River-crossing and sea-crossing shield tunnel double-doped fiber concrete durability detection system

Abstract

The invention relates to the technical field of concrete detection, and discloses a durability detection system for double-doped fiber concrete of a river-crossing and sea-crossing shield tunnel. A durability data acquisition unit of the system obtains multi-source monitoring data in a segment concrete service environment, wherein the multi-source monitoring data comprises material performance, environmental erosion and structural response parameters; the feature fusion processing unit integrates multi-source data according to a time sequence, extracts performance degradation features and constructs a multi-dimensional durability feature space; the damage mode recognition unit establishes a damage state classification model based on the feature space and recognizes a current damage development stage; a critical state prediction unit calculates a performance degradation rate according to the damage change track and predicts a time node for reaching a critical state; and the detection scheme generation unit automatically matches an optimal detection scheme according to a prediction result, and outputs an execution instruction containing a detection position, a detection method and a detection period. The system can comprehensively master the durability state of the concrete, and the detection scientificity and effectiveness are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of concrete detection, in particular to a durability detection system for double-mixed fiber concrete of a river-crossing and sea-crossing shield tunnel. BACKGROUND

[0002] As an important part of water-crossing transportation infrastructure, the river-crossing and sea-crossing shield tunnel has the characteristics of high humidity, high salinity, and frequent temperature fluctuations. The tunnel segment concrete is long-term exposed to such a complex erosion environment, which is prone to durability damage problems such as chloride ion penetration, steel corrosion, and surface crack propagation, thereby affecting the overall safety and service life of the tunnel structure. Double-mixed fiber concrete is widely used in the construction of tunnel segments of river-crossing and sea-crossing shield tunnels due to its excellent crack resistance and impermeability. However, even with such concrete materials, the durability will still degrade over time due to environmental erosion. If the durability of the concrete cannot be accurately monitored in a timely manner, structural damage may be overlooked, ultimately leading to safety accidents or increased maintenance costs.

[0003] The detection method for the durability of shield tunnel concrete is mainly periodic manual detection. The detection personnel need to enter the tunnel and use handheld detection equipment to sample and detect the surface state and steel corrosion of the segment concrete. This detection method has obvious limitations. On the one hand, manual sampling detection cannot achieve full coverage of the tunnel segments, and may miss some areas that have already deteriorated. On the other hand, manual detection relies on the experience and professional level of the detection personnel, and the detection results are easily affected by human factors, making it difficult to ensure the accuracy and consistency of the data. Moreover, the traditional detection method cannot realize real-time data collection and continuous monitoring of the durability of the concrete, and can only obtain discrete data at a specific time point, making it difficult to reflect the dynamic degradation process of the concrete performance over time and providing incomplete data support for subsequent damage prediction and maintenance decisions.

[0004] Although the existing part of the detection system attempts to introduce sensors for data collection, most of them can only process a single type of monitoring data, such as collecting only the chloride ion concentration data inside the concrete or monitoring only the steel corrosion potential data, and cannot effectively integrate multi-source monitoring data. Since the durability degradation of concrete is the result of the combined action of material properties, environmental erosion, structural response and other factors, a single type of data cannot fully reflect the actual durability state of the concrete, resulting in low accuracy of the damage judgment model established based on these data, making it difficult to accurately identify the damage development stage of the concrete, and even more difficult to effectively predict the time node when the concrete reaches the critical durability state. In terms of detection scheme development, existing systems mostly use fixed detection modes and cannot dynamically adjust the detection position, detection method and detection cycle according to the actual durability state and degradation trend of the concrete, resulting in unreasonable allocation of detection resources, over-detection in some areas and under-detection in some key areas, affecting the efficiency and effectiveness of the detection work. SUMMARY

[0005] The purpose of the present application is to provide a durability detection system for double-mixed fiber concrete of a river-crossing and sea-crossing shield tunnel, in order to solve the problems raised in the background art.

[0006] To achieve the above-mentioned purpose, the present application provides a durability detection system for double-mixed fiber concrete of a river-crossing and sea-crossing shield tunnel, which comprises: a durability data acquisition unit for acquiring multi-source monitoring data of the concrete of the shield tunnel segment under service environment, wherein the multi-source monitoring data includes material property parameters, environmental erosion parameters and structural response parameters; a feature fusion processing unit for aligning and integrating the multi-source monitoring data according to time sequence, extracting performance degradation features of the concrete at different erosion stages, and constructing a multi-dimensional durability feature space; a damage mode recognition unit for establishing a damage state classification model based on the multi-dimensional durability feature space, and identifying the current damage development stage of the concrete; a critical state prediction unit for calculating the performance degradation rate according to the change trajectory of the damage development stage, and predicting the time node when the concrete reaches the critical durability state; a detection scheme generation unit for automatically matching the optimal detection scheme according to the critical state prediction result, and outputting an execution instruction including detection position, detection method and detection cycle.

[0007] Preferably, the specific way of acquiring multi-source monitoring data by the durability data acquisition unit is: acquiring the chloride ion concentration distribution data and the steel corrosion potential data inside the concrete through a buried sensor array; acquiring the morphological feature parameters and distribution density parameters of the apparent cracks on the concrete surface by using a surface scanning device; The environmental monitoring device is used to record the temperature and humidity change curve and the erosion medium contact frequency data inside the tunnel.

[0008] Preferably, the specific steps for the feature fusion processing unit to construct a multi-dimensional durability feature space are: The chloride ion concentration distribution data and the steel corrosion potential data are grid-matched according to the spatial coordinates to generate a material performance degradation feature matrix; The morphological feature parameters of the apparent cracks are analyzed for topological structure, and the crack network connectivity feature and the extension direction feature are extracted; The temperature and humidity change curve and the erosion medium contact frequency data are analyzed for time domain correlation to establish an environmental erosion intensity feature vector; The material performance degradation feature matrix, the crack network features and the environmental erosion intensity feature vector are mapped to a unified time coordinate system to form a durability feature space containing spatial and time dimensions.

