GRPC structure service life evaluation method and system based on ensemble learning

Abstract

This invention relates to the field of concrete life assessment technology, and discloses a method and system for assessing the lifespan of GRPC structures based on ensemble learning. The method includes: calculating the creep coefficient of GRPC under compression; constructing a creep clock based on the development trend of the creep coefficient, and constructing a damage clock based on the creep clock; outputting lifespan decay parameters characterizing the influence of dual-clock coupling on lifespan decay; performing sensitivity analysis on the coupling parameters of the creep clock and the damage clock to obtain a set of coupling influence parameters; optimizing the coupling influence parameter set to obtain a lifespan feature set for ensemble learning; constructing a GRPC structure lifespan prediction model; and using the GRPC structure lifespan prediction model to predict the lifespan of the GRPC structure under assessment, obtaining the GRPC structure lifespan assessment result. This invention provides the lifespan decay law of GRPC under the condition of co-evolution of compressive creep and damage, improving the accuracy of lifespan assessment.

Description

Technical Field

[0001] This invention relates to the field of concrete life assessment technology, and more specifically, to a method and system for assessing the lifespan of GRPC structures based on ensemble learning. Background Technology

[0002] Green Reactive Powder Concrete (GRPC) structures are typically subjected to the combined effects of loads, environmental factors, and internal material creep and damage evolution during long-term service. Their performance degradation is not a simple linear process but exhibits significant time-varying, nonlinear, and coupled characteristics. With increasing service time, stress redistribution, microcrack propagation, and material interface deterioration accumulate, leading to decreased load-bearing capacity, stiffness degradation, and shortened service life. Therefore, conducting service life assessments for GRPC structures is not only crucial for structural health monitoring and safety assurance but also a key foundation for operational and maintenance decisions, repair scheduling, and risk warning.

[0003] Existing methods for structural life assessment mainly include life estimation methods based on empirical formulas, statistical analysis methods based on single degradation indices, and prediction methods based on machine learning. While empirical formula methods are easy to implement, they typically rely on specific experimental conditions or engineering experience, making it difficult to accurately reflect the true degradation patterns of GRPC structures under complex operating conditions. Single degradation index methods often focus only on a single type of response, ignoring the combined effects of creep, damage, and their coupling on life attenuation. Although machine learning methods have strong nonlinear fitting capabilities, if the input features are not selected appropriately, redundant information or noise interference can easily be introduced, leading to insufficient model generalization ability, poor prediction stability, and difficulty in adapting to the differences in samples from different batches of materials, different loading levels, and different service stages.

[0004] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0005] To address the problems in related technologies, this invention proposes a method and system for evaluating the lifespan of GRPC structures based on ensemble learning, in order to overcome the aforementioned technical problems existing in the prior art.

[0006] Therefore, the specific technical solution adopted by the present invention is as follows: According to one aspect of the present invention, a method for assessing the service life of a GRPC structure based on ensemble learning is provided, comprising: constructing an expression for calculating the compressive creep coefficient of a GRPC by modifying the parameters of a pre-acquired basic creep prediction model of a GRPC, and calculating the creep coefficient of the GRPC under compression; constructing a creep clock based on the development trend of the creep coefficient, and constructing a damage clock based on the creep clock; establishing a state-space coupling relationship between the creep clock and the damage clock, and combining it with the life decay observation modeling process to output a life decay parameter characterizing the influence of the dual-clock coupling on service life decay; performing sensitivity analysis on the coupling parameters of the creep clock and the damage clock based on Latin hypercube sampling technology, combined with the life decay parameter, to obtain a set of coupling influence parameters; optimizing the parameters of the set of coupling influence parameters using a genetic algorithm to obtain a set of life features for ensemble learning; constructing a GRPC structure service life prediction model using the life feature set as input; and using the GRPC structure service life prediction model to predict the service life of the GRPC structure to be assessed, thereby obtaining the GRPC structure service life assessment result.

[0007] Another aspect of the present invention provides a GRPC structure service life assessment system based on ensemble learning, comprising: a creep correction module, used to construct a calculation expression for the GRPC compressive creep coefficient by correcting the parameters of a pre-acquired GRPC basic creep prediction model, and to calculate the creep coefficient of the GRPC under compressive conditions; a dual-clock coupling module, used to construct a creep clock based on the development trend of the creep coefficient, and to construct a damage clock based on the creep clock; to establish a state-space coupling relationship between the creep clock and the damage clock, and to output a service life decay parameter characterizing the impact of dual-clock coupling on service life decay by combining the service life decay observation modeling process; a sensitivity analysis module, used to perform sensitivity analysis on the coupling parameters of the creep clock and the damage clock based on Latin hypercube sampling technology, combined with the service life decay parameter, to obtain a set of coupling influence parameters; a parameter optimization module, used to optimize the parameters of the coupling influence parameter set using a genetic algorithm, to obtain a service life feature set for ensemble learning; and a service life prediction module, used to construct a GRPC structure service life prediction model using the service life feature set as input; and to use the GRPC structure service life prediction model to predict the service life of the GRPC structure to be evaluated, to obtain the GRPC structure service life assessment result.

[0008] The beneficial effects of this invention are as follows: 1. This invention combines dual-clock coupling mechanism modeling, parameter sensitivity analysis, genetic optimization screening and ensemble learning prediction to provide the lifetime decay law of GRPC under the conditions of compressive creep and damage co-evolution, improve the accuracy and stability of lifetime assessment, and enhance robustness to complex working conditions and sample noise.

[0009] 2. By establishing a dual-clock coupled state space of creep clock and damage clock, the creep duration and damage accumulation process are unified into the same dynamic framework, which can simultaneously reflect the bidirectional coupling mechanism of creep-driven damage growth and damage reaction on lifetime decay. This allows lifetime assessment to no longer be limited to a single time variable or a single degradation amount, but to reveal the formation mechanism of GRPC structure lifetime decay from the internal state evolution level.

[0010] 3. This invention efficiently covers coupled parameters through Latin hypercube sampling, conducts sensitivity analysis based on the statistical response characteristics of lifetime decay parameters, and utilizes a genetic algorithm to optimize and screen coupled parameters, effectively identifying the key parameters that contribute most to lifetime decay and reducing interference from redundant parameters. This invention reduces the subjectivity of parameter selection and improves the discriminativeness and stability of the feature set, allowing model training to focus more on the core factors truly affecting GRPC lifetime decay, thereby improving prediction accuracy and reducing computational complexity.

[0011] 4. Multiple life prediction base learners are constructed based on the life feature set, and the final prediction model is formed by weighted fusion. This fully utilizes the complementary advantages of different learners in terms of nonlinear fitting, local generalization, and noise resistance. Ensemble learning can effectively reduce the risk of overfitting, enhance the adaptability to samples from different operating conditions, material batches, and service stages, and make the life assessment results more stable and reliable. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, 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.

[0013] Figure 1 This is a flowchart of a GRPC structure lifetime assessment method based on ensemble learning according to an embodiment of the present invention; Figure 2 This is a block diagram of a GRPC structure life assessment system based on ensemble learning according to an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the change of ambient temperature and humidity over time according to an embodiment of the present invention; Figure 4 This is a graph showing the change of GRPC shrinkage value over time according to an embodiment of the present invention; Figure 5 This is a graph showing the relationship between GRPC creep and holding time according to an embodiment of the present invention; Figure 6This is a diagram showing the effect of recycled micro powder and steel fiber on creep according to an embodiment of the present invention; Figure 7 This is a graph showing the relationship between the GRPC creep coefficient and the holding time according to an embodiment of the present invention; Figure 8 This is a graph showing the relationship between the GRPC creep function and the holding time according to an embodiment of the present invention; Figure 9 This is a comparison chart of the GRPC0 test results and model prediction results according to an embodiment of the present invention; Figure 10 This is a comparison chart of the GRPC3 experimental results and model prediction results according to an embodiment of the present invention; Figure 11 This is a comparison chart of the experimental results and model prediction results of GRPC3-STF according to an embodiment of the present invention; Figure 12 This is a comparison chart of the experimental results of GRPC0 according to an embodiment of the present invention and the prediction results of the modified model MC1990(99)R1; Figure 13 This is a comparison chart of the experimental results of GRPC3 according to an embodiment of the present invention and the prediction results of the modified model MC1990(99)R2; Figure 14 This is a comparison chart of the experimental results of GRPC3-STF and the prediction results of the modified model MC1990(99)R3 according to the embodiments of the present invention; Figure 15 This is a comparison chart of the experimentally measured values ​​of Xu Wenxue and the calculated values ​​of the MC1990 (99) R model according to embodiments of the present invention. Figure 15 (a) shows a comparison between the measured values ​​from the experiment in the Xu C16SF0 literature and the calculated values ​​from the MC1990 (99)R model. Figure 15 (b) is a comparison between the measured values ​​from the Xu 16SF2 literature experiment and the calculated values ​​from the MC1990(99)R model. Figure 15 (c) is a comparison between the measured values ​​of the XuC22SF2 literature experiment and the calculated values ​​of the MC1990(99)R model; Figure 16 This is a comparison chart of the experimentally measured values ​​of the Tristansen RPC (SC) literature and the calculated values ​​of the MC1990 (99) R model according to the embodiments of the present invention; Figure 17 This is a comparison chart of the experimentally measured values ​​of Chen's literature and the calculated values ​​of the MC1990 (99) R model according to embodiments of the present invention. Figure 17 (a) shows a comparison between the measured values ​​of Chen RPC-Ⅰ and the calculated values ​​of the MC1990(99)R model. Figure 17(b) is a comparison between the measured value of ChenRPC-V and the calculated value of the MC1990(99)R model; Figure 18 This is a graph showing the annual increase in the GRPC creep coefficient as a function of the load holding time, according to an embodiment of the present invention.

[0014] In the picture: 1. Creep correction module; 2. Dual-clock coupling module; 3. Sensitivity analysis module; 4. Parameter optimization module; 5. Lifetime prediction module. Detailed Implementation

[0015] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention. The components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0016] According to embodiments of the present invention, a method and system for evaluating the lifespan of GRPC structures based on ensemble learning are provided.

[0017] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, according to an embodiment of the present invention, a method for evaluating the lifespan of a GRPC structure based on ensemble learning is provided, comprising: S1. By correcting the parameters of the pre-obtained GRPC basic creep prediction model, construct the GRPC compressive creep coefficient calculation expression, and calculate the GRPC creep coefficient under compressive conditions.

[0018] In one embodiment, by modifying the parameters of a pre-obtained GRPC foundation creep prediction model, an expression for calculating the compressive creep coefficient of GRPC is constructed, and the creep coefficient of GRPC under compression is calculated. This includes: obtaining a GRPC foundation creep prediction model, which includes a nominal creep coefficient and a creep time development function; determining a micronized correction factor based on the micronized powder incorporation state of GRPC, and a steel fiber correction factor based on the steel fiber incorporation state; calculating a strength influence correction factor using the micronized powder correction factor and the steel fiber correction factor, combined with the average compressive strength of GRPC; correcting the nominal creep coefficient using the strength influence correction factor to obtain the nominal creep coefficient of GRPC; obtaining a corrected creep time development function by reconfiguring the exponent of the creep time development function; constructing an expression for calculating the compressive creep coefficient of GRPC based on the nominal creep coefficient of GRPC and the corrected creep time development function; and calculating the creep coefficient of GRPC under compression using the expression for calculating the compressive creep coefficient of GRPC. GRPC is a type of reinforced concrete.

[0019] S2. Construct a creep clock based on the development trend of the creep coefficient, and construct a damage clock based on the creep clock; establish the state-space coupling relationship between the creep clock and the damage clock, and combine it with the life decay observation modeling process to output life decay parameters that characterize the impact of dual-clock coupling on life decay.

[0020] In one embodiment, constructing a creep clock based on the development trend of the creep coefficient includes: identifying the monotonicity and fitting the trend of the creep coefficient variation sequence along the loading duration to obtain a creep coefficient development function characterizing the compression creep of GRPC over time; performing long-term limit analysis on the creep coefficient development function and determining the limit value when the loading duration tends to infinity as the creep ultimate value; using the creep ultimate value as a normalization benchmark to make the creep development state under any loading duration dimensionless relative to the creep ultimate value, and outputting a normalized creep evolution variable; using the normalized creep evolution variable as a state measure of the creep process, and mapping the actual loading duration to an equivalent time variable controlled by the degree of creep development, in order to construct a creep clock.

[0021] This invention constructs a creep clock based on the development trend of the creep coefficient. During long-term service under pressure, the creep coefficient of GRPC typically increases gradually with the duration of loading, eventually approaching a stable limit. Therefore, the creep coefficient is considered a time response quantity characterizing the material's compressive creep process. By identifying the monotonicity and fitting the trend of the creep coefficient sequence, a continuous creep coefficient development function is obtained to describe the degree of creep development at any loading duration. Long-term limit analysis is performed on the creep coefficient development function to obtain the ultimate creep value. This ultimate value is then used as a normalization benchmark to construct a normalized creep evolution variable, representing the proportion of the current creep development degree to the final creep development degree. Thus, the actual loading time is no longer solely characterized by physical time but is mapped to an equivalent time variable controlled by the degree of creep development, thereby forming a creep clock that reflects the rate of advancement and evolution stages of the creep state within the material.

