Concrete permeability multi-parameter comprehensive detection system and evaluation method

Abstract

The invention discloses a concrete permeability multi-parameter comprehensive detection system and an evaluation method. The system comprises a data acquisition module, a data processing module, a comprehensive evaluation module and a man-machine interaction module, and synchronous acquisition and calibration of multiple parameters such as electric flux, a chloride ion diffusion coefficient, water absorption and an air permeability coefficient are realized through a unified synchronous trigger mechanism and an environment compensation algorithm; carrying out intelligent processing on the data by utilizing an improved data cleaning, multi-level and multi-modal feature extraction and fusion strategy; and finally, through dynamic fusion of fuzzy comprehensive evaluation, grey correlation analysis and a machine learning classification model, outputting a quantitative grade evaluation result of the concrete permeability. The method can comprehensively and accurately evaluate the permeability of the concrete, remarkably improves the detection precision and efficiency, adapts to a complex environment and a long-term aging process, and provides a scientific basis for durability evaluation and maintenance of a concrete structure.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of building structure disease detection, in particular to a permeability detection method for concrete structures of railway passenger stations. BACKGROUND

[0002] As one of the most widely used materials in modern building structures, the permeability of concrete is directly related to the durability and safety of the structure. The intrusion of harmful substances such as water and chloride ions can cause a series of diseases such as steel corrosion and freeze-thaw damage, which seriously shortens the service life of concrete structures. Therefore, accurate detection and scientific evaluation of concrete permeability has become a key technical link to ensure the long-term performance of engineering structures.

[0003] Currently, the detection of concrete permeability mainly relies on single parameter detection methods, such as electric flux method and chloride ion diffusion coefficient method. These methods can reflect the permeability characteristics of concrete to some extent, but they cannot comprehensively and systematically evaluate the overall permeability behavior. In addition, although existing technologies attempt to achieve multi-parameter synchronous detection, there are still problems such as data collection out of sync, large environmental factor interference, low data processing efficiency, lack of effective multi-source information fusion mechanism, etc. Especially, the existing evaluation models are mostly static models, which are difficult to adapt to the dynamic influence brought by environmental changes and material aging in actual engineering, resulting in insufficient accuracy and reliability of the evaluation results. SUMMARY

[0004] In view of the above problems, the present application aims to provide a concrete permeability multi-parameter comprehensive detection system and evaluation method to solve the problems of multi-parameter detection out of sync, low data processing efficiency, single fusion method, and static evaluation model. By realizing multi-parameter synchronous acquisition, environmental dynamic compensation, intelligent data processing and dynamic fusion evaluation, the present application can comprehensively and accurately evaluate the permeability of concrete, adapt to complex environmental changes and long-term aging process, not only significantly improve the detection accuracy and efficiency, but also enhance the on-site applicability and long-term monitoring capability of the system, provide scientific basis for the durability evaluation and maintenance management of concrete structures, and promote the development of concrete permeability detection technology towards intelligentization and systematization.

[0005] In order to achieve the above purpose, the present application provides a concrete permeability multi-parameter comprehensive detection system and evaluation method, which specifically comprises:

[0006] A concrete permeability multi-parameter comprehensive detection system, comprising:

[0007] A data acquisition module for synchronously acquiring multiple permeability parameter data and environmental parameter data of a concrete test piece;

[0008] The data processing module is in communication connection with the data acquisition module, and is used for cleaning, normalizing and feature extraction of the collected data.

[0009] The comprehensive evaluation module is in communication connection with the data processing module, and is used for comprehensive evaluation of the extracted features based on a dynamic fusion strategy, to generate a permeability level evaluation result.

[0010] The human-computer interaction module is in communication connection with the comprehensive evaluation module, and is used for configuration of system parameters, visual display of data and evaluation results.

[0011] Further, the data acquisition module comprises:

[0012] An electric flux test unit, a chloride ion diffusion coefficient test unit, a water absorption rate test unit and an apparent permeability coefficient test unit;

[0013] An environment sensing unit is used for collecting environmental temperature, relative humidity and air pressure data.

[0014] A synchronous control unit is used for setting a unified time window and a time delay compensation mechanism for each test unit and the environment sensing unit, to realize synchronous acquisition and time stamp alignment of multi-parameter data.

[0015] Further, the synchronous control unit synchronizes data streams of different parameters through a time interpolation resampling algorithm, and is provided with a time correction mechanism to realign data when device sampling deviates.

[0016] Further, the data processing module comprises:

[0017] A data cleaning submodule is configured to identify abnormal values using a Z-score and IQR (interquartile range) joint method, and to fill in missing values using a time weighted interpolation method, a multi-mode sliding window method or a space-time collaborative reconstruction method according to the size of a missing rate r.