[0009] Preferably, the specific method for the damage mode recognition unit to establish a damage state classification model is: Four damage development stages are divided in the durability feature space, respectively corresponding to the initial stable period, the local deterioration period, the accelerated degradation period and the critical failure period; The density clustering algorithm is used to identify the clustering areas of the data points in the feature space, and the typical modes of each damage stage are determined according to the feature parameter combinations of the center points of the clustering areas; The discriminant function based on Mahalanobis distance is established to calculate the distance values between the current monitoring data and each typical mode in real time, and the damage stage corresponding to the minimum distance value is taken as the recognition result.

[0010] Preferably, the specific process for the critical state prediction unit to calculate the performance degradation rate is: The feature parameter change trajectory corresponding to the current damage development stage is extracted, and the local change rate of each feature parameter is calculated using a sliding time window; The mapping relationship between the feature parameter change rate and the remaining service time is established, and the comprehensive degradation rate index is obtained through weighted fusion; According to the development duration of the same damage stage in the historical data, the predicted remaining time of the current stage is corrected.

[0011] Preferably, the specific logic for the detection scheme generation unit to match the optimal detection scheme is: When the predicted remaining time is greater than a preset threshold, a conventional detection scheme mainly based on non-damage detection is generated, and the detection period is set to a fixed proportion of the predicted remaining time; When the predicted remaining time is less than a preset threshold, a special detection scheme including core drilling sampling and laboratory analysis is generated, and the detection period is shortened to half of the conventional scheme; For the pipe piece area in the accelerated degradation period, crack propagation monitoring and reinforcement corrosion rate monitoring items are added in the detection scheme.

[0012] Preferably, the system further comprises a data quality verification unit, and the working process of the data quality verification unit is as follows: The integrity and consistency of the data collected by each sensor are monitored in real time, and spatial interpolation is performed on the missing data by using the data of adjacent sensors; When the monitoring data of different sensors at the same position differ by more than an allowable range, a redundant measurement mechanism is started to verify the data; Abnormal data points are marked and a credibility weight is calculated, and the proportion of data participating in calculation is adjusted according to the weight value in the feature fusion processing stage.

[0013] Preferably, the data quality verification unit further comprises a drift compensation module, and the compensation method of the drift compensation module is as follows: Standard test data of the concrete standard test block are collected periodically, and a deviation curve of the sensor measurement value and the reference value is established; The drift compensation coefficient of each sensor is calculated according to the change trend of the deviation curve, and the original monitoring data is automatically corrected in the data processing stage; When the drift compensation coefficient exceeds a critical threshold, a sensor calibration reminder signal is generated and the data weight of the sensor is adjusted.

[0014] Preferably, the system further comprises a detection result feedback unit, and the operation mechanism of the detection result feedback unit is as follows: The on-site detection result is compared and analyzed with the system prediction result, and a prediction accuracy index is calculated; The discrimination threshold of the damage state classification model and the weight parameter of the critical state prediction unit are dynamically adjusted according to the accuracy index; When the deviation of the detection result of three consecutive times exceeds an allowable range, a model retraining process is triggered and the feature space division standard is updated.

[0015] Preferably, the detection result feedback unit further comprises a scheme optimization module, and the optimization strategy of the scheme optimization module is as follows: The actual execution effect of each detection scheme is recorded, and an evaluation index system of detection efficiency and detection accuracy is established; When the improvement amplitude of the newly added detection item on the diagnosis accuracy is lower than a preset value, the detection item is excluded in the subsequent scheme; According to the importance level and accessibility condition of the pipe piece area, a spatial allocation scheme of detection resources is optimized.

[0016] Compared with the prior art, the present application has the following beneficial effects: The durability detection system of the double-fiber mixed concrete of the cross-river shield tunnel can comprehensively obtain multi-source monitoring data of the shield tunnel segment concrete under the service environment, including material performance parameters, environmental erosion parameters and structural response parameters, break the limitation of traditional detection methods that can only obtain single type data or discrete data, realize all-round and multi-dimensional collection of concrete durability related data, and enable the staff to understand the initial state and change of the concrete from multiple angles, providing a rich and comprehensive data basis for subsequent data analysis and state judgment.

[0017] The feature fusion processing unit aligns and integrates the multi-source monitoring data according to the time sequence, extracts the performance degradation features of the concrete at different erosion stages, and constructs a multi-dimensional durability feature space, effectively solving the problem of scattered multi-source data in the existing detection system and difficult to analyze collaboratively. Through time sequence alignment, the consistency of data from different sources in the time dimension is ensured, and the analysis error caused by data time misalignment is avoided; and the extraction of performance degradation features and the construction of multi-dimensional feature space can convert scattered and seemingly unrelated data into feature information with clear physical meaning, intuitively present the performance change rule of the concrete at different erosion stages, and enable the staff to clearly master the dynamic process of the concrete from the initial state to gradual degradation, providing a structured and systematic feature basis for subsequent damage mode identification.

[0018] The damage mode identification unit establishes a damage state classification model based on the multi-dimensional durability feature space, which can accurately identify the damage development stage of the concrete. Compared with traditional manual judgment or single data judgment method, the classification model fully utilizes the rich information in the multi-dimensional feature space, accurately distinguishes the different damage stages of the concrete through comprehensive analysis of feature parameters, avoids misjudgment or omission caused by one-sided information, and enables the staff to timely and accurately master the damage degree of the concrete, providing a reliable state basis for subsequent targeted treatment measures.

