A method, system, device and medium for damage assessment of reinforced concrete members

Abstract

The application provides a reinforced concrete component damage evaluation method, system, device and medium, and belongs to the technical field of structural damage evaluation, and comprises the following steps: constructing a fire response characteristic knowledge base containing reinforced concrete component fire physical test data and simulation data; taking component design parameters and fire working condition parameters of the reinforced concrete component in the knowledge base as input parameters, taking post-fire component response data as output parameters, training a machine learning agent model, and obtaining a prediction model for predicting direct physical damage indexes of the reinforced concrete component after the fire; constructing a fire vulnerability curve representing the relationship between failure probability and fire intensity based on the prediction model; receiving parameters of a component to be evaluated, and generating and outputting damage evaluation results of the component based on the prediction model and the fire vulnerability curve. Through data and model collaborative optimization, efficient and accurate post-fire reinforced concrete component damage evaluation is realized.

Description

Technical Field

[0001] This invention belongs to the field of structural damage assessment technology, specifically relating to a method, system, equipment, and medium for assessing damage to reinforced concrete components. Background Technology

[0002] Fire is a major hazard threatening the safety of urban buildings. Reinforced concrete structures account for more than 65% of the global building stock, and their post-fire safety assessment has become a core issue in the international disaster prevention field. Fire can cause a series of damages to reinforced concrete components, including material degradation, internal crack development, and reduced load-bearing capacity. Accurately assessing the extent of damage to components is a crucial prerequisite for subsequent repair and reinforcement, safety rating, and decision-making.

[0003] Currently, damage assessment of reinforced concrete beams after a fire still relies primarily on traditional methods, which mainly include three categories: on-site visual inspection, material sampling tests, and load tests. On-site visual inspection makes a preliminary judgment of damage by observing appearance features such as spalling, cracks, and color changes on the surface of the component; material sampling tests require taking material samples from the damaged component and determining key performance parameters such as concrete strength and steel yield strength through laboratory tests; load tests assess the load-bearing capacity and damage state by applying loads to the component and observing its deformation, cracking, and other responses.

[0004] These traditional methods have several insurmountable drawbacks: First, they are inefficient, with cumbersome on-site inspection and testing procedures that often take days or even weeks from preparation and implementation to result analysis, failing to meet the needs of rapid response and emergency decision-making after a fire. Second, they are highly subjective, with visual inspection results heavily reliant on the experience and expertise of the assessors, leading to potential biases in judgment and making it difficult to guarantee the objectivity and consistency of the assessment. Third, they are highly destructive, as material sampling and load testing can cause secondary damage to already damaged structures, even affecting the remaining load-bearing capacity, and the transportation and setup costs of testing equipment are high, making them unsuitable for large-scale assessments of existing buildings. Fourth, they have limitations in assessment, as traditional methods often focus on surface damage and local performance of components, failing to penetrate to the interior and thus unable to comprehensively reflect the impact of fire on key indicators such as the overall mechanical properties and internal crack distribution of components, resulting in insufficient comprehensiveness and accuracy of the assessment results.

[0005] To overcome the shortcomings of traditional methods, the industry is gradually exploring non-contact, non-destructive assessment technologies to achieve efficient, non-destructive, and comprehensive damage assessment. Currently available technologies fall into two categories: one is analysis methods based on multi-attribute decision-making, such as the grey relational analysis-modified approximation of the ideal solution (MTOPSIS-GRA), which integrates multiple damage-related indicators to assess post-fire damage to concrete T-beams; the other is non-destructive testing technologies based on physical detection, typically including infrared thermography, electrochemical diagnostics, and ultrasonic diagnostics. Infrared thermography identifies internal defects by detecting differences in surface temperature distribution within components; electrochemical diagnostics uses the correlation between steel corrosion and electrochemical signals to determine the degree of damage; and ultrasonic diagnostics analyzes damage such as cracks and porosity based on changes in the propagation characteristics of ultrasonic waves within components.

[0006] However, these non-contact, non-destructive methods have limited applicability. For example, the MTOPSIS-GRA method is currently mostly applicable to specific types of components (such as T-beams) and lacks versatility. Infrared thermography, ultrasonic technology, and other technologies are easily affected by factors such as ambient temperature and humidity and the surface condition of components, and have poor adaptability in complex fire scenes. At the same time, the accuracy of the assessment needs to be optimized. Some methods can only achieve a qualitative judgment of the damage level and are difficult to quantify the residual performance and failure risk of components, thus failing to provide sufficient support for accurate repair decisions. Summary of the Invention

[0007] To address the aforementioned problems, this invention provides a method for assessing damage to reinforced concrete components.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] A method for assessing damage to reinforced concrete members, comprising:

[0010] Construct a fire response feature knowledge base that includes component design parameters, fire condition parameters, and post-fire damage response data for reinforced concrete components;

[0011] Using the component design parameters and fire condition parameters of reinforced concrete components as input parameters and the damage response data after fire as output parameters, a machine learning proxy model is trained to obtain a predictive model for predicting the direct physical damage index of reinforced concrete components after fire.