[0018] A normalization submodule is configured to automatically select an improved Sigmoid normalization method or a Z-score normalization method according to parameter characteristics to process data.

[0019] A feature extraction submodule is configured to extract at least one of time sequence dynamic features and spatial distribution features from data.

[0020] Further, the feature extraction submodule is configured to:

[0021] The electric flux, the chloride ion diffusion coefficient, the water absorption rate and the apparent permeability coefficient use time sequence dynamic feature extraction, specifically using a sliding window incremental method combined with wavelet transform to extract average increment, maximum increment and volatility index of parameters.

[0022] Further, the comprehensive evaluation module adopts a multi-mechanism synergistic fusion strategy, the fusion strategy including adopting a weighted average fusion method to fuse the time sequence dynamic characteristics of the electric flux and the water absorption rate, adopting a neural network fusion method to fuse nonlinear coupling characteristics (such as the chloride ion concentration gradient and the water absorption rate), and adopting a physical-data dual driving model fusion method to establish a permeation parameter coupling equation according to the electric flux, the chloride ion concentration gradient and the water absorption rate gradient.

[0023] Further, the comprehensive evaluation module further includes a model selection unit configured to dynamically select an optimal evaluation model based on a model scoring function, the scoring function being: , wherein, is the comprehensive score of the model m; is the root mean square error of the model on the validation set; is a trade-off coefficient for punishing high complexity models; is a model complexity index; the evaluation model includes a fuzzy comprehensive evaluation model, a grey correlation analysis model and a machine learning classification model.

[0024] Further, the comprehensive evaluation module is further provided with a dynamic weight adjustment mechanism, the weight adjustment mechanism dynamically adjusting the weight of each parameter in the fusion according to an aging attenuation coefficient of the parameter and a current environmental excitation factor, the weight function being: .

[0025] The application also relates to a concrete permeability multi-parameter comprehensive evaluation method, comprising:

[0026] synchronously collecting multiple permeability parameters and environmental parameters of a concrete test piece;

[0027] based on environmental temperature and humidity data, compensating and correcting the collected permeability parameters;

[0028] cleaning, normalizing and extracting multi-dimensional features of the collected data;

[0029] based on a dynamic fusion strategy, fusing and comprehensively evaluating the extracted features to generate a concrete permeability grade evaluation result; wherein the comprehensive evaluation includes dynamically selecting an evaluation model based on a model scoring function and updating the fusion weight by using a dynamic weight adjustment mechanism;

[0030] outputting and displaying the evaluation result.

[0031] Further, in the comprehensive evaluation step, a rolling update mechanism is adopted, the model error is evaluated based on a sliding validation window at regular time intervals, and the reselection and training of the model are triggered to adapt to data changes; the method further includes a cross-cycle fusion step of fusing the evaluation results of the current cycle and the historical cycles according to a time decay factor to obtain a comprehensive score .

[0032] Compared with the prior art, the beneficial effects of the present invention are:

[0033] 1. It can simultaneously detect multiple permeability parameters of concrete, overcoming the limitations of traditional methods that only detect a single parameter, and can more comprehensively and accurately reflect the permeability performance of concrete.

[0034] 2. By employing data fusion technology, multiple parameter information is combined, which improves the accuracy and reliability of the evaluation results.

[0035] 3. Establish a comprehensive evaluation model that can quantitatively evaluate the permeability of concrete and provide a scientific basis for assessing the durability of concrete structures. Attached Figure Description

[0036] Figure 1 This is a block diagram of a multi-parameter integrated testing system for concrete permeability. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this disclosure. All other embodiments obtained by those skilled in the art based on the described embodiments of this disclosure without creative effort are within the scope of protection of this disclosure.

[0038] Example

[0039] This embodiment provides a specific implementation of a comprehensive testing system and evaluation method for multi-parameter permeability of concrete.

[0040] 1. Parameter detection and data acquisition

[0041] This system integrates various specialized testing equipment to achieve automated or semi-automated testing of multiple permeability parameters in concrete specimens. The testing process is coordinated under a unified synchronous control unit to ensure data consistency in time and space. Specific parameter testing methods include:

[0042] Electrical flux detection: The RCM (Rapid Chloride Ion Migration) method is used. Using a PROCEQ Torrent RCM tester or equivalent device, a 60V DC voltage is applied to both ends of a pre-treated saturated concrete specimen for 6 hours, and the amount of charge passing through (unit: coulomb C) is measured. The results are automatically stored. The measurement error of this detection unit is controlled within ±5C. Current response calibration is performed using national standard samples, and calibration is recommended every 30 days or 500 data acquisition cycles.

