Intelligent durability monitoring and life evaluation method for new structure of port engineering

Abstract

The invention relates to the technical field of structure monitoring, and discloses a port engineering new structure durability intelligent monitoring and life evaluation method, which comprises the following steps: identifying and monitoring a key degradation area; fusing multi-modal data and quantifying structural damage; predicting residual life based on transfer learning; and perspective optimization of a monitoring strategy and self-adaptive correction of an evaluation model are realized through double feedback closed loops of prediction monitoring and a decision model. The system comprises an acquisition module which identifies a key area and acquires multi-modal data through a stress field, chemical field and geometric field coupling model and an optimization algorithm; the fusion module executes dynamic credibility fusion to generate a structure damage comprehensive index; the prediction module is used for predicting the residual service life of the structure through a transfer learning model; the re-evaluation module dynamically re-evaluates the key area according to the prediction result; and the self-correction module responds to the maintenance record to update each model in the system. According to the invention, the problem of insufficient monitoring data of a new structure is solved, and the durability monitoring and prospective life evaluation of the structure are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of structural monitoring, in particular to a new port structure durability intelligent monitoring and life assessment method. BACKGROUND

[0002] Port engineering structures, as key infrastructure for marine transportation and economic activities, have long served in harsh marine environments with chloride salt corrosion, high humidity, and complex loads. Their structural durability and service life are the core issues of concern in the engineering community. In order to ensure their operational safety, it is crucial to effectively monitor the health and assess the life of port structures.

[0003] However, the existing technology has inherent limitations in practical application. Traditional structural health monitoring methods rely heavily on the measurement of a single physical quantity (such as strain or potential). This approach cannot fully and accurately reveal the true damage state of the structure under the complex coupling of stress, chemical corrosion, and material aging. At the same time, the selection and placement of monitoring points largely depend on engineering design experience or existing damage signs. This static and subjective monitoring strategy may not capture new and unexpected critical deterioration areas over time, lacking sufficient foresight and dynamic adaptability.

[0004] In addition, the high dynamics and strong corrosion characteristics of marine environments often result in significant noise and signal drift in sensor data, which seriously interferes with the extraction of true damage information, limiting the spatial and temporal accuracy of damage identification. For newly built structures that lack long-term monitoring data accumulation, traditional life prediction models often struggle to provide reliable assessment results due to insufficient data samples, and cannot effectively draw on and utilize the deterioration evolution knowledge of a large number of similar old structures, resulting in significant constraints on the accuracy and reliability of life assessment.

[0005] Therefore, the present application proposes a new port structure durability intelligent monitoring and life assessment method to address the shortcomings of the existing technology. SUMMARY

[0006] To address the shortcomings of the existing technology, the present application provides a new port structure durability intelligent monitoring and life assessment method, which solves the problem of inaccurate damage assessment and low reliability of life prediction for new structures due to the static monitoring strategy, single data dimension, and lack of historical experience transfer in port structure monitoring technology.

[0007] To achieve the above purpose, the present application realizes the following technical solutions: a new port structure durability intelligent monitoring and life assessment method, the method comprising the following steps:

[0008] S1, identifying a key deterioration area on a port structure and collecting multi-modal monitoring data from the key deterioration area;

[0009] S2, performing a dynamic credibility fusion process on the multi-modal monitoring data to generate a structural damage comprehensive index, the dynamic credibility fusion process comprising dynamically adjusting credibility weights corresponding to the multi-modal monitoring data based on spatiotemporal continuity of the multi-modal monitoring data;

[0010] S3, inputting the structural damage comprehensive index into a transfer learning life prediction model to predict a remaining useful life of the port structure;

[0011] S4, using the predicted remaining useful life to perform a predictive analysis-based dynamic reevaluation of key deterioration zones to identify future drift risk zones, and using the future drift risk zones to update a model used to identify the key deterioration zones;

[0012] S5, and in response to an externally input maintenance activity record, performing a decision feedback-based model self-correction, the model self-correction comprising updating a deterioration analysis model used to identify key deterioration zones using the maintenance activity record, updating a credibility weight calculation model for a maintenance area in the dynamic credibility fusion process, and marking post-maintenance data in the transfer learning life prediction model.

[0013] Preferably, in step S1, the step of identifying key deterioration zones on the port structure comprises:

[0014] constructing a stress field, chemical field, and geometric field triple-coupling deterioration analysis model of the port structure to obtain an initial damage field distribution;

[0015] using a particle swarm optimization algorithm to identify key deterioration zones from candidate areas within the initial damage field distribution using the initial damage field distribution as input.

[0016] Preferably, in step S2, the step of dynamically adjusting credibility weights corresponding to the multi-modal monitoring data comprises:

[0017] based on a normalized data variation amplitude |ΔV i (t)| of the multi-modal monitoring data at time t, using the following formula to calculate a dynamically corrected credibility W corr,i (t) of sensor i at time t:

[0018] W corr,i (t) = W init,i · exp(-λ i · |ΔV i (t)|);

[0019] where W init,i is an initial credibility weight of sensor i, and W corr,i(t) represents the reliability of sensor i after dynamic correction at time t, λ i Let |ΔV be the penalty coefficient for sudden changes in sensor i's data. i (t)| represents the normalized data change magnitude of sensor i at time t.

