those skilled in the art can make several modifications to the present invention without departing from the principle of the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention. Example 1:

[0029] The durability of bridge concrete is related to the chloride ion concentration within the concrete. Inside the concrete, the chloride ion concentration distribution trend is as follows: it gradually increases in concentration along the surface, reaching a peak at a certain location, and then gradually decreases in concentration along the depth direction, eventually stabilizing. Specifically, concrete can be divided into a reaction layer and a diffusion layer from the surface to the interior. The reaction layer refers to the area on the concrete surface where chloride ions undergo chemical reactions and reach their peak concentration; it is the core area where chloride ions combine with hydration products and accumulate. The diffusion layer refers to the area where chloride ions migrate into the concrete interior via diffusion, and their concentration gradually decreases with depth. The chloride ion concentration at the peak location within the reaction layer (i.e., the maximum chloride ion concentration value) is a key parameter for durability assessment.

[0030] Due to conservative safety considerations, most studies and regulations use the maximum chloride ion concentration as the upper limit to establish a series of chloride ion evaluation models. For example, GB / T 51355 specifies that "powder samples for chloride ion concentration on the concrete surface should be taken from within 5 mm of the component surface," a regulation based on domestic and international research. Numerous domestic and international studies indicate that the maximum chloride ion concentration inside concrete often occurs in the region approximately 3 mm to 10 mm from the concrete surface. Therefore, the essential intention of GB / T 51355's requirement to "determine the chloride ion concentration on the concrete surface at 5 mm" is to measure the maximum chloride ion concentration inside the concrete and use this to infer the maximum limit of bridge concrete durability. However, existing methods have relatively low accuracy in determining the maximum chloride ion concentration and its distribution inside concrete.

[0031] To accurately measure the maximum concentration and distribution of chloride ions inside bridge concrete, Example 1 of this application constructs a rapid, non-destructive, and highly accurate method for determining chloride ion concentration on-site, optimizes the accuracy of inferring the service life of steel reinforcement corrosion, and ultimately improves the reliability of bridge concrete durability assessment.

[0032] Specifically, such as Figure 1 As shown, Embodiment 1 of this application provides a bridge durability testing method including: S1. Prepare standard concrete test specimens and collect chloride ion concentration data at multiple discrete depths for concrete specimens with different mix proportions, different chloride salt erosion concentrations, and different service ages as discrete observation data. Also collect chloride ion concentration data at continuous depths within the corresponding depth range as complete label data.

[0033] S1 includes: S101. Prepare standard test specimens.

[0034] In the embodiments of this application, concrete specimens with different mix proportions, different chloride erosion concentrations, and different service ages were prepared in S101.

[0035] For example, in terms of mix design, the embodiments of this application cover four common types of bridge concrete: "ordinary concrete, single-admixture fly ash, single-admixture slag, and double-admixture composite material," and design a total of 6 mix proportions, as shown in Table 1 below: Table 1 Mix Proportion Design of Standard Test Specimens It should be noted that the mix design can be based on GB 50164 "Standard for Quality Control of Concrete" and JTG / TF50 "Technical Specification for Construction of Highway Bridges and Culverts". The application ratio of admixtures (fly ash and slag) in bridge concrete is concentrated between 15% and 50%, and there are two mainstream forms: "single admixture" and "double admixture". Furthermore, the lower limit of the admixture ratio should be ≥15%. When it is below 15%, the effect on the chloride ion diffusion coefficient is extremely weak, and admixtures below 15% are rarely used in engineering and have no research value. In addition, this application requires upper limit control: fly ash ≤40%, slag ≤50%; according to GB 50164, the total amount of admixtures should not exceed 50% of the total amount of cementitious materials. High admixture amounts easily lead to insufficient early strength, and proportions above 50% are rarely used in engineering.

[0036] For specimen aging treatment, salt spray tests can be used to simulate the natural chloride salt corrosion environment.

[0037] Specifically, six gradients of chloride salt erosion concentrations were set for NaCl solutions at concentrations of {0%, 2%, 5%, 7%, 10%, and 12%}. Among these: The 0% concentration test group served as a natural exposure blank control group, simulating the background state of bridge concrete without external chloride salt erosion. It only required standard curing under the same conditions (without salt spray treatment) and maintained the same temperature, humidity, curing age, and specimen size as the other 5 salt spray erosion groups. The 2% concentration test group corresponds to mild chloride salt erosion (coastal offshore areas, roads and bridges with very little salt application).

[0038] The 5% concentration test group corresponds to moderate to mild erosion (bridges in northern areas where snow melts and salt is spread in winter, and coastal areas with mild salt spray).

[0039] The 7% concentration test group corresponds to moderate erosion (coastal nearshore areas and urban main road bridges with a large amount of salt).

[0040] The 10% concentration test group corresponds to moderate to severe erosion (extreme scenarios, such as coastal high-salt fog areas and near-shore sections of cross-sea bridges).

[0041] The 12% concentration test group corresponds to severe extreme erosion (seawater splash zone, salt field perimeter / coastal industrial area bridge).

[0042] Specifically, the simulation simulates service life at three different tiers: 1 year, 3 years, and 5 years.

[0043] For bridges in the short-term (<5 years) and medium-term (5-10 years) service phases, chloride ion diffusion is in a critical period of non-steady state.

[0044] In addition, three parallel specimens of 100mm×100mm×30mm were prepared for each type of condition.

[0045] Specifically, 6 mix proportions × 3 age gradients × 6 erosion concentrations = 108 states; therefore, a total of 108 × 3 = 324 specimens were prepared.

[0046] Specifically, specimen preparation and curing were carried out in accordance with the relevant methods and requirements of GB / T 50082 "Standard for Test Methods of Long-Term Performance and Durability of Concrete".