[0022] By constructing a creep clock, discrete creep coefficient data, which is affected by experimental fluctuations, is transformed into a continuous, stable, and physically meaningful creep process variable. Since the creep clock does not simply use the actual loading time but rather uses a normalized creep development degree to characterize the service condition, it can more accurately reflect the differences in the rate of creep development of GRPC under different stress levels, environmental conditions, ages, or material compositions. For example, under the same actual loading time, if the creep coefficient increases faster under a certain condition, its creep clock also advances faster, indicating that the structure is closer to the long-term creep stabilization stage or the potential damage accumulation stage. This avoids the errors that arise from judging the service life state solely based on physical time.

[0023] The creep clock is: ; In the formula, Indicates the first The creep clock variable corresponding to each loading moment Indicates the first Each observation time or calculation time For GRPC during loading age Calculation time The coefficient of compressive creep under the following conditions This represents the final creep value as the loading duration approaches infinity. This indicates a small positive number that prevents the denominator from being zero.

[0024] In one embodiment, constructing a damage clock based on a creep clock includes: outputting a creep clock variable to characterize the equivalent duration of GRPC compressive creep based on the creep clock; performing incremental decomposition on the growth process of the creep clock variable along the loading duration, extracting the creep clock increment between adjacent loading times, and constructing a creep clock increment sequence; constructing a damage evolution increment based on the creep clock increment sequence, converting the creep clock increment within each loading time step into the damage growth of the structural material, and obtaining the damage state variable induced by compressive creep through accumulation; and constructing a damage clock based on the evolution ratio of the damage state variable relative to the failure damage threshold.

[0025] The creep clock variable converts the actual loading duration into an equivalent time period characterizing the degree of creep development. Therefore, its increment between adjacent time steps reflects the creep evolution rate of the material within that time step. By decomposing the creep clock increment sequence into evolutionary quantities at discrete time steps and mapping each increment to the corresponding damage growth, a monotonic mapping relationship between creep evolution and damage accumulation can be established. Then, using the failure damage threshold as a reference, a normalized damage ratio is constructed and defined as the damage clock. This allows the damage clock to characterize the evolution process of the structure from initial damage-free to near-failure. The failure damage threshold is determined based on the failure criteria of the GRPC structure, accelerated aging test results, long-term loading test results, or historical service monitoring data, and can be set to 0.75-0.90.

[0026] Among them, let the first The creep clock increment for each loading time step is: ; The damage increment is then expressed as: ; in, For damage state variables, The damage growth coefficient can be selected as... ; The creep increment nonlinear exponent can be selected as... ; To sustain compressive stress, Average compressive strength; The stress sensitivity index can be selected as... ; The environmental impact factor can be selected as... The damage growth coefficient, variable increment nonlinearity index, stress sensitivity index, and environmental impact coefficient were obtained by calibration using long-term compression creep test, accelerated aging test, and damage evolution data under different stress levels / environmental conditions of GRPC specimens.

[0027] The damage state variables are recursively calculated as follows: ; And it can be constrained as follows: ; The damage clock is: ; in, This is the failure damage threshold.

[0028] By constructing a damage clock based on a creep clock, the material damage process, which is difficult to observe directly, is transformed into a time-series variable driven by the creep clock. This allows damage modeling to rely not solely on actual time, but on the equivalent evolution process induced by creep, thus better aligning with the degradation mechanism of creep-induced damage accumulation under long-term pressure service of GRPC. Compared to directly accumulating damage based on physical time, this invention can more accurately distinguish the damage growth rate under different stress levels, different temperature and humidity environments, or different material compositions, avoiding modeling biases where the damage degree varies even with the same service time.

[0029] In one embodiment, a state-space coupling relationship between the creep clock and the damage clock is established. Combined with the lifespan decay observation modeling process, a lifespan decay parameter characterizing the impact of dual-clock coupling on lifespan decay is output. This includes: establishing a dual-clock synchronization variable sequence corresponding to the creep clock and the damage clock, including creep clock variables and damage clock variables; using the creep clock variables and damage clock variables as the first and second state components in the state space to form a dual-clock state vector sequence; performing differential processing on the dual-clock state vectors of adjacent loading time steps to obtain creep clock state increments and damage clock state increments; using the creep clock state increment as the equivalent duration advancement of GRPC compressive creep, and the damage clock state increment as the equivalent duration advancement of GRPC compressive creep damage, to form a dual-clock state increment sequence; and constructing a dual-clock coupling strength variable based on the driving relationship between the creep clock state increment and the damage clock state increment to form the dual-clock coupling strength. The variable sequence is used as follows: The state-space coupling equation between the creep clock and the damage clock is established, with the dual-clock state vector as the state variable and the dual-clock coupling strength variable as the coupling input variable. This ensures that the dual-clock state at the next loading time step is jointly determined by the dual-clock state at the current loading time step and the dual-clock coupling strength at the current loading time step. Based on the joint contribution relationship between the creep clock variable, the damage clock variable, and the dual-clock coupling strength variable to the GRPC lifetime decay, a lifetime decay observation equation is established. Based on the state-space coupling equation between the creep clock and the damage clock, the state vector evolved by dual-clock coupling at each loading time step is obtained recursively. The evolved state vector and the dual-clock coupling strength variable are mapped to lifetime decay response quantities through the lifetime decay observation equation. The lifetime decay response quantities are dimensionless using the initial lifetime as a normalization benchmark to obtain lifetime decay parameters characterizing the impact of dual-clock coupling on lifetime decay.

[0030] The creep and damage processes, previously two independent temporal descriptors, are unified into a state-space framework for coupled recursive modeling. Specifically, the creep clock variable at the same loading moment is... and damage clock variables Consolidate the dual-clock synchronization variables and construct the dual-clock state vector. ,in, This indicates the equivalent duration of GRPC propagation under pressure creep. This represents the equivalent time-progression state of compression-induced creep damage; by differencing adjacent loading time steps, we obtain... and and according to right The driving relationship constructs a dual-clock coupling strength variable, which can be represented as follows: ,in, To prevent tiny positive numbers with a denominator of zero, and to avoid abnormal amplification of coupling strength due to excessively small creep clock increments, a limiting process is used.

[0031] With dual clock state vectors As a state variable, with dual-clock coupling strength variable As coupled input variables, state-space coupled state equations are established, such as: ,in, This represents the self-evolutionary relationship of the dual-clock state. This indicates the effect of coupling strength on the dual-clock state progression. This represents the state perturbation term; simultaneously, a lifetime decay observation equation is established, such as: ,in, This is the lifetime decay response quantity. This indicates the weighting of the creep clock and damage clock contributions to lifetime decay. This indicates the additional contribution of coupling strength to lifetime decay. This is the observed perturbation term.

[0032] The self-evolutionary relationship of the dual-clock state refers to the continuation, maintenance, and natural advancement of the creep clock state and damage clock state at the current loading time step on the corresponding state at the next loading time step, without considering external abrupt changes. It reflects the inertial characteristics of the GRPC compression creep development and damage accumulation process. This relationship is characterized by state self-evolution parameters. The influence of coupling strength on the advancement of the dual-clock state refers to the moderating effect of the dual-clock coupling strength variable on the advancement speed of the creep clock and damage clock at the next loading time step. That is, when the creep clock increment can significantly drive the damage clock increment, the dual-clock state will exhibit a stronger co-evolutionary trend. This influence is determined by the coupling input influence parameters. and Characterization. The contribution weights of creep clock and damage clock to lifetime decay refer to the proportions of influence of creep development state and damage accumulation state on the degree of reduction in the lifetime of GRPC structure in the lifetime decay observation model, respectively, derived from lifetime decay mapping parameters. and Characterization. The additional contribution of coupling strength to lifetime decay refers to the additional accelerating effect of the driving force of creep development on damage accumulation, beyond the influence of creep clock state and damage clock state itself on lifetime decay. This additional contribution is typically determined by the direct coupling influence parameters. Characterization.

[0033] Based on initial service life As a normalized baseline, the lifetime degradation response quantity Convert to dimensionless lifetime decay parameter or Thus, lifetime degradation parameters that can characterize the impact of dual-clock coupling on the lifetime degradation of GRPC are obtained.

[0034] It should be noted that unifying creep development, damage accumulation, and their mutual driving relationship into a state-space model can improve the dynamics and interpretability of long-term service life degradation modeling for GRPC. Specifically, a dual-clock state vector describes the synchronous evolution of creep and damage processes, a dual-clock coupling strength variable describes the driving strength between them, and a life degradation observation equation maps the internal state to an externally assessable life degradation response. This approach reflects the internal degradation process of the material and outputs directly usable life degradation parameters for engineering applications.

[0035] S3. Based on the Latin hypercube sampling technique and combined with the lifetime decay parameter, a sensitivity analysis is performed on the coupling parameters of the creep clock and the damage clock to obtain the set of coupling influence parameters.

[0036] In one embodiment, based on Latin hypercube sampling technology and combined with lifetime decay parameters, sensitivity analysis is performed on the coupling parameters of creep clock and damage clock to obtain a set of coupling influence parameters. This set includes: using the state self-evolution parameters, dual-clock interactive coupling parameters, and coupling input parameters in the state-space coupling state equations of creep clock and damage clock, as well as the lifetime decay mapping parameters and direct coupling influence parameters in the lifetime decay observation equations, as the set of coupling parameters to be analyzed, and determining the sampling interval for each coupling parameter; performing stratified sampling of each coupling parameter sampling interval based on Latin hypercube sampling technology, and forming a coupling parameter sample matrix through random combination; and obtaining the lifetime decay parameter sample sequence. The statistical response characteristics of the lifetime decay parameter under different coupling parameter sample conditions are calculated. These characteristics include the mean, variance, and magnitude of change of the lifetime decay parameter, used to characterize the impact of coupling parameter perturbations on the GRPC lifetime decay results. The correspondence between the change of each coupling parameter sample and the change of the lifetime decay parameter is obtained, and the sensitivity index of each coupling parameter to the lifetime decay parameter is calculated to quantify the contribution of that coupling parameter to the dual-clock coupled lifetime decay results. The coupling parameters are sorted from largest to smallest according to their sensitivity index, and coupling parameters with sensitivity indices exceeding a preset sensitivity threshold are included in the coupling influence parameter set. The preset sensitivity threshold is determined based on the contribution of the coupling parameter to the lifetime prediction results.

[0037] Specifically, the state self-evolution parameters, dual-clock interactive coupling parameters, and coupling input parameters in the state-space coupled state equations, as well as the lifetime decay mapping parameters and direct coupling influence parameters in the lifetime decay observation equations, are first unified into a set of coupling parameters to be analyzed. For example, it can be represented as follows: ,in, These represent the evolutionary inertia of the creep clock and the damage clock themselves (which can be determined as 0.80-1.05 and 0.80-1.10, respectively). These parameters represent the interactive coupling relationship between the two clocks (which can be determined as 0-0.30 and 0.10-0.80), respectively. These parameters are obtained through state-space identification using long-term creep tests, damage monitoring data, or historical service data. For example, the least squares method and Kalman filter parameter estimation and inversion are used to minimize the error between the recursively obtained dual clock states and the measured creep / damage evolution curves. This indicates the effect of the coupling strength input on the state evolution. This indicates the degree to which coupling strength affects creep clock propagation, and can be taken as 0-0.30. This indicates the degree to which coupling strength affects the advancement of the damaged clock, and can be taken as 0.10-0.80. , Regression identification was performed using dual-clock sequences under different stress levels and environmental conditions; This represents the mapping weights of the dual-clock state to the lifetime decay response. This represents the mapping weight of creep clock states to lifetime decay, and can be set between 0.10 and 0.50. The weight representing the mapping of damage clock state to lifetime decay can be set between 0.30 and 0.90. This weight is obtained through supervised regression based on measured remaining lifetime, residual bearing capacity, stiffness degradation rate, or failure time label. This represents the direct impact of coupling strength on lifetime decay, and can be taken as 0.02-0.35.

[0038] Set sampling interval for each parameter Latin hypercube sampling is used to divide each parameter interval into... The system divides the system into three equal-probability subintervals and randomly selects a sample point from each subinterval. These samples are then randomly combined to form a coupling parameter sample matrix. Substituting each set of parameter samples into the dual-clock coupled state equation and the lifetime decay observation equation yields the corresponding lifetime decay parameter sample sequence. And calculate their mean. ,variance and range of change Calculate the sensitivity index between changes in each coupling parameter and changes in the lifetime decay parameter, for example, using a correlation coefficient-type index. Regression coefficient indicators or variance contribution rate indicators are used to quantify parameters. The strength of the impact on lifetime degradation results, according to Sort by size from largest to smallest, and satisfy The parameters are included in the coupling effect parameter set. Set it to 0.10-0.50.