[0043] Chloride ion diffusion coefficient detection: Electromigration or free diffusion methods are employed. Using the NT Build 443 chloride ion migration test system, under set temperature and humidity conditions, the chloride ion concentration gradient is measured, and the diffusion coefficient is calculated. For the free diffusion method, based on the results of unsteady-state migration tests, the formula is used... Calculate the chloride ion diffusion coefficient D (unit: ×10⁻¹² m² / s). The accuracy of this detection unit is better than ±0.1×10⁻¹² m² / s, and it is calibrated using a comparison method.

[0044] Water absorption rate testing: An automatic water absorption rate tester or a constant weight weighing system is used. The dried concrete specimen is weighed first, then immersed in water. It is removed and weighed at specific time points, and the water absorption rate (unit: %) is obtained by calculating the rate of change in mass. The measurement accuracy of this testing unit is ±0.2%, calibrated by comparison with standard saturated samples.

[0045] Apparent air permeability coefficient testing: A Torrent air permeability tester was used. A vacuum cavity was formed on the surface of the specimen, the gas permeation rate was measured, and the apparent air permeability coefficient (unit: 10⁻) was calculated. 4 (cm / s). The accuracy of this detection unit is controlled within ±0.01×10⁻ 4 cm / s, and is periodically calibrated using standard mortar specimens.

[0046] Environmental parameter acquisition: During each permeation parameter detection process, the system automatically and synchronously records ambient temperature (T, °C), relative humidity (RH, %), and air pressure (P, kPa) data through the integrated environmental sensing unit. The environmental sensor accuracies are ±0.5°C, ±2%RH, and ±0.1kPa, respectively, and it automatically self-calibrates once a month by default.

[0047] The collected data will be embedded as tags into the original permeability data and corrected using the following environmental compensation formula:

[0048]

[0049] in, These are the corrected parameter values; These are the original measured values; =20°C, =50%, α is the temperature sensitivity coefficient; β is the humidity sensitivity coefficient; T is the ambient temperature during the actual test; The standard reference temperature is usually taken as 20°C; This refers to the relative humidity during the actual test. The standard relative humidity is usually taken as 50%.

[0050] Table 1. Specific values ​​of α and β

[0051]

[0052] Under the command of the synchronization control unit, all detection units trigger data acquisition by setting a unified time window τ and Δt delay compensation mechanism. The acquired data is automatically timestamped, and the synchronization of different data streams is ensured by a time interpolation resampling algorithm. All parameters are timestamped and merged into the same batch of records to ensure spatiotemporal consistency. If there is a deviation in the sampling of the equipment, the system will trigger a time correction mechanism to realign.

[0053] 2. Data Processing and Feature Extraction

[0054] The collected raw data is transmitted to the data processing module for intelligent processing, including data cleaning and normalization. To meet the needs of processing multi-parameter, high-frequency, and large-volume detection data, an efficient parallel data preprocessing workflow has been constructed, including the following steps:

[0055] 2.1 Data Cleaning

[0056] To eliminate outliers caused by environmental noise or testing errors, the system uses a combined Z-score and IQR (interquartile range) method to mark outliers and identify anomalies for each set of detection parameters.

[0057] Outlier detection range: The default value is k=1.5, which can be adjusted adaptively based on historical data. If a value is determined to be outlier, it will be removed.

[0058] For missing values, the system automatically selects an imputation strategy based on the missing value rate r.

[0059] First, a missing data diagnosis is performed on the raw observation data. The missing rate r for each parameter data stream is calculated using the following formula:

[0060]

[0061] Based on the missing rate, missing information is classified into three categories: when r < 5%, it is considered a minor missing information; when 5% ≤ r < 20%, it is considered a moderate missing information; and when r ≥ 20%, it is considered a severe missing information.

[0062] For minor missing data, mainly isolated missing points, time-weighted interpolation is used for imputation. This method considers both linear interpolation results and the slope trend of the time series. The imputation value is calculated using the following formula:

[0063]

[0064] in, The reconstructed value for the missing point t is calculated by interpolation and is used to fill in missing data. For valid observations adjacent to the missing point, adjacent non-missing data are directly read from the time series. The effective observations before and after the missing point are extracted from the time series.

[0065] `linear()` is a linear interpolation function, and its calculation formula is:

[0066]

[0067] slope() is a trend-based slope prediction value, defined as:

[0068]

[0069] λ is the weighting coefficient, representing the degree of influence of the trend term on the interpolation, based on the volatility (e.g., variance) within the time window where the missing points are located. Dynamic calculation:

[0070]

[0071] Where c is the smoothing factor, and its empirical value is 1e-3.