[0047] It should be noted that 100mm×100mm×30mm is a commonly used size in standard specifications. According to the current research consensus, the main diffusion area of ​​chloride ions in concrete is 0~30mm below the surface (peak value 3~10mm, tending to stabilize after 15mm). A thickness of 30mm just covers the entire diffusion range, which can completely capture the full distribution law of "increasing, peak value, decreasing, and stable". This size belongs to the "adaptive size" recommended by the specification and meets the test requirements of GB / T 50082.

[0048] S102. Obtain experimental data for standard test specimens.

[0049] In S102, the first step is to divide the experimental data into pairs.

[0050] Specifically, the data of a single specimen is divided into discrete observation data pairs and complete label data pairs.

[0051] The discrete observation data pairs include at least four depths, with one concentration value at each depth. The complete label data pairs include 30 depths (0 mm, 1 mm, 2 mm...29 mm), with one concentration value at each depth.

[0052] For example, discrete observation data includes concentration values ​​at four depths (2mm, 5mm, 8mm, and 12mm), which cover the core diffusion layer: 1) Surface peak region (2-5mm): The erosion peak of chloride ions on the concrete surface is usually concentrated on the surface. 2mm is close to the surface, and 5mm corresponds to the standard sampling point. These two points are key to confirming whether the concentration exceeds the standard; 2) Effective diffusion region (8-12mm): The chloride ion concentration decreases with increasing depth. 8mm and 12mm are located in the main attenuation segment of the diffusion curve, and these two points can effectively backfit the diffusion coefficient; 3) Global fitting: The four points of 2mm, 5mm, 8mm, and 12mm are distributed in the first 40% of the 0-29mm range. After reconstruction by the CFM model, they are sufficient to support the continuity of the entire depth field. Missing even one point may lead to the loss of peak values. Therefore, 2mm, 5mm, 8mm, and 12mm are key depths, and only diffusion models constructed based on these depths have practical significance. Based on these four depths, this application can also increase the number of sampling points as needed, which will not be elaborated further.

[0053] For example, when the discrete observation data pairs include 4 depths, the total number of data pairs that can be obtained for a single specimen is 4 + 30 = 34. For all specimens (324 specimens), a total of 34 × 324 = 11016 data pairs can be obtained, of which: discrete observation data pairs = 4 / specimen × 324 = 1296, and complete label data pairs = 30 / specimen × 324 = 9720.

[0054] After dividing the experimental data into pairs, the experimental data is obtained.

[0055] Specifically, a commonly used micro sampler (diameter ≤ 5 mm) can be used to drill a small amount of powder sample (about 5 g) at the specified depth of the standard test piece. The chloride ion concentration is determined according to the GB / T 50344 method and recorded as experimental data pairs: [depth xi, concentration Ci], i = 1, 2, 3, 4...

[0056] S2. Based on the discrete observation data and complete label data, construct and train a diffusion flow model based on condition-flow matching; the model takes the physical parameters of concrete and the observed chloride ion concentration at discrete depths as input conditions, and the chloride ion concentration distribution field within a continuous depth range as the output target, and learns the transformation law of chloride salt erosion flow from the initial distribution to the target concentration distribution field.

[0057] S2 includes: S201. Construct a standardized training dataset.

[0058] Specifically, based on the standard test specimen of S1, as shown in Table 2, the raw data of a single specimen contains three types of core information, which are finally integrated into a single sample data unit.

[0059] Table 2. Raw data for single specimens Next, data preprocessing is performed, including data normalization and dimensional integration.

[0060] The data normalization involves normalizing depth features, concentration features, and physical parameters separately, mapping all data to the [0, 1] interval to improve model convergence speed, as shown in the following formula: (Equation 1) In the formula, C norm (i) represents the normalized concentration value at the i-th depth.

[0061] It should be noted that the concentration characteristic must retain its non-negativity (chloride ion concentration cannot be negative). If a small negative value appears after normalization (floating point error), it should be set to 0 directly.

[0062] The purpose of this dimensional integration is to unify the model's input and output formats, integrating the three types of preprocessed data into a tensor format readable by the model (e.g., a PyTorch tensor with dimensions of [number of samples, number of features]), clarifying the correspondence between each tensor and the conditional features and target distribution field. Specific dimensional design includes: conditional input tensor X... cond Discrete observation tensor X obs The total input to the model is X, and the output label of the model is Y.

[0063] Among them, the conditional input tensor X cond ∈ N×4 N represents the number of valid samples, and the 4 dimensions correspond to the normalized physical parameters, which are the global conditional features of the distribution field generation. Discrete observation tensor X obs ∈ N×8 The 8-dimensional normalized depth and concentration corresponding to discrete sampling points are local observation condition features of the distribution field generation. Model total conditional input X∈ N×12 : X cond With X obs Concatenation, i.e., X = concat(X cond X obs () represents the comprehensive conditional characteristics of the distribution field generation; Model output label Y∈ N×30 The 30-dimensional normalized discrete representation of the concentration diffusion distribution field from 0 to 29 mm is the label tensor for model training. The model learns the mapping relationship from X to Y (from comprehensive conditional features to the concentration diffusion distribution field).

[0064] After data preprocessing, the dataset is divided. Specifically, a stratified random partitioning strategy based on distribution characteristics is adopted, with partitions based on "mixture ratio group + age + erosion concentration" to ensure that the chloride ion concentration diffusion distribution type (such as peak position 3-10mm, stable zone gradient, concentration trend) is evenly distributed in the training set, validation set, and test set. The partitioning ratio is: 70% for training set, 20% for validation set, and 10% for test set.

[0065] S202, Model Architecture Design.

[0066] This invention adopts a conditional flow matching (CFM) model architecture.