[0039] By combining Latin hypercube sampling with statistical response analysis of lifetime decay parameters, we can identify which parameters have the greatest impact on lifetime prediction results and which parameters have a weaker impact, thereby reducing the parameter dimensionality of subsequent genetic algorithm optimization and ensemble learning modeling and lowering the model training complexity.

[0040] For example, using Latin hypercube sampling to generate Set up parameter samples and calculate lifetime decay parameters for each sample set. If statistics are obtained right The sensitivity index is , for , for ,and Only When the preset sensitivity threshold is At that time, it can be , and Include the set of coupling effect parameters. The number of samples in Latin hypercube sampling is between 100 and 1000.

[0041] S4. Using a genetic algorithm, the parameters of the coupling influence parameter set are optimized to obtain a lifetime feature set for ensemble learning.

[0042] In one embodiment, a genetic algorithm is used to optimize the coupling effect parameter set to obtain a lifetime feature set for ensemble learning. This includes: constructing a fitness function based on the lifetime decay parameter prediction error, model stability error, or cross-validation error; performing genetic evolution optimization based on the fitness function to obtain an optimized coupling effect parameter set; and combining the optimized coupling effect parameter set with the lifetime decay parameter and the statistical response features of the lifetime decay parameter to form a lifetime feature set for subsequent ensemble learning model training.

[0043] In one embodiment, genetic evolution optimization is performed based on the fitness function to obtain an optimized set of coupling influence parameters, including: generating an initial population based on the value range of the coupling influence parameters; evaluating the fitness of parameter combinations in the initial population sequentially based on the fitness function; and updating the population iteratively to obtain candidate optimal coupling parameter combinations; when the iteration terminates, selecting the coupling influence parameter combination with the highest fitness from the final population as the optimized set of coupling influence parameters.

[0044] The optimization problem of coupling effects on the parameter set is transformed into a global optimization problem with lifetime prediction performance as the objective, and a genetic algorithm is used for adaptive search within the parameter space. A fitness function is constructed based on lifetime decay parameter prediction error, model stability error, or cross-validation error, for example... ,in, This represents the set of coupling effect parameters to be optimized. This indicates the prediction error of the lifetime degradation parameter. This indicates the model stability error. Indicates the cross-validation error. Different weighting coefficients are determined based on the relative importance of lifetime prediction accuracy, model stability, and cross-validation generalization ability in the optimization objective, such as... =0.50-0.60, =0.20-0.30, =0.15-0.25.

[0045] An initial population is generated with the range of values ​​for the coupling effect parameters as the boundary. Each set of parameters is substituted into the dual-clock coupled state equation and lifetime decay observation equation to calculate fitness. The population is continuously updated through genetic operations such as selection, crossover, and mutation, allowing the parameter combinations to evolve towards smaller prediction errors, stronger stability, and better generalization ability. When the iteration termination condition is met (e.g., reaching the maximum number of iterations), the parameter combination with the highest fitness is selected from the final population as the optimized set of coupling effect parameters. This set is then combined with the lifetime decay parameters and their statistical response characteristics to form a lifetime feature set.

[0046] The genetic algorithm uses real-number encoding, with each individual representing a set of coupled influence parameters. The initial population size is 30-100; the maximum number of iterations is 50-300; the crossover probability is 0.70-0.95; the mutation probability is 0.01-0.10; and the elite retention rate is 5%-10%. Wherein: ; ; ; in, or ; The number of repetitions is set to 5-10. Indicates the first The true lifetime decay parameters of each sample Indicates the first Lifetime decay parameters predicted from a sample; Indicates the first The second training session or the first During the second prediction, for the th Predicted lifetime decay parameters for each sample output; Indicates the first The root mean square error obtained from cross-validation.

[0047] S5. Using the life feature set as input, construct a GRPC structure life prediction model; use the GRPC structure life prediction model to predict the life of the GRPC structure to be evaluated, and obtain the GRPC structure life evaluation results.

[0048] In one embodiment, a GRPC structure lifespan prediction model is constructed using a lifespan feature set as input, including: constructing a lifespan prediction sample set using the lifespan feature set as input features and the corresponding actual lifespan of the GRPC structure as supervision labels; training multiple lifespan prediction base learners based on the sample set, and using weighted fusion to fuse the prediction results of the multiple lifespan prediction base learners to establish a GRPC structure lifespan prediction model.

[0049] The monitoring labels are obtained as follows: the actual time from the commissioning of the component to the failure criterion in long-term service monitoring; the equivalent service life corresponding to the specimen reaching the failure damage threshold in accelerated aging or long-term loading tests; and the remaining service life calculated by reliability analysis.

[0050] Multiple lifespan prediction base learners are trained based on a sample set, such as regression trees, support vector regression, random forest sub-models, gradient boosting sub-models, or neural network sub-models. Each base learner learns the nonlinear mapping relationship between creep clock, damage clock, and their coupling effect on lifespan decay from different perspectives. The outputs of multiple base learners are combined into the final prediction result through a weighted fusion method, represented as... ,in, For the first The output of each base learner For the corresponding weight and satisfying The model established in this way is an ensemble learning regression model that utilizes the complementarity of multiple weak or moderate strength predictors to form a comprehensive estimate of the service life of the GRPC structure and outputs the service life assessment results of the component to be evaluated.

[0051] Regarding lifespan prediction base learners, for example, regression tree base learners use lifespan feature vectors as input. During training, the maximum depth of the regression tree... Set to 5-8, preferably Minimum number of split samples Set to 8-20, preferably Minimum number of leaf node samples Setting it to 4-10 is preferred. The node partitioning criterion adopts the mean square error minimization criterion. If the number of training samples is... It can be found in That is, 224 sets of samples were used as the training set. That is, 48 ​​samples are used as the validation set, and the remaining 48 samples are used as the test set. During training, each node tries candidate thresholds for each input feature and selects the feature and threshold that maximizes the reduction in the lifetime variance of the child nodes. When the actual lifetime of samples in a node is mainly concentrated in 42-48 years, while the two child nodes after splitting are concentrated in 38-42 years and 46-50 years respectively, the split is considered to effectively improve the lifetime discrimination ability.

[0052] For example, the support vector regression model uses the radial basis function kernel. Because different lifetime features have different dimensions and ranges, each input feature is standardized before training to ensure that its mean is... Standard deviation is Distribution, lifetime tags Normalization can be performed as needed. SVR penalty coefficient Select an insensitive loss parameter from (10, 30, 50, 100, 150). Select the kernel parameters from (0.02, 0.05, 0.08, 0.10). Select from (0.05, 0.10, 0.20, 0.50, 1.00). The preferred configuration is (80, 0.05, 0.20). If the actual lifespan label range is 25-70 years, then... In the normalized lifetime space, this corresponds to an error tolerance range of approximately 2.25 years, which can be used to suppress the impact of small measurement noise on the model. During training, 5-fold cross-validation can be used to evaluate different parameter combinations. A set of parameters with a mean absolute error below 2.8 years and a root mean square error below 3.6 years on the validation set can be considered a candidate model; if... Too large, for example The model may over-track individual outlier samples, leading to increased validation errors. Therefore, it is advisable to optimize the model. The range of moderate penalties.

[0053] Based on the established GRPC structure service life prediction model, the service life feature set parameters corresponding to the structure to be evaluated are input into the model, and the service life of the GRPC structure is predicted by the model.

[0054] like Figure 2 As shown, according to another embodiment of the present invention, a GRPC structure lifespan assessment system based on ensemble learning is also provided, comprising: Creep correction module 1 is used to construct the GRPC compression creep coefficient calculation expression by correcting the parameters of the pre-acquired GRPC basic creep prediction model, and to calculate the GRPC creep coefficient under compression. The dual-clock coupling module 2 is used to construct a creep clock based on the development trend of the creep coefficient, and to construct a damage clock based on the creep clock; to establish the state-space coupling relationship between the creep clock and the damage clock, and to output a lifetime decay parameter characterizing the impact of dual-clock coupling on lifetime decay in combination with the lifetime decay observation modeling process. Sensitivity analysis module 3 is used to perform sensitivity analysis on the coupling parameters of creep clock and damage clock based on Latin hypercube sampling technology and combined with lifetime decay parameters, so as to obtain a set of coupling influence parameters; Parameter optimization module 4 is used to optimize the parameters of the coupling influence parameter set using a genetic algorithm to obtain a lifetime feature set for ensemble learning; The life prediction module 5 is used to construct a life prediction model for GRPC structures using a life feature set as input; the life prediction model is then used to predict the life of the GRPC structure to be evaluated, and the life evaluation results of the GRPC structure are obtained.

[0055] This invention also includes the following creep prediction and testing process: The study on the compressive creep performance of green reactive powder concrete includes the generation mechanism, influencing factors, and common prediction models of concrete creep. Compressive creep tests were conducted on GRPC0, GRPC3, and GRPC3-STF according to standard experimental methods. The effects of recycled micropowder and steel fibers on the shrinkage and creep performance of GRPC were investigated. The calculation results of commonly used creep prediction models were compared with the experimental results, and the reasons for the prediction deviations of each model were analyzed. Based on the influencing factors of creep and the differences between GRPC and ordinary concrete, the MC1990(99) model was modified, and the MC1990(99)R model suitable for GRPC creep prediction was proposed. The prediction accuracy of the model was evaluated using the B3 coefficient of variation method, and the applicability of the model to RPC (reactive powder concrete) was verified using literature data. Finally, the ultimate value of GRPC creep was suggested based on this prediction model.

[0056] I. Creep of Concrete: Creep refers to the phenomenon where the deformation of a concrete structure increases over time under sustained load; it is an inherent time-varying characteristic of concrete. Creep increases with the duration of load holding, and the rate of increase gradually decreases. Creep development can last for decades, but most of it is completed within 1-2 years. Generally, the deformation caused by creep in concrete is 1-3 times greater than the instantaneous elastic deformation during loading. Creep itself does not affect the strength of plain concrete, but it does affect the stress and deformation of reinforced concrete structures. Creep can cause prestress loss in prestressed concrete structures; for large-span reinforced concrete members, creep increases mid-span deflection, leading to increased curvature; for high-rise buildings, if the creep effect of concrete is not considered during construction, the redistribution of internal forces caused by creep may lead to unforeseen conditions. If creep causes excessive deformation, it can render the structure unusable, and some members may even buckle, crack, or fail, losing their load-bearing capacity. These are the adverse effects of creep on concrete structures. However, creep can also reduce internal stress caused by uneven shrinkage in concrete, thus reducing cracking. In statically indeterminate structures, creep can eliminate stress concentration caused by shrinkage, temperature changes, or support movement. For large-volume concrete structures, creep can reduce internal temperature stress and decrease cracks caused by shrinkage. These are the beneficial effects of creep on concrete structures. As an important property of concrete, creep is a crucial factor that must be considered in engineering design and construction; therefore, research on the creep performance of concrete is necessary and meaningful.

[0057] Concrete creep typically includes two types: loading creep and drying creep. Generally, the total creep is calculated by subtracting the unloaded deformation (compensation deformation for comparative specimens) from the creep deformation under loading. The creep produced under load under sealed conditions is called basic creep. Subtracting basic creep from total creep yields drying creep. In actual engineering projects, concrete is mostly subjected to simultaneous loading and drying, thus yielding the total creep.

[0058] 1.1 Creep Mechanism: Since the discovery of concrete creep, researchers have conducted extensive experimental and theoretical studies on the creep properties of concrete, and have proposed many theories to explain the creep mechanism of concrete. Among them, the most widely accepted theories include the exudation theory, the viscous flow theory, the plastic flow theory, the microcrack theory, the internal force equilibrium theory, and the hysteretic elasticity theory.

[0059] 1.1.1 Exudation Theory: The exudation theory posits that under external loads, the gel water within the cement paste (including water adsorbed on the surface of gel particles and interlayer water between particles) slowly flows out. As some water is squeezed out, the stress on the gel particles gradually increases, while the stress on the water gradually decreases, thus slowing the water exudation rate. The deformation of the paste during this process macroscopically manifests as concrete creep, a phenomenon that occurs when the gel and the surrounding medium reach a new moisture equilibrium under external pressure. Under continuous external load, the distance between gel particles decreases, the distance between water molecules decreases, the surface energy between particles decreases, leading to chemical bonding of some particles and increasing the stability of the gel. Therefore, after unloading, the gel does not return to its pre-loading state; the residual deformation is the non-recoverable creep of the concrete.

[0060] The rate of water seepage depends on the external compressive stress and the capillary resistance within the concrete. Greater compressive stress leads to a faster seepage rate and greater creep. Conversely, higher density cement paste results in greater capillary resistance, slower water seepage, less deformation, and consequently, less creep. Therefore, under the same conditions, high-strength concrete exhibits less creep.

[0061] 1.1.2 Viscous Flow Theory: The viscous flow theory views concrete as consisting of two parts: a cement paste that flows viscously under load and inert aggregate that does not flow viscously under load. When concrete is loaded, the cement paste flows under pressure, gradually transferring the pressure to the aggregate. As the load duration increases, the stress in the cement paste decreases while the stress in the aggregate increases. Greater stress leads to greater elastic deformation in the aggregate. The deformation of the cement paste is proportional to the applied stress; over time, the flow rate of the cement paste decreases. Macroscopically, this manifests as a gradual decrease in the creep rate of concrete. From the microscopic structure of the cement paste, the tangential points of the gel particles are covered with adsorbed water. When the applied stress reaches a certain level, the viscous flow of the adsorbed water causes slippage at these points, leading to irreversible creep in the concrete.