[0012] Key uncertainty parameters in the fire response feature knowledge base are identified, and probability distribution models for each parameter are established. Latin hypercube sampling is used to generate multiple random sample combinations within the probability distribution interval of each parameter. These sample combinations are then input into the prediction model for batch prediction. The prediction results are combined with preset multi-level damage limit state criteria to statistically determine the failure probability of components exceeding each level of damage state under different fire intensities. Fire intensity and failure probability are correlated to construct a fire vulnerability curve. The key uncertainty parameters refer to input parameters that, in the post-fire damage assessment of reinforced concrete components, have a greater impact on the fire response than a preset value and exhibit inherent fluctuations due to objective factors, making them impossible to completely fix.

[0013] The system receives the design parameters and fire intensity of the reinforced concrete component to be evaluated after the fire, calls the prediction model to output the direct physical damage index, and queries the vulnerability curve under the corresponding fire intensity to obtain the failure probability of each damage level.

[0014] Preferably, the fire response feature knowledge base includes a fire physical test dataset of reinforced concrete components and a fire simulation dataset of reinforced concrete components;

[0015] The construction of the physical test dataset for fire-exposed reinforced concrete components includes: collecting publicly available information on the pre-fire parameters of reinforced concrete components and their response data after fire exposure; standardizing and cleaning the pre-fire parameters and response data after fire exposure; extracting the initial geometric design parameters, fire condition parameters, and corresponding post-fire damage response data of the components before fire exposure; and forming the physical test dataset.

[0016] The construction of a fire simulation dataset for reinforced concrete components includes: establishing a refined finite element model of the reinforced concrete components calibrated with physical test data; performing numerical simulation calculations by parametrically combining and analyzing the input parameters of the model to generate a simulation dataset covering various working conditions; the input parameters include the initial geometric design parameters and fire condition parameters of the reinforced concrete components.

[0017] Preferably, the initial geometric design parameters include the cross-sectional dimensions of the reinforced concrete member, material property parameters, reinforcement parameters, concrete strength, and protective layer thickness.

[0018] Preferably, the construction of the fire vulnerability curve specifically includes the following steps:

[0019] Identify uncertain parameters that have an impact greater than a threshold on the post-fire damage response from the initial geometric design parameters of the components and the fire condition parameters;

[0020] Based on relevant design specifications and engineering statistics, the optimal probability distribution model is matched for each uncertainty parameter, and the key parameters of each probability distribution are identified.

[0021] Based on the Latin hypercube sampling method, the sampling quantity is determined, and the probability distribution interval of each uncertainty parameter is divided into non-overlapping sub-intervals with equal probability according to the sampling quantity. One sample point is randomly selected in each sub-interval of each parameter, and then the sample points of all parameters are combined in order to form multiple complete sets of input random parameter sample combinations.

[0022] All generated random parameter samples are combined and input into the prediction model, and the direct physical damage index corresponding to each group of samples is output in batches.

[0023] Based on engineering specifications and practical experience, the criteria for determining multi-level damage are clarified. The direct physical damage index of each sample group is compared with the limit state criteria of each level to determine the damage level corresponding to the sample. With the key fire intensity index as the horizontal axis, a continuous gradient interval is set. For each fire intensity gradient, the number of samples that exceed the limit state of each level of damage is counted and divided by the total number of samples to obtain the failure probability of the corresponding damage level.

[0024] By plotting fire intensity on the horizontal axis and failure probability on the vertical axis, curve fitting is performed on the failure probability data of each damage level to form a fire vulnerability curve corresponding to each damage level.

[0025] Preferably, the direct physical damage indicators include one or more of the following: residual deflection, residual bearing capacity, residual stiffness, and concrete spalling depth of the component.

[0026] Preferably, the machine learning agent model is the Gradient Boosting Decision Tree (GBDT) algorithm; the training of the machine learning agent model specifically includes:

[0027] The fire response feature knowledge base is divided into a training set, a validation set, and a test set. The proxy model is trained using the training set, and the key hyperparameters of the model are optimized using grid search and 5-fold cross-validation until the model converges and performs optimally on the validation set.

[0028] Use the test set to evaluate the prediction accuracy and generalization ability of the trained surrogate model.

[0029] Preferably, the method further includes outputting the direct physical damage index output by the prediction model and the failure probability of each damage level obtained by query in a visual form.

[0030] This invention also proposes a damage assessment system for reinforced concrete components, comprising:

[0031] The knowledge base construction module is used to build a fire response feature knowledge base containing component design parameters, fire condition parameters, and post-fire damage response data of reinforced concrete components.

[0032] The model training module is used to train the machine learning proxy model with the component design parameters and fire condition parameters of the reinforced concrete component as input parameters and the damage response data after fire as output parameters, so as to obtain a predictive model for predicting the direct physical damage index of the reinforced concrete component after fire.