[0067] Specifically, the model has 5 layers: input layer, conditional feature embedding layer, stream transformation core module, distribution sampling layer, and output layer.

[0068] The input layer is used to receive the total conditional input X∈ of the S201 model. B×12 (B is batch_size), the feature dimension is mapped to a high-dimensional feature space through linear transformation, which is suitable for the subsequent fusion of conditional features and distribution field. The formula is as follows: (Equation 2) In the formula: This is a linear transformation operation; For batch normalization processing; The activation function outputs a high-dimensional integrated conditional feature X. in ∈ B×64 .

[0069] The conditional feature embedding layer is used to achieve deep fusion of comprehensive conditional features and concentration diffusion distribution field. It assigns adaptive conditional features to each depth position of the distribution field, so that the distribution field generation has depth position specificity and adapts to the depth dependence law of chloride ion diffusion. It mainly includes two steps: deep sequence embedding and hybrid feature fusion.

[0070] Specifically, for depth sequence embedding, the 0-29mm depth sequence (after normalization) is denoted as x. depth ∈ 30×1 The depth location feature X is mapped to a 1D convolution. depth ∈ 30×32 The formula is: (Equation 3) Specifically, hybrid feature fusion includes: combining high-dimensional comprehensive conditional features X in With depth location feature X depth Perform broadcast stitching to generate the distribution field location adaptive conditional feature tensor X. fusion ∈ B×30×96 , which is the input of the stream transform core module; in this tensor, 30 corresponds to the depth position of the concentration diffusion distribution field, and 96 is the adaptive condition feature at this position, realizing "the condition feature is dynamically adjusted with the depth position of the distribution field".

[0071] Furthermore, the diffusion flow model includes a flow transformation module (the core of CFM). Specifically, this module is the core execution module for generating the concentration diffusion distribution field, and its core is the velocity field v under conditional feature constraints. t (x t ,X fusionThe initial Gaussian distribution is gradually evolved into the target concentration diffusion distribution field through the velocity field. This module adopts 6 layers of alternating masking affine coupling layers. Specific details include: basic definition of conditional flow matching, velocity field prediction, and flow transformation reversibility.

[0072] Basic definition of conditional flow matching: Adaptive conditional feature X at the location of the distribution field fusion Under constraints, define the continuous stream transformation x t ∈ 30×1 (t∈[0,T], T=1 is the total number of stream transformation steps), satisfying: (Equation 4) In the formula, v t (·) represents the velocity field function learned by the model (with parameter θ), x0~N(0,I) is the initial Gaussian distribution, and x1 is the concentration diffusion distribution field generated by the model, which is consistent with the target distribution field Y.

[0073] Velocity field prediction: The velocity field v is modeled using 6 layers of alternating affine coupling layers. t Each layer will have X fusion Based on the channel, it is divided into fixed parts X fix and the transformation part X trans By fixing the scale s and offset t of the predicted transformation part, the affine transformation of the feature is achieved, according to the formula: (Equation 5) (Equation 6) Wherein, the scaling function s( ) and offset function t ( Both are 3-layer MLPs, with fixed features as input and scale / offset coefficients of the transformed part as output, as shown in the formula: (Equation 7) Stream transform reversibility: Ensuring the log-likelihood calculation of the stream transform improves the stability of model training, while allowing physical constraints to be reversibly embedded in the stream transform process; the affine coupling layer is a reversible transformation, as shown in the formula: (Equation 8) Distributed sampling layer: Based on the Euler method, the continuous flow transformation of Equation 4 is numerically solved to achieve rapid sampling from the initial Gaussian distribution to the target concentration diffusion distribution field. Sampling formula: (Equation 9) Output layer: The feature tensor after the stream transformation is linearly transformed, and the output value is forced to fall within the [0,1] interval by the Sigmoid activation function, matching the normalized concentration diffusion distribution field while ensuring the global nonnegativity of the distribution field. (Equation 10) In the formula, Sigmoid(·) is the activation function, and the normalized concentration diffusion distribution field generated by the output model is Y. pred ∈ B×30 Z T This is the feature tensor output by the core module of the stream transform (T=1, stream transform terminates).

[0074] The flow matching model of this invention has a total of about 850,000 parameters, which is much lower than that of conventional deep learning models. Specifically, it is distributed as follows: input layer (12×64=768) + embedding layer (about 120,000) + flow transformation module (about 700,000) + output layer (30×30=900), ensuring that the model is lightweight and easy to deploy.

[0075] S203. Establish physical constraints.

[0076] In S203, this application transforms the one-dimensional diffusion physics of chloride ions in concrete into model process constraints, deeply integrating the velocity field, distribution field output, and loss function of the conditional flow matching model. This forms a triple constraint system of "velocity field evolution constraint," "distribution field output constraint," and "loss function penalty," ensuring that the concentration diffusion distribution field generated by the model not only closely matches the measured data but also strictly conforms to the physical laws of chloride ion diffusion. It should be noted that all constraints are designed for the entire concentration diffusion distribution field, not for a single data point.

[0077] Specifically, a triple physical constraint is established, including: hard constraint, soft constraint, and peak enhancement constraint.

[0078] The hard constraints, embedded in the model architecture, enforce the physical rationality of the constraint flow transformation process and the distribution field output without the need for a loss function. There are two hard constraints (hard constraint 1 and hard constraint 2).

[0079] Hard constraint 1: Global non-negative constraint on chloride ion concentration diffusion distribution field.

[0080] Based on physical principles, the chloride ion concentration in concrete is a non-negative value; therefore, the model output layer uses the Sigmoid activation function to ensure the generated distribution field Y. pred The concentration values ​​at all depth locations are ∈ [0,1], and after inverse normalization, they are naturally non-negative (the minimum concentration before normalization is 0).