[0062] 1.1.3 Plastic Flow Theory: The lattice slip theory posits that when the applied stress is less than the internal thermal energy of the molecular structure, the material exhibits viscous behavior; when the applied stress exceeds the internal thermal energy, the material exhibits plasticity. Concrete creep, in some aspects, displays properties similar to the plastic deformation of metallic materials. Under low stress, concrete creep can be explained as viscous flow, while under high stress, it can be explained as plastic deformation (lattice slip). The theory also points out that due to the different chemical bonding forces between concrete and metallic materials, the creep exhibited by concrete under high stress is a form of pseudo-plasticity. The bonding force of metallic materials is mainly metallic chemical bonds, which are easily formed, broken, and recovered; therefore, metals exhibit high fluidity and plasticity with crystalline extension properties. In contrast, the bonding force of cement paste is mainly rigid chemical ionic bonds, which can only undergo brittle fracture and do not produce true plasticity. The macroscopic nonlinearity of concrete is caused by the propagation of microcracks existing between different material interfaces within the concrete, and is not true plasticity.

[0063] 1.1.4 Microcrack Theory: The microcrack theory explains concrete creep using the theory of crack generation and development. Concrete is a composite material composed of multiple phases, including solid, liquid, and gas. Before loading, microcracks exist within it due to various reasons. Within the normal working stress range, the load is transmitted through friction between the concrete crack interfaces. Frictional displacement under pressure causes small deformations in the concrete. When the load exceeds the normal working stress, existing microcracks gradually expand, new cracks continuously appear, and excessive loads may even produce a small number of cracks penetrating the aggregate, at which point the concrete will exhibit significant deformation. The microcrack theory posits that microcracks only have a significant impact on creep when the applied stress exceeds the crack resistance strength of the concrete. Under long-term loading, when the stress is less than the crack resistance strength of the concrete, the external load makes the concrete denser. When the stress exceeds the crack resistance strength, with the generation and development of microcracks, the concrete will undergo additional deformation, macroscopically manifested as a nonlinear relationship between creep and stress.

[0064] 1.1.5. The Theory of Internal Force Equilibrium: The internal force equilibrium theory posits that concrete creep is a process in which the original internal force equilibrium state of the cement paste is disrupted and a new equilibrium state gradually forms under the influence of external forces. These internal forces include the surface tension generated by the shrinkage of gel particles, the van der Waals forces between particles, the separation force generated by adsorbed water on the surface of gel particles at the tangent point of the particles, and hydrostatic pressure. Among these, the separation force of adsorbed water is the most important. According to this theory, any change in the environmental conditions of the concrete, such as temperature, humidity, or load, will disrupt the original internal force equilibrium of the concrete. The process of establishing a new internal force equilibrium state is what causes the shrinkage or creep of the concrete.

[0065] 1.1.6 The hysteretic elasticity theory shares similarities with the plastic flow theory, but it views cement paste as a two-component combination. One part is an elastic framework composed of hydrated products of crystalline phases, and the other part is a viscous component containing adsorbed water and amorphous substances. When an external load is applied, the elastic framework and the viscous component in the cement paste share the load according to their stiffness. As the load-bearing time increases, the viscous component flows and gradually transfers the load it bears to the elastic framework. The elastic framework then undergoes greater elastic deformation, ultimately leading to increased deformation of the cement paste (concrete creep), exhibiting a hysteretic elastic behavior.

[0066] The main mechanisms of concrete creep are summarized as follows: (1) Viscous deformation of cement stone caused by sliding or shearing of cement gel under stress and lubrication of adsorbed water layer; (2) Compression caused by seepage of adsorbed water or transfer of interlayer water under stress; (3) Hysteretic elastic deformation caused by the constraint of cement gel on the elastic deformation of skeleton; (4) Permanent deformation caused by local cracking and recrystallization and new bonding.

[0067] From the above theories of creep, it is clear that no single theory can fully and reasonably explain all creep phenomena; rather, several theories must be combined. For example, concrete creep is characterized by a high growth rate in the early stages of loading and a gradual decrease in the later stages, and some creep is recoverable. This can be well explained by the viscous flow theory and the hysteretic elasticity theory. The partially irreversible creep of concrete can be explained by the exudation theory. When the applied stress exceeds the normal working capacity, the rapid increase in the creep rate can be explained by the plasticity theory and the microcrack theory.

[0068] 1.2 Influencing Factors: Factors influencing concrete creep can be categorized into internal and external factors. Internal factors mainly refer to concrete raw materials, mix proportions, and the shape and dimensions of the specimen itself; external factors mainly refer to loading conditions and environmental conditions. Numerous experiments have shown that these factors are all related to the creep characteristics of concrete, and the parameter levels of these factors provide important basis for the quantitative assessment and prediction of creep.

[0069] 1.2.1 Cement type and fineness: The influence of cement type on concrete creep is mainly reflected in its effect on concrete strength development. Cement that promotes faster strength development in concrete can reduce creep. Therefore, based on early-age loading, the order of increasing creep is: early-strength cement, ordinary cement, and low-heat cement. However, if the stress-strength ratio at loading is the same as the comparison benchmark, concrete with a relatively larger increase in strength after loading has smaller creep. The order of increasing creep then becomes: low-heat cement, ordinary cement, and early-strength cement. Furthermore, concrete creep decreases with increasing tricalcium silicate (C3S) content in cement clinker and increases with increasing dicalcium silicate (C2S) content.

[0070] The effect of cement fineness on concrete creep is mainly manifested in two aspects: Firstly, the finer the cement, the greater the early strength and the smaller the creep of the concrete; secondly, finer cement requires the addition of more gypsum to prevent retardation, which in turn increases concrete creep. Creep tests under the same strength conditions show that from the start of loading to one year, the finer the cement, the greater the concrete creep, but the rate of creep decreases over time; after one year of loading, creep decreases with increasing fineness. This may be because finer cement concrete has a higher strength gain, leading to a rapid decrease in the stress-to-strength ratio.

[0071] 1.2.2 Aggregates: The influence of aggregates on concrete creep is mainly reflected in two aspects: aggregate content and aggregate type. It is generally believed that aggregates do not creep and play a restraining role in the deformation of cement paste. In ordinary concrete, creep decreases with increasing aggregate content. Different aggregate types have different elastic moduli, hardness, and porosity. Under the same conditions, creep decreases with increasing elastic modulus and hardness of aggregates in concrete, but when the aggregate hardness exceeds a certain value, its influence on concrete creep tends to stabilize. Concrete with higher aggregate porosity exhibits greater creep. 1.2.3 Water-cement ratio and mortar ratio: The water-cement ratio (W / C) of concrete is the main factor affecting creep. Concrete with a high water-cement ratio has more internal pores, larger pore sizes, and lower strength, resulting in greater creep. Studies have shown that when W / C is in the range of 0.4-0.6, creep is linearly related to W / C, but in a larger water-cement ratio range (0.35-0.85), concrete creep exhibits a curvilinear relationship with W / C. The cement paste content per unit volume of concrete is called the mortar ratio, which is a comprehensive indicator reflecting the amount of cement, water, and aggregate in concrete. Concrete creep increases with increasing mortar ratio, showing a linear relationship. Simultaneously, with the increase of cement paste content, the water content per cubic meter of concrete increases, leading to an increase in the total porosity of hardened concrete and thus increasing concrete creep.

[0072] 1.2.4 Mineral admixtures and water-reducing agents: Commonly used mineral admixtures for concrete include silica fume, fly ash, and slag. The influence of admixtures on concrete creep is mainly reflected in their contribution to the strength of concrete at different ages. Silica fume can exert high pozzolanic activity and a micro-filling effect in concrete, improving the strength of concrete in the early stages, thus reducing concrete creep. Due to its low early-age activity and high later-age activity, fly ash concrete has the characteristics of low early-age strength and high later-age strength compared with ordinary concrete. Under early-age loading, fly ash concrete exhibits larger creep; however, in the long term, fly ash has a reducing effect on concrete creep. The mechanism of slag's influence on concrete creep is more complex, and the conclusions of different researchers are inconsistent. Slag can reduce concrete creep; low slag dosage has little effect on creep, while high dosage increases concrete creep; studies have found that when exposed to air, the early-age creep of finely ground slag concrete increases slightly but decreases slightly in the later stages, while under sealed conditions, the creep of finely ground slag concrete is significantly reduced.

[0073] There are many types of concrete admixtures, most of which are used to improve certain properties of concrete. Different functional admixtures have different effects on concrete creep. If the goal is to increase strength, creep will be reduced; if the goal is to save cement and maintain strength, creep will be largely unaffected; if the goal is to increase fluidity, creep will increase. The use of air-entraining agents will cause air bubbles to form in the concrete, and these air bubbles are equivalent to aggregates with a zero elastic modulus, leading to increased creep. Water-reducing agents are the most widely used concrete admixtures. Studies on water-reducing agents have found that, compared to naphthalene-based water-reducing agents, several polycarboxylate water-reducing agents tested had a stronger effect on reducing creep.

[0074] 1.2.5, Steel Fiber: Concrete has low tensile strength, making it prone to cracking and lacking toughness. It has been found that adding fibers can solve this problem. Steel fibers, with their high strength and high modulus of elasticity, are the most widely used in engineering. The bond between steel fibers and the concrete matrix can inhibit creep. Creep tests on cement paste, mortar, and concrete incorporating steel fibers show that steel fibers can suppress creep at all stress levels, with the best suppression effect at lower stress levels and higher fiber content.

[0075] Steel fibers can effectively reduce concrete creep and also mitigate the effects of temperature on creep. Studies on the influence of steel fiber length and dosage on concrete creep revealed that proper selection of fiber length and dosage is essential to reduce creep; otherwise, it may even increase creep. The effect of steel fiber dosage on concrete creep was investigated, finding that steel fibers effectively inhibit concrete creep, but the inhibitory effect initially increases and then decreases with increasing fiber dosage, indicating the existence of an optimal dosage value. Therefore, adding a reasonable dosage of steel fibers can effectively reduce concrete creep.

[0076] 1.2.6 Component Shape and Dimensions: The rate of moisture evaporation from concrete is related to the shape and size of the structural member. Therefore, concrete creep is affected by the size effect, the magnitude of which is related to the volume-to-surface area ratio (V / S). Generally, a higher V / S value indicates a lower creep rate. However, when the member thickness exceeds 90 cm, the concrete's ability to achieve humidity equilibrium with the environment reaches its limit, and the size effect on creep becomes very small and can be essentially ignored. Furthermore, the final creep value of concrete is also affected by the size effect; a higher V / S value results in a smaller final creep value.

[0077] 1.2.7 Load Conditions: Loading age, loading stress, and holding time all affect concrete creep. The degree of cement hydration and concrete strength increase with age, therefore concrete creep decreases with increasing loading age. Under the same conditions, concrete loaded at earlier ages exhibits greater creep than concrete loaded at later ages. Using 28-day-old concrete as a reference standard, the creep values ​​of concrete loaded at 3 days and 7 days are approximately 1.6-2.3 times and 1.5 times that of 28-day-old concrete, respectively; the creep values ​​of concrete loaded at 90 days and 1 year are approximately 0.7 times and 0.34-0.5 times that of 28-day-old concrete, respectively. The magnitude of the loading stress determines the ratio of the stress to the ultimate strength of the concrete during creep (stress-strength ratio). It is generally believed that when the stress-intensity ratio does not exceed 0.4, the creep of concrete is linearly related to the stress-intensity ratio; when the stress-intensity ratio is greater than 0.4, creep develops rapidly with the increase of the stress-intensity ratio, exhibiting a significant nonlinear relationship; if the stress-intensity ratio exceeds 0.75, existing cracks inside the concrete will accelerate their propagation and continuously generate new cracks until various cracks widen and interconnect, ultimately leading to concrete failure. As the holding time increases, concrete creep gradually increases, but the rate of increase gradually decreases. Most concrete creep deformation is completed within 1-2 years of holding the load, but the development of creep can continue for decades. Experimental results show that, using 1 year of holding load as a reference, after holding loads for 2, 5, 10, 20, and 30 years, the creep of concrete is 1.14 times, 1.20 times, 1.26 times, 1.33 times, and 1.36 times the reference value, respectively.

[0078] 1.2.8 Environmental conditions: Ambient temperature and humidity affect cement hydration and moisture exchange between concrete and the surrounding environment, making them key factors influencing concrete creep. Lower relative humidity generally leads to greater concrete creep, and the initial creep rate increases with decreasing relative humidity. After a certain period, once the internal humidity of the concrete reaches equilibrium with the ambient humidity, ambient humidity no longer affects creep. It is generally believed that increased temperature decreases the viscosity and modulus of elasticity of concrete, thus increasing creep. When the ambient temperature is between 20°C and 60°C, the creep rate increases with temperature; however, when the ambient temperature exceeds 60°C, the creep rate decreases rapidly with increasing temperature. This may be because, at a certain temperature, moisture is desorbed from the surface of the gel particles, transforming the cement paste into a structure with stronger resistance to shear deformation.