[0033] The evaluation curve construction module is used to identify key uncertainty parameters in the fire response characteristic knowledge base and establish probability distribution models for each parameter. Latin hypercube sampling is used to generate multiple random sample combinations within the probability distribution interval of each parameter. These sample combinations are then input into the prediction model for batch prediction. The prediction results are combined with preset multi-level damage limit state criteria to statistically determine the failure probability of components exceeding each level of damage state under different fire intensities. Fire intensity and failure probability are correlated to construct a fire vulnerability curve. The key uncertainty parameters refer to input parameters that, in the post-fire damage assessment of reinforced concrete components, have a greater impact on the fire response than a preset value and exhibit inherent fluctuations due to objective factors, making them impossible to completely fix.

[0034] The evaluation module is used to receive the design parameters and fire intensity of the reinforced concrete component to be evaluated after a fire, call the prediction model to output the direct physical damage index, and query the vulnerability curve under the corresponding fire intensity to obtain the failure probability of each damage level.

[0035] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of any one of the methods for assessing damage to reinforced concrete components.

[0036] The present invention also provides a computer-readable storage medium storing a computer program that, when loaded by a processor, can execute any of the steps in the method for assessing damage to reinforced concrete components.

[0037] The method for assessing damage to reinforced concrete components provided by this invention has the following beneficial effects:

[0038] This invention first constructs a feature knowledge base integrating component design, fire conditions, and damage response data, covering various component types and conditions. This overcomes the specific applicability limitations of the MTOPSIS-GRA method, enhancing its versatility. A machine learning surrogate model is trained using this knowledge base; inputting standard design parameters allows for the prediction of direct physical damage indicators, avoiding the dependence of technologies like infrared thermography on the environment and component surface conditions, thus enhancing adaptability to complex scenarios. Simultaneously, by identifying key uncertainty parameters and establishing a probability distribution model, combined with Latin hypercube sampling to generate sample combinations, and after batch prediction by the surrogate model, the failure probability is statistically analyzed to construct a fire vulnerability curve. This upgrades qualitative assessment to quantitative analysis, accurately outputting direct physical damage indicators and failure probabilities at each level, solving the pain point of traditional non-contact methods' difficulty in quantifying residual performance and failure risk. Finally, by inputting only the design parameters of the component to be evaluated and the fire intensity, comprehensive quantitative assessment results can be quickly obtained, maintaining the advantages of non-contact and non-destructive methods while achieving universal adaptability and precise quantification, providing reliable support for repair decisions. Attached Figure Description

[0039] To more clearly illustrate the embodiments and design schemes of the present invention, the accompanying drawings required for this embodiment will be briefly described below. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a flowchart of a damage assessment method for reinforced concrete components according to Embodiment 1 of the present invention;

[0041] Figure 2 A flowchart illustrating the phased process for damage assessment of reinforced concrete components provided in an embodiment of the present invention;

[0042] Figure 3 Example graphs of a set of fire vulnerability curves for different damage levels generated by the method of the present invention;

[0043] Figure 4 This is a schematic diagram of a visual evaluation result interface of the user interface module of the system of the present invention. Detailed Implementation

[0044] To enable those skilled in the art to better understand and implement the technical solutions of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be construed as limiting the scope of protection of the present invention.

[0045] Current methods for post-fire safety assessment of reinforced concrete structures lack an intelligent approach that can directly utilize readily available component design parameters (such as geometric dimensions, material strength, and reinforcement information), specifically tailored to post-fire assessment scenarios, and output direct, quantitative, and scientifically reflective assessment results reflecting the impact of uncertainty. In particular, existing technologies have failed to provide advanced tools like fire vulnerability curves, which can comprehensively assess component failure risk in a probabilistic manner, leaving post-disaster assessment decisions lacking sufficiently scientific and efficient support.

[0046] Therefore, the purpose of this invention is to provide a rapid assessment method and system for post-fire damage of reinforced concrete components based on machine learning and vulnerability analysis. This addresses the limitations of existing assessment methods, such as long processing times, high subjectivity, and significant destructiveness, as well as the limitations of existing intelligent technologies that either focus on real-time disaster warnings and rely on pre-installed sensors or on dynamic modal parameters that are difficult to obtain post-disaster. This invention focuses on post-disaster scenarios, proposing a method that uses only the component's conventional design parameters as input, rapidly predicts direct damage indicators through a machine learning model, and scientifically quantifies the component's safety risk in the form of failure probability using vulnerability analysis. This method requires no pre-installed sensors or professional dynamic testing, has wider applicability and stronger engineering practicality, and aims to provide rapid, objective, and scientific intelligent support for the safety rating and repair decisions of buildings after a fire.

[0047] Example 1

[0048] This invention provides a method for assessing damage to reinforced concrete components. This embodiment uses a reinforced concrete beam as an example, specifically a method and system for assessing post-fire damage to a simply supported reinforced concrete beam based on machine learning and vulnerability analysis. (Refer to...) Figure 1 and Figure 2 The specific implementation process of this method is as follows:

[0049] Step S1 (i.e.) Figure 2 Phase 1): Constructing a fire response feature knowledge base that integrates multi-source data.