[0081] Hard constraint 2: The correlation constraint between diffusion coefficient D and admixture (Fb / Sb).

[0082] According to GB / T 50344 and the concrete durability specification, the chloride ion diffusion coefficient D is negatively correlated with the proportion of fly ash (Fb) and the proportion of slag (Sb). The core correlation formula is as follows: (Equation 11) Where: D0 is the diffusion coefficient of ordinary concrete, α F (Fly ash influence coefficient) = 0.015, α S (Slag Influence Coefficient) = 0.02 (Slag has better resistance to chloride salt erosion than fly ash).

[0083] Embedding this formula into the velocity field of the flow transform, and using correction coefficients... The velocity field is weighted depthwise; first, based on the input F... b S b Calculate the correction factor ,formula: (Equation 12) Depth-wise weighted formula for velocity field: (Equation 13) In the formula, v t 'This is the weighted velocity field,' Scaling the global velocity field for high-admixture specimens Smaller size results in a slower velocity field evolution and a smoother diffusion of the generated distribution field concentration, thus constraining the diffusion law from the root of the flow transformation.

[0084] In this way, the model can automatically adjust the diffusion law according to the proportion of admixtures during the flow transformation process (e.g., the feature weight of high Sb specimens is lower, which corresponds to slower concentration diffusion and lower concentration value), forcing the model output to conform to the physical law that "the more admixtures, the stronger the resistance to chloride salts".

[0085] The soft constraints guide the overall trend of the concentration diffusion distribution field to conform to the chloride ion diffusion law. The loss penalty makes the distribution field generated by the model satisfy "the diffusion region first increases and then decreases, and the gradient in the stable region approaches 0". There are two terms (soft constraint 1 and soft constraint 2).

[0086] Soft constraint 1: Concentration gradient approaches zero in the stable region (depth > 15 mm). (Global gradient constraint across the distribution field) According to physical rules, the core principle of chloride ion diffusion in concrete is: 0-15mm is the diffusion zone (concentration initially increases then decreases, with a peak value), and 15-29mm is the stable zone (concentration gradient approaches 0, concentration remains essentially constant); this relates to the generated distribution field Y. pred ∈ B×30 Calculate the discrete first-order gradient tensor g∈ in the stable region. B×14 , (Equation 14) In the formula, g iY represents the first-order concentration gradient between depths of the i-th mm and (i+1)-th mm; pred [i] represents the normalized concentration at depth i predicted by the model.

[0087] Gradient loss: (Equation 15) If the concentration fluctuation (large gradient) in the 15-29mm range of the model output is large, then L grad Increase the penalty model to guide its output of a distribution where the concentration in the stable region remains basically unchanged.

[0088] Soft constraint 2: The concentration trend of first increasing and then decreasing in the 0-29mm range is constrained (peak area constraint).

[0089] According to physical rules, chloride ions diffuse from the surface of concrete into its interior. Influenced by the concrete's density and capillary distribution, the concentration exhibits a pattern of initial increase followed by decrease: "lower at the surface, higher in the peak region, and lower at deeper layers." It cannot monotonically increase / decrease. The discrete second-order gradient tensor of the generated distribution field is calculated. ∈ B×28 (Reflecting the concavity and convexity of the distribution field), formula: (Equation 16) In the formula, The concentration second-order gradient at the i-th mm depth (reflecting concavity / convexity); Let be the first-order gradient at the i-th mm depth.

[0090] Trend loss : (Equation 17) The greater the loss, the more the distribution field trend deviates from "increase first and then decrease". The model reduces this loss by optimizing and guides the distribution field to show the trend characteristics that conform to chloride ion diffusion.

[0091] The peak enhancement constraint and weighted loss improve the fitting accuracy in the 3-10mm peak range.

[0092] Based on physical laws, the 3-10mm range is the core region for the peak chloride ion concentration distribution, requiring the model to be more sensitive to prediction errors in this region. First, for the predicted concentration in the 3-10mm range (a total of 8 depths), a weighting coefficient λ=2 (with a weight of 1 for other depths) is added to the reconstruction loss to enhance the fitting accuracy in this region; then, the peak-weighted reconstruction loss is calculated. In the formula, This represents the peak-weighted reconstruction loss, a weighted loss value used during model training to evaluate the difference between the generated concentration field and the true concentration field. The smaller this value, the closer the reconstructed concentration distribution field is to the actual measured value. `i` is the concrete depth subscript, ranging from 0, 1, 2, ..., 29, corresponding to the complete diffusion depth range of 0~29 mm. For example, `i=3:10` means a continuous depth subscript range, representing 8 depth points: `i=3, 4, 5, 6, 7, 8, 9, 10`. `i≠3:10` means the depth subscript includes all points not within the 3~10 mm range, i.e., `i=0, 1, 2, 11, 12, ..., 29`, a total of 22 depth points. Y pred [i] represents the predicted normalized concentration at depth i mm; Y true [i] represents the measured normalized concentration at the i-th depth. MSE(·,·) represents the mean square error.

[0093] In summary, as stated in S203, all physical constraints do not exist independently, but are deeply integrated into the "feature layer-loss layer-output layer" of the conditional flow matching model, forming a triple constraint system of "structural mandatory constraints + loss-guided constraints + peak-weighted fitting". The overall fusion logic of constraints and model is as follows: {Physical parameters (Fb / Sb), velocity field weighting (hard constraint 2), stream transformation core module (constrained feature guidance), distribution field output (hard constraint 1: non-negativity), loss layer (soft constraint + peak enhancement constraint), model parameter θ optimization, velocity field update} S204, Model Training and Optimization.