[0019] The critical state prediction unit calculates the performance degradation rate according to the change trajectory of the damage development stage, predicts the time node when the concrete reaches the critical durability state, changes the situation that the traditional detection method can only passively find the damage that has occurred, and realizes the active prediction of the durability state of the concrete. Through the calculation of the performance degradation rate and the prediction of the critical time node, the staff can know when the concrete may reach the critical state that needs emergency treatment, so as to have enough time to develop a response plan, avoid the structural safety risk caused by sudden damage, and also provide a scientific reference for the time arrangement of maintenance work, facilitating the reasonable planning of maintenance period and resource allocation.

[0020] The detection scheme generation unit automatically matches the optimal detection scheme according to the critical state prediction result, and outputs an execution instruction containing a detection position, a detection method and a detection cycle, so as to realize individualization and dynamic adjustment of the detection scheme. According to the actual durability state of the concrete and the critical state prediction result, the unit can determine the position that needs to be detected, select the detection method that is most suitable for the current state, and reasonably set the detection cycle, so as to avoid the problems of waste of detection resources or insufficient detection in the traditional fixed detection mode. By optimizing the detection position, it can ensure that the key damage area is paid attention to; by selecting the appropriate detection method, the accuracy of the detection data and the efficiency of the detection work can be improved; by adjusting the detection cycle, the state of the concrete can be grasped in time while unnecessary detection times are reduced, the detection cost is reduced, and the scientificity and effectiveness of the detection work are improved as a whole, which provides a strong guarantee for long-term stable service of the double-mixed fiber concrete structure of the river-crossing and sea-crossing shield tunnel. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 A timing diagram of the durability detection system of the double-mixed fiber concrete of the river-crossing and sea-crossing shield tunnel is described. Figure 2 A flowchart of a multi-source monitoring data acquisition process of the durability data acquisition unit is described. Figure 3 A flowchart of a damage state classification model establishment of the damage mode recognition unit is described. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0023] Please refer to Figure 1The application provides a durability detection system for a cross-river and cross-sea shield tunnel double-mixed fiber concrete, which comprises dynamic evaluation and prediction of concrete durability through multi-source data fusion and machine learning algorithm. The core of the system is composed of a durability data acquisition unit, a feature fusion processing unit, a damage mode recognition unit, a critical state prediction unit and a detection scheme generation unit, which are connected in sequence to form a closed-loop processing procedure. The durability data acquisition unit is responsible for collecting material performance parameters, environmental erosion parameters and structural response parameters in real time from the service environment of the shield tunnel segment concrete. These parameters include but are not limited to chloride ion concentration, steel potential, crack morphology and temperature and humidity data. The feature fusion processing unit aligns and integrates the collected multi-source monitoring data in time sequence, extracts the performance degradation characteristics of concrete at different erosion stages, and constructs a multi-dimensional durability feature space containing time and space dimensions. The damage mode recognition unit uses the feature space to establish a damage state classification model to identify the current damage development stage of the concrete, such as the initial stable period or the accelerated degradation period. The critical state prediction unit calculates the performance degradation rate according to the change trajectory of the damage development stage and predicts the time node when the concrete reaches the critical durability state. The detection scheme generation unit automatically matches the optimal detection scheme according to the prediction results and outputs the execution instructions including detection position, detection method and detection cycle, so as to realize the full automation from data acquisition to decision output.

[0024] Example 1: see Figure 2 The implementation of the durability data acquisition unit depends on a multi-level and multi-type sensor network deployment, which is pre-planned and embedded in the key parts of the shield tunnel segment concrete. The embedded sensor array is integrated into the concrete during the segment precast stage. Its distribution strategy is determined based on fluid mechanics simulation and previous erosion experience, focusing on vulnerable parts such as segment joint areas, stress concentration areas and backwater surfaces. The chloride ion sensor in the array uses Ag / AgCl electrode principle, which can continuously monitor the concentration gradient of free chloride ions in the concrete pore solution. The data is stored in the form of two-dimensional grid coordinates, and each grid point is associated with depth information. The steel corrosion potential sensor is connected to the steel network through a pre-embedded reference electrode, which measures the change of steel half-cell potential. Its data has a one-to-one correspondence with the chloride ion concentration data in space, and the time stamps are strictly synchronized to ensure the consistency of the data time sequence. The surface scanning equipment includes a high-resolution line array camera and a laser scanner, which periodically move on the preset track in the tunnel to perform panoramic image acquisition on the inner wall of the segment.

[0025] After pre-processing, the image data is recognized and quantified by computer vision algorithms. The morphological feature parameters include the maximum width, average width, total length, and tortuosity of the cracks. The distribution density parameter is obtained by calculating the total length of the cracks per unit area. The environmental monitoring device consists of temperature and humidity sensors and chemical medium sensors. The temperature and humidity sensors are installed on the surface and at different depths inside the segment, respectively, to record the dynamic variation curves of temperature and relative humidity. The sampling frequency is configured according to the environmental differences inside and outside the tunnel. The chemical medium sensor is used to detect the contact frequency of aggressive media, such as monitoring the carbon dioxide concentration by an electrochemical sensor or monitoring the sulfate ion concentration by an ion-selective electrode. The contact frequency is quantified by counting the number of times the medium concentration exceeds the threshold value per unit time. All these multi-source monitoring data are aggregated through wired or wireless transmission networks to the central processor of the data acquisition unit. The data packets are accompanied by precise timestamps and spatial location labels, which are ready for subsequent time alignment.