[0072] For moderate missing data, i.e., short-term consecutive missing segments, a multi-modal sliding window method is used for imputation. Predictive reconstruction is performed by combining the mean, median, trend line, and periodic signals of the time series. The imputed value is:

[0073]

[0074] in: The reconstructed value of the missing point t; The weight coefficients for the i-th mode are dynamically adjusted based on the residuals of the reconstruction results for each mode, satisfying the following conditions: Initially, the weights of each mode can be set to be equal, and the residuals are reconstructed based on the history of each mode. Adjustments will be made:

[0075]

[0076] These are real observations, derived from complete historical records, stored in the system in time series, and can be directly extracted; The predicted value of reconstruction mode i at time point t is obtained by reconstructing mode iii after simulating missing data; N is the number of validation samples, which is usually taken from the most recent historical data. The results represent different reconstruction modes: mean mode is the mean within the current window; median mode is the median within the current window; trend mode is the linear regression fit value within the window; and periodic mode is the value at the corresponding time point of the historical periodic signal.

[0077] For severe deficiencies, i.e., long periods of time or entire segments of data missing, a spatiotemporal collaborative reconstruction method is used to fill the gaps. This includes:

[0078] Time dimension: Train a Long Short-Term Memory (LSTM) network or a Transformer-based time series prediction model to predict missing segments;

[0079] Spatial dimension: Spatial reconstruction is performed using Kriging interpolation algorithm based on adjacent spatial measurement point data;

[0080] The results of the joint reconstruction are as follows:

[0081]

[0082] in, The reconstructed value of the missing point t; For time-dimensional predictions, use a trained LSTM, ARIMA, or Transformer time series model. The spatial dimension interpolation value is calculated using the observation values ​​of adjacent measurement points through the Kriging or IDW (inverse distance weighting) method; The spatiotemporal fusion weighting coefficient is dynamically adjusted based on the comparison of temporal and spatial reconstruction residuals, and its calculation formula is as follows:

[0083]

[0084] In the formula: , These are the residuals from temporal prediction and spatial interpolation, respectively.

[0085] If redundant measurement points exist, a voting mechanism for these points is used to improve the robustness of filling severely missing segments. When redundant sensors exist in the same area, weights are assigned based on the historical reliability and real-time stability of each redundant measurement point, and the filling value is calculated as follows:

[0086]

[0087] in, The reconstructed value of the missing point t; Let be the observation value of the i-th redundant measuring point at time t; Let be the weight of the i-th redundant measurement point, dynamically allocated based on the historical stability, real-time reliability, and correlation with the target measurement point of each redundant measurement point. The calculation formula is as follows:

[0088]

[0089] In the formula, The historical correlation between the measuring point and the target point is calculated using the historical correlation coefficient matrix; The real-time signal stability (variance) at the measurement point is calculated through the fluctuation of real-time data. The historical reliability score for the measurement points is determined by statistical analysis of historical operation logs.

[0090] After the imputation is completed, the system verifies the imputation quality through the following indicators: residual sum of squares (RSS), consistency with historical distribution (such as Kolmogorov–Smirnov test), temporal smoothness, spatial continuity, and compliance with physical constraints.

[0091] 2.2 Data Normalization

[0092] Since different permeability parameters have different dimensions and large numerical differences, the system needs to process them uniformly and normalize all parameters to the [0,1] interval.

[0093] An improved Sigmoid normalization is used for nonlinear parameters such as chloride ion diffusion coefficient and air permeability coefficient to enhance robustness to extreme values.

[0094]

[0095] Z-score normalization was applied to linear parameters such as water absorption rate and electrical flux.

[0096]

[0097] The system has a built-in normalization selection module that automatically matches the optimal method based on parameter volatility, autocorrelation, and distribution type, thereby improving the consistency of standardized data.

[0098] 2.3 Feature Extraction

[0099] To fully extract useful information from multi-dimensional concrete permeability testing data, a multi-level, multi-modal feature extraction algorithm framework is proposed to jointly extract temporal dynamic features and spatial distribution features. This framework not only considers the saliency of individual features but also integrates the coupling relationships between features and multi-scale information, improving the accuracy and robustness of permeability evaluation. This invention employs a feature extraction algorithm to extract temporal dynamic features of electrical flux, chloride ion diffusion coefficient, water absorption rate, and air permeability coefficient, forming a time series. Short-term fluctuations (sliding window increments) and long-term trends (wavelet transform) are extracted to determine the evolution trend of permeability performance.

[0100] Feature extraction algorithms consist of the following two levels:

[0101] (1) Temporal dynamic feature extraction

[0102] To reflect the variation of permeability parameters over time, a sliding window incremental method combined with frequency domain analysis was used. The specific steps are as follows:

[0103] Construct a time series with a length of Using a sliding window, calculate the increment within the window:

[0104]

[0105] in, The increment at time point t represents the rate of parameter change and is used to detect abrupt changes. It is calculated from the observations at two adjacent time points. The observation value at time point t represents the real-time data collected by the sensor. These are the observations from the previous time point, and the historical data stored in the system.