[0050] The goal of this step is to prepare a high-quality dataset for the subsequent training of machine learning models. This requires integrating two types of data: physical experiments and numerical simulations. Specifically, it consists of three sub-steps:

[0051] S110: Collect and clean physical test data: Systematically collect and organize publicly available fire-exposed physical test data of reinforced concrete components from both domestic and international sources. This data specifically includes the component's parameters before exposure to fire and its response data after exposure. The data is standardized and cleaned to extract the initial geometric design parameters, material property parameters, reinforcement parameters, fire condition parameters, and corresponding post-fire damage response data of the components before exposure to fire, forming a physical test dataset.

[0052] Specifically, this invention, through literature review, systematically collected detailed data from 150 publicly published fire resistance tests on reinforced concrete beams both domestically and internationally. For each set of physical test data, this invention extracted and recorded the following key information:

[0053] Geometric design parameters: beam span, cross-sectional width, and cross-sectional height.

[0054] Material property parameters: design strength grade of concrete (e.g., C30, C40), grade of steel reinforcement (e.g., HRB400) and yield strength.

[0055] Reinforcement parameters: reinforcement ratio of longitudinal tensile reinforcement, reinforcement ratio of longitudinal compressive reinforcement, stirrup spacing, and concrete cover thickness.

[0056] Fire condition parameters: load ratio during loading, type of temperature rise curve (such as ISO-834 standard temperature rise curve), fire exposure time, and fire intensity.

[0057] Post-fire damage response data (output labels): The residual deflection value at mid-span of the beam after the fire, as measured by the test.

[0058] The collected raw data was standardized and cleaned, units were standardized, and outliers were removed, resulting in a physical experiment dataset containing 150 samples.

[0059] S120: Generating Parametric Simulation Data: Establishing a refined finite element model of reinforced concrete components calibrated with physical test data; through systematic parametric combination and analysis of the model's input parameters (including but not limited to cross-sectional dimensions, reinforcement ratio, concrete strength, protective layer thickness, fire exposure time, etc.), large-scale numerical simulation calculations are performed to generate a simulation dataset covering a wide range of working conditions. Because training a surrogate simulation capable of rapidly calculating residual deflection at mid-span of beams requires a large amount of data, and the currently available physical test dataset is insufficient to begin training the surrogate model, the finite element simulation method is adopted. After calibrating the model, fire response analyses of beams with numerous different design parameters are conducted using finite element methods to expand the data sample and train the surrogate model.

[0060] Specifically, to compensate for the limitations of physical test data, this invention uses the finite element software MSC.Marc to establish a refined finite element model of the RC beam with fire-force coupling, calibrated using the aforementioned physical test data. Based on this model, this invention conducts large-scale parametric numerical simulation analysis. The range of design parameters covers common engineering scenarios, such as:

[0061] Section width: 200mm~400mm; Section height: 400mm~800mm; Concrete strength grade: C30~C60; Steel strength grade: HPB300, HRB335, HRB400, HRB500; Tensile reinforcement ratio: 0.8%~2.5%; Fire exposure time: 30min~180min.

[0062] By orthogonally combining these parameters, a total of 2,000 sets of simulated working conditions were generated. These 2,000 sets are finite element models to be calculated. After performing finite element analysis of the fire response of the components, the corresponding residual deflection value at mid-span after the fire was obtained for each working condition. The residual deflection value at mid-span was paired with the previous design parameters to form a simulation dataset containing 2,000 samples.

[0063] S130: Integrating and constructing a fire response feature knowledge base: Integrating the physical test dataset and the simulation dataset to obtain 2150 high-quality samples, constructing a structured fire response feature knowledge base with component design parameters and fire condition parameters as input features and direct physical damage indicators after fire (such as residual deflection and residual bearing capacity) as output labels, and storing it in a structured table format, with each row representing a sample, containing 10 input features and 1 output label.

[0064] Step S2 (i.e.) Figure 2 Phase 2): Training and validating the fire response machine learning agent model.

[0065] This step utilizes the established fire response feature knowledge base to train a fast prediction model that can replace time-consuming finite element calculations. It consists of three sub-steps:

[0066] S210: Constructing a proxy model: Based on the fire response feature knowledge base constructed in step 1, select at least one machine learning algorithm (such as deep neural network, support vector regression, gradient boosting decision tree GBDT, random forest, etc.) to construct a proxy model that can represent the complex nonlinear relationship between input features and output labels.

[0067] Specifically, this embodiment selects the Gradient Boosting Decision Tree (GBDT) algorithm as the surrogate model and implements it using the Python language and the Scikit-learn machine learning library.

[0068] S220: Training and Optimizing the Model: Divide the fire response feature knowledge base into a training set, a validation set, and a test set. Use the training set to train the surrogate model and use methods such as cross-validation to optimize the model's hyperparameters until the model converges and performs optimally on the validation set.