[0094] Specifically, in conjunction with the triple physical constraints of this invention S203, the training and optimization of the conditional flow matching model includes: loss function design (fusion of CFM basic loss and physical constraint loss), setting of basic training parameters, two-stage training process, model optimization strategy, and model evaluation.

[0095] Specifically, the total loss function of the model is a weighted fusion of the conditional flow matching basic loss (velocity field matching loss) and the physical constraint loss of this invention (peak weighted reconstruction loss + gradient constraint loss + trend constraint loss), as shown in the following formula: (Equation 19) In the formula: , , , , where represents the loss weight parameter. In this example, the initial loss weight is set to . =1.0, (Main loss) , ; The total loss of the model, The peak-weighted reconstruction loss, For gradient constraint loss, For trend-constrained losses, The conditional flow matching base loss (velocity field matching loss) is the core loss for training the CFM model. It is used to learn the matching relationship between the velocity field and the real flow, and the formula is as follows: (Equation 20) In the formula, The expected value is represented by t ~ U(0,1); t ~ U(0,1) represents random time step sampling, where t follows a [0,1] distribution; x0 represents the initial Gaussian distribution sample; x1 ~ P data This indicates that the target state x1 follows the target data distribution P. data P data This represents the probability distribution corresponding to the complete label data (concentration distribution across the entire depth from 0-29mm) of the standard specimen, i.e., the probability distribution of the actual chloride ion concentration diffusion field that the model needs to learn; x t =t·x1+(1-t)·x0 represents the intermediate state of the stream transformation, connecting the initial Gaussian distribution and the target data distribution P. data ;v t (·) represents the velocity field function learned by the model; X fusion For the adaptive conditional feature tensor of the distribution field location; It is the square of the L2 norm, which measures the difference between the predicted velocity field and the actual flow (x1-x0).

[0096] Regarding the basic training parameter settings, the initial hyperparameter settings are shown in Table 3.

[0097] Table 3 Initial settings for hyperparameters Regarding the two-stage training process, specifically, a two-stage training strategy of "low-weight pre-training to full-weight formal training" is adopted, first allowing the model to learn the basic distribution reconstruction rules, and then incorporating physical constraints.

[0098] The first stage is low-weight pre-training (epochs 0-100), with the following parameter settings: {loss weight settings:} =1.0, , , Learning rate: 1e-3}. This stage aims to enable the model to learn the basic laws of the velocity field of conditional flow matching, and to achieve a basic reconstruction of the concentration diffusion distribution field. The training set L is required to be [value missing]. cfm It dropped below 0.01.

[0099] The second stage is full-weight formal training (epochs 101-500), with the following parameter settings: {Loss weights restored to initial values:} =1.0, , , Incorporating complete physical constraints; Learning rate strategy: employing cosine annealing learning rate decay; Early stopping strategy: to verify lumped loss L. total As a monitoring metric, if the loss on the validation set does not decrease for 20 consecutive epochs, training is stopped and the current optimal model weights are saved.

[0100] In addition, the model optimization strategies include a triple optimization strategy of hyperparameter grid search, data augmentation, and lightweight model pruning. Hyperparameter grid search: A grid search is performed on the core hyperparameters, with the peak localization error of the validation set as the core indicator, to select the optimal combination, as shown in Table 4.

[0101] Table 4 Hyperparameter grid search combinations Data augmentation: Apply mild Gaussian noise augmentation to the discrete observation data of the training set (simulating sampling / detection errors in the engineering field), with the noise intensity controlled within ±3% (meeting the detection error requirements of GB / T 50344), as shown in the following formula: (Equation 21) Lightweight model pruning: After training, channel pruning (pruning rate 10%-20%) is performed on the model's stream transform module to remove feature channels with low contribution. Without losing accuracy, this further reduces the number of model parameters and improves inference speed.

[0102] In terms of model evaluation, a triple evaluation system of "CFM-specific indicators + general fitting indicators + engineering core indicators" is constructed to comprehensively evaluate the overall accuracy, physical rationality, and engineering applicability of the concentration diffusion distribution field generated by the model.

[0103] Specifically, the model evaluation metrics include: CFM metrics, distribution field metrics, general fit metrics, and engineering core metrics.

[0104] CFM Specification: Velocity Field Matching Error L cfm This reflects the degree of matching between the velocity field learned by the model and the real flow; the smaller the value, the better.

[0105] Distribution field indices include Earth Movement Distance (EMD) and Distribution Field Pearson Correlation Coefficient (PCC). EMD measures the overall similarity between the generated and measured distribution fields; the smaller the value, the more accurate the reconstruction. EMD should be ≤ 5 mg / kg. PCC measures the trend consistency between the generated and measured distribution fields; a value closer to 1 is better. PCC should be ≥ 0.95.

[0106] Common fitting metrics: used to evaluate the accuracy of overall distribution reconstruction, mainly including mean squared error (MSE) and coefficient of determination (R²). 2 A smaller MSE is better, as it reflects the overall concentration prediction bias; R 2 The closer to 1, the better, reflecting the degree of fit between the predicted distribution and the true distribution. R0 is required to be... 2 ≥0.95.

[0107] Key engineering indicators include peak location error (mm), peak concentration relative deviation (%), and mean gradient in the stable region (mg / kg·mm). Peak location error (mm) = [model predicted peak depth - measured peak depth] ≤ 0.5 mm; Peak concentration relative deviation (%) = [(predicted peak concentration - measured peak concentration) / measured peak concentration × 100%] ≤ 5%; Mean gradient in the stable region (mg / kg·mm) ≤ 0.5 mg / kg·mm.

[0108] It should be noted that all core engineering indicators must meet preset thresholds for the model to be considered as successfully trained.