[0106] The average increment, maximum increment, and volatility indicators are extracted to characterize short-term mutation trends.

[0107] By introducing wavelet transform, the signal is decomposed into different frequency bands to extract long-term trends and short-term anomalies:

[0108]

[0109] in, This is the original time series; Let be the wavelet basis function, and be the translation and scaling form of the mother wavelet. Its calculation formula is: ψ(t) is the mother wavelet, which is a finite energy function that satisfies the wavelet condition; j is the scaling factor, which determines the degree of scaling of the wavelet; k is the translation factor, which determines the position of the wavelet on the time axis. The normalization factor ensures energy conservation; wavelet coefficients The projection intensity of the original signal at the corresponding scale j and position k is represented by the formula: ; y(t) is the original time series.

[0110] (2) Extraction of spatial distribution features

[0111] Only the spatial distribution characteristics of the chloride ion diffusion coefficient are collected. Chloride ion diffusion tests are often conducted by setting measurement points at different depths (or by layering core samples), and the data naturally possesses spatial distribution characteristics, allowing for the extraction of depth gradient vectors and tensor norms. To characterize the distribution of the chloride ion diffusion coefficient within concrete at different depths or locations, a depth gradient tensor feature extraction method is proposed:

[0112] For depth sequences Calculate the gradient vector:

[0113]

[0114] in, The chloride ion concentration gradient between adjacent depths is used to characterize the uniformity of permeability along depth and is calculated by the concentration difference between measuring points at different depths. In depth The chloride ion concentration measured at the location was obtained through a depth sensor or a stratified sampling experiment; In the previous depth Concentration at that location; , The depth position of two adjacent measuring points.

[0115] Multi-point and depth information is constructed into a tensor form, and features such as tensor norm and principal direction are calculated to quantitatively characterize the uniformity of distribution and the location of outliers. In summary, the applicability of feature extraction methods for each parameter is shown in Table 2. "-" in the table indicates that data for that parameter does not need to be collected.

[0116] Table 2: Overview of the Applicability of Parameter Feature Extraction

[0117]

[0118] 3. Data fusion and comprehensive evaluation

[0119] The extracted multi-dimensional features (as shown in Table 2) are input into the comprehensive evaluation module for dynamic fusion and evaluation.

[0120] 3.1 Data Fusion

[0121] To comprehensively reflect the complex coupling relationships among multiple parameters of concrete permeability, a multi-mechanism collaborative data fusion strategy is proposed. This strategy integrates three fusion methods: weighted average fusion, neural network fusion, and a physics-data dual-driven model, supporting both static evaluation and dynamic prediction. It is worth noting that the three fusion strategies can be flexibly selected according to the scenario (as shown in Table 3), depending on the complexity of data coupling. This fusion strategy aims to unify the modeling of permeability characteristics (such as electrical flux, water absorption rate, and chloride ion diffusion coefficient) with different physical meanings, dimensions, and frequencies, thereby forming a comprehensive permeability performance index with strong interpretability, wide adaptability, and high accuracy. When the fusion object has a clear physical coupling relationship (such as the coupling of water absorption rate gradient and electrical flux evolution) or multi-scale information linkage, the physics-data dual-driven fusion model is used to replace schemes a and b, achieving unified modeling.

[0122] a. For linearly correlated parameters (such as electrical flux and water absorption rate), a weighted average fusion method is used:

[0123]

[0124] in, For the normalized first A permeability parameter; For the first The fusion weights of the parameters satisfy ; This refers to the integrated characterization features after fusion.

[0125] b. For parameters with complex nonlinear coupling (such as chloride ion concentration gradient and water absorption rate), a neural network fusion method (including ReLU and Sigmoid activation functions) is used:

[0126]

[0127] Where x is a multidimensional parameter input vector; , This is the weight matrix of the neural network, learned through backpropagation; , σ is the bias term; ReLU() is the activation function used to enhance nonlinear modeling capabilities; σ() is the output layer activation function, such as Sigmoid or Tanh, used to normalize the fusion results.

[0128] c. The physical-data dual-driven model includes: performing spatiotemporal feature encoding and establishing permeation parameter coupling equations.

[0129]

[0130] In the formula, k1 and k2 are coupling coefficients, representing the relationship between electrical flux and chloride ion concentration gradient and water absorption rate gradient; Q is electrical flux; W is water absorption rate; and C is chloride ion concentration. The water diffusion coefficient; The permeation velocity is denoted as ρ.