[0079] 1.3 Creep Representation: The creep of concrete is usually expressed using the creep coefficient, creep degree, or creep function. The creep coefficient refers to the creep rate of concrete at the calculated time (…). t The creep strain generated by the loading time (t The ratio of elastic strain to elastic strain φ is given by the symbol φ( t,t' )express: ; Where: ε c ( t,t' ) is concrete t The creep strain generated at any moment; σ c for t Stress applied at any given time; E c This refers to the elastic modulus of concrete.

[0080] There are two methods for determining the elastic modulus of concrete when calculating creep. One method is to select the elastic modulus of concrete at 28 days, and the other method is to select the value at the time of concrete loading. t The elastic modulus of creep. The CEB-FIB series and GL2000 creep prediction models use the first type, while the ACI209 series and RILEMB3 models use the latter. Creep refers to the degree of creep from the loading time (…). t ') to calculation time ( t The creep strain produced by concrete under unit stress is denoted by the symbol ( ). C ( t,t' )express: ; Where: ε c ( t,t' ) is concrete t The creep strain generated at any moment; σ c for t 'Stress applied at any given time.'

[0081] Creep function refers to the instantaneous elastic strain of concrete at the moment of loading under unit stress. t The sum of creep and strain generated at every moment, represented by symbols J ( t,t' )express: ; Three methods of representing creep describe the creep of concrete from different perspectives and are used in different creep prediction models. To facilitate the analysis of experimental data and comparison between different creep prediction models, the conversion formulas between the creep coefficient, creep degree, and creep function are provided: ; In the formula, The creep coefficient is . For Xu Biandu, This is the creep function.

[0082] 1.4 Commonly used creep prediction models: There are many existing models for calculating, analyzing, and predicting the long-term creep of concrete. These models can be divided into two categories: rheological models (material constitutive models) and mathematical models. Rheological models explain the nature of concrete creep from the perspective of energy conversion and can derive corresponding mathematical expressions for creep. However, these models are often quite complex and difficult to apply directly in practical engineering. Mathematical models are empirical formulas established using mathematical methods based on a large amount of accumulated experimental data, relating creep to material mix design, environmental conditions, and load conditions. Based on known conditions, the future creep deformation of concrete can be predicted, providing important reference for engineering design and construction, and these models have been widely adopted by standards.

[0083] Currently, the most influential creep prediction models include ACI209 (92) model, RILEMB3 (1995) model, CEB-FIP (90) model, MC1990 (99) model, GZ (1993) model, GL2000 model, and China Academy of Building Research (1986) model.

[0084] 1.4.1 ACI209(92) model: The ACI209(92) model uses the creep coefficient to characterize concrete creep, which is expressed as the product of the ultimate creep coefficient and the time factor. The calculation expression is as follows: ; In the formula, The limiting creep coefficient: ; In the formula, This is a correction factor for the load age; This is a correction factor for relative humidity. This is a component size correction factor; This is the slump correction factor; This is a correction factor for the fine aggregate content; This is a correction factor for air content. ; ; ; In the formula: V / S is the ratio of the component's volume to its surface area exposed to air (mm); RH is the ambient relative humidity (%).

[0085] ; In the formula: S is the slump of fresh concrete (mm).

[0086] ; In the formula: This refers to the percentage (%) of fine aggregate in the total aggregate content of concrete.

[0087] ; In the formula, This represents the volume (%) of air content in the concrete.

[0088] Each coefficient in this model has a clear mathematical expression and can be directly applied to calculations. It also fully considers the influence of internal factors such as mix proportion, workability, and curing conditions on concrete creep. However, some studies have shown that its predictive accuracy is not high.

[0089] 1.4.2, RILEMB3 (1995) model: The B3 model is a simpler creep model based on the theory of concrete curing and improved upon the BP and BP-KX models. This model uses a creep function to characterize concrete creep and considers creep in two parts: basic creep and drying creep. Its calculation expression is as follows: J ( t , t ′)= q 1+ C 0( t , t ′)+ C d ( t , t ′, t 0); In the formula, The strain produced under unit stress; The basic creep degree; C d ( t , t ′, t 0) represents the drying creep degree, when The time value is 0; This is the drying start time.

[0090] ; In the formula, The standard cylinder has a 28-day compressive strength (MPa).

[0091] ; In the formula, , , These are coefficients to be determined; It is a time function.

[0092] ; ; ; ; ; ; ; In the formula, c is the cement content (Kg / m3); w / c is the water-cement ratio; a / c is the ratio of aggregate to cement content; m and n are constants, with values ​​of 0.5 and 0.1 respectively. If other cementing materials are added, the above c is replaced by b, which is the cementing material content. , , These are different intermediate functions in the time function fitting process; m、n These represent different material constants for creep calculations, but for concrete, they are taken as 0.5 and 0.1 respectively. t 'For the loading age, d; t To calculate the age during creep, d; t 0 represents the drying start time (when the ambient relative humidity is below 100%), d.

[0093] ; ; ; ; ; ; ; In the formula, α 1 represents the cement type coefficient, with 1.0 for ordinary Portland cement (Class I cement), 0.85 for sulfate-resistant Portland cement (Class II cement), and 1.1 for rapid-hardening Portland cement (Class III cement). α 2 represents the curing condition coefficient, with 0.75 for steam curing, 1.2 for standard curing, and 1.0 for underwater curing; w Water consumption for concrete (Kg / m3); k s The cross-sectional shape factor is 1.0 for plates, 1.15 for cylinders, 1.25 for prisms, 1.30 for spheres, and 1.55 for cubes. This model comprehensively considers the effects of cement type and dosage, water-cement ratio, coarse and fine aggregate content, component shape and dimensions, concrete mechanical properties, environmental factors, and load factors on concrete creep, making the model closer to reality and achieving high calculation accuracy. For drying creep, This is the shrinkage effect coefficient. The constant is the shrinkage strain. It is a function of relative humidity. The coefficient representing the effect of drying shrinkage time. k t The first intensity influence coefficient, This is the size effect coefficient. V For the volume of the specimen, S This represents the surface area of ​​the specimen exposed to air.

[0094] 1.4.3, CEB-FIP(90) model and MC1990-99 model: The CEB-FIP (90) model is a concrete creep prediction model. The MC1990 (99) model is an improvement on the CEB-FIP (90) model, making it applicable to high-strength concrete. This model uses the creep coefficient to characterize concrete creep, expressed as the product of the nominal creep coefficient and the time development function. The calculation expression is as follows: ; In the formula, The nominal creep coefficient; This is a time evolution function. t 'This refers to the loading age.'

[0095] ; ; ; ; ; ; ; ; ; In the formula, RH represents the compressive strength of concrete at 28 days (MPa); RH represents the relative humidity of the surrounding environment (%). h Equivalent section height (mm); Ac The cross-sectional area of ​​the component is (mm²). u The perimeter of the component is in mm. α 1. α 2. α 3 is Strength correction factor for pressure greater than 35 MPa; Loading age (d) to account for temperature adjustment; The temperature is T Number of days (d); for The average temperature over a given time period; α The cement type coefficients are taken as -1 (32.5N), 0 (32.5R, 42.5N), and 1 (42.5R, 52.5N, 52.5R), respectively. This model considers the effects of component size, ambient temperature and humidity, concrete strength, and cement type on creep, but does not consider curing methods. Studies on CEB-FIP (90) have demonstrated its high accuracy. This is the second intensity influence coefficient. This is the temperature influence coefficient. The nominal creep coefficient, Humidity influence coefficient For time evolution function, This is a comprehensive correction factor for humidity, intensity, and size.

[0096] 1.4.4, GZ (1993) model and GL2000 model: Both the ACI and CEB-FIP models employ the concept of ultimate creep, which contradicts many long-term experimental results. Based on the analysis of extensive long-term concrete creep data, the GZ model was proposed. This model also uses the creep coefficient to characterize concrete creep and considers factors such as relative humidity, age, member size, 28-day compressive strength, and loading, but it does not account for the effects of mix proportions and aggregates. In 1999, based on the shrinkage and creep prediction model criteria adopted by the ACI209 committee, the GL2000 model was proposed as an improvement on the GZ model. The calculation expression is: ; In the formula, t , t 'and t0 Calculate the age (d), loading age (d), and drying age (d) respectively; RH is the ambient relative humidity (%); V / S is the component surface area ratio (mm). t '= t0 hour, =1; when t '> t0 When the time is right, the calculation formula is: ; The GL2000 model is simple with few parameters, only needing to consider the effects of age, relative humidity, and component size on creep. However, it introduces a separate correction term to account for the effect of pre-loading drying. This model can adapt to different concrete materials and shows good agreement with experimental data, but it has a large error in predicting early-age loading conditions.

[0097] 1.4.5. Model from the China Academy of Building Research (1986): This model primarily addresses the current state of concrete usage. Through statistical analysis of experimental data, it establishes the relationship between the creep coefficient and various key influencing factors. The calculation expression is as follows: ; In the formula, The basic equation for creep of ordinary concrete or lightweight aggregate concrete; k 1 represents the influence coefficient of ambient relative humidity (RH), which is 1.30 when RH is 40%, 1.00 when RH is 60%, and 0.75 when RH is 80%. k 2 is the influence coefficient of the component cross-sectional dimensions, which can be taken from Table 1; k 3 is the influence coefficient of curing conditions, which is 1.0 for standard curing and 0.85 for steam curing; The loading age influence coefficient is set to 1.40, 1.20, 1.00, and 0.70 for loading ages of 7d, 14d, 28d, and 90d, respectively. k 5 represents the influence coefficient of the amount of cement replaced by fly ash. It is taken as 1.00 when the replacement rate is 0% and 0.85 when the replacement rate is between 10% and 20%. k 6 is the influence coefficient of concrete strength grade, selected according to Table 2.

[0098] ; ; Table 1 Influence coefficient of cross-sectional dimensions Table 2 Influence coefficient of concrete strength grade The coefficients are simple to determine and easy to use, taking into account the effects of environmental conditions, component size effects, curing methods, loading age, fly ash content, and concrete strength grade on creep. V / S is the ratio of volume to surface area.

[0099] 1.5 Analysis of factors affecting creep in GRPC: The summary of the above analysis of factors influencing concrete creep and the prediction models shows that these theories are all proposed for ordinary (high-strength) concrete and have certain applicable ranges (see Table 3). The mix proportion of reactive powder concrete (RPC) differs from that of ordinary concrete. Coarse aggregates, which have a significant impact on creep, are removed from the composition, the proportion of cementitious materials is greatly increased, and the water-cement ratio is extremely low. These factors determine that the creep performance of RPC will differ significantly from that of ordinary concrete. The incorporation of recycled micropowder and steel fibers are two other important factors affecting creep. Therefore, experimental research is needed to understand the creep characteristics of RPC and establish a corresponding creep prediction model.

[0100] 2.1 Specimen Preparation: The creep test considered the effects of recycled micro-powder and steel fiber on the creep performance of GRPC. The mix proportions used are shown in Table 3. The cubic compressive strength test used 100mm×100mm×100mm specimens, three for each mix proportion; the axial compressive strength and elastic modulus tests used 100mm×100mm×300mm specimens, six for each mix proportion; the creep test used 100mm×100mm×400mm specimens, five for each mix proportion, three of which were used to test drying shrinkage deformation and two to test creep deformation.

[0101] Table 3 Mix proportions for GRPC compression creep test 2.2 Mechanical property test: Mechanical property tests were conducted on a compression testing machine, such as the SYE-2000 model, for a testing age of 28 days. The loading rate for the cube and prism compressive strength tests was 1.0 MPa per second, and the arithmetic mean of the test results for three specimens was taken as the strength value. Elastic modulus tests were performed based on the completed axial compressive strength test results. The loading rate was also 1.0 MPa per second, and three pre-compression processes were performed. The elastic modulus was calculated as the arithmetic mean of the test results for three specimens. If the difference between the axial compressive strength measured in the elastic modulus test of one specimen and the previous axial compressive strength exceeded 20%, the average of the measurements from the other two specimens was taken.