[0109] S3. Obtain chloride ion concentration detection data at multiple discrete depths for representative engineering parts of the concrete of the bridge under test, and collect the corresponding physical parameters.

[0110] S4. Input the detection data and physical parameters obtained in S3 into the trained diffusion flow model. The model generates a chloride ion concentration diffusion flow matching field for the engineering part under test within a continuous depth range.

[0111] S5. Extract the parameters required for bridge durability assessment from the chloride ion concentration diffusion flow matching field, and substitute them into the preset standard calculation formula to obtain the bridge durability assessment result. Example 2:

[0112] Based on Example 1, Example 2 of this application provides a more specific method for bridge durability testing, including: S1. Prepare standard concrete test specimens and collect chloride ion concentration data at multiple discrete depths for concrete specimens with different mix proportions, different chloride salt erosion concentrations, and different service ages as discrete observation data. Also collect chloride ion concentration data at continuous depths within the corresponding depth range as complete label data.

[0113] S2. Based on the discrete observation data and complete label data, construct and train a diffusion flow model based on condition-flow matching; the model takes the physical parameters of concrete and the observed chloride ion concentration at discrete depths as input conditions, and the chloride ion concentration distribution field within a continuous depth range as the output target, and learns the transformation law of chloride salt erosion flow from the initial distribution to the target concentration distribution field.

[0114] S3. Obtain chloride ion concentration detection data at multiple discrete depths for representative engineering parts of the concrete of the bridge under test, and collect the corresponding physical parameters.

[0115] S3 includes: S301. Select representative test objects.

[0116] Specifically, based on the bridge's structural type, service environment, and service life, representative components are selected for core sampling, prioritizing areas with high chloride corrosion risk. These primarily include: cross-sea / coastal bridges and bridges in areas where salt is applied during winter.

[0117] For cross-sea / coastal bridges, representative parts include: the bottom of the main beam, the pier water level fluctuation zone, and the bearing pad; for bridges in areas where salt is applied in winter, representative parts include: the lower layer of the bridge deck pavement, the side edge of the main beam, and the lower part of the pier column.

[0118] S302, On-site sampling and data acquisition.

[0119] Specifically, in accordance with relevant standards (such as GB / T 51355) regarding sampling size and quantity requirements, representative components were selected for core sampling. Following the S102 method, the discrete sampling procedure using a miniature electric sampler for the standard specimen was employed to ensure that the sampling depth of the four core samples was consistent with that of the standard specimen. Subsequently, the chloride ion concentration data of the sample was determined according to the S102 method.

[0120] In addition, four-dimensional physical parameters (Fb, Sb, t0, chloride salt erosion concentration level) consistent with those of the standard specimen are collected as input conditions for the flow model; the parameters can be obtained from design data / on-site relevant tests.

[0121] Specifically, the key physical parameters and their acquisition methods are shown in Table 5 below: Table 5 Key physical parameters and their acquisition methods It should be noted that the data collected by S302, serving as the raw basis for subsequent data processing, contains two types of core information: Core Information 1. Discrete Observation Data: Two-dimensional data pairs at four depths (depth, measured concentration), such as (2mm, 325.6mg / kg), (5mm, 489.2mg / kg), etc. Core Information 2: Conditional Physical Parameters: Specific values ​​for Fb, Sb, t0, and chloride salt corrosion concentration levels.

[0122] S303, Standardization of engineering sample data.

[0123] Specifically, referring to the S201 method, the data collected in S302 is normalized and dimensionally integrated, and transformed into a standardized tensor that can be directly read by the flow matching model.

[0124] First, feature classification and normalization are performed.

[0125] For the three types of features—physical parameters, depth, and concentration—the data are calculated according to the normalization formula of the training set, and then mapped to the [0,1] interval: (Equation 22) In the formula, X 工程 The original parameters (physical parameters / depth / concentration) of the sample; X 训练集Min X 训练集Max These are the minimum and maximum values ​​of the parameters counted in the S21 training set, respectively.

[0126] Next, dimensional integration is performed: following the format used during model training, the normalized parameters are concatenated into a 12-dimensional row tensor: {first 4 dimensions: normalized physical parameters (Fb, Sb, t0, chloride erosion concentration); last 8 dimensions: "normalized depth + normalized concentration" of 4 discrete depths are concatenated sequentially (e.g., 2mm_norm, C2mm_norm, 5mm_norm, C5mm_norm...)}.

[0127] S4. Input the detection data and physical parameters obtained in S3 into the trained diffusion flow model. The model generates a chloride ion concentration diffusion flow matching field for the engineering part under test within a continuous depth range.

[0128] S4 includes: S401, Model Reconstruction and Global Restoration of Flow Matching Field.

[0129] First, the engineering input tensor ([1,12]) is input into the flow matching model. The model is based on the chloride erosion flow transformation law learned by S2. It evolves from the initial Gaussian distribution to generate the normalized chloride ion concentration diffusion flow matching field tensor [1,30] of 0-29mm. This tensor is a discrete digital representation of the flow matching field, corresponding to the normalized concentration value of 1mm depth in 0-29mm, reflecting the global distribution characteristics of chloride erosion. Then, the normalized flow matching field tensor is denormalized globally using the denormalization formula S21 to restore it to the actual chloride ion concentration value (unit: mg / kg) in the engineering field, thus obtaining the actual chloride ion concentration diffusion flow matching field of the component under test, denoted as F. eng =[C0,C1,...,C 29 ] T C i Let be the actual chloride ion concentration at depth i mm in the flow matching field, so as to achieve accurate reconstruction of the flow matching field from the model space to the engineering space.

[0130] S402. Verification of the physical rationality of the flow matching field across the entire domain.