[0131] Table 3 Applicable Scenarios for Different Fusion Strategies

[0132]

[0133] 3.2 Dynamic Evaluation

[0134] To achieve comprehensive judgment and classification of multiple parameters of concrete permeability, this invention constructs a multi-model collaborative evaluation model system. This system integrates three types of models: fuzzy comprehensive evaluation model, grey relational analysis model, and machine learning classification model. It dynamically selects the optimal evaluation model through a "model scoring function," achieving adaptive and accurate evaluation for different data structures, engineering backgrounds, and working conditions.

[0135] a. Fuzzy Comprehensive Evaluation Model

[0136] Fuzzy comprehensive evaluation is suitable for handling complex systems with multiple factors, multiple levels, and fuzzy boundaries. In this embodiment, it is used to perform fuzzy membership mapping and comprehensive evaluation of multiple indicators such as electric flux, chloride ion diffusion coefficient, and water absorption rate. The technical process is as follows:

[0137] Construct a membership function, assign four levels to each indicator (e.g., excellent, good, average, poor), and calculate the membership degree using the Gaussian membership function:

[0138]

[0139] in, Membership degree; This is the value of the i-th indicator; The center value of the j-th level; The standard deviation corresponding to this level.

[0140] Constructing the fuzzy membership matrix R

[0141]

[0142] The weight vector W is determined using the adaptive inverse residual method:

[0143]

[0144] in, The evaluation weight for the i-th permeability parameter; is the prediction residual of the i-th parameter during the training / validation phase; n is the total number of parameters used in the fusion or evaluation.

[0145] Perform comprehensive calculations

[0146]

[0147] Where B is the comprehensive evaluation vector, corresponding to the comprehensive membership degree of each level.

[0148] Each permeability parameter (such as chloride ion diffusion coefficient, water absorption rate, etc.) needs to be calculated independently, and the final result is a fuzzy relation matrix R.

[0149] Table 4 uses three parameters as an example.

[0150]

[0151] b. Grey Relational Analysis Model (GRA)

[0152] This is used to calculate the degree of similarity between the tested sample and the "ideal permeability sample." The technical process is as follows:

[0153] Set an ideal reference sequence x 0， The best level value of each indicator is selected as the ideal sample.

[0154] The correlation coefficient between the i-th indicator and the reference sample is calculated as follows:

[0155]

[0156] in, ζ is the correlation coefficient; ζ is the discrimination coefficient, usually taken as 0.5; min, max: the minimum and maximum absolute differences under all indicators.

[0157] Calculate the overall correlation degree

[0158]

[0159] Where Γ is the overall correlation degree. If Γ approaches 1, it means that the overall performance of the sample is closer to the ideal state. The correlation coefficient; is the evaluation weight for the i-th permeability parameter.

[0160] c. Data-driven classification models (such as SVM, random forest, decision tree, neural network) automatically determine the penetration level through model training. For evaluation methods using neural network models, incremental data fine-tuning is supported to avoid full retraining, and a freeze-thaw strategy is used to control weight transfer. The technical process is as follows:

[0161] Construct a training sample set, including input features x: multi-parameter normalized data; output labels y: manually labeled or historical data rankings.

[0162] Training and cross-validation are performed using Support Vector Machines (SVM), Random Forests (RF), or Multilayer Perceptrons (MLP); 5-fold cross-validation is used to determine the optimal model parameters.

[0163] It outputs probability predictions or direct classification results, outputting the probability of each sample belonging to a certain level, and can also output the Top-1 level; it also supports outputting model credibility and feature contribution (such as SHAP analysis).

[0164] In the current system design, dynamic evaluation modules (such as fuzzy comprehensive evaluation and grey relational analysis) are mainly based on the original physical index data (such as electric flux, chloride ion diffusion coefficient, etc.) for evaluation, rather than using comprehensive indicators that have been fused or dimensionality reduced. This is mainly due to the following reasons:

[0165] 1. The original indicators usually correspond to the evaluation thresholds specified in the clear national or industry standards (e.g., "chloride ion diffusion coefficient ≤2×10⁻¹²m² / s" is an excellent level), which makes it easy to perform intuitive level mapping and membership calculation in fuzzy comprehensive evaluation; however, the features obtained after fusion or dimensionality reduction often lack a unified standard interpretation and are difficult to use directly in such evaluation systems based on physical thresholds.

[0166] 2. The original parameters have clear physical meaning and engineering interpretability. For example, "excessive electrical flux" can directly point to abnormal conductivity of concrete, making it easier for engineers to identify the source of risk and take targeted measures. Conversely, mathematically constructed features such as "high score of principal component 1" are not conducive to actual engineering diagnosis and communication.

[0167] 3. Grey relational analysis and fuzzy evaluation require each input to have physical meaning; after dimensionality reduction, the features are mathematical combinations, making it inconvenient to define hierarchical mapping and physical correlation.