[0020] Preferably, in step S2, the step of performing dynamic reliability fusion processing on the multimodal monitoring data to generate a comprehensive structural damage index includes:

[0021] With the aforementioned dynamic correction and credibility W corr,i Based on (t), calculate the fusion weight ω corresponding to each physical quantity. j (t);

[0022] And calculate the position using the following formula. Structural damage comprehensive index at time t :

[0023]

[0024] In the formula, ω is the comprehensive index of structural damage. j (t) represents the fusion weight of physical quantity j at time t. Let physical quantity j be at position The normalized monitoring value at time t, where j represents different physical quantities in the multimodal monitoring data.

[0025] Preferably, in step S3, the step of inputting the structural damage comprehensive index into a transfer learning lifetime prediction model to predict the remaining service life of the port structure includes:

[0026] Construct a source domain feature database containing historical data with similar structures;

[0027] Extract the target domain feature set from the comprehensive structural damage index;

[0028] The source domain feature database and the target domain feature set are input into the transfer learning lifetime prediction model to calculate the failure probability curve, and the remaining lifetime is determined based on the failure probability curve.

[0029] Preferably, step S4, and the step of updating the model used to identify the critical degradation zone using the future drift risk zone, includes:

[0030] The objective function in the particle swarm optimization algorithm is updated using the identified future drift risk region.

[0031] Preferably, in step S5, the step of performing model self-correction based on decision feedback in response to externally input maintenance activity records includes:

[0032] adjusting a local material physical parameter of a corresponding maintenance area in the deterioration analysis model according to a maintenance measure type recorded in the maintenance activity record;

[0033] suspending a spatiotemporal continuity penalty on sensor data in the maintenance area for a preset time period after completion of maintenance indicated by the maintenance activity record to correct the credibility weight calculation model;

[0034] labeling data collected after maintenance as post-intervention data in the transfer learning life prediction model to enable the model to distinguish between data evolution law changes caused by artificial intervention and natural deterioration in subsequent prediction.

[0035] Preferably, the method further comprises:

[0036] applying an environment-coupled dynamic compensation coefficient to the structure damage comprehensive index to obtain a corrected structure damage comprehensive index before inputting the structure damage comprehensive index into the transfer learning life prediction model;

[0037] the environment-coupled dynamic compensation coefficient is determined based on real-time collected tide level and temperature data;

[0038] wherein the corrected structure damage comprehensive index is input into the transfer learning life prediction model.

[0039] Preferably, the method further comprises:

[0040] using the failure probability curve as input, using reachability analysis method to generate a maintenance decision tree containing maintenance timing and measure suggestions.

[0041] The application also provides a harbor engineering new structure durability intelligent monitoring and life evaluation system, the system comprising:

[0042] a collection module for identifying key deterioration areas on a harbor engineering structure and collecting multi-modal monitoring data from the key deterioration areas;

[0043] a fusion module for performing dynamic credibility fusion processing on the multi-modal monitoring data collected by the collection module to generate a structure damage comprehensive index, the dynamic credibility fusion processing performed by the fusion module comprising dynamically adjusting credibility weights corresponding to the multi-modal monitoring data based on spatiotemporal continuity of the multi-modal monitoring data;

[0044] a prediction module for inputting the structure damage comprehensive index generated by the fusion module into a transfer learning life prediction model to predict remaining service life of the harbor engineering structure;

[0045] a reevaluation module configured to perform a predictive analysis-based dynamic reevaluation of the critical deterioration zone using the remaining useful life predicted by the prediction module to identify a future drift risk zone and update the model used by the acquisition module to identify the critical deterioration zone using the future drift risk zone;

[0046] a self-correction module configured to perform a decision feedback-based model self-correction in response to an externally input maintenance activity record, the model self-correction performed by the self-correction module including updating the deterioration analysis model used by the acquisition module using the maintenance activity record, updating the confidence weight calculation model for the maintenance zone used by the fusion module, and labeling post-maintenance data in the transfer learning life prediction model used by the prediction module.

[0047] The application provides a port engineering new structure durability intelligent monitoring and life evaluation method.

[0048] 1、The application can scientifically and quantitatively identify the critical region with the most severe deterioration on the port engineering structure by constructing a stress field, a chemical field and a geometric field triple-coupled deterioration analysis model and combining a particle swarm optimization algorithm for optimization.

[0049] 2、The application proposes a dynamic confidence fusion processing method for the problem of multi-source heterogeneous monitoring data fusion.

[0050] 3、The application uses a transfer learning life prediction model to solve the technical problem that a new service structure is difficult to perform long-term life prediction due to insufficient historical monitoring data.

[0051] 4、The present application has the foresight and self-adaptive ability by constructing the prediction monitoring and decision model double feedback closed loop, on the one hand, the system utilizes the life prediction result to dynamically reevaluate the key deterioration area, realizes the foresight optimization of monitoring strategy, on the other hand, the system can respond to external maintenance record, automatically corrects the internal deterioration analysis, data fusion and life prediction model. This double closed loop mechanism makes the method not a static evaluation tool, but an intelligent evaluation system that can run through the whole life cycle of structure and self-evolve. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 The port engineering new structure durability intelligent monitoring and life evaluation method flow chart of the present application;

[0053] Figure 2 The port engineering new structure durability intelligent monitoring and life evaluation system structure block diagram of the present application;

[0054] Figure 3 The key deterioration area identification and multi-modal data acquisition flow chart of the present application;

[0055] Figure 4 The health state grading atlas generation and application schematic diagram of the present application.