[0131] To ensure that the reconstructed flow matching field closely matches the actual chloride erosion patterns in the engineering project, it is necessary to adjust F... eng A full-domain physical rationality verification is required. Parameters can only be extracted after the verification is passed. If the verification fails, the sampling and data processing procedures of S3 need to be reviewed and data needs to be collected again. The verification indicators include: peak feature verification, trend feature verification, and input fitting verification.

[0132] Peak feature verification: Peak depth x of the flow-matched field peak Within the 3~10mm range, the peak concentration Cs is the maximum value across the entire flow matching field, which conforms to the peak distribution law of chloride salt erosion; Trend characteristic verification: The flow matching field should show a global trend of "low concentration at the surface, high concentration in the peak region, low concentration in the deep layer, and stable concentration in the stable region". The concentration in the 0~15mm diffusion region should first increase and then decrease. The average concentration gradient in the 15~29mm stable region should be ≤0.5mg / kg・mm, and the coefficient of variation should be ≤10%. Input fitting verification: The relative deviation between the predicted concentration values ​​at the four discrete sampling depths in the flow matching field and the measured values ​​is ≤8%, ensuring that the flow matching field fits the actual engineering data.

[0133] S403, Extraction of core parameters for durability assessment of flow matching field.

[0134] Global feature parameters were extracted from the validated chloride ion concentration diffusion flow matching field. All parameters are inherent features of the flow matching field, rather than values ​​of a single discrete point, ensuring the accuracy and representativeness of the parameters. The core parameters and extraction methods are shown in Table 6 below: Table 6. Explanation of core parameter extraction for durability assessment The extracted durability assessment parameters, combined with the calculation formulas in bridge standards and specifications, lead to the required durability assessment conclusions.

[0135] S5. Extract the parameters required for bridge durability assessment from the chloride ion concentration diffusion flow matching field, and substitute them into the preset standard calculation formula to obtain the bridge durability assessment result.

[0136] Specifically, based on Cs and C(x,t0) extracted from the flow matching field, durability evaluation parameters are calculated strictly in accordance with relevant standards.

[0137] In this example, following GB / T 51355, k is calculated sequentially according to the formula specified therein (as shown in Table 7). s 、D0、D、K、ti.

[0138] Table 7. Formulas for assessing the durability of concrete structures in chloride-erosion environments according to GB / T 51355. For example, S5 includes: S501, Calculate the chloride ion aggregation coefficient k on the concrete surface s With the revised C s .

[0139] Specifically, calculate according to formula 1-2 in Table 7. Where C se The global peak concentration C extracted for the flow-matched field s That is, the measured peak concentration.

[0140] S502, Calculate the chloride ion diffusion coefficient D0.

[0141] Specifically, calculate according to formula 1-4 of GB / T 51355. Where x is the selected diffusion depth, and C(x,t0) is extracted from the model results; select C(x,t0) at three typical depths of 5mm, 8mm and 10mm from the flow matching field, calculate D0 for each, and take the arithmetic mean as the final D0 value.

[0142] S503. Calculate the time-dependent chloride ion diffusion coefficient D.

[0143] Specifically, calculate according to formulas 1-3 in Table 7. Where α is the influence coefficient obtained from the table, and Fb and Sb are the physical parameters of the engineering sample.

[0144] S504. Calculate the chloride corrosion coefficient K and the service life ti of the steel reinforcement before it begins to rust.

[0145] Among them, ti is the core indicator for evaluating the durability of bridge concrete under chloride salt environment, reflecting the remaining durability of the component from the time of testing to the time when the steel reinforcement begins to corrode.

[0146] Specifically, calculate according to Formula 1-1 in Table 7. Where c is the thickness of the concrete cover, C cr The critical concentration obtained from the table, erf -1 It is the inverse function of the error function.

[0147] S505, Durability rating.

[0148] Specifically, as shown in Table 8, based on the calculated service life ti of the steel reinforcement to begin corroding, combined with the target service life t of the bridge... target (According to the bridge design documents, the service life is generally 50 years / 100 years). The durability level is classified according to the provisions of GB / T 51355, and the level classification adopts a four-level evaluation method.

[0149] Table 8. Durability Grade Classification Standards Furthermore, the durability assessment conclusions not only include the standard grade classification, but also, in conjunction with the global distribution characteristics of the flow matching field, propose targeted engineering protection and reinforcement suggestions, specifically: (1) If the peak depth of the flow matching field is shallow (3~5mm) and the peak concentration is high, it indicates that chloride salt erosion is concentrated on the concrete surface. It is recommended to give priority to the use of light protective measures such as coating the surface with an anti-chloride salt coating to block the continued penetration of external chloride salt. (2) If the peak depth of the flow matching field is relatively deep (8~10mm) and the average concentration in the stable zone is relatively high, it indicates that chloride salts have penetrated into the deep layer of concrete, close to the thickness of the steel reinforcement protective layer. It is recommended to adopt a moderate protection measure of surface repair + penetrating rust inhibitor to inhibit steel reinforcement corrosion. (3) If the peak depth of the flow matching field exceeds 10 mm and the gradient characteristics show that the deep concentration is still increasing, it indicates that chloride erosion has penetrated into the concrete and the durability level is level four (inferior). It is recommended to immediately adopt heavy reinforcement measures such as replacing the deteriorated concrete, installing rebar, and attaching steel plates. At the same time, the flow matching field of the component should be retested regularly to monitor the chloride erosion trend.