[0168] It is worth noting that although fuzzy evaluation and grey analysis require the use of original indicators, the results of feature extraction and fusion remain crucial for machine learning classification models (such as SVM, random forest, and neural networks) in dynamic evaluation systems. These models typically use extracted features (such as temporal dynamic indicators, spatial distribution tensors, and dimensionality-reduced features), fused comprehensive indicators (such as weighted fusion outputs or nonlinear fusion features), and dynamically weighted parameter combinations as inputs to capture and classify complex nonlinear relationships. Therefore, the system as a whole adopts a hybrid evaluation strategy: traditional methods rely on original indicators to ensure interpretability and standard compliance, while machine learning methods depend on high-level features to improve prediction accuracy and adaptability.

[0169] To ensure model adaptability, a scoring function for evaluating the model is constructed, and the current optimal model is dynamically selected during system operation.

[0170]

[0171] in, The overall score for model m; The root mean square error of the model on the validation set; This is a tradeoff coefficient used to penalize highly complex models; This refers to metrics related to model complexity (such as the number of parameters and training time).

[0172] The system automatically sets the parameters based on the complexity of the real-time evaluation task. Finally, the model with the lowest score was selected for determining the penetration performance level of the current sample.

[0173] Therefore, the specific implementation of the overall dynamic fusion evaluation includes:

[0174] To adapt to the changing weights of various permeability parameters of concrete (such as water absorption, electrical flux, and chloride ion diffusion coefficient) with environmental conditions and aging during long-term service, a dynamic weighting function is constructed:

[0175]

[0176] in, Let be the fusion weight of the i-th parameter at time t; Initial weights (e.g., inverse residual ratio); The aging degradation coefficient represents the rate at which the importance of the i-th indicator decreases during aging, and is obtained by fitting historical monitoring data or accelerated aging tests; t is the current service time or monitoring cycle number. The current environmental stimulus factor reflects the influence of external variables such as temperature, humidity, and carbonization depth on the weights; it can be described by the following function:

[0177]

[0178] in, , Current temperature and humidity; , Temperature and humidity of the standard environment; , This is the temperature and humidity sensitivity coefficient (determined by gradient experiments).

[0179] Rolling update model:

[0180] Using a sliding verification window mechanism, the system performs a sliding window error assessment on the past N samples every fixed monitoring period T. If the RMSE of the currently used model exceeds a preset threshold (e.g., year-on-year growth > 15%), the system will automatically enter the "model review" stage, triggering model replacement. At this point, the system will call the model scoring function. Retrain or tune the models in the current candidate model library, retain the old models for comparison, and select the one with the best score to replace the old model.

[0181] Considering the significant time-dependent nature of concrete permeability, such as abrupt changes or trend shifts, a cross-period integrated evaluation mechanism was designed:

[0182]

[0183] in, The fusion evaluation score for period t; This represents the historical single-cycle evaluation results for period t-k. The time decay factor satisfies Generally, an exponential decay distribution is adopted:

[0184]

[0185] In the formula, The sliding time window length is typically 3 to 7 periods; This is the information attenuation coefficient.

[0186] The system also utilizes an intelligent triggering mechanism in the auxiliary module to set a fusion trigger threshold. The dynamic fusion module is automatically activated when any of the following conditions are met:

[0187] The volatility of the detected data exceeds the threshold (e.g., a sudden increase in the standard deviation).

[0188] The magnitude of current environmental change is greater than the threshold (e.g., temperature fluctuation is greater than 5°C).

[0189] Abnormal model prediction errors (e.g., RMSE year-on-year increase >20%).

[0190] User-forced update command.

[0191] 3.3 Classification

[0192] The final output permeability rating (Levels I to IV) corresponds to "Excellent," "Good," "Medium," and "Poor" permeability performance of concrete, respectively. The classification criteria are based on a comprehensive reference to the "Code for Design of Concrete Durability," "JT / T 617," "ASTM C1202," and extensive engineering statistical data. The main parameter classification boundaries are shown in the table below:

[0193] Table 5 Parameter Classification Table

[0194]

[0195] The final grade of the evaluation results shall be determined based on one of the following methods:

[0196] Maximum membership principle (used in fuzzy evaluation models): Output the membership matrix and select the level with the highest membership degree as the final judgment;

[0197] The maximum probability principle (used in machine learning models): take the classification label with the highest predicted probability as the level;

[0198] The comprehensive score segmentation method (applicable to weighted summation scoring) maps the weighted score results into four level intervals.

[0199] It is worth noting that users can manually change feature weights or adjust level ranges. The system records manual operations and updates the explanation module and weight recommendations accordingly.