[0056] Wherein, 10, acquisition module;20, fusion module;30, prediction module;40, reevaluation module;50, self-correction module. DETAILED DESCRIPTION

[0057] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the specification of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0058] Referring to the drawings Figure 1 , Figure 1 It is the flow chart of the port engineering new structure durability intelligent monitoring and life evaluation method according to an embodiment of the present application. The method can include the following steps:

[0059] Step S1, identifying the key deterioration area on the port engineering structure, and collecting multi-modal monitoring data from the key deterioration area.

[0060] Step S2, performing dynamic credibility fusion processing on the multi-modal monitoring data to generate a structure damage comprehensive index.

[0061] Step S3, inputting the structure damage comprehensive index into a transfer learning life prediction model to predict the remaining service life of the port structure.

[0062] Step S4, using the predicted remaining service life to perform a dynamic reevaluation of the key deterioration zone based on predictive analysis to update the model used to identify the key deterioration zone.

[0063] Step S5, in response to an externally input maintenance activity record, performing a model self-correction based on decision feedback to update the relevant model used in the method.

[0064] In an embodiment of the present application, the above steps constitute a complete technical process comprising a double feedback closed loop. Steps S1 to S3 constitute the core forward analysis link, i.e. from data acquisition to life prediction. Step S4 constitutes the first feedback closed loop, feeding back the prediction result of step S3 to step S1 for prospectively adjusting the monitoring area. Step S5 constitutes the second feedback closed loop, feeding back external maintenance intervention information to steps S1, S2 and S3 for correcting the physical model, fusion model and prediction model inside the system, so that the system can adapt to human changes in the structure state.

[0065] Referring to the drawings Figure 2 , Figure 2 is a structural diagram of a port structure durability intelligent monitoring and life evaluation system according to an embodiment of the present application. The system is used to perform any of the methods described above, and the system comprises an acquisition module 10, a fusion module 20, a prediction module 30, a reevaluation module 40 and a self-correction module 50.

[0066] The acquisition module 10 is used to perform step S1, identify the key deterioration zone on the port structure, and acquire multi-modal monitoring data from the key deterioration zone. The output data of the acquisition module 10 is transmitted to the fusion module 20.

[0067] The fusion module 20 is used to perform step S2, and its input end is connected to the output end of the acquisition module 10. The fusion module 20 performs dynamic credibility fusion processing on the multi-modal monitoring data to generate a structure damage comprehensive index, and transmits the index to the prediction module 30.

[0068] The prediction module 30 is used to perform step S3, and its input end is connected to the output end of the fusion module 20. The prediction module 30 inputs the structure damage comprehensive index into a transfer learning life prediction model to predict the remaining service life of the port structure. The remaining service life output by the prediction module 30 is transmitted to the reevaluation module 40.

[0069] A reevaluation module 40, for performing step S4, is connected to the output of the prediction module 30. The reevaluation module 40 uses the predicted remaining service life to perform a dynamic reevaluation of the critical deterioration zone based on predictive analysis, and transmits the updated model parameters to the acquisition module 10 to adjust the identification model of the critical deterioration zone thereof.

[0070] A self-correction module 50, for performing step S5, receives an externally input record of maintenance activities. The self-correction module 50 transmits instructions and data to the acquisition module 10, the fusion module 20 and the prediction module 30, respectively, according to the record, to update the respective internally used deterioration analysis model, the reliability weight calculation model and the parameters for marking the post-maintenance data.

[0071] Referring to the accompanying drawings Figure 3 , Figure 3 is a detailed flowchart of step S1, i.e. critical deterioration zone identification and multi-modal data acquisition, according to an embodiment of the present application. This step of the present application is performed by the acquisition module 10, and aims to achieve accurate locking and data acquisition of the critical position of structural damage by means of a combination of physical models and optimization algorithms, instead of the traditional experience-dependent monitoring point layout method.

[0072] In this embodiment, the acquisition module 10 first determines the types of multi-modal monitoring data to be acquired. In order to comprehensively represent the damage state of the port concrete structure under the multi-dimensional coupling action of mechanics, chemistry and material degradation, three types of core monitoring data are selected: micro-strain, chloride concentration gradient and resistivity mutation. The micro-strain data are used to represent the local stress response state of the structure under constant and variable loads; the chloride concentration gradient data are used to represent the erosion degree and spatial distribution of chloride ions in the concrete, which is a key inducement for steel bar corrosion; the resistivity mutation data are used as a comprehensive indicator of the degree of material degradation. The change of the concrete resistivity can reflect the internal microstructure, such as the development state of porosity, water content and micro-cracks. A rapid and significant change in the value of the resistivity in a specific region directly indicates the aging, internal cumulative damage or deterioration of the material in that region.