[0026] The core task of the feature fusion processing unit is to integrate monitoring data from different sources and different formats into a unified analysis framework with spatial and temporal dimensions. Data processing begins with time series alignment. Due to differences in sampling periods and start times of various sensors, the system uses interpolation algorithms to resample all data sequences to a unified time axis, such as aligning all data to one data point per minute. Spatial alignment relies on the accurate coordinate mapping relationship established during the layout stage, mapping the physical location of each sensor to a virtual three-dimensional grid model. Based on the spatial and temporal alignment, for material performance parameters, the chloride ion concentration distribution data and the reinforcement corrosion potential data are grid-matched according to the same spatial coordinates. In each grid cell, the chloride ion concentration value and the reinforcement potential value are combined into a data pair. All grid cell data pairs collectively form a feature matrix reflecting material performance degradation. This matrix not only contains numerical information but also implies the spatial correlation between ion migration and electrochemical corrosion. For apparent crack data, topology analysis is a key step. The algorithm identifies cracks as a network graph, where the intersection points and endpoints of the cracks constitute nodes, and the crack segments constitute edges. By calculating the number of connected components, node degree distribution, and average length of crack segments of the network, the connectivity features describing the overall structure of the crack network are extracted.

[0027] By comparing the crack images at consecutive time points, the algorithm tracks the movement trajectory of each crack tip, thereby extracting the crack propagation direction features, which can be represented by principal direction vectors. For the environmental erosion parameter, the time-domain correlation analysis of the temperature and humidity variation curves and the erosion medium contact frequency data uses the sliding window method to calculate the cross-correlation function, to find the coupling relationship in time between the sharp fluctuations in temperature and humidity and the peak values of the erosion medium concentration, thereby establishing a feature vector that comprehensively represents the environmental erosion intensity, which can identify the period of most intense environmental erosion. Finally, the material performance degradation feature matrix, the crack network features, and the environmental erosion intensity feature vector are collectively mapped into a unified multi-dimensional durability feature space, which can be understood as a high-dimensional database, where each data point represents the comprehensive state of the concrete at a specific time point and a specific spatial location, with dimensions including physical and chemical parameters, structural morphology parameters, and environmental history parameters, and the time dimension as the main line throughout all parameters, enabling the degradation process of concrete performance to be completely traced and quantitatively analyzed. The multi-dimensional durability feature space output by the feature fusion processing unit is not a simple data set, but an ordered structure that can depict the entire process from micro material degradation to macro structural response of concrete. The data flow process is automated, but each processing step includes data quality verification steps, such as judging the reasonableness of the interpolated data after spatio-temporal alignment and evaluating the significance of the features after feature extraction, which ensures the credibility of the fused features.

[0028] Example 2: see Figure 3 The implementation of the damage mode recognition unit begins with an in-depth analysis of the multi-dimensional durability feature space, which is constructed by the feature fusion processing unit and contains comprehensive information about the degradation of concrete performance. The core task of the unit is to establish a classification model in this high-dimensional space that can automatically classify the concrete state at any time into the pre-set damage development stages. The implementation process first defines four representative damage development stages, namely the initial stable period, the local degradation period, the accelerated degradation period, and the critical failure period, in the feature space based on a large amount of historical engineering data, laboratory accelerated aging test data, and theoretical models. Each stage is not simply divided by a single threshold, but is defined by a specific range of values of a group of feature parameters and the pattern of their mutual relationship. For example, the features of the initial stable period may show that the chloride ion concentration is below the critical value, the crack network connectivity is zero, and the environmental erosion intensity vector is in low fluctuation; while the accelerated degradation period may correspond to the chloride ion concentration reaching the critical threshold on the surface of the steel bar, the crack appearing active expansion and forming a connected path, and the steel corrosion potential significantly negative shift.

[0029] After the theoretical framework of phase division is established, the accumulated historical and real-time monitoring data are subjected to unsupervised learning by using the density clustering algorithm to discover the natural clustering of data points in the feature space. This algorithm can adapt to the high dimensionality of the feature space and the complexity of data distribution. It does not pre-set the shape and number of clusters, but identifies dense regions according to the density distribution of data points themselves. The algorithm scans the entire feature space, marks those data points with enough neighbors within a specified radius as core points, and groups the core points and their neighbors that are connected to each other into the same cluster. These clusters represent different damage phases in the feature space as specific "cloud clusters". The center point of each cluster is obtained by calculating the average value of the feature parameters of all data points in the cluster. The combination of feature parameters of this center point is regarded as the "typical mode" or "prototype" of the damage phase. For example, the center point of a cluster may contain a chloride ion concentration median, an average crack length, and a typical corrosion potential value. This combination quantifies the typical state of the "local deterioration period". Based on the above clustering results, the next step is to establish a discriminant function that can perform real-time classification. Mahalanobis distance is chosen as the basis of the discriminant function because it takes into account the correlation between feature parameters, overcoming the inaccuracy of Euclidean distance when parameter scales are not the same and are correlated. The system will pre-compute and store the feature vectors of the four typical modes representing the four damage phases and their covariance matrices. When new monitoring data is fused to form a new feature vector, the discriminant function will calculate the Mahalanobis distance from this new vector to the four typical mode feature vectors. The calculation of Mahalanobis distance is essentially a kind of weighted distance measurement, and the weights are determined by the inverse matrix of the covariance matrix of the feature parameters. This makes the distance compressed in the direction of strong correlation and stretched in the direction of strong independence, thus more accurately reflecting the true "similarity" between data points.

[0030] The four calculated distance values represent the difference between the current state and the four typical damage states, and the identification logic follows the "nearest neighbor" principle, that is, the damage stage with the smallest Mahalanobis distance is selected as the current identification result. This identification process is dynamic and continuous, and the system obtains the latest feature vector from the feature fusion processing unit at fixed time intervals (for example, every hour or every day), performs a classification and discrimination once, and thus generates a continuous time series label about the health state of the concrete structure. This time series not only indicates the current stage, but also reveals the progress of damage development through the jumps between stages. The damage state classification model is not fixed, and it has an update mechanism built in. When the system runs for a period of time and accumulates enough new monitoring data, especially when the detection result feedback unit provides label data verified in the field, the system will start the retraining process of the model. The retraining process may include fine-tuning of the cluster centers, or recalculating the covariance matrix based on new data, so that the discriminant function can better adapt to the particularity of the actual tunnel environment. The successful operation of this unit depends on the rationality of the feature space construction, the robustness of the clustering algorithm, and the scientificity of the distance measure.