[0069] Specifically, in this embodiment, 2150 samples in the fire response feature knowledge base are randomly divided into a training set (1720 samples) and a test set (430 samples) in an 8:2 ratio. The GBDT model is trained using the training set, and the key hyperparameters of the model (such as learning rate, number of trees, maximum tree depth, etc.) are optimized using grid search and 5-fold cross-validation.

[0070] S230: Evaluate model performance: After training, use an independent test set to evaluate the prediction accuracy and generalization ability of the trained surrogate model. The core metrics include the coefficient of determination (R²) and the mean absolute percentage error (MAPE) to ensure that it can accurately and quickly predict the direct physical damage indicators of the component.

[0071] The evaluation results show that the model has a determination coefficient (R²) of 0.96 on the test set and a mean absolute percentage error (MAPE) of less than 8%, which proves that the surrogate model has high prediction accuracy and good generalization ability. It can quickly (less than 0.01 seconds for a single prediction) and accurately predict the residual deflection of RC beams after a fire, meeting the requirements for fast and accurate evaluation.

[0072] Step S3 (i.e.) Figure 2 Phase 3): Constructing a fire vulnerability model based on a proxy model.

[0073] This step aims to achieve the probabilistic quantification of component failure risk, realizing the transformation from deterministic prediction to probabilistic risk assessment, and is specifically divided into 4 sub-steps:

[0074] S310: Uncertainty Modeling: Identify and model random variables for probabilistic analysis.

[0075] The material properties of components (such as concrete strength and steel yield strength), geometric dimensions (such as protective layer thickness), and fire conditions (such as actual fire temperature) are affected by factors such as production processes, construction errors, and the randomness of fire scenarios. These parameters cannot perfectly match the design values ​​and inevitably contain uncertainties. If these uncertainties are ignored and damage indicators are calculated using only fixed design values, the assessment results will deviate from reality and fail to reflect the differences in failure risk of components under different parameter combinations (e.g., some extreme parameter combinations may significantly increase the probability of component failure). Therefore, this invention identifies key uncertainty parameters in a fire response characteristic knowledge base and establishes probability distribution models for each parameter. These key uncertainty parameters refer to input parameters that have a greater impact on the fire response of reinforced concrete components than a preset value in post-fire damage assessment and exhibit inherent fluctuations due to objective factors, making them impossible to completely fix. By identifying key influencing parameters (such as concrete strength and protective layer thickness, which are sensitive to fire response) and establishing their probability distribution models (such as normal distribution and log-normal distribution) based on standards and statistical data, the fluctuation patterns of the parameters can be accurately characterized, providing a mathematical basis for subsequent risk calculations.

[0076] Specifically, in post-fire assessment, uncertainty mainly stems from three aspects: uncertainty in material properties, uncertainty in geometric dimensions, and uncertainty in the fire's effects (loads). Based on engineering consensus, this embodiment selects concrete compressive strength, steel yield strength, and concrete cover thickness—which significantly affect the fire resistance of components—as random variables representing the inherent uncertainty of the structure, and establishes a probability distribution model for them. Considering the complexity of the fire scene, the estimation of the equivalent fire intensity also involves uncertainty. Therefore, this invention allows the equivalent fire exposure time to also be set as a random variable. Its probability distribution parameters can be determined based on damage traces at the scene, eyewitness accounts, or fire records. In future implementations, fire condition parameters can also be more complex combinations of parameters that better reflect fire uncertainty, such as fire load density, ventilation factor, and temperature rise curve type. As long as these parameters are used as input features for training the surrogate model, the method of this invention is equally applicable.

[0077] Specifically, the random variables identified in this invention include concrete compressive strength, steel bar yield strength, concrete cover thickness, and fire exposure time. These parameters are the main uncertainties affecting the structural fire response, and probability distribution models have been established for each parameter based on relevant design codes and statistical literature. For example, concrete strength follows a log-normal distribution, and cover thickness follows a normal distribution.

[0078] S320: Efficient Sampling: Employs efficient sampling methods such as Latin hypercube sampling to generate a large number of random parameter sample combinations within the probability space of the parameters.

[0079] Latin hypercube sampling divides the probability distribution interval of each parameter with equal probability, extracting only one sample from each sub-interval, and then combining them to form a sample set. This allows for comprehensive coverage of the fluctuation range of all parameters with a smaller sample size, avoiding the problems of sample concentration and incomplete coverage that may occur with traditional random sampling. This invention uses the Latin hypercube sampling method to generate 10,000 sets of random parameter sample combinations within the probability space of the aforementioned key uncertainty input parameters, ensuring that the samples cover the range of parameter uncertainty. Simultaneously, Latin hypercube sampling significantly reduces computational load while ensuring sample representativeness through efficient sampling, and combined with machine learning surrogate models, it can achieve batch prediction within seconds.

[0080] S330: Rapid simulation and statistics: Combine a large number of random parameter samples and input them into the surrogate model trained in step S2 for batch and rapid prediction calculation; based on one or more preset performance limit states (e.g., minor damage, moderate damage, severe damage, collapse), statistically calculate the failure probability of the component reaching or exceeding each level of limit state under different fire intensities (e.g., equivalent fire exposure time).