[0073] Before physical deployment, the acquisition module 10 constructs a three-dimensional coupled deterioration analysis model of the stress field, chemical field, and geometric field of the port structure. The model can be established based on the finite element method. First, the geometric model of the structure, material physical parameters, boundary conditions, and known environmental loads and service loads are input to perform stress field simulation analysis, and the overall micro-strain distribution field of the structure is output. Subsequently, a chemical field analysis model is established based on Fick's-second-law to simulate the migration and diffusion process of chloride ions in the concrete pores under the set marine environmental conditions, and the chloride ion concentration gradient distribution field of the structure at different depths is output. At the same time, a geometric field deterioration analysis model is constructed, which is used to simulate the deterioration of the material microstructure. Specifically, a continuous medium damage model based on damage mechanics can be introduced to equivalently reduce the local elastic modulus of the material due to the initiation and propagation of micro-cracks in the material. In addition, the change in material resistivity can be associated with the change in pore structure, for example, a function relationship between resistivity p and porosity φ, saturation S w and crack parameters can be established according to the extended form of Archie's-Law, thereby establishing a quantitative correlation between the geometric field (micro-crack damage) and the resistivity mutation field. The distribution of these three fields constitutes the initial damage field distribution.

[0074] The acquisition module 10 takes the above initial damage field distribution as input to identify the key deterioration area. Specifically, a particle swarm optimization (PSO) algorithm is used to search for the optimal solution in the entire candidate area covered by the initial damage field distribution. The objective function of the algorithm is set to find the spatial position point p that maximizes the comprehensive damage index F(p) defined as follows:

[0075] F(p) = a · S ∈ (p) + β · G Cl (p) + γ · R ρ (p);

[0076] where p is the three-dimensional spatial position vector in the structure; F(p) is the comprehensive damage index at position p; S ∈ (p) is the normalized micro-strain sensitivity at position p; G Cl (p) is the normalized chloride ion concentration gradient amplitude at position p; R ρ (p) is the normalized resistivity mutation rate at position p; a, β, γ are the preset weighting coefficients of the three terms respectively, and a + β + γ = 1. The particle swarm optimization algorithm iteratively searches and finally outputs one or more regions with maximum comprehensive damage index F(p), which are determined as the key deterioration area, and outputs the accurate three-dimensional spatial coordinates and boundary range.

[0077] According to the key deterioration zone position and range output by the particle swarm optimization algorithm, the acquisition module 10 performs physical deployment of the multi-modal sensor array. The multi-modal sensor array including micro-strain sensors, chloride ion concentration sensors and resistivity sensors is deployed inside the key deterioration zone. The deployment density of the sensors is proportional to the spatial gradient of the damage field calculated by the deterioration analysis model in the key deterioration zone, that is, dense deployment is performed at the predicted position of the sharp change in damage. After deployment, the acquisition module 10 collects data from the sensors according to the preset sampling frequency f s The data collected by each sensor is synchronized, and the raw data stream is preprocessed, which includes the removal of abnormal data points based on statistical methods (such as the 3-sigma principle) and the interpolation of missing data based on spatial and temporal correlation. Finally, standardized multi-modal monitoring data with accurate time stamps and spatial coordinates are generated and transmitted to the fusion module 20.

[0078] In a specific embodiment of the present application, a dynamic credibility fusion process is performed on the multi-modal monitoring data to generate a structural damage comprehensive index in step S2, which is performed by the fusion module 20. The module receives the standardized multi-modal monitoring data with accurate spatial and temporal labels transmitted by the acquisition module 10.

[0079] The fusion module 20 first parses the received data packet, and constructs a space-time data sequence corresponding to the spatial distribution of the physical sensor array according to the spatial coordinates and time stamp information contained therein. The sequence aligns the data of sensors of different types and different positions on a unified time axis, providing a structured data basis for subsequent spatio-temporal continuity analysis and fusion processing.

[0080] Subsequently, the fusion module 20 performs dynamic credibility weight calculation. This process first sets an initial credibility weight W init,i for each sensor i in the array, which is determined according to the sensor factory calibration parameters or historical operation stability data, reflecting the baseline measurement accuracy of the sensor. To overcome the influence of environmental noise, sensor drift or sudden failure on data quality, the fusion module 20 further introduces a dynamic correction mechanism. This mechanism evaluates the credibility of each sensor data point at the current time by analyzing the deviation of each sensor data point from its historical data in the time dimension and the deviation of adjacent sensor data in the spatial dimension. When the data of a sensor is significantly inconsistent with its spatio-temporal neighborhood data, the data variation amplitude will increase.

[0081] The fusion module 20 calculates the dynamic corrected credibility W i of sensor i at time t based on the normalized data variation amplitude |ΔV corr,i (t) of the multi-modal monitoring data at time t using the following formula:

[0082] W corr,i (t)=W init,i ·exp(-λ i ·|ΔV i (t)|);

[0083] In the formula, W init,i W represents the initial confidence weight for sensor i. corr,i (t) represents the reliability of sensor i after dynamic correction at time t, λ i Let |ΔV| be the penalty coefficient for sudden changes in sensor i's data. The magnitude of this coefficient determines the sensitivity of the reliability to data fluctuations. i (t)| represents the normalized data change magnitude of sensor i at time t, which incorporates the inconsistencies between the time and spatial dimensions.