[0031] Example 3: The implementation of the critical state prediction unit is based on the current damage development stage output by the damage pattern recognition unit. The core task of this unit is to predict the time when the concrete reaches the critical state by quantifying the performance degradation dynamics. Its operation begins with extracting the feature parameter change trajectory associated with the current identified stage from the multi-dimensional durability feature space. These parameters include but are not limited to chloride ion diffusion coefficient, crack propagation rate, and steel corrosion current density. When extracting the trajectory, the system will backtrack the historical data within a certain time period, and the length of this time period matches the typical duration of the current damage stage to ensure the integrity and representativeness of the analyzed trajectory. The local change rate of each feature parameter is calculated using the sliding time window method, and the window size is dynamically adjusted according to the data acquisition frequency and parameter fluctuation characteristics. For example, a larger window can be used for slowly changing chloride ion concentration to smooth random fluctuations, while a smaller window can be used for rapidly responding crack propagation data to capture transient changes. The method for calculating the local change rate is based on linear regression analysis of data points within the window, and the slope of the parameter value change with time in this time interval is obtained as the degradation rate of the parameter in this window.

[0032] After obtaining the local rate of change of each characteristic parameter, a mapping relationship between these rates and the remaining service life of the concrete needs to be established. This relationship is obtained by training a prediction model based on analysis of a large number of historical case data. The model takes into account the differences in the impact of the rate of change of each parameter on overall durability at different damage stages. For example, in the initial stable period, the chloride ion permeation rate may be the dominant factor, while in the critical failure period, the change in crack connectivity may be more critical. In order to obtain a comprehensive degradation rate index, the system weights and integrates the rates of change of each parameter. The weight coefficients are determined through expert knowledge base and historical data statistical analysis, reflecting the relative importance of each parameter in a specific environment. The weighting and integration process can be represented as a linear combination, expressed as:

[0033] wherein the symbol represents the calculated comprehensive degradation rate index, with a larger value indicating faster performance deterioration; the symbol represents the total number of characteristic parameters considered; the symbol represents the weight coefficient of the th characteristic parameter, satisfying ; the symbol represents the local rate of change of the th characteristic parameter calculated within the current sliding window. This comprehensive index condenses multi-dimensional degradation information into a single, interpretable quantitative value for subsequent time prediction.

[0034] When predicting the remaining time, the system compares the current comprehensive degradation rate with the pre-set critical state threshold to make a preliminary estimate of the remaining time. However, the preliminary estimate is uncertain, so a correction mechanism needs to be introduced. The correction process relies on a historical database that stores the actual time taken by concrete in similar tunnel projects to develop from the current state to the critical state at the same damage stage. The system retrieves the most similar cases from the historical data based on the current characteristic parameter combination, calculates the average duration of these cases at this stage, and adjusts for environmental condition differences to update the preliminary prediction value using Bayesian updating, resulting in a more reliable corrected remaining time. The entire prediction process is iterative, with the sliding window moving forward as new monitoring data continuously flows in, the local rate of change is recalculated, the comprehensive degradation rate index is updated, and the predicted value of the remaining time is dynamically adjusted, achieving online learning and self-adaptation of the prediction model.

[0035] The detection scheme generation unit then automatically triggers the corresponding detection strategy decision logic according to the predicted remaining time value output by the critical state prediction unit. The core basis of the decision is to compare the predicted remaining time with one or more preset time thresholds, which are set based on engineering specifications, risk assessment, and economic benefit analysis. When the predicted remaining time is greater than a higher threshold (for example, five years), the system judges that the structure is in a relatively safe state, and therefore generates a conventional detection scheme mainly using non-damage detection methods. The conventional detection scheme focuses on macroscopic and rapid general inspection, and the detection items may include visual inspection of the entire tunnel, knocking method inspection of hollowing, and comprehensive radar scanning of the steel reinforcement layer thickness. The detection period is set in proportion to the predicted remaining time, for example, set to a certain fixed percentage of the remaining time, such as 20%, which means that if the predicted remaining time is ten years, the detection period is set to two years, and this setting makes the detection frequency gradually increase with the deterioration of the structure state.

[0036] When the predicted remaining time decreases to below the higher threshold mentioned above but is still higher than a lower threshold (for example, two years), the system may maintain the conventional detection scheme but appropriately shorten the detection period, for example, adjust the period to a smaller percentage of the remaining time to increase the monitoring density. Once the predicted remaining time falls below the lower threshold, the system determines that the structure enters a high-risk period and immediately switches to a special detection scheme. The special detection scheme includes more detailed and invasive detection methods, and the core item is core sampling and detailed physical and chemical analysis in the laboratory, such as measuring the compressive strength of concrete, the chloride ion content profile, and microscopic structure observation. The period of special detection is significantly shortened, usually set to half or less of the period of the conventional scheme, to ensure that the deterioration process can be closely tracked. In addition, the scheme generation logic also deeply integrates damage mode information, especially when the damage mode recognition unit indicates that it is currently in the accelerated degradation period, the generated scheme will forcibly increase targeted monitoring items, such as installing deformation meters at key locations of cracks to continuously monitor the crack propagation rate, or laying linear polarization resistance sensor arrays to directly measure the instantaneous corrosion rate of steel bars.