[0081] This invention combines these 10,000 sets of random samples and inputs them into the GBDT surrogate model trained in step S2 to perform batch prediction of the mid-span residual deflection of the beam. Due to the extremely high computational efficiency of the surrogate model, the entire process can be completed within seconds. Simultaneously, based on engineering practice and relevant standards, this invention defines performance limit states for four damage levels, with the criterion being the mid-span residual deflection (…). d ) and beam span ( L The ratio of )

[0082] Minor damage (DS1): d / L >1 / 200; Moderate damage (DS2): d / L >1 / 100; Severely damaged (DS3): d / L >1 / 75; Collapse (DS4): d / L >1 / 50.

[0083] The failure probability (number of failure samples / total number of samples) of components exceeding each level of limit state is statistically analyzed under different fire intensities (such as fire exposure time).

[0084] Specifically, for each fire intensity (for example, taking the fire exposure time as a variable, from 30 min to 180 min, taking a point every 15 min), the present invention counts the number of samples in 10,000 samples whose predicted residual deflection exceeds the above-mentioned limit states, and divides it by the total number of samples 10,000 to obtain the failure probability of the fire intensity reaching or exceeding each damage level.

[0085] S340 (i.e.) Figure 2 Stage 4): Constructing the fragility curve. For example... Figure 3 As shown, the calculated series of failure probabilities are correlated with the corresponding fire intensity to construct a set of fire vulnerability curves, which intuitively reflect the evolution of component failure probability with fire intensity. For example, for the same damage level, the longer the fire exposure time, the higher the probability of reaching severe damage. At the same time (1) in the figure, querying yields the failure probability of the same component reaching each damage level, for example: P(DS1)=99%, P(DS2)=85%, P(DS3)=55%, P(DS4)=20%. This clearly demonstrates how the probability of the beam reaching each damage level increases with the increase of fire exposure time.

[0086] Step S4: Generate and visualize the post-fire damage assessment results.

[0087] This step demonstrates the final application of the invention's results, providing intuitive decision support for engineering applications. It consists of three sub-steps:

[0088] S410: Receive Assessment Request: Receives the deterministic design parameters of the reinforced concrete member to be assessed after a fire and the assessment fire intensity from the user, and outputs the direct physical damage index.

[0089] Specifically, the deterministic design parameters of the component to be evaluated (such as cross-section 300×600mm, concrete C40) and the fire intensity (such as fire exposure for 90 minutes) are received through the user interface module.

[0090] S420: Output deterministic and probabilistic results: Simultaneously call the surrogate model of step S2 and the vulnerability model of step S3 to calculate and generate evaluation results, which include at least: (i) the deterministic value of the direct physical damage index predicted by the surrogate model (such as the predicted residual deflection of 65mm in the document); (ii) the failure probability of the component reaching each level of damage state under the current fire intensity calculated by the vulnerability model.

[0091] S430: Visual presentation: The deterministic values ​​of direct physical damage indicators and the failure probability of each level of damage are presented to users in a graphical user interface (such as dashboards, damage cloud maps, vulnerability curves, etc.), which makes it convenient for engineering technicians to intuitively obtain the degree of damage and risk level, and make engineering decisions such as auxiliary repair, reinforcement or demolition.

[0092] Compared with the prior art, the present invention has the following significant advantages:

[0093] (1) Fast, efficient, objective and accurate: By replacing complex finite element calculations or time-consuming physical experiments with a well-trained machine learning proxy model, the evaluation time is shortened from several days or weeks to seconds. The evaluation process is data-driven, avoiding the subjectivity of manual evaluation, and the results are more objective and reproducible.

[0094] (2) Convenient input and wide applicability: This invention only requires input of the conventional design parameters of the components that can be obtained from the design drawings or through simple on-site measurement, completely eliminating the dependence on pre-set real-time sensor networks and professional power testing equipment, making it widely applicable to post-fire assessment scenarios of various existing buildings and having strong engineering universality.

[0095] (3) Scientific Quantification and Visible Risk: The innovative introduction of vulnerability analysis elevates the assessment of component damage after a fire from a deterministic "point" prediction to a probabilistic "surface" assessment that considers uncertainty. The failure probability and vulnerability curves output by the system can scientifically quantify the failure risk of components, providing an unprecedented scientific basis for risk-based performance evaluation and decision-making.

[0096] (4) Convenient application and decision support: By encapsulating complex models within a user-friendly visual interface, engineers can quickly obtain intuitive and comprehensive assessment reports without delving into algorithmic details. The report includes both deterministic damage levels and probabilistic risk levels, providing strong intelligent support for major engineering decisions such as the repair, reinforcement, or demolition of buildings after a fire.

[0097] Based on the same inventive concept, this invention also proposes a damage assessment system for reinforced concrete components, including a knowledge base construction module, a model training module, an assessment curve construction module, and an assessment module.

[0098] Specifically, the knowledge base construction module is used to build a fire response feature knowledge base containing component design parameters, fire condition parameters, and post-fire damage response data of reinforced concrete components.