[0150] It should be noted that the parts in this embodiment that are the same as or similar to those in Embodiment 1 can be referred to each other, and will not be repeated in this application. Example 3:

[0151] Based on Example 2, Example 3 of this application provides a bridge durability testing system, such as... Figure 2 As shown, it includes: The preparation module is used to prepare standard concrete test specimens, collect chloride ion concentration data at multiple discrete depths of concrete specimens with different mix proportions, different chloride salt erosion concentrations, and different service ages as discrete observation data, and collect chloride ion concentration data at continuous depths within the corresponding depth range as complete label data. The construction module is used to build and train a diffusion flow model based on condition-flow matching based on the discrete observation data and the complete label data. The model takes the physical parameters of concrete and the observed chloride ion concentration at discrete depths as input conditions, and the chloride ion concentration distribution field within a continuous depth range as the output target, and learns the transformation law of chloride salt erosion flow from the initial distribution to the target concentration distribution field. The acquisition module is used to acquire chloride ion concentration detection data at multiple discrete depths of representative engineering parts of the concrete of the bridge under test, and to collect the corresponding physical parameters. The input module is used to input the detection data and physical parameters acquired by the acquisition module into the trained diffusion flow model. The model generates a chloride ion concentration diffusion flow matching field for the engineering part under test within a continuous depth range. The extraction module is used to extract the parameters required for bridge durability assessment from the chloride ion concentration diffusion flow matching field, and substitute them into the preset standard calculation formula to obtain the bridge durability assessment result.

[0152] It should be noted that the system provided in this embodiment corresponds to the system of the method provided in Embodiment 2. Therefore, in this embodiment... The parts of the embodiments that are the same as or similar to those of Embodiment 2 can be referred to each other, and will not be repeated in this application.

Claims

1. A method for testing the durability of bridges, characterized in that, include: S1. Prepare standard concrete test specimens and collect chloride ion concentration data; S2. Based on the chloride ion concentration data, construct and train a diffusion flow model based on condition-flow matching; the model is used to learn the transformation law of chloride erosion flow from the initial distribution to the target concentration distribution field; S3. Obtain chloride ion concentration detection data at multiple discrete depths for representative engineering parts of the concrete of the bridge under test, and collect the corresponding physical parameters; S4. Input the detection data and physical parameters obtained in S3 into the trained diffusion flow model. The model generates a chloride ion concentration diffusion flow matching field for the engineering part under test within a continuous depth range. S5. Extract the parameters required for bridge durability assessment from the chloride ion concentration diffusion flow matching field, and substitute them into the preset standard calculation formula to obtain the bridge durability assessment result.

2. The bridge durability testing method according to claim 1, characterized in that, In S1, chloride ion concentration data of concrete specimens with different mix proportions, different chloride salt erosion concentrations, and different service ages are collected at multiple discrete depths as discrete observation data, and chloride ion concentration data at continuous depths within the corresponding depth range are collected as complete label data.

3. The bridge durability testing method according to claim 2, characterized in that, In S1, the multiple discrete depths include multiple different depth points within the depth range of the diffusion zone; the continuous depth range is the complete diffusion interval from the concrete surface to the interior where the chloride ion concentration tends to stabilize.

4. The bridge durability testing method according to claim 3, characterized in that, In S2, the physical parameters include the proportion of fly ash, the proportion of slag, the service life, and the chloride erosion concentration; the diffusion flow model includes a flow transformation module, which is used to learn the velocity field under the constraints of physical parameters and discrete observations, driving the initial Gaussian distribution to gradually evolve into the target concentration diffusion distribution field.

5. The bridge durability testing method according to claim 4, characterized in that, In S2, the training process of the model incorporates physical constraints, including hard constraints, soft constraints, and peak enhancement constraints. The hard constraints include: ensuring non-negative concentration through the activation function of the model output layer, and introducing the influence coefficient of admixtures into the flow transformation velocity field to make the chloride ion diffusion coefficient negatively correlated with the proportion of fly ash and slag. The soft constraints include: guiding the gradient of the concentration distribution field output by the model through the loss function to approach zero in the depth range of the stable region, and showing a trend of first increasing and then decreasing in the depth range of the diffusion region; The peak enhancement constraint includes: assigning higher weight to the prediction error of the peak region depth range in the reconstruction loss, thereby improving the fitting accuracy of peak concentration and peak depth.

6. The bridge durability testing method according to claim 5, characterized in that, In S2, the total loss function of the model is a weighted sum of the conditional flow matching basic loss, peak weighted reconstruction loss, gradient constraint loss, and trend constraint loss.

7. The bridge durability testing method according to claim 6, characterized in that, The parameters extracted from the flow-matched field in S5 include at least: the maximum chloride ion concentration and its corresponding depth; the chloride ion concentration at any depth; the average chloride ion concentration within the depth range of the stable region; and the gradient characteristics of the flow-matched field.

8. A bridge durability testing system, characterized in that, For performing the method according to any one of claims 1 to 7, comprising: The preparation module is used to prepare standard concrete test specimens and collect chloride ion concentration data. A construction module is used to construct and train a diffusion flow model based on condition-flow matching based on the chloride ion concentration data; the model is used to learn the transformation law of chloride erosion flow from the initial distribution to the target concentration distribution field; The acquisition module is used to acquire chloride ion concentration detection data at multiple discrete depths of representative engineering parts of the concrete of the bridge under test, and to collect the corresponding physical parameters. The input module is used to input the detection data and physical parameters acquired by the acquisition module into the trained diffusion flow model. The model generates a chloride ion concentration diffusion flow matching field for the engineering part under test within a continuous depth range. The extraction module is used to extract the parameters required for bridge durability assessment from the chloride ion concentration diffusion flow matching field, and substitute them into the preset standard calculation formula to obtain the bridge durability assessment result.

9. A computer storage medium, characterized in that, The computer storage medium stores a computer program; when the computer program is run on the computer, it causes the computer to perform the method described in any one of claims 1 to 7.

10. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the method as described in any one of claims 1 to 7.

Images