[0037] The output of the detection scheme generation unit is a series of executable instructions that explicitly specify the specific location of the detection task (high-risk grid area identified based on feature space), the recommended list of detection methods, and the detailed implementation period. The scheme generation process is not rigidly executed, and it embeds a resource optimization algorithm that takes into account the constraints of tunnel operation, such as scheduling detection tasks during traffic off-peak periods and preferentially allocating resources to the pipe section area with the highest predicted risk and the most important structure. The entire process from prediction to scheme generation forms a closed loop, and the implementation effect of the detection scheme is recorded and evaluated by the detection result feedback unit for continuous optimization of the prediction model and decision threshold, enabling the system to have the ability of self-improvement. The collaborative work of critical state prediction and detection scheme generation realizes the seamless connection from state monitoring to maintenance decision-making, providing a precise action guide for preventive maintenance of the tunnel.

[0038] Embodiment 4: The implementation of the data quality verification unit runs throughout the entire process from raw data collection to feature fusion. Its operating mechanism can be illustrated by a hypothetical tunnel pipe section monitoring scenario, which is equipped with four chlorine ion sensors numbered C1 to C4 and four steel bar potential sensors numbered P1 to P4. The unit continuously monitors the integrity and internal consistency of the data stream, for example, in a certain data collection period, the system finds that the transmission data packet of sensor C3 is missing, resulting in a data gap. At this time, the integrity verification module will start the spatial interpolation program, which estimates the value of the C3 position based on the spatial correlation principle using the readings of adjacent sensors C2 and C4 at the same time. The interpolation algorithm may use the inverse distance weighting method, which considers that the closer the sensors, the higher the correlation of their readings, and therefore gives them greater weight, thus generating an estimated value for the C3 position, which is marked as "interpolated" and stored separately from the original data.

[0039] Data consistency check is performed after interpolation, the system compares the logical relationship of different sensor data at the same monitoring point or adjacent monitoring points. For example, at the grid coordinate G-07 position, the chloride ion sensor C1 reading shows a concentration value of 0.15% (by mass of cement), while according to the historical model, the corresponding steel bar potential sensor P1 reading at this concentration level should theoretically be in the range of -250mV to -350mV (relative to Cu / CuSO4 electrode). However, the actual received P1 data is -120mV, which deviates significantly from the expected range of correlation with chloride ion concentration, and the difference exceeds the preset allowed threshold. At this time, the consistency check module determines that the data pair is abnormal, and immediately starts the redundant measurement mechanism. Redundant measurement may be performed by activating the backup sensor in this area (if available), or generating an instruction to request maintenance personnel to go to the G-07 position and use a portable potentiostat for on-site manual review measurement. The potential value obtained by manual review is -320mV, which is significantly different from the initial P1 sensor reading of -120mV.

[0040] Based on the review results, the system labels and processes the original data points. The initial P1 reading of -120mV is labeled as "suspected abnormality", and the manual review value of -320mV is labeled as "verified". The system will calculate the confidence weight of the abnormal data point, and the weight of the initial abnormal data will be significantly reduced, for example, set to 0.2, while the weight of the verified data remains 1.0. In the subsequent feature fusion processing stage, when the data at this time point is used to calculate the feature vector, these weight values will participate in the calculation, for example, in the calculation of the average potential, the abnormal data will greatly reduce the contribution to the final result due to its low weight. This weight-based data processing method avoids interference caused by a few abnormal points to the overall judgment, and enhances the robustness of the system.

[0041] Table 1: Sensor data consistency check and labeling

[0042] The data quality verification unit also integrates an important drift compensation module, which is dedicated to solving the inevitable performance degradation or calibration offset problem of sensors in long-term operation. The module regularly (such as every quarter) starts a calibration process, in which a batch of standard test blocks with the same mix proportion and curing conditions as the tunnel segment concrete and placed in the same environment are measured. The reference physical and chemical parameters of these test blocks are known, such as the accurate measurement of their chloride ion content by laboratory titration method. During calibration, the monitoring sensors installed near or integrated with the test blocks will simultaneously read the measurement values. The drift compensation module collects the reference values and measurement values of all sensors in a period, and establishes the "measurement value-reference value" deviation curve of each sensor. For an ideal sensor, its deviation curve should fluctuate slightly around zero. However, a chloride ion sensor in actual operation may show a continuous upward or downward drift trend. The module fits this deviation curve by linear regression analysis, calculates the average drift of the sensor in the current period and its trend, i.e. the drift compensation coefficient. In subsequent daily data processing, each raw reading of the sensor will be automatically corrected by subtracting this compensation coefficient before entering the subsequent feature fusion and other links. For example, a chloride ion sensor with a calculated +0.02% positive drift, when its reading is 0.10%, the corrected value will be 0.08%. When the drift compensation coefficient of a sensor continues to increase and exceeds the set critical threshold (for example, the drift reaches 5% of the full scale), the drift compensation module will determine that the performance of the sensor has been seriously degraded. At this time, the module will not only continue to perform data correction, but also generate a high-level sensor calibration reminder signal to the system administrator.

[0043] Within the data quality verification unit, the overall credibility weight of the sensor will be systematically lowered, and its influence in fusion calculation will decrease until the maintenance personnel completes the on-site calibration or replacement of the sensor and resets its weight and compensation coefficient. This drift compensation mechanism effectively maintains the accuracy and comparability of long-term monitoring data, providing a reliable time series basis for durability assessment. The entire data quality verification unit works like a precise filter, layer by layer screening and correcting the data flowing into the system. It first ensures the completeness of the data, then verifies the internal consistency and reasonableness of the data, and finally counteracts the measurement error caused by time through drift compensation. Through a series of combined strategies such as real-time monitoring, redundant verification, weight allocation and drift compensation, the unit provides high-quality and reliable data raw materials for the entire durability detection system.

[0044] Example 5: The implementation of the detection result feedback unit constitutes a closed loop of system self-learning and optimization, whose operation begins with the comparative analysis of the field detection results and the system prediction results. This unit is provided with a dedicated database for storing the prediction records generated by the system each time and the results records verified by the subsequent actual detection actions. For example, the system may predict that the segment A area of the Xth ring of the tunnel is in the "local deterioration period" and the remaining service life is eight years. When the maintenance team carries out core sampling and laboratory analysis on this area according to the instructions generated according to the detection scheme, the feedback unit will collect the actual detection report. The report may indicate that the chloride ion concentration of the concrete in this area has approached the critical value, the steel bar has appeared slight corrosion, the actual state is evaluated as the later stage of the "local deterioration period", and the expert estimates that the remaining service life is about seven years. The system compares the predicted "local deterioration period" with the measured "local deterioration period", and determines that the stage prediction is correct; at the same time, the relative error between the predicted remaining time (eight years) and the actual evaluation time (seven years) is calculated, which is about 14%. This error value will be recorded as a data point of the accuracy index.

[0045] The calculation of the accuracy index is periodic, and an average accuracy or error distribution will be calculated after a certain number of detection feedbacks (for example, ten times). According to this statistical result, the feedback unit will dynamically adjust the parameters of the upstream model. If it is found that the system's prediction of the "accelerated deterioration period" is consistently optimistic (i.e. the predicted remaining time is always longer than the actual time), the feedback unit may trigger the adjustment of the discrimination threshold of the "accelerated deterioration period" in the damage state classification model. For example, the crack width greater than 0.3 mm was originally used as one of the indicators of the "accelerated deterioration period", and now this threshold may be revised to 0.25 mm, so that the model can identify this stage earlier. At the same time, the weight parameter used to calculate the comprehensive deterioration rate in the critical state prediction unit may also be adjusted. If it is found that the actual importance of a certain characteristic parameter (such as crack propagation rate) is higher than the initial setting, the weight coefficient in the weighted fusion will be appropriately increased.

[0046] One important trigger mechanism is continuous bias monitoring. Suppose the system's prediction results for a certain tunnel segment deviate significantly from the actual detection results for three consecutive times, for example, the predicted damage phase does not match the actual phase for three consecutive times, or the error in the remaining time exceeds the allowed range (e.g., 25%). This systematic deviation indicates that the current model may no longer be suitable for the current environmental conditions or material state of the tunnel segment. At this time, the feedback unit will not only fine-tune the parameters, but will trigger a model retraining process. This process will use all the new data accumulated in the recent period (including monitoring data and corresponding detection results) as a new training set, possibly using incremental learning or batch learning, to re-execute the feature space partitioning and damage mode clustering process. This means that the typical mode feature vectors and classification boundaries representing each damage phase will be updated to better adapt to the latest evolution of the structure state.

[0047] The scheme optimization module embedded in the detection result feedback unit focuses on improving the efficiency and effectiveness of the detection action itself. This module records detailed metadata of each executed detection scheme, including the types of detection means used, the time spent, the equipment and personnel costs invested, and the role of the final detection in confirming or correcting the system's judgment. Based on these records, the module establishes an evaluation index system, with core indicators including detection efficiency (such as average time per detection and unit length tunnel detection cost) and detection accuracy (such as defect detection rate and false positive rate). The scheme optimization module will perform cost-benefit analysis based on these data. For example, suppose in a detection scheme for a tunnel in the "local deterioration period", the conventional items include full-face laser scanning and knock inspection, and the new item is "infrared thermal imaging detection". The scheme optimization module will analyze historical data and compare the diagnostic accuracy of detection schemes with and without infrared thermal imaging. If the analysis finds that adding infrared thermal imaging can only improve the diagnostic accuracy by less than 1%, but the cost increases by 15%, and this means does not provide unique information that cannot be replaced by other methods in this damage phase, then the optimization module may decide to automatically exclude the infrared thermal imaging item in future detection schemes for similar conditions, and focus resources on more effective detection means.

[0048] The optimization module also considers the importance level and on-site accessibility of tunnel segments. Importance level is determined based on the segment's load-bearing role in the tunnel structure (e.g., main load-bearing segments, connecting segments) and the critical equipment attached to it (e.g., cables, pipelines). Accessibility involves factors such as inspection window, traffic organization difficulty, and safety risks. The module uses an optimization algorithm to prioritize the allocation of limited inspection resources to areas with high importance, high predicted risk, and good accessibility. For example, for a critical load-bearing segment under a pedestrian tunnel, even if its predicted risk is the same as a non-load-bearing segment in a ventilation shaft, the system will allocate more frequent and refined inspection resources. Conversely, for a segment with low predicted risk located in a difficult maintenance area, the system may appropriately extend its inspection cycle or adopt remote monitoring as the primary method. Through this continuous feedback, comparison, adjustment, and optimization, the inspection result feedback unit transforms the entire durability inspection system from a static tool into an intelligent entity that learns from practice and evolves over time.

[0049] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0050] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A durability testing system for cross-river and cross-sea shield tunnels using double-fiber-coated concrete, characterized in that, include: The durability data acquisition unit is used to acquire multi-source monitoring data of the concrete of the shield tunnel segment under service environment. The multi-source monitoring data includes material performance parameters, environmental erosion parameters and structural response parameters. The feature fusion processing unit is used to align and integrate multi-source monitoring data according to time series, extract the performance degradation characteristics of concrete at different erosion stages, and construct a multi-dimensional durability feature space. The damage pattern recognition unit establishes a damage state classification model based on a multidimensional durability feature space to identify the current damage development stage of the concrete. The critical state prediction unit calculates the performance degradation rate based on the change trajectory of the damage development stage and predicts the time node when concrete reaches the critical durability state. The detection scheme generation unit automatically matches the optimal detection scheme based on the critical state prediction results and outputs execution instructions including detection location, detection method and detection cycle.

2. The durability testing system for cross-river and cross-sea shield tunnels with double-fiber-coated concrete as described in claim 1, characterized in that, The specific method by which the durability data acquisition unit acquires multi-source monitoring data is as follows: Data on chloride ion concentration distribution inside concrete and steel corrosion potential are collected using an embedded sensor array. The morphological characteristics and distribution density parameters of apparent cracks in concrete were obtained using surface scanning equipment. Environmental monitoring devices were used to record temperature and humidity variation curves and contact frequency data of corrosive media inside the tunnel.

3. The durability testing system for cross-river and cross-sea shield tunnels with double-fiber-coated concrete as described in claim 2, is characterized in that, The specific steps for the feature fusion processing unit to construct the multidimensional durable feature space are as follows: Chloride ion concentration distribution data and steel corrosion potential data are matched by gridding according to spatial coordinates to generate a material performance degradation characteristic matrix. Topological structure analysis was performed on the morphological characteristic parameters of the apparent cracks to extract the crack network connectivity and propagation direction characteristics. A time-domain correlation analysis was performed between the temperature and humidity change curves and the contact frequency data of the corrosive medium to establish a characteristic vector of environmental erosion intensity. The material performance degradation feature matrix, crack network features, and environmental erosion intensity feature vector are mapped to a unified time coordinate system to form a durability feature space that includes spatial and temporal dimensions.

4. The durability testing system for cross-river and cross-sea shield tunnels with double-fiber-coated concrete as described in claim 3, characterized in that, The specific method for the damage pattern recognition unit to establish a damage state classification model is as follows: Four damage development stages are divided in the durability feature space, corresponding to the initial stabilization period, local degradation period, accelerated degradation period and critical failure period, respectively. Density clustering algorithm is used to identify clustered regions of data points in the feature space, and typical patterns of each damage stage are determined based on the combination of feature parameters of the center point of the clustered region. A discrimination function based on Mahalanobis distance is established to calculate the distance between the current monitoring data and each typical pattern in real time, and the damage stage corresponding to the minimum distance value is taken as the identification result.

5. The durability testing system for cross-river and cross-sea shield tunnels with double-fiber-coated concrete as described in claim 4, characterized in that, The specific process by which the critical state prediction unit calculates the performance degradation rate is as follows: Extract the trajectory of characteristic parameter changes corresponding to the current stage of damage development, and calculate the local rate of change of each characteristic parameter using a sliding time window; Establish a mapping relationship between the rate of change of characteristic parameters and the remaining service time, and obtain a comprehensive degradation rate index through weighted fusion; The predicted remaining time for the current stage is adjusted based on the duration of development of the same damage stage in historical data.

6. The durability testing system for cross-river and cross-sea shield tunnels with double-fiber-coated concrete as described in claim 5, characterized in that, The specific logic of the detection scheme generation unit matching the optimal detection scheme is as follows: When the predicted remaining time is greater than a preset threshold, a conventional inspection scheme based on non-destructive testing is generated, and the inspection cycle is set to a fixed proportion of the predicted remaining time. When the predicted remaining time is less than a preset threshold, a special testing plan including core sampling and laboratory analysis is generated, shortening the testing cycle to half that of the conventional plan. For tunnel segments in the accelerated degradation phase, crack propagation monitoring and steel corrosion rate monitoring should be added to the testing plan.

7. The durability testing system for cross-river and cross-sea shield tunnels with double-fiber-coated concrete as described in claim 1, characterized in that, It also includes a data quality verification unit, whose workflow is as follows: Real-time monitoring of the integrity and consistency of data collected by each sensor, and spatial interpolation of missing data using data from adjacent sensors; When the difference in monitoring data from different sensors at the same location exceeds the allowable range, a redundant measurement mechanism is activated to verify the data. Abnormal data points are marked and their credibility weights are calculated. During the feature fusion processing stage, the proportion of data participating in the calculation is adjusted according to the weight values.

8. The durability testing system for cross-river and cross-sea shield tunnels with double-fiber-coated concrete as described in claim 7, characterized in that, The data quality verification unit also includes a drift compensation module, the compensation method of which is as follows: Regularly collect benchmark test data of standard concrete test blocks and establish a deviation curve between sensor measurements and benchmark values; The drift compensation coefficient of each sensor is calculated based on the trend of the deviation curve, and the original monitoring data is automatically corrected during the data processing stage. When the drift compensation coefficient exceeds the critical threshold, a sensor calibration reminder signal is generated and the data weight of the sensor is adjusted.

9. The durability testing system for cross-river and cross-sea shield tunnels with double-fiber-coated concrete as described in claim 1, characterized in that, It also includes a test result feedback unit, whose operating mechanism is as follows: The on-site detection results are compared and analyzed with the system prediction results to calculate the prediction accuracy index. The discrimination threshold and weight parameters of the critical state prediction unit of the damage state classification model are dynamically adjusted based on the accuracy index. When the deviation of three consecutive detection results exceeds the allowable range, the model retraining process is triggered and the feature space partitioning criteria are updated.

10. The durability testing system for cross-river and cross-sea shield tunnels with double-fiber-coated concrete according to claim 9, characterized in that, The detection result feedback unit also includes a scheme optimization module, whose optimization strategy is as follows: Record the actual implementation effect of each testing plan and establish an evaluation index system for testing efficiency and accuracy; If the improvement in diagnostic accuracy by a newly added test item is less than the preset value, the test item will be removed from the subsequent plan. Optimize the spatial allocation scheme of testing resources based on the importance level and accessibility of the pipeline area.

Images