[0099] The model training module is used to train a machine learning proxy model with the component design parameters and fire condition parameters of reinforced concrete components as input parameters and the damage response data after fire as output parameters, so as to obtain a predictive model for predicting the direct physical damage index of reinforced concrete components after fire.

[0100] The evaluation curve construction module is used to identify key uncertainty parameters in the fire response characteristic knowledge base and establish probability distribution models for each parameter. Latin hypercube sampling is used to generate multiple random sample combinations within the probability distribution interval of each parameter. These sample combinations are then input into the prediction model for batch prediction. The prediction results are combined with preset multi-level damage limit state criteria to statistically determine the failure probability of components exceeding each level of damage state under different fire intensities. Fire intensity and failure probability are correlated to construct a fire vulnerability curve. Key uncertainty parameters refer to input parameters that, in the post-fire damage assessment of reinforced concrete components, have a greater impact on the fire response than preset values ​​and exhibit inherent fluctuations due to objective factors, making them impossible to completely fix.

[0101] The assessment module receives the design parameters and fire intensity of the reinforced concrete component to be assessed after a fire, calls the prediction model to output direct physical damage indicators, queries the vulnerability curve under the corresponding fire intensity, and obtains the failure probability of each damage level.

[0102] Finally, this invention constructs an operating system for a fire damage assessment system for reinforced concrete components. The specific interface of the operating system is as follows: Figure 4 As shown. For example, when engineers need to evaluate a RC beam after a fire, they first enter the deterministic design parameters and fire exposure parameters of the beam in the parameter input window on the left side of the interface. Specific parameters include component name, component type, component height or span, section height, concrete strength grade, reinforcement ratio, protective layer thickness, and fire exposure time, but do not include vibration mode parameters such as frequency and mode shape. For example: span 6000mm, section 300×600mm, concrete C40, protective layer thickness 25mm, and fire exposure time of 90min.

[0103] After clicking the "Start Prediction" button, the predicted residual disturbance value and risk level assessment results will be displayed in the evaluation results window on the right. At the same time, they will be presented to the user in a graphical user interface (such as a dashboard and vulnerability curve) and a list of specific damage level failure probabilities will be displayed.

[0104] The various modules in the aforementioned reinforced concrete component damage assessment system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can call and execute the corresponding operations of each module.

[0105] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the steps in an embodiment of a method for assessing damage to reinforced concrete components. Specific implementation methods can be found in the method embodiments, and will not be repeated here.

[0106] Furthermore, the present invention also provides a non-transitory computer-readable storage medium containing instructions on which a computer program is stored. For example, a memory containing instructions that can be executed by a processor of a computer device to perform the above-described method. For example, the non-transitory computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc. When the computer program is executed by the processor, it can implement the steps in an embodiment of a method for assessing damage to reinforced concrete components. Specific implementation methods can be found in the method embodiments, which will not be repeated here.

[0107] Those skilled in the art will understand that embodiments of the present invention can provide methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0108] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0109] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0110] These computer program instructions can also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0111] It should be noted that the specific embodiments described above enable those skilled in the art to more fully understand the present invention, but do not limit the present invention in any way. Therefore, although the present invention has been described in detail in this specification and embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention; and all technical solutions and improvements that do not depart from the spirit and scope of the present invention are covered within the protection scope of the present invention patent. No reference numerals in the claims should be construed as limiting the scope of the claims. Any simple variations or equivalent substitutions of technical solutions that can be readily obtained by those skilled in the art within the scope of the technology disclosed in the present invention are within the protection scope of the present invention.

Claims

1. A method for assessing damage to reinforced concrete members, characterized in that, include: Construct a fire response feature knowledge base that includes component design parameters, fire condition parameters, and post-fire damage response data for reinforced concrete components; Using the component design parameters and fire condition parameters of reinforced concrete components as input parameters and the damage response data after fire as output parameters, a machine learning proxy model is trained to obtain a predictive model for predicting the direct physical damage index of reinforced concrete components after fire. Key uncertainty parameters in the fire response feature knowledge base are identified, and probability distribution models for each parameter are established. Latin hypercube sampling is used to generate multiple random sample combinations within the probability distribution interval of each parameter. These sample combinations are then input into the prediction model for batch prediction. The prediction results are combined with preset multi-level damage limit state criteria to statistically determine the failure probability of components exceeding each level of damage state under different fire intensities. Fire intensity and failure probability are correlated to construct a fire vulnerability curve. The key uncertainty parameters refer to input parameters that, in the post-fire damage assessment of reinforced concrete components, have a greater impact on the fire response than a preset value and exhibit inherent fluctuations due to objective factors, making them impossible to completely fix. The system receives the design parameters and fire intensity of the reinforced concrete component to be evaluated after the fire, calls the prediction model to output the direct physical damage index, and queries the vulnerability curve under the corresponding fire intensity to obtain the failure probability of each damage level. The construction of the fire vulnerability curve specifically includes the following steps: Identify uncertain parameters that have an impact greater than a threshold on the post-fire damage response from the initial geometric design parameters of the components and the fire condition parameters; Based on relevant design specifications and engineering statistics, the optimal probability distribution model is matched for each uncertainty parameter, and the key parameters of each probability distribution are identified. Based on the Latin hypercube sampling method, the sampling quantity is determined, and the probability distribution interval of each uncertainty parameter is divided into non-overlapping sub-intervals with equal probability according to the sampling quantity. One sample point is randomly selected in each sub-interval of each parameter, and then the sample points of all parameters are combined in order to form multiple complete sets of input random parameter sample combinations. All generated random parameter samples are combined and input into the prediction model, and the direct physical damage index corresponding to each group of samples is output in batches. Based on engineering specifications and practical experience, the criteria for determining multi-level damage are clarified. The direct physical damage index of each sample group is compared with the limit state criteria of each level to determine the damage level corresponding to the sample. With the key fire intensity index as the horizontal axis, a continuous gradient interval is set. For each fire intensity gradient, the number of samples that exceed the limit state of each level of damage is counted and divided by the total number of samples to obtain the failure probability of the corresponding damage level. By plotting fire intensity on the horizontal axis and failure probability on the vertical axis, curve fitting is performed on the failure probability data of each damage level to form a fire vulnerability curve corresponding to each damage level.