[0200] 4. Results Output and Human-Computer Interaction

[0201] The final evaluation results are presented in a visual format through the human-computer interaction module. The system generates a structured evaluation report, which includes basic information, raw and preprocessed parameter data, comprehensive evaluation results (grade, model score, parameter contribution), trend prediction charts (such as LSTM prediction curves), and engineering recommendations. The report can be exported in PDF, Excel, and other formats. All data is stored in a local or cloud database indexed by "project number / sample number / timestamp".

[0202] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A comprehensive testing system for multiple parameters of concrete permeability, characterized in that, include: The data acquisition module is used to simultaneously acquire multiple permeability parameters and environmental parameters of concrete specimens; The data processing module is communicatively connected to the data acquisition module and is used to clean, normalize, and extract features from the acquired data. The comprehensive evaluation module is communicatively connected to the data processing module and is used to comprehensively evaluate the extracted features based on a dynamic fusion strategy to generate a penetration level evaluation result. The human-computer interaction module is communicatively connected to the comprehensive evaluation module and is used to configure system parameters and visualize data and evaluation results.

2. The system according to claim 1, characterized in that, The data acquisition module includes: Electrical flux testing unit, chloride ion diffusion coefficient testing unit, water absorption rate testing unit, and air permeability testing unit; An environmental sensing unit is used to collect ambient temperature, relative humidity, and air pressure data. The synchronization control unit is used to set a unified time window and delay compensation mechanism for each test unit and environmental sensing unit to achieve synchronous acquisition of multi-parameter data and timestamp alignment.

3. The system according to claim 2, characterized in that, The synchronization control unit synchronizes data streams of different parameters through a time interpolation resampling algorithm and is equipped with a time correction mechanism to realign data when there is a deviation in the device sampling.

4. The system according to claim 1, characterized in that, The data processing module includes: The data cleaning submodule is configured to identify outliers using a combined Z-score and IQR interquartile range method, and to fill in missing values ​​using time-weighted interpolation, multi-modal sliding window method, or spatiotemporal collaborative reconstruction method, depending on the missing value rate r. The normalization submodule is configured to automatically select the improved Sigmoid normalization or Z-score normalization method to process the data based on the parameter characteristics; The feature extraction submodule is configured to extract at least one of temporal dynamic features and spatial distribution features from the data.

5. The system according to claim 4, characterized in that, The feature extraction submodule is configured as follows: Electrical flux, chloride ion diffusion coefficient, water absorption rate, and apparent air permeability coefficient were extracted using time-series dynamic features. Specifically, the sliding window incremental method combined with wavelet transform was used to extract the average increment, maximum increment, and fluctuation index of the parameters.

6. The system according to claim 1, characterized in that, The comprehensive evaluation module adopts a multi-mechanism collaborative fusion strategy. The fusion strategy includes using a weighted average fusion method to fuse the time-series dynamic characteristics of electrical flux and water absorption rate, using a neural network fusion method to fuse the time-series dynamic characteristics of chloride ion concentration gradient and water absorption rate, and using a physical-data dual-drive model fusion method to establish a osmotic parameter coupling equation based on electrical flux, chloride ion concentration gradient, and water absorption rate gradient.

7. The system according to claim 6, characterized in that, The comprehensive evaluation module also includes a model selection unit, configured to dynamically select the optimal evaluation model based on a model scoring function, wherein the scoring function is: ,in, The overall score for model m; The root mean square error of the model on the validation set; This is a tradeoff coefficient used to penalize highly complex models; The evaluation model is a metric for model complexity; the evaluation model includes fuzzy comprehensive evaluation model, grey relational analysis model and machine learning classification model.

8. The system according to claim 7, characterized in that, The comprehensive evaluation module also includes a dynamic weight adjustment mechanism. This mechanism dynamically adjusts the weights of each parameter in the fusion process based on the aging decay coefficients of the parameters and the current environmental incentive factors. The weight function is as follows: .

9. A multi-parameter comprehensive evaluation method for concrete permeability, characterized in that, The method, applied to the system according to any one of claims 1-8, comprises: Multiple permeability parameters and environmental parameters of concrete specimens were collected simultaneously. Based on ambient temperature and humidity data, the collected permeability parameters are compensated and corrected. The collected data is cleaned, normalized, and multi-dimensional features are extracted. Based on a dynamic fusion strategy, the extracted features are fused and comprehensively evaluated to generate concrete permeability grade evaluation results; wherein, the comprehensive evaluation includes dynamically selecting the evaluation model based on the model scoring function and updating the fusion weights using a dynamic weight adjustment mechanism. Output and display the evaluation results.

10. The method according to claim 9, characterized in that, In the comprehensive evaluation step, a rolling update mechanism is adopted to periodically evaluate model error based on a sliding validation window and trigger model reselection and retraining to adapt to data changes. The method also includes a cross-period fusion step, which fuses the evaluation results of the current period and historical periods according to a time decay factor to obtain a comprehensive score. .

Images