[0102] 2.3 Creep Test: This test uses a concrete creep meter, such as YC-XB50S, for creep testing. The instrument is a spring-type compression creep meter with a maximum test pressure of 500kN. The creep meter spring height is 300mm and the compression stroke is greater than 20mm. The creep meter is equipped with a pressure sensor and digital display controller, such as JLBU-50T, which can display the loading pressure in real time. The creep loading is done with a manual jack. Shrinkage and creep deformation are measured using a Harbin 8-drill anti-vibration dial gauge with a range of 0-1mm and an accuracy of 0.001mm. The dial gauge is used with a special fixed bracket, and the measuring gauge distance between the brackets is 200mm. The creep test is conducted in an indoor room to simulate natural conditions without temperature and humidity control. A temperature and humidity meter, such as WSB-5-H2, is used to monitor the ambient temperature and relative humidity. The specific steps of the GRPC creep test are as follows: (1) The loading age of the creep test is 28d. One day before the creep test, the creep meter, digital display controller, jack and other test equipment were installed and debugged, and the dial gauge was symmetrically fixed on the side of the specimen. Since the deformation of the specimen may be repeated, the initial value of the dial gauge was adjusted to 0.200 mm. On the day of loading, the axial compressive strength test of the prism was carried out first to determine the creep loading control stress. (2) Before the creep test, the position of the specimen was adjusted so that the axis of the specimen, the loading device and the creep meter were aligned and coincided, and then the jack was used for loading. When loading, the load was first increased to 20% of the creep control stress, and the readings of the dial gauges on both sides of the specimen were checked. If the difference in deformation on both sides did not meet the requirement of less than 10% of its average value, the specimen needed to be unloaded and the position of the specimen readjusted, and the load was reloaded until the deformation difference met the requirements. After the calibration was qualified, the load was continued to the creep control stress, and the readings of the dial gauges on both sides were read in time. The average value was taken as the initial deformation value under the stress. At the same time, the load value displayed by the digital display controller was recorded. The time between the completion of calibration and the initial deformation value should not exceed 1 minute. Considering the high strength of GRPC and the working range of the creep tester's spring, 30% of the axial compressive strength was taken as the creep control stress in this test. (3) After completing the creep loading, the dial gauge readings of each comparative shrinkage specimen in the creep chamber and the temperature and humidity values ​​of the creep chamber were recorded simultaneously. (4) The creep test lasted for one year, and the creep data were recorded regularly. Creep data were recorded 2 hours and 6 hours after loading on the day of the test; once a day between 1 day and 14 days, and once every 7 days thereafter.

[0103] 2.4 Processing of creep test results: The shrinkage and creep deformation of the specimens were obtained through creep tests. The creep strain, creep degree, and creep coefficient of GRPC were calculated as follows: ; In the formula, The creep strain of the creep specimen after loading for time t (d) is given in mm / m. The total deformation of the creep specimen after loading for time t (d) is expressed in mm. The initial deformation value measured during loading is in mm; For measuring gauge length, mm, accurate to 1 mm; Shrinkage values ​​for specimens of the same age, in mm / m, accurate to 0.001 mm / m. Creep values ​​for concrete after loading time t (d), in 1 / MPa, are calculated using the following formula: ; The creep stress is expressed in MPa. ; ; In the formula: The creep coefficient of the creep specimen after loading for time t (d); The initial strain value measured during loading, in mm / m, accurate to 0.001 mm / m. For each mix proportion, the average creep strain (creep degree or creep coefficient) of two specimens is used as the creep value of that group; for each mix proportion, the arithmetic mean of the shrinkage strain of three shrinkage specimens is used as the shrinkage value of that group.

[0104] 3. GRPC creep test results: 3.1 Mechanical property test results: The 28-day mechanical property test results of each mix proportion of GRPC are shown in Table 4.

[0105] Table 4 Test results of mechanical properties of GRPC As shown in Table 4, the incorporation of recycled micro powder reduces the mechanical properties of GRPC, while the addition of steel fibers improves the mechanical properties of GRPC.

[0106] 3.2 Changes in ambient temperature and humidity: The changes in temperature and relative humidity over time during creep, such as Figure 3 As shown. By Figure 3 It can be seen that the temperature and humidity fluctuations in the creep chamber were relatively small during the test period. The temperature varied between 13°C and 35°C, and the relative humidity varied between 51% and 71%. Based on the pore structure test results mentioned earlier, GRPC itself has high density, and the majority of its pores are small-diameter gel pores. Therefore, it can be considered that the ambient temperature and humidity have little effect on the shrinkage of GRPC, and the resulting drying creep can be ignored.

[0107] 3.3 Shrinkage test results: The shrinkage strain of GRPC changes over time, such as Figure 4As shown. It should be noted that the shrinkage here does not refer to the total shrinkage after GRPC molding, but only to the shrinkage value after 28 days of curing, mainly due to drying shrinkage caused by internal moisture loss of the material. Figure 4 It can be seen that, overall, the shrinkage of GRPC increases with age, exhibiting a rapid growth rate in the early stages and a slower growth rate in the later stages. The one-year shrinkage strains of GRPC0, GRPC3, and GRPC3-STF are 136.7µε, 86.7µε, and 61.0µε, respectively, far lower than the shrinkage values ​​of ordinary concrete (200-1000µε). Comparing the curves of GRPC0 and GRPC3 shows that recycled micronized powder can reduce the shrinkage of GRPC. The drying shrinkage of concrete is essentially the shrinkage of its internal hydration products. Aggregates and unreacted cementitious materials play a role in limiting shrinkage. After recycled micronized powder partially replaces cement, the cement content in GRPC is relatively reduced, and the hydration products are correspondingly reduced, resulting in less shrinkage. At the same time, with the increase of age, the pozzolanic activity of recycled micronized powder gradually develops, and the secondary hydration reaction refines the pore size of GRPC, which can reduce the rate of water loss and also inhibit shrinkage. Comparing the curves of GRPC3 and GRPC3-STF, it can be seen that steel fibers can also reduce the shrinkage of GRPC. In terms of the rate of shrinkage development, steel fibers can reduce the early shrinkage rate and delay the occurrence of shrinkage; this effect persists throughout the entire shrinkage process. On the one hand, steel fibers can disperse the capillary shrinkage stress within the material, alleviating localized stress concentration caused by shrinkage; on the other hand, there is a certain bonding force between the steel fibers and the GRPC matrix, which can, like aggregate, hinder the deformation of the slurry, thus inhibiting shrinkage. 3.4. Creep Degree, Creep Coefficient, and Creep Function: The creep strain, creep degree, and creep coefficient of GRPC can be calculated from the creep test results. Since GRPC0, GRPC3, and GRPC3-STF have different strengths and different creep loading control stresses, creep strain is not suitable for comparing their performance differences. The creep coefficient is related to the elastic modulus of the material and cannot be directly used for comparing creep performance. Therefore, this invention uses the creep degree to analyze and evaluate the influence of recycled micro-powder and steel fiber on the creep performance of GRPC. The relationship between the creep degree and holding time for each mix proportion is shown below. Figure 5 As shown. By Figure 5It can be seen that the creep of GRPC increases with the duration of loading, with a faster growth rate in the early stage, gradually slowing down and converging in the later stage. After one year of loading, the creep values ​​of GRPC0, GRPC3, and GRPC3-STF are 16.4 MPa / µε, 16.0 µε / MPa / µε, and 23.3, respectively, which are classified as low-creep concrete. By calculating the difference in creep values ​​between GRPC0 and GRPC3, and between GRPC3 and GRPC3-STF, the influence of recycled micro-powder and steel fibers on the development of GRPC creep can be observed. Figure 6 As shown.

[0108] contrast Figure 5 The creep curves of GRPC0 and GRPC3 show that the incorporation of recycled micronized powder increases the creep of GRPC. Figure 6 It is evident that the effect of regenerated micronized powder on GRPC creep varies with loading time. In the early stages of creep development, regenerated micronized powder leads to rapid creep development in GRPC, exhibiting an accelerating effect (0-91 days). After loading time exceeds 91 days, the regenerated micronized powder gradually reduces the creep development rate of GRPC, decreasing the later creep growth. This is because the regenerated micronized powder has low activity in the early stages, resulting in lower early strength of GRPC3. After 28 days of loading, the initial creep growth rate is rapid, leading to significant creep. As the age increases, the regenerated micronized powder gradually becomes more active, and GRPC3 exhibits greater strength growth than GRP0. The creep development rate gradually decreases, manifested as a reduction in the later creep growth, eventually stabilizing. Figure 5 The creep curves of GRPC3 and GRPC3-STF show that steel fibers can significantly reduce the creep of GRPC. Figure 6 It is evident that steel fibers significantly inhibit the creep development of GRPC in the early stages. This inhibitory effect gradually decreases and eventually stabilizes with increasing loading time. For ordinary concrete, the strength of the interface layer is insufficient to limit concrete shrinkage and creep in the early stages. As the concrete strength increases in later stages, the bonding force of the interface layer increases, allowing steel fibers to effectively limit the deformation of the paste. For GRPC, the extremely low water-cement ratio and high density result in a smaller interface layer volume, and GRPC can achieve much higher strength than ordinary concrete in the early stages. Therefore, the randomly distributed steel fibers initially inhibit paste deformation, hinder the occurrence and development of internal microcracks, and reduce creep deformation. GRPC mixed with recycled micropowder exhibits rapid creep development and a large creep increment in the early stages of loading, demonstrating a significant fiber inhibition effect. As the creep increment gradually decreases and converges in later stages, the inhibitory effect of steel fibers also gradually decreases and stabilizes. Creep coefficients and creep functions are commonly used in creep prediction models to characterize concrete creep. The relationships between the creep coefficient, creep function, and loading time of GRPC are shown below. Figure 7 and Figure 8 As shown, the creep coefficient reflects the ratio of creep deformation to elastic deformation at the moment of loading. Due to the incorporation of steel fibers, GRPC3-STF has a high elastic modulus and relatively small elastic deformation, thus exhibiting a large creep coefficient. The creep coefficients of GRPC0 and GRPC3 are more similar. The creep function reflects the sum of creep deformation and elastic deformation, i.e., the total deformation of the material under load. A comprehensive comparison reveals that GRPC3-STF has the smallest total deformation and the fastest convergence speed, followed by GRPC0, while GRPC3 exhibits the largest deformation.

[0109] 4. GRPC compressive creep prediction model: Creep is a crucial time-varying characteristic of concrete. Creep deformation significantly impacts the normal use and safety of concrete structures. However, concrete creep typically continues for years or even decades. Therefore, various concrete creep prediction models have been proposed to provide important references for the design and construction of concrete structures. However, most existing creep prediction models are based on statistical analysis of creep data for ordinary concrete, and their applicability to GRPC needs further verification.

[0110] 4.1 Comparison of GRPC creep test results with existing model predictions: Most commonly used creep prediction models are based on experimental results of ordinary concrete. Considering the influencing factors of creep, prediction expressions are proposed through statistical analysis of a large amount of data. The relevant parameters are mostly obtained by regression fitting of existing data, so the scope of application is limited. With the development of modern concrete technology, various high-strength and high-performance concretes with low water-cement ratio, high strength, and high density have emerged. The changes in material composition, mix proportion and curing method have also led to significant differences in their creep characteristics. Creep develops rapidly in the early age period but quickly stabilizes. Long-term creep and creep final value are significantly lower than those of ordinary concrete. In order to study the applicability of existing commonly used creep prediction models to predict the creep performance of GRPC, the experimental results of this creep test were compared with the prediction results of the ACI209 (92) model, RILEMB3 (1995) model, MC1990 (99) model, GL2000 model and JKY1986 model. The parameters used in each model were selected based on the actual mix proportions, curing conditions, component geometry, mechanical properties, loading age, and environmental conditions of GRPC. Some parameters were obtained through conversion. All models uniformly used the creep coefficient as the reference index. Since the RILEMB3 (1995) model uses a creep function expression, conversion between the creep coefficient, creep degree, and creep function is required. The comparison results of GRPC0, GRPC3, and GRPC3-STF are as follows: Figure 9 , Figure 10 and Figure 11As shown.

[0111] Depend on Figures 9-11 It can be seen that the comparison results of the various models are similar for all mix proportions. Among them, the MC1990 (99) model is closest to the experimental results, while the prediction results of the ACI209 (92) model, RILEMB3 (1995) model, GL2000 model and JKY1986 model all have large deviations and overestimate the creep of GRPC. The ACI209 (92) model adopts the form of time function and ultimate creep product, which comprehensively considers the internal and external factors affecting concrete creep. The reason for overestimating the creep of GRPC is that its creep limit coefficient is too large (2.35). Based on the existing RPC data, the coefficient was adjusted (0.82), and a better prediction accuracy was achieved. Although the RILEMB3 (1995) model has good prediction accuracy, it has the largest deviation from the experimental results in this paper. The analysis found that it was caused by the excessive value of the undetermined coefficient 4q in the creep expression of the model, which reflects the ratio of aggregate and cement (cementing material) content. Therefore, it is judged that for GRPC with coarse aggregate removed, the model overestimates the effect of aggregate on creep limitation. The GL2000 model has the best prediction accuracy, but it is also not applicable to GRPC. The GL2000 model only considers the effects of age, relative humidity and component size on creep, and its calculation expression is more applicable to predicting the creep of low-strength ordinary concrete. However, since it fails to reflect the influence of mix proportion and strength on creep, it will have significant prediction deviations for high-strength GRPC that exceeds its applicable range, and it is difficult to make targeted corrections to the model. The JKY1986 model and the ACI209 (92) model have the same calculation expression, and the function used for the limit creep in the two models is also the same, only the parameter values ​​are slightly different, and the adjustment coefficients used are also very similar. The only difference is that the JKY1986 model introduces the influence factor of fly ash content. The reason why the model predicts GRPC too high is that its applicable range is limited to concrete below C40 grade, and the strength grade influence coefficient k is too high (the minimum value is 0.80). The MC1990 (99) model provides relatively accurate predictions for GRPC. This is because, based on the original CEBFIP (1990) model, it introduces three strength correction coefficients, expanding its applicability to high-strength concrete, thus resulting in better consistency in the predictions. Comparing the prediction results of GRPC0, GRPC3, and GRPC3STF, it can be found that the model's predicted value for creep decreases with increasing material strength. The predicted values ​​for GRPC0 and GRPC3 are higher than the experimental values, while the predicted value for GRPC3-STF is lower than the experimental values, therefore, the model still needs to be corrected.