2. The method for assessing damage to reinforced concrete components according to claim 1, characterized in that, The fire response feature knowledge base includes a physical test dataset of reinforced concrete components subjected to fire and a simulation dataset of reinforced concrete components subjected to fire. The construction of the physical test dataset for fire-exposed reinforced concrete components includes: collecting publicly available information on the pre-fire parameters of reinforced concrete components and their response data after fire exposure; standardizing and cleaning the pre-fire parameters and response data after fire exposure; extracting the initial geometric design parameters, fire condition parameters, and corresponding post-fire damage response data of the components before fire exposure; and forming the physical test dataset. The construction of a fire simulation dataset for reinforced concrete components includes: establishing a refined finite element model of the reinforced concrete components calibrated with physical test data; performing numerical simulation calculations by parametrically combining and analyzing the input parameters of the model to generate a simulation dataset covering various working conditions; the input parameters include the initial geometric design parameters and fire condition parameters of the reinforced concrete components.

3. The method for assessing damage to reinforced concrete components according to claim 2, characterized in that, The initial geometric design parameters include the cross-sectional dimensions, material properties, reinforcement parameters, concrete strength, and protective layer thickness of the reinforced concrete member.

4. The method for assessing damage to reinforced concrete components according to claim 1, characterized in that, The direct physical damage indicators include one or more of the following: residual deflection, residual bearing capacity, residual stiffness, and concrete spalling depth of the component.

5. The method for assessing damage to reinforced concrete components according to claim 1, characterized in that, The machine learning agent model is the Gradient Boosting Decision Tree (GBDT) algorithm; the training of the machine learning agent model specifically includes: The fire response feature knowledge base is divided into a training set, a validation set, and a test set. The proxy model is trained using the training set, and the key hyperparameters of the model are optimized using grid search and 5-fold cross-validation until the model converges and performs optimally on the validation set. Use the test set to evaluate the prediction accuracy and generalization ability of the trained surrogate model.

6. The method for assessing damage to reinforced concrete components according to claim 1, characterized in that, It also includes outputting the direct physical damage index output by the prediction model and the failure probability of each damage level obtained by query in a visual form.

7. An assessment system for implementing the damage assessment method for reinforced concrete members according to claim 1, characterized in that, include: The knowledge base construction module is used to build a fire response feature knowledge base containing component design parameters, fire condition parameters, and post-fire damage response data of reinforced concrete components. The model training module is used to train a machine learning proxy model with the component design parameters and fire condition parameters of reinforced concrete components as input parameters and the damage response data after fire as output parameters, so as to obtain a predictive model for predicting the direct physical damage index of reinforced concrete components after fire. The evaluation curve construction module is used to identify key uncertainty parameters in the fire response characteristic knowledge base and establish probability distribution models for each parameter. Latin hypercube sampling is used to generate multiple random sample combinations within the probability distribution interval of each parameter. These sample combinations are then input into the prediction model for batch prediction. The prediction results are combined with preset multi-level damage limit state criteria to statistically determine the failure probability of components exceeding each level of damage state under different fire intensities. Fire intensity and failure probability are correlated to construct a fire vulnerability curve. The key uncertainty parameters refer to input parameters that, in the post-fire damage assessment of reinforced concrete components, have a greater impact on the fire response than a preset value and exhibit inherent fluctuations due to objective factors, making them impossible to completely fix. The evaluation module is used to receive the design parameters and fire intensity of the reinforced concrete component to be evaluated after a fire, call the prediction model to output direct physical damage indicators, and query the vulnerability curve under the corresponding fire intensity to obtain the failure probability of each damage level.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is loaded by the processor, it is able to perform the steps of the method according to any one of claims 1 to 6.

Images