A Safety Assessment and Optimization Design Method for High-Flow Aqueduct Structures

Abstract

This invention relates to the field of hydraulic engineering, and more particularly to a method for safety assessment and optimization design of large-flow aqueduct structures. The method includes: collecting displacement, strain, and temperature monitoring sequences based on a triple probe, completing grouping, coordinate and time registration, and establishing a monitoring baseline; constructing a spatial model including support nonlinearity and fluid-structure boundary conditions, applying loading conditions, and obtaining response predictions; aligning monitoring and prediction in time and location, performing hierarchical Bayesian filtering inversion, and obtaining parameter posteriors and structural health; under the integration of in-service monitoring and accelerated degradation characteristics, using multi-task learning to generate material evolution and lifespan intervals, and automatically generating and simulating optimization schemes under reliability and maintainability constraints. This application forms a closed loop from assessment to design, effectively enhancing data comparability, traceability of uncertain information, and adaptability of the scheme to engineering projects.

Description

Technical Field

[0001] This invention relates to the field of water conservancy engineering, and in particular to a method for safety assessment and optimization design of large-flow aqueduct structures. Background Technology

[0002] In the field of structural safety assessment for large-scale water conveyance projects, existing solutions typically rely on on-site sensors to collect structural response data, which is then compared and analyzed using finite element numerical models to determine the structural safety status and formulate maintenance strategies. However, existing methods often depend on simple comparisons between single-type sensor data and offline numerical models. In real-world operating scenarios involving the combined effects of multiple factors such as water level fluctuations, temperature changes, and support wear, these methods are prone to problems such as ineffective correspondence between monitoring data and model responses, deviations in parameter identification results from reality, and assessment conclusions lagging behind changes in structural condition. Consequently, they fail to meet the requirements for accurate safety assessment and optimized design of aqueducts during long-term operation. Existing technologies for the joint processing of displacement, strain, and temperature data acquired by triple probes with the output of fluid-structure interaction numerical models generally suffer from common shortcomings, such as insufficient matching accuracy between data and models in terms of time scale and spatial location, difficulty in quantifying and fusing multi-source uncertain information during parameter identification, and inability to effectively constrain parameter drift trends. These shortcomings make it difficult to form a consistent process from data acquisition, model alignment, parameter inversion to durability prediction in the entire life cycle operation and maintenance scenario of aqueducts. As a result, the assessment results deviate from the actual health status of the structure, the optimization scheme lacks verifiability, and it cannot support integrated closed-loop decision-making from safety assessment to optimization design. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a method for safety assessment and optimization design of high-flow-rate aqueduct structures, comprising:

[0004] S100: Based on the triple probe deployment scheme, perform probe group configuration, data access and raw monitoring sequence processing to generate monitoring baseline data packets;

[0005] S200. Based on the monitoring baseline data packet, perform observation point mapping, model configuration and solution control processing to generate a fluid-solid response prediction packet;

[0006] S300. Based on the fluid-solid response prediction package, perform time location alignment, parameter inversion and multi-factor evaluation to generate parameter posterior distribution and structural health indicator.

[0007] S400. Based on the posterior distribution of the parameters and the structural health indicator, perform durability feature construction, durability modeling and automated optimization processing to generate a set of solutions and a safety assessment report.

[0008] Furthermore, the process of configuring the probe group and accessing data includes:

[0009] The probe group configuration process includes: generating initial groups based on three candidate locations—mid-span section, support neighborhood, and temperature gradient sensitive area—according to the three-way combination rule of displacement gauge-strain gauge-temperature probe; calculating the spatial overlap and channel occupancy of each initial group to form a simplified group set; and recording the combination number, group role, relative position vector, installation attitude, wiring port, and acquisition channel mapping for each simplified group to constitute the probe group configuration. The data access process includes: establishing a coordinate mapping table from the structural reference coordinate system to the local probe coordinate system according to the probe group configuration; setting the sampling frequency, range, resolution, and trigger threshold of the acquisition hardware; recording the reference time and timestamp source using the same time anchor point registration strategy; and performing continuity detection and static noise sampling on each probe channel to generate a channel health registration table.

[0010] Furthermore, the process of processing the original monitoring sequence includes:

[0011] The three types of observation sequences acquired are identified and labeled for missing data, and the missing data index is recorded without interpolation. Robust denoising processing is performed on candidate anomaly segments, and reference values ​​are output and retained along with the original values. Displacement, strain and temperature entries are uniformly resampled and divided into time windows to form a structured entry table arranged with fixed time steps and accompanied by missing data markers, anomaly segment markers, reference value links, probe group numbers, correspondence between local coordinates and structural reference coordinates, and time axis anchor point summaries. This table is then encapsulated into a monitoring baseline data package.

[0012] Furthermore, the process of mapping observation points and configuring the model includes:

[0013] The observation point mapping process includes: resolving three types of labels—mid-span section, support neighborhood, and temperature gradient sensitive area—in the structural reference coordinates; aggregating similar entries by probe group number; verifying attitude correction and mapping error estimation; establishing a candidate list of observation points for each probe group; mapping displacement entries to beam-slab element deflection observation points; mapping strain entries to stress-strain observation points of support perimeter components; and mapping temperature entries to material temperature field observation points; generating an observation mapping table containing observation point number, model node or element identifier, observation category, time window availability, spare mapping list, and unit list. The model configuration process includes: constructing a three-dimensional geometric skeleton based on the aqueduct geometry; determining the mesh density of locally refined areas based on the observation point distribution; configuring placeholder parameters for support nonlinear constitutive parameter groups and material temperature coupling parameters; establishing a hydrodynamic boundary list associated with the working conditions; and generating a model configuration list containing geometric skeleton parameters, mesh generation parameters, support constitutive parameter groups, hydrodynamic boundary list, observation channel mapping copies, and working condition set placeholders.

[0014] Furthermore, the solution control process includes:

[0015] A set of working conditions covering conventional water flow, fluctuation transition, seismic coupling, and temperature difference superposition is established and loaded. The solution control step size and nonlinear iteration threshold are set, the observation channel mapping copy is activated, and numerical robustness verification including pre-verification, process verification, and post-verification is performed. When encountering iterative convergence problems or key quantity transitions during process verification, the step size is automatically backed up and reduced. The nodal displacement sequence, element strain sequence, and material temperature field sequence under each working condition are summarized and aligned to the channel and position consistent with the monitored quantities according to the observation channel mapping copy. The result is encapsulated into a fluid-structure response prediction package containing a working condition index, time axis, observation channel number, numerical sequence body, and verification mark. At the time of encapsulation, the same-site alignment label and time window alignment label are written.

[0016] Furthermore, the process of aligning the time positions includes:

[0017] The time-position alignment process includes: establishing a mapping relationship between the fixed time step on the monitoring side and the output time history on the model side based on the time axis anchor point summary and time window alignment label; performing same-site pairing for monitoring entries and model entries with the same probe group number and the same observation channel number; calling the observation point index and the observation channel mapping copy to verify the consistency of the three-dimensional position, observation category and unit list; if the consistency is still not met after the backup mapping is replaced, register the time window as a reference window only and remove it from the same-site comparison set; constructing the same-site comparison set by partitioning according to the three types of labels; performing consistency judgment to generate the same-site comparison fragments that enter statistical fusion and the bypass reference fragments that do not enter statistical fusion; performing uncertainty registration processing; generating three types of uncertainty identifiers and comprehensive uncertainty descriptions according to the source on the monitoring side, the source on the model side and the source of the alignment process; and concatenating the fragments that enter statistical fusion into a fusion fragment set, which is encapsulated together with the bypass reference fragments, the three types of uncertainty identifiers, the comprehensive uncertainty description and alignment metadata into a fusion input set.

[0018] Furthermore, the parameter inversion process includes:

[0019] The parameter inversion process includes: constructing a hierarchical inversion task list based on the response characteristics of mid-span section classes, support neighborhood classes, and temperature gradient sensitive area classes. The upper level focuses on slow variables of the overall structure and support system, while the lower level focuses on fast variables of local components and material temperature fields. Displacement responses and support neighborhood strain responses corresponding to the upper-level tasks, and temperature responses corresponding to the lower-level tasks, are extracted from the fused input set. Confidence levels are assigned to different segments based on the comprehensive uncertainty description. The nonlinear constitutive parameter groups of supports and the spaced parameters of material temperature coupling parameters in the model configuration list are read as updatable parameter slots. A binding relationship is established between the observation sequence, prediction sequence, and uncertainty weights. A hierarchical filtering update process is executed. At the upper level, displacement responses and support neighborhood strain responses are used to update the parameter slots related to the overall structure and supports. At the lower level, temperature responses and local strain responses are used to update the parameter slots related to material temperature coupling in one step. When bypass reference segments exist, only circumstantial evidence is recorded. Instead of generating update actions, parameter drift constraints and robust suppression strategies are introduced during the filtering update process. The parameter drift constraints include parameter change range control based on the prior change boundary of the hierarchical task. The robust suppression strategies include suppression based on the statistics of the update residual neighborhood and the physical consistency penalty function applied when the parameter change direction is contrary to the temperature gradient and water pressure change direction. After the cross-time step cyclic update is completed, a time series update trajectory is formed for each parameter slot. The time slots are segmented and aggregated according to the working condition index and time window. For time periods with multiple backup mapping replacements or multiple unit conversions, the expansion range of the candidate interval is appropriately increased. For time periods surrounded by a large number of bypass reference segments, they are only registered and not used for downstream health determination. The candidate intervals of the upper and lower layer tasks are summarized to form a parameter posterior candidate set that covers the parameters related to the coupling of concrete equivalent elastic modulus and temperature, as well as the parameters related to the hysteresis of friction pendulum support, and retains the source composition, time coverage and uncertainty annotation of each candidate interval.

[0020] Furthermore, the multi-factor assessment process includes:

[0021] The multi-factor assessment process includes: acquiring a set of factors for water pressure, temperature, support status, and constraints, and comparing them item by item with the operating condition index; constructing a three-dimensional assessment unit of operating condition-factor-candidate interval by aggregating candidate intervals that overlap with the time period of each operating condition; calculating the factor importance and factor uncertainty weights and binding them to the candidate intervals to form a weighted candidate set; performing interval screening and delimitation on the weighted candidate set; adhering to the principle that data from different sources within the weighted candidate set should be comparable at the same location and within the same time window; reducing the weight of candidate intervals that have undergone multiple backup mapping replacements or unit conversions and expanding them towards the safety side; grouping the weighted candidate intervals of each parameter item by operating condition, performing weighted segmentation and merging to obtain the posterior center and extended