[0112] 4.2. GRPC creep prediction model based on the modified MC1990(99) model: As can be seen from the above analysis, MC1990 (99) is suitable for high-strength concrete, and its predicted value has the smallest deviation from the measured value of GRPC. Therefore, this invention is based on this model and proposes a creep prediction model suitable for GRPC. GRPC has the characteristics of low water-cement ratio, high cementitious material content, use of silica fume, and no coarse aggregate, so its mechanical properties and long-term performance are quite different from ordinary mix concrete. The creep influencing factors mainly considered by the MC1990 (99) model are environmental conditions, cross-sectional shape and size, and concrete strength. It is precisely the introduction of three strength influencing factors that makes it suitable for high-strength concrete. Since environmental conditions and component cross-sectional shape and size do not reflect the difference between GRPC and ordinary concrete, the improvement of the model is mainly reflected in the correction of the strength factors. First, the model is corrected based on the data of GRPC0. Comparing the original model prediction curve and the experimental measured curve, it can be seen that the two have similar shapes, but the predicted value of MC1990 (99) is too large. Therefore, the time development function needs to be corrected here. Adjustments are made. The exponential term (0.3) has a significant impact on the function value, so it is modified and set as an undetermined exponent (α). In the time evolution function, only the intensity correction coefficient is related to intensity. α 3. Therefore, an intensity correction factor is introduced. γ 3. Corrected time evolution function and intensity correction coefficient α 3r is calculated using the following formulas.

[0113] ; ; In the formula, The environmental influence factors are related to curing humidity conditions and component cross-sectional dimensions. A new GRPC creep calculation expression is obtained using the modified parameters, and numerical fitting is performed based on GRPP0 experimental data. The results are as follows: Figure 12 As shown in the figure, the fitted value of the exponent α is 0.5, and the intensity correction factor... γ The fit value is 1.6. The modified model has a high goodness of fit, with a coefficient of determination (R²) greater than 0.96 and a residual sum of squares (RSS) of only 0.0012.

[0114] When the modified model was used to predict GRPC3, some bias was still found. This is because 30% recycled micronized powder was used to replace cement in the cementitious material. The difference in hydration properties between recycled micronized powder and cement determines their different effects on creep. A strength correction factor was introduced into the new model. γ 1. Intensity Influence Factors α1. Make corrections to obtain a new intensity influence factor. α 1r It is calculated using the following formula: ; Using the modified time development function and different intensity correction coefficients α 1r and α 3r A new GRPC creep calculation expression was obtained, and numerical fitting was performed based on the experimental data of GRPP3. The results are as follows: Figure 13 As shown. By Figure 13 It can be seen that the intensity correction factor γ The fit value for 1 is 0.46. The fit is good, and the coefficient of determination R² is [value missing]. 2 (COD) is greater than 0.99, and the residual sum of squares (RSS) is only 0.0136.

[0115] Depend on Figure 11 It is evident that the unmodified model predicted lower GRPC-STF values ​​than the experimental values. The modified model further amplifies this bias. Previous creep models did not consider the influence of fibers on creep independently. However, numerous studies have shown that steel fibers have a significant impact on creep. Therefore, a strength correction factor was introduced into the modified model. γ 2 factors affecting intensity α 2. After correction, a new intensity influence factor is obtained. α 2r It is obtained from the following calculation formula: ; Using the modified time development function and intensity correction coefficient α 1r , α 2r and α 3r A new GRPC creep calculation expression was obtained, and numerical fitting was performed based on the experimental data of GRPP3-STF. The results are as follows: Figure 14 As shown. By Figure 14 It can be seen that the intensity correction factor γ The fit value for 2 is 1.57. The final corrected model fits... γ The goodness of fit for parameter 2 is slightly worse than that for the previous parameters, indicating that the influence of steel fibers on creep is complex. However, the coefficient of determination R² obtained from the fitting is still relatively good. 2 The COD is still greater than 0.97, and the residual sum of squares (RSS) is 0.10363.

[0116] By adopting the modified time development function exponent ( α =0.5), different intensity correction factors γ1. γ 2 and γ 3. The compressive creep prediction model MC1990(99)R suitable for GRPC is obtained, and the calculation expression is as follows: ; Nominal creep coefficient and parameters Referring to the formula above, the strength influence coefficient is calculated using the following formula: ; γ 2 is the micro powder correction factor, which is 1.0 when no regenerated micro powder is added and 0.46 when micro powder is added; γ 3 is the steel fiber correction factor, which is 1.0 when there is no steel fiber and 1.57 when there is steel fiber.

[0117] 4.3 Evaluation and verification of the prediction accuracy of model MC1990(99)R: To more objectively evaluate the applicability of the modified model, the prediction accuracy of the MC1990(99)R model needs to be assessed. Currently, common methods for evaluating the accuracy of creep prediction models include the residual method, the mean and standard deviation method, the CEB coefficient of variation method, the CEB root mean deviation method, the CEB mean deviation method, and the B3 coefficient of variation method. Among these, the B3 coefficient of variation method (…) was proposed when evaluating the prediction accuracy of the B3(1995) model. The most widely accepted method is the B3 coefficient of variation method, which divides the experimental data into segments based on age (0-10d, 11-100d, 101-1000d, and 1001-10000d) and calculates the overall coefficient of variation (%) using a weighted average. ,%).by As an evaluation indicator The smaller the value, the higher the degree of agreement between the model's calculated value and the experimental results, that is, the higher the calculation accuracy. It is calculated by the following formula: ; ; ; ; In the formula, N This represents the total number of data sets selected. For the first j coefficient of variation of the group data; n This represents the total number of data points in the j-th data group; For the first j Group of data i Weights of each data point; For the first jThe first set of data i The absolute difference between the experimental value and the model prediction value; Jij For the first j The first set of data i Point test value; nd For the first j The dataset includes the number of age segments; nk For the first k Data points for each age group.

[0118] According to the above calculation method, the three groups of data, GRPC0, GPRC3, and GRPC3-STF, are divided into three age ranges: 0-10d, 11-100d, and 101-1000d. All measured data points are selected for each group for calculation. The calculation results of the B3 coefficient of variation of the MC1990(99)R model are shown in Table 5. Existing research results show that the prediction accuracy of the creep model is 20%-30%, which is considered excellent. As shown in Table 5, the B3 coefficient of variation of the MC1990(99)R model is only 12.5%, indicating that the model has good prediction accuracy for GRPC. In particular, the prediction is most accurate for GRPC3 mixed with recycled micro powder. Only 4.6%.

[0119] Table 5. Coefficient of variation of B3 in MC1990 (99) R model ( Calculation results To verify the applicability of the MC1990(99)R model to RPC (UHPC), creep test results with similar materials and test conditions were selected as verification data to validate the prediction accuracy. The loading age (t') of the creep tests was 28 days, the ambient humidity (RH) was 60%, and the cross-sectional shape characteristics were 2. A c / u All values ​​were 50. Other relevant experimental parameters are shown in Table 6. The measured values ​​of Xu's literature experiment, Cui Cunsen's RPC (SC) literature experiment, and Chen's literature experiment are compared with the calculated values ​​of the MC1990 (99) R model. Figures 15-17 ,in, Figure 15 (a) shows a comparison between the measured values ​​from the experiment in the Xu C16SF0 literature and the calculated values ​​from the MC1990 (99)R model. Figure 15 (b) is a comparison between the measured values ​​from the Xu 16SF2 literature experiment and the calculated values ​​from the MC1990(99)R model. Figure 15 (c) is a comparison between the measured values ​​of the experiment in the Xu C22SF2 literature and the calculated values ​​of the MC1990 (99)R model; Figure 17 (a) shows a comparison between the measured values ​​of ChenRPC-Ⅰ and the calculated values ​​of the MC1990(99)R model. Figure 17 Table (b) compares the measured value of Chen's RPC-V with the calculated value of the MC1990(99)R model. The calculated B3 coefficient of variation values ​​for each literature data are shown in Table 5. As can be seen from the above tables, the MC1990(99)R model has high prediction accuracy for the experimental results of Xu and Cuicunsen, but it has a large deviation in predicting Chen's experimental results, especially for RPC-V. The creep rate was as high as 128.2%. The reason for this is that the curing conditions of the first two are the same as those in this paper, while the RPC in Chen was steam-cured at 90℃. It can be seen that the curing conditions have a very important influence on the creep of RPC, and the steam curing conditions amplify the influence of steel fibers on creep. Overall, the modified MC1990(99)R model has good prediction accuracy for RPC (UHPC), so it can be generally applied to the creep prediction of this type of material. At the same time, if the model is to be extended to a wider range of applications, the influence factor of curing conditions on creep should be introduced.

[0120] Table 6. Relevant parameters from literature experiments used to validate the prediction accuracy of the MC1990(99)R model. V f This refers to the volumetric content of steel fibers. σ / f cm The stress intensity ratio is used; the original data from Cui Cunsen's literature is the creep degree, which is converted into the creep coefficient through the conversion formula between the creep coefficient, creep degree and creep function.

[0121] 4.4. Ultimate creep value of GRPC: Most of the creep of concrete is completed within 1-2 years of holding load, after which the growth rate decreases significantly, but it can still continue for a long time. Excessive creep deformation can cause serious prestress loss and stress redistribution in statically indeterminate structures, which may endanger the safety of the structure. Therefore, the ultimate creep value of concrete is an important reference indicator that must be considered when designing a structure. In the prior art, the 10-year creep coefficient of concrete is taken as the ultimate creep value, while considering the effects of loading age, relative humidity and component cross-sectional properties. According to the previous analysis, the drying creep caused by changes in environmental humidity can be ignored, so it is considered that the creep coefficient of GRPC can ignore the effects of relative humidity and component shape. Based on the GRPC creep prediction model MC1990(99)R proposed in this invention, the creep coefficients of GRPC with a loading age of 28 days and holding loads of 1, 3, 5, 10, 15 and 20 years were calculated, and the results are shown in Table 7. Table 7 shows that the creep coefficient of GRPC gradually increases with the duration of loading, but even after 20 years of loading, none of the creep coefficient values ​​exceed 1. Taking the creep value at one year of loading as the baseline, the annual growth rate of the creep coefficient with the duration of loading is as follows: Figure 18 As shown. By Figure 18 It is known that the creep growth rate decreases significantly after 5 years of load holding. Therefore, the creep coefficient at 5 years of load holding can be regarded as the ultimate creep value of GRPC. This can be used as a reference indicator to calculate the impact of GRPC creep on the structure in structural design.

[0122] Table 7 Creep coefficient and ultimate creep value of GRPC This invention provides the generation mechanism, influencing factors, and common prediction models of concrete creep. Compressive creep tests were conducted on GRPC0, GRPC3, and GRPC3-STF according to standard testing methods. The effects of recycled micropowder and steel fibers on GRPC shrinkage and creep were investigated. The reasons for prediction deviations of each model were analyzed by comparing the calculated values ​​of commonly used prediction models with experimentally measured data. Based on the MC1990 (99) model, a modified model suitable for GRPC was proposed. The prediction accuracy of the modified model was evaluated using the B3 coefficient of variation method, and the applicability of the model was verified using literature data. The main conclusions are as follows: (1) The shrinkage test results show that the shrinkage deformation of GRPC increases with age, and has the characteristics of rapid early development and slow later development. The shrinkage deformation after 1 year is much lower than that of ordinary concrete. Both recycled powder and steel fiber can inhibit the shrinkage of GRPC.

[0123] (2) Creep test results show that the creep degree of GRPC increases with the duration of loading, with a faster growth rate in the early stage, gradually slowing down in the later stage, and tending to converge. After one year of loading, the creep degrees of GRPC0, GRPC3, and GRPC3-STF are 16.4. µ ε / MPa 23.3 µε / MPa and 16.0 µε / MPa This type of concrete is classified as low-creep concrete. The incorporation of recycled micropowder increases the creep of GRPC, and the accelerating effect on creep first increases and then decreases with the duration of load holding. Steel fibers, on the other hand, can significantly reduce the creep of GRPC, and this inhibitory effect first increases and then decreases with the duration of load holding, eventually stabilizing.

[0124] (3) A comparison of the calculated values ​​of several concrete creep prediction models with the measured values ​​of GRPC creep tests shows that most models overestimate the creep value of GRPC. Among them, the calculation results of the RILEMB3 (1995) model have the largest deviation, which is because the model overestimates the limiting effect of aggregate on creep. The calculation results of the MC1990 (99) model are closest to the measured values. This is because the model introduces a strength correction coefficient, which expands its applicable range to high-strength concrete, and therefore has better prediction accuracy for high-strength GRPC.

[0125] (4) Due to the significant difference in mix proportions between GRPC and ordinary (high-strength) concrete, and the use of recycled micro-powder to partially replace cement as a binder, the original prediction model is not suitable for predicting the creep of GRPC. Furthermore, the existing model does not consider the inhibitory effect of steel fibers on the creep performance of concrete. Based on the above analysis, the MC1990(99) model was modified by changing the exponent of the time development function and introducing three strength correction factors. The MC1990(99)R model, suitable for predicting the creep of GRPC, was obtained by fitting experimental measured data.

[0126] (5) The evaluation results of the model's prediction accuracy show that the B3 coefficient of variation of the MC1990(99)R model is relatively low, indicating that the model has good prediction accuracy for GRPC. The prediction accuracy of the model was evaluated using creep test results under similar conditions from domestic and foreign literature, and the results also verified that the model has good applicability for RPC creep prediction. At the same time, it was found that maintenance conditions have a very important influence on RPC creep, so the influence of this factor should be considered in the prediction model.

[0127] (8) The calculation results of the MC1990(99)R model show that the creep coefficient of GRPC gradually increases with the increase of the holding time, but even the maximum creep coefficient value after 20 years of holding time does not exceed 1. The growth rate of the creep coefficient of GRPC decreases rapidly with the increase of the holding time, and the creep growth rate decreases significantly after 5 years of holding time. Therefore, it is recommended that in structural design, the creep coefficient after 5 years of holding time can be regarded as the ultimate creep value of GRPC.

[0128] In summary, the creep test results in this invention show that the creep degree of GRPC increases with the duration of loading, with a rapid increase in the early stage, gradually slowing down in the later stage, and tending to converge. After one year of loading, the creep degrees of GRPC0, GRPC3, and GRPC3-STF are 16.4. µε / MPa 23.3 µε / MPa and 16.0 µε / MPa This type of concrete is classified as low-creep concrete. The incorporation of recycled micropowder increases the creep of GRPC, with the accelerating effect on creep initially increasing and then decreasing over time. Steel fibers, on the other hand, significantly reduce the creep of GRPC; this inhibitory effect initially increases and then decreases over time, eventually stabilizing.

[0129] Significant differences exist between the creep values ​​of GRPC calculated using existing concrete creep prediction models and the experimentally measured values, with most models overestimating the creep value of GRPC. This is due to the significant difference in mix proportions between GRPC and ordinary (high-strength) concrete. Considering the influence of recycled micropowder and steel fibers on creep performance, based on the MC1990(99) prediction model, the MC1990(99)R model was modified by changing the exponent of the time development function and introducing three strength correction factors, and by fitting the parameters with experimentally measured data. This modified model is suitable for predicting GRPC creep.

[0130] The evaluation of the prediction accuracy of the modified model using the B3 coefficient of variation method shows that the MC1990(99)R model has good prediction accuracy for GRPC. The prediction accuracy of the model was verified using creep test results of reactive powder concrete under similar conditions from domestic and international literature, and the results show that the model is applicable to the creep prediction of RPC. It was also found that curing conditions have a very important influence on the creep of RPC, therefore, this factor should be considered in the prediction model.

[0131] The creep coefficient of GRPC gradually increases with the increase of the load holding time, and its growth rate decreases rapidly with the increase of the load holding time. The creep coefficient value of GRPC calculated using the MC1990(99)R model shows that the creep of GRPC increases rapidly in the first 3 years of the load holding period, and after 5 years, the creep growth rate decreases significantly. The maximum creep coefficient value at 20 years of load holding time does not exceed 1. Therefore, it is recommended that in structural design, the creep coefficient at 5 years of load holding time can be regarded as the ultimate creep value of GRPC.

[0132] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for evaluating service life of a GRPC structure based on ensemble learning, characterized in that, include: By correcting the parameters of the pre-obtained GRPC basic creep prediction model, an expression for calculating the GRPC compressive creep coefficient is constructed, and the creep coefficient of GRPC under compressive conditions is calculated. A creep clock is constructed based on the development trend of the creep coefficient, and a damage clock is constructed based on the creep clock. Establish the state-space coupling relationship between the creep clock and the damage clock, and combine it with the lifetime decay observation modeling process to output lifetime decay parameters characterizing the impact of dual-clock coupling on lifetime decay, specifically including: A dual-clock synchronization variable sequence corresponding to the creep clock and the damage clock is established, including creep clock variables and damage clock variables. The creep clock variables and damage clock variables are used as the first and second state components in the state space to form a dual-clock state vector sequence. The dual-clock state vectors of adjacent loading time steps are differentially processed to obtain the creep clock state increment and the damage clock state increment. The creep clock state increment is used as the equivalent duration advancement of GRPC compressive creep, and the damage clock state increment is used as the equivalent duration advancement of GRPC compressive creep damage to form a dual-clock state increment sequence. Based on the driving relationship between the creep clock state increment and the damage clock state increment, a dual-clock coupling strength variable is constructed to form a dual-clock coupling strength variable sequence. The dual-clock state... Using vectors as state variables and dual-clock coupling strength variables as coupling input variables, a state-space coupled state equation for creep clock and damage clock is established. Based on the combined contribution of creep clock variables, damage clock variables, and dual-clock coupling strength variables to GRPC lifetime decay, a lifetime decay observation equation is established. Based on the state-space coupled state equation for creep clock and damage clock, the state vector evolved by dual-clock coupling at each loading time step is obtained recursively. The lifespan decay response quantity is mapped to the evolved state vector and dual-clock coupling strength variables through the lifetime decay observation equation. The lifetime decay response quantity is dimensionless using the initial lifetime as a normalization benchmark to obtain the lifetime decay parameter characterizing the impact of dual-clock coupling on lifetime decay. Based on the Latin hypercube sampling technique and combined with lifetime decay parameters, sensitivity analysis is performed on the coupling parameters of creep clock and damage clock to obtain a set of coupling influence parameters; Using a genetic algorithm, the parameters of the coupling effect parameter set are optimized to obtain a lifetime feature set for ensemble learning; Using a lifespan feature set as input, a lifespan prediction model for GRPC structures is constructed. The lifespan prediction model is then used to predict the lifespan of the GRPC structure to be evaluated, resulting in a lifespan assessment of the GRPC structure. 2.The ensemble learning based GRPC structure service life evaluation method according to claim 1, wherein, The process involves modifying the parameters of a pre-obtained GRPC basic creep prediction model to construct an expression for calculating the GRPC compressive creep coefficient, and then calculating the GRPC creep coefficient under compressive conditions, including: Obtain the GRPC basic creep prediction model, which includes the nominal creep coefficient and the creep time development function; The micro powder correction factor is determined based on the micro powder incorporation state of GRPC, and the steel fiber correction factor is determined based on the steel fiber incorporation state; using the micro powder correction factor and the steel fiber correction factor, combined with the average compressive strength of GRPC, the strength influence correction coefficient is calculated. The nominal creep coefficient is corrected using the intensity influence correction factor to obtain the GRPC nominal creep coefficient; the corrected creep time development function is obtained by reconfiguring the exponent of the creep time development function. Based on the nominal creep coefficient of GRPC and the modified creep time development function, an expression for calculating the compressive creep coefficient of GRPC is constructed; using this expression, the creep coefficient of GRPC under compression is calculated. 3.The ensemble learning based GRPC structure service life evaluation method according to claim 1, wherein, The creep clock constructed based on the development trend of the creep coefficient includes: Monotonicity identification and trend fitting were performed on the change sequence of creep coefficient along the loading duration to obtain the creep coefficient development function characterizing the compression creep of GRPC over time; A long-term limit analysis was performed on the creep coefficient development function, and the limit value when the loading duration tends to infinity was determined as the ultimate creep value. Using the creep endpoint as a normalization benchmark, the creep development state under any loading duration is represented in a dimensionless manner relative to the creep endpoint, and the normalized creep evolution variable is output. Normalized creep evolution variables are used as state measures of the creep process, and the actual loading duration is mapped to an equivalent time variable controlled by the creep development degree to construct a creep clock. 4.The ensemble learning based GRPC structure service life evaluation method according to claim 1, wherein, The construction of a damage clock based on a creep clock includes: Based on the creep clock, output the creep clock variable used to characterize the equivalent duration of GRPC compression creep; The creep clock variable is incrementally decomposed along the loading duration, the creep clock increment between adjacent loading moments is extracted, and a creep clock increment sequence is constructed. Damage evolution increments are constructed based on the creep clock increment sequence. The creep clock increments within each loading time step are converted into the damage growth of the structural material, and the damage state variables induced by compressive creep are obtained by accumulation. A damage clock is constructed based on the evolution ratio of the damage state variables relative to the failure damage threshold. 5.The ensemble learning based GRPC structure service life evaluation method according to claim 1, wherein, The method, based on Latin hypercube sampling and combined with lifetime decay parameters, performs sensitivity analysis on the coupling parameters of creep clock and damage clock to obtain a set of coupling influence parameters, including: The state self-evolution parameters, dual-clock interactive coupling parameters, and coupling input parameters in the state-space coupling state equation of the creep clock and the damage clock, as well as the lifetime decay mapping parameters and direct coupling influence parameters in the lifetime decay observation equation, are taken as the set of coupling parameters to be analyzed, and the sampling interval of each coupling parameter is determined. Based on the Latin hypercube sampling technique, the sampling intervals of each coupling parameter are sampled in a stratified manner, and a coupling parameter sample matrix is ​​formed by random combination. Obtain the lifetime decay parameter sample sequence; calculate the statistical response characteristics of the lifetime decay parameter under different coupling parameter sample conditions to obtain the correspondence between the change of each coupling parameter sample and the change of the lifetime decay parameter, and calculate the sensitivity index of each coupling parameter to the lifetime decay parameter. The coupling parameters are sorted from largest to smallest according to their sensitivity index, and the coupling parameters whose sensitivity index exceeds the preset sensitivity threshold are included in the coupling influence parameter set.

6. The method for evaluating the lifespan of a GRPC structure based on ensemble learning according to claim 1, characterized in that, The genetic algorithm is used to optimize the parameters of the coupling influence parameter set to obtain a lifetime feature set for ensemble learning, including: Construct a fitness function based on the lifetime decay parameter prediction error, model stability error, or cross-validation error; Based on the fitness function, genetic evolutionary optimization is performed to obtain the optimized set of coupling influence parameters. The optimized set of coupling effect parameters is combined with the lifetime decay parameters and the statistical response characteristics of the lifetime decay parameters to form a lifetime feature set.

7. The method for evaluating the lifespan of a GRPC structure based on ensemble learning according to claim 6, characterized in that, The optimization of the coupling effect parameter set, based on the fitness function and performing genetic evolutionary optimization, includes: An initial population is generated based on the range of values ​​of the coupling effect parameters. The fitness of the parameter combinations in the initial population is evaluated sequentially based on the fitness function, and the population is updated iteratively. When the iteration terminates, the combination of coupling influence parameters with the highest fitness is selected from the final population as the optimized set of coupling influence parameters.

8. The method for evaluating the lifespan of a GRPC structure based on ensemble learning according to claim 1, characterized in that, The step of constructing a GRPC structure lifespan prediction model using a lifespan feature set as input includes: Using the lifetime feature set as input features and the actual lifetime of the corresponding GRPC structure as supervision labels, a lifetime prediction sample set is constructed. Multiple life prediction base learners are trained based on the sample set, and a weighted fusion method is used to fuse the prediction results of multiple life prediction base learners to establish a GRPC structure life prediction model.

9. A GRPC structure lifespan assessment system based on ensemble learning, used to implement the GRPC structure lifespan assessment method based on ensemble learning as described in any one of claims 1-8, characterized in that, include: The creep correction module is used to construct the GRPC compression creep coefficient calculation expression by correcting the parameters of the pre-acquired GRPC basic creep prediction model, and to calculate the GRPC creep coefficient under compression. A dual-clock coupling module is used to construct a creep clock based on the development trend of the creep coefficient, and to construct a damage clock based on the creep clock; Establish the state-space coupling relationship between the creep clock and the damage clock, and combine it with the lifetime decay observation modeling process to output lifetime decay parameters that characterize the impact of dual-clock coupling on lifetime decay. The sensitivity analysis module is used to perform sensitivity analysis on the coupling parameters of creep clock and damage clock based on Latin hypercube sampling technology and combined with lifetime decay parameters, so as to obtain the set of coupling effect parameters. The parameter optimization module is used to optimize the parameters of the coupling effect parameter set using a genetic algorithm to obtain a lifetime feature set for ensemble learning. The life prediction module is used to construct a life prediction model for GRPC structures using a life feature set as input; the life prediction model is then used to predict the life of the GRPC structure to be evaluated, and the life evaluation results of the GRPC structure are obtained.

Images