[0084] In this embodiment, the normalized data variation amplitude |ΔV i (t) can be calculated using the following formula:

[0085]

[0086] In the formula, V i (t) represents the measurement value of sensor i at time t; V i (t-1) represents the measurement value of the sensor at the previous moment; V N(i) (t) represents the average value of the measurements taken by the set N(i) of sensors spatially adjacent to sensor i at time t; V range This is the normalization factor for the range of this type of sensor; w T and w S Let w be the weighting coefficients that characterize the differences in the time and spatial dimensions, respectively. T +w S =1.

[0087] After obtaining the confidence levels of each sensor after dynamic correction, the fusion module 20 performs data fusion to generate a comprehensive structural damage index. The confidence level W after dynamic correction is then used to... corr,i Based on (t), the fusion weight ω corresponding to each physical quantity (micro-strain, chloride ion concentration gradient, resistivity abrupt change) at time t is first calculated. j (t). Then, the following formula is used to calculate the position at the sensor location. Structural damage comprehensive index at time t :

[0088]

[0089] In the formula, ω is the comprehensive index of structural damage. j (t) represents the fusion weight of physical quantity j at time t. is the normalized monitoring value of physical quantity j at location and time t, j represents different physical quantities in the multi-modal monitoring data.

[0090] The above calculation obtains damage indices at discrete sensor location points. To form a continuous damage assessment view, the fusion module 20 further performs spatio-temporal smoothing on these discrete indices. In the spatial dimension, a Kriging interpolation method is used to perform spatial interpolation on the entire key degradation area according to the damage index values of each sensor location and their spatial correlation, to generate a continuous damage distribution field covering the entire area. In the time dimension, a Kalman filtering algorithm is applied to the time series data of each grid point on the damage distribution field to filter out high-frequency noise and obtain a damage state that evolves smoothly over time. Finally, the fusion module 20 outputs a structural damage comprehensive index matrix with spatio-temporal continuity and transmits it to the prediction module 30.

[0091] In embodiments of the present application, the generated structural damage comprehensive index is subjected to environmental compensation, health state assessment and feature extraction to link the subsequent life prediction step S3. These processing steps can be performed by the fusion module 20 or an independent processing unit whose input is connected to the fusion module 20 and whose output is connected to the prediction module 30.

[0092] The process receives the structural damage comprehensive index matrix with spatio-temporal continuity output by the fusion module 20 To separate the interference caused by environmental factor fluctuations on the structure response and make the damage index more reflect the degradation state of the structure itself, this embodiment introduces an environmental coupling dynamic compensation model. This model synchronously collects real-time tide level data L(t) and temperature data Θ(t) of the area where the structure is located, and calculates a dynamic compensation coefficient based on these data. First, the tide level influence correction factor and the temperature influence correction factor are constructed respectively, and then the dynamic compensation coefficient K env (t) is generated by coupling calculation and applied to the structural damage comprehensive index matrix to obtain the corrected structural damage comprehensive index This calculation process is defined by the following formula:

[0093]

[0094] K env (t) = C T (L(t)) · C Θ (Θ(t));

[0095] In the formula, is the corrected structural damage comprehensive index; is the uncorrected structural damage comprehensive index; K env(t) is the environmental coupling dynamic compensation coefficient at time t; C T (L(t)) is the tide level influence correction factor calculated based on real-time tide level L(t); C Θ (Θ(t)) is the temperature influence correction factor calculated based on real-time temperature Θ(t).

[0096] In the present embodiment, the two correction factors can be determined by an empirical function established based on historical data or physical experiments. For example, the temperature influence correction factor C Θ (Θ(t)) can be defined in the form of Arrhenius equation to describe the change of chemical corrosion rate and material resistivity with temperature. The tide level influence correction factor C T (L(t)) can be constructed as a piecewise function related to tide level height L(t) to represent the different influences of different dry-wet cycle regions and full immersion regions on corrosion and material saturation.

[0097] Referring to the accompanying Figure 4 , Figure 4 is a schematic diagram of health state grading atlas generation and application according to an embodiment of the present application. After obtaining the corrected structure damage comprehensive index , a maintenance urgency evaluation model is constructed. The model establishes a piecewise mapping function to map the continuous damage index value to the preset discrete maintenance urgency level. For example, two damage thresholds T1 and T2 (T1 < T2) are set, wherein: if , the maintenance urgency level is low; if , the maintenance urgency level is medium; if , the maintenance urgency level is high.

[0098] Based on the maintenance urgency level, color gradient mapping technology is used to present the damage index of different spatial positions in a graded visual manner. For example, the low, medium and high levels are mapped to green, yellow and red respectively, and these colors are assigned to the corresponding grid elements on the structure three-dimensional model, and finally a health state grading atlas is generated, which intuitively reflects the damage degree and maintenance urgency of each part of the structure.

[0099] To provide input to the migration learning life prediction model of step S3, features need to be extracted from the current structure (target domain). With the health state grading atlas as a guide, a set of target domain features consistent with the source domain feature database is extracted from the spatial distribution areas of different maintenance urgency levels in the atlas. Specifically, for the areas labeled as medium or high urgency in the atlas, the corresponding original multi-modal monitoring data time series are traced back, and the following features are calculated therefrom: micro-strain evolution trend features, which can be obtained by calculating the linear regression slope of the micro-strain data within a specified time window; chloride concentration gradient cumulative change features, which can be obtained by integrating the chloride concentration gradient values over time; and resistivity mutation frequency features, which can be obtained by counting the number of times the resistivity change rate exceeds a predetermined threshold within a specified time window. These calculated quantitative features collectively constitute the target domain feature set input to the prediction module 30.

[0100] In an embodiment of the present application, step S3 of inputting the structure damage comprehensive index into a migration learning life prediction model to predict the remaining service life of the port structure is performed by the prediction module 30. The core function of this module is to solve the problem that new service structures are difficult to accurately predict the life due to insufficient accumulation of their own monitoring data.

[0101] The prediction module 30 first builds and stores a source domain feature database internally. This database is established by collecting a large amount of full-life cycle data of historical structures of the same type (e.g., similar design specifications, material grades, service environments) as the current target structure. Specifically, the monitoring data of these historical structures, including micro-strain, chloride concentration, resistivity, etc., are collected, and their evolution trends, cumulative changes, mutation frequencies, etc. are extracted; at the same time, the database also contains explicit failure records of these structures, such as the time when a predetermined damage threshold is reached, the time when cracking occurs, or the time when major maintenance is performed. These historical features and their corresponding failure times collectively constitute the source domain feature database, providing rich prior knowledge for the prediction model.

[0102] The prediction module 30 receives the target domain feature set extracted from the current target structure generated by the preceding steps, which includes micro-strain evolution trend features, chloride concentration gradient cumulative change features, and resistivity mutation frequency features. The prediction module 30 inputs this target domain feature set together with the source domain feature database into a life prediction model based on cross-domain feature migration optimization. This model aims to learn a mapping function from features to failure time, and optimizes this function through migration learning.

[0103] Within the model, the knowledge transfer is guided by calculating the difference between the target domain feature distribution and the source domain feature distribution. For example, the Maximum-Mean-Discrepancy (MMD) can be used as a measure, and its calculation formula is as follows:

[0104]

[0105] In the formula, X s is the source domain feature sample set, X t is the target domain feature sample set; n s and n t are the sample numbers of the source domain and the target domain, respectively; and are the feature vectors of the source domain and the target domain, respectively; φ(·) is a mapping function that maps the feature vector to the reproducing kernel Hilbert space HH. One of the optimization objectives of the model in the training process is to minimize the MMD 2 (X s ,X t ), so that the data characteristics of the target domain are aligned with the data characteristics of the similar degradation stage in the source domain in the mapped space, and the cross-domain feature weight optimization is realized.

[0106] Through transfer learning, the model finally outputs the failure probability curve P f (t) of the target structure in the future period of time. The curve represents the cumulative probability of the structure reaching the failure state at time t. Based on this curve, and by pre-setting an engineering-acceptable failure probability threshold P th (for example, P th = 0.5), the predicted failure time t fail of the structure can be determined, that is, the time point that satisfies P f (t fail ) = P th . The remaining useful life (RUL) of the structure is calculated as follows:

[0107] RUL = t fail -t current ;

[0108] In the formula, t current is the current time.

[0109] Further, the prediction module 30 also outputs the failure probability curve P f (t) as input, and uses the reachability analysis method to generate a maintenance decision tree containing maintenance time and measure recommendations. The analysis method evaluates the maintenance time t maintAfter taking different maintenance measures, how will the failure probability curve evolve, thus generating a tree structure. Each node of the decision tree represents a decision point (whether to maintain), each branch represents a decision option (specific maintenance measures or no maintenance), and the expected remaining useful life and cost under the option are marked, providing quantitative basis for the final maintenance decision. The prediction module 30 finally transmits the calculated remaining useful life to the reevaluation module 40.

[0110] In an embodiment of the present application, with the predicted remaining useful life, the step S4 of dynamic reevaluation of the critical deterioration zone based on predictive analysis is performed by the reevaluation module 40. This step receives the harbor structure remaining useful life (RUL) and the failure probability curve P f (t) output by the prediction module 30, and aims to form a forward-looking feedback closed loop to dynamically optimize the configuration of monitoring resources.

[0111] The reevaluation module 40 first generates a future potential risk distribution map using the life prediction model optimized by transfer learning in step S3. Specifically, the module 40 selects one or more future time nodes t future (for example, t future = current time + 0.2 * RUL), and uses the prediction model to calculate the evolution values of various physical quantity characteristics (such as micro-strain, chloride ion concentration, etc.) at each spatial position of the structure at this time node. Then, these predicted characteristic values are substituted into the calculation process of the structure damage comprehensive index to obtain a predicted structure damage comprehensive index distribution at the future time t future , which is the future potential risk distribution map.

[0112] After obtaining the future potential risk distribution map, the reevaluation module 40 identifies the drift risk zone by comparing the current state with the future prediction. The drift risk zone is defined as: the region that does not belong to the critical deterioration zone determined by the acquisition module 10 at present, but in the future potential risk distribution map, its predicted damage comprehensive index exceeds the preset high-risk threshold T risk . This identification process is completed through the set operation of spatial regions, and the difference set obtained by subtracting the current actual critical deterioration zone set from the future high-risk region set is the drift risk zone.

[0113] The reevaluation module 40 feeds back the spatial coordinates and boundary range information of the identified drift risk zone to the acquisition module 10, which is used to update the optimization objective function in the particle swarm optimization (PSO) algorithm used by the acquisition module 10 to perform critical deterioration zone identification. The original optimization objective function F(p) is updated to F'(p), and the updating method is as follows:

[0114] F'(p) = F(p) + δ * I D(p) ;

[0115] where p is a three-dimensional spatial position vector within the structure; F'(p) is the updated comprehensive damage index; F(p) is the pre-updated comprehensive damage index, i.e. a · S ∈ (p) + β · G Cl (p) + γ · R ρ (p) ; I D (p) is an indicator function, if position p belongs to the identified drift risk zone, then I D (p) = 1, otherwise I D (p) = 0; δ is a pre-set drift risk weighting coefficient, which is positive.

[0116] By making the above modification to the optimization objective function, when the collection module 10 performs the next key deterioration zone identification, the search process of its particle swarm optimization algorithm will not only be biased towards the current most deteriorated region, but also towards the drift risk zone with high deterioration potential. In this way, the forward-looking adjustment of the monitoring strategy is realized, so that the monitoring resources such as sensors can be arranged in advance to the area where serious deterioration may occur in the future, thus forming a feedback closed loop of predictive monitoring.

[0117] In the embodiment of the present application, the step S5 of model self-correction based on decision feedback is performed in response to the externally input maintenance activity record, which is performed by the self-correction module 50. This step constitutes a feedback closed loop that integrates external decision-making behavior into the internal evaluation system, ensuring the synchronization of the model and the physical actual state of the structure.

[0118] This step is triggered by an externally input standardized maintenance activity record. The record contains the three-dimensional spatial coordinates of the maintenance area, the specific maintenance measure type (e.g. concrete overlay, crack grouting, replacement of components, etc.), the physical parameters of the new material used, and the completion time of the maintenance activity. After receiving the record, the self-correction module 50 performs a series of linked model self-correction operations.

[0119] Firstly, the self-correction module 50 sends instructions to the collection module 10 to update its internal deterioration analysis model. According to the maintenance measure type and new material parameters recorded in the maintenance activity record, the local material physical parameters of the corresponding maintenance area in the deterioration analysis model are adjusted. For example, if the record shows that concrete overlay repair is performed in a certain area, then in the finite element model of three-dimensional coupling of stress field, chemical field and geometric field, the chloride ion diffusion coefficient, resistivity, elastic modulus, etc. of the grid element corresponding to the area are updated to the numerical values of the newly repaired material.

[0120] Secondly, the self-revision module 50 sends instructions to the fusion module 20 to update its credibility weight calculation model for the maintenance area. After the maintenance completion time indicated by the maintenance activity record, within a preset time period T stable , the spatiotemporal continuity penalty for the sensor data located in the maintenance area is suspended. Specifically, for sensor i in the area, the data mutation penalty coefficient λ corr,i (t) in the calculation formula of the dynamic revised credibility W i (t) is temporarily set to zero. This is intended to allow the system to smoothly adapt to the new data baseline after structural repair without mistakenly determining benign data mutations caused by maintenance as sensor abnormalities.

[0121] Finally, the self-revision module 50 processes the data stream input to the prediction module 30. In the transfer learning life prediction model, the data collected from the maintenance area after the maintenance completion time indicated by the maintenance activity record is marked as post-intervention data. This marking is achieved by adding a binary intervention identification feature bit to the feature vector input to the prediction model. When this feature bit is activated, the transfer learning life prediction model can recognize that the change in data evolution law is due to artificial intervention rather than natural degradation when making predictions. This allows the model to treat the data sequence before and after maintenance as different stages, thereby avoiding the erroneous extrapolation of performance improvement brought about by repair, and ensuring the accuracy of the remaining service life prediction after structural repair.

[0122] Although embodiments of the present application have been shown and described, it is to be understood that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for port engineering new structure durability intelligent monitoring and life evaluation, characterized in that, The method comprises the following steps: S1, identifying a key deterioration zone on a port structure, and collecting multi-modal monitoring data from the key deterioration zone; S2, performing dynamic credibility fusion processing on the multi-modal monitoring data to generate a structure damage comprehensive index, the dynamic credibility fusion processing comprising dynamically adjusting credibility weights corresponding to the multi-modal monitoring data based on the spatio-temporal continuity of the multi-modal monitoring data; S3, inputting the structure damage comprehensive index into a transfer learning life prediction model to predict the remaining service life of the port structure; S4, using the predicted remaining service life to perform key deterioration zone dynamic reevaluation based on predictive analysis to identify future drift risk zones, and updating the model for identifying the key deterioration zone using the future drift risk zones; S5, and in response to an externally input maintenance activity record, performing model self-correction based on decision feedback, the model self-correction comprising updating the deterioration analysis model for identifying the key deterioration zone, updating the credibility weight calculation model for the maintenance area in the dynamic credibility fusion processing, and marking post-maintenance data in the transfer learning life prediction model.

2. The port structure durability intelligent monitoring and life evaluation method according to claim 1, characterized in that, In step S1, the step of identifying a key deterioration zone on a port structure comprises: constructing a deterioration analysis model of the triple coupling of stress field, chemical field and geometric field of the port structure to obtain an initial damage field distribution; using a particle swarm optimization algorithm to identify a key deterioration zone from candidate areas in the initial damage field distribution, taking the initial damage field distribution as input.

3. The port structure durability intelligent monitoring and life evaluation method according to claim 1, characterized in that, In step S2, the step of dynamically adjusting credibility weights corresponding to the multi-modal monitoring data comprises: based on the normalized data variation amplitude |AV i (t) of the multi-modal monitoring data at time t, the following formula is used to calculate the dynamic corrected reliability W corr,i (t) of sensor i at time t: W corr,i (t) = W init,i • exp(-λ i • |ΔV i (t) |) where W init,i is the initial confidence weight of sensor i, W corr,i (t) is the dynamic modified confidence of sensor i at time t, λ i is the penalty coefficient of data mutation of sensor i, |ΔV i (t) is the normalized data variation amplitude of sensor i at time t.

4. The port structure durability intelligent monitoring and life evaluation method according to claim 3, characterized in that, In step S2, the step of performing dynamic credibility fusion processing on the multi-modal monitoring data to generate a structure damage comprehensive index comprises: The dynamic correction credibility W corr,i (t) is calculated on the basis of each physical quantity corresponding to the fusion weight ω j (t) And the following formula calculates the position and the structural damage comprehensive index at time t wherein is the structural damage comprehensive index, ω j (t) is the fusion weight of physical quantity j at time t, is the normalized monitoring value of physical quantity j at position and time t, j represents different physical quantities in the multi-modal monitoring data.

5. The port structure durability intelligent monitoring and life evaluation method according to claim 1, characterized in that, In step S3, the step of inputting the structure damage comprehensive index into a transfer learning life prediction model to predict the remaining service life of the port structure comprises: constructing a source domain feature database containing historical data of similar structures; extracting a target domain feature set from the structure damage comprehensive index; inputting the source domain feature database and the target domain feature set into the transfer learning life prediction model, calculating a failure probability curve, and determining the remaining service life based on the failure probability curve.

6. The port structure durability intelligent monitoring and life evaluation method according to claim 1, characterized in that, In step S4, the step of updating the model for identifying the key deterioration zone using the future drift risk zones comprises: updating the optimization objective function in the particle swarm optimization algorithm using the identified future drift risk zones.

7. The port structure durability intelligent monitoring and life evaluation method according to claim 1, characterized in that, In step S5, the step of performing model self-correction based on decision feedback in response to an externally input maintenance activity record comprises: adjusting the local material physical parameters of the deterioration analysis model corresponding to the maintenance area according to the type of maintenance measures recorded in the maintenance activity record; suspending the spatio-temporal continuity penalty for sensor data in the maintenance area for a preset period of time after the maintenance indicated in the maintenance activity record is completed to correct the credibility weight calculation model. The post-maintenance collected data is marked as post-intervention data in the transfer learning life prediction model, so that the model can distinguish the data evolution law change caused by artificial intervention and natural degradation in subsequent prediction. 8.The port structure durability intelligent monitoring and service life evaluation method according to claim 1, characterized in that, The method further comprises: Before inputting the structure damage comprehensive index into the transfer learning life prediction model, an environment coupling dynamic compensation coefficient is applied to the structure damage comprehensive index to obtain a revised structure damage comprehensive index; The environment coupling dynamic compensation coefficient is determined based on real-time collected tidal level and temperature data; Wherein, the input into the transfer learning life prediction model is the revised structure damage comprehensive index.

9. The port structure durability intelligent monitoring and life evaluation method according to claim 5, characterized in that, The method further comprises: Taking the failure probability curve as input, an accessibility analysis method is used to generate a maintenance decision tree containing maintenance time and measure recommendations.

10. The system for intelligent monitoring and life assessment of durability of new port structures, applied to the method according to any one of claims 1-9, characterized in that, The system comprises: A collection module for identifying key degradation zones on the port structure and collecting multi-modal monitoring data from the key degradation zones; A fusion module for performing dynamic credibility fusion processing on the multi-modal monitoring data collected by the collection module to generate a structure damage comprehensive index, wherein the dynamic credibility fusion processing performed by the fusion module includes dynamically adjusting the credibility weight corresponding to the multi-modal monitoring data based on the spatio-temporal continuity of the multi-modal monitoring data; A prediction module for inputting the structure damage comprehensive index generated by the fusion module into a transfer learning life prediction model to predict the remaining useful life of the port structure; A reevaluation module for performing predictive analysis-based dynamic reevaluation of key degradation zones using the remaining useful life predicted by the prediction module to identify future drift risk zones, and updating the model used by the collection module to identify the key degradation zones using the future drift risk zones; A self-correction module for performing decision feedback-based model self-correction in response to externally input maintenance activity records, wherein the model self-correction performed by the self-correction module includes updating the degradation analysis model used by the collection module, updating the credibility weight calculation model for the maintenance area used by the fusion module, and marking the post-maintenance data in the transfer learning life prediction model used by the prediction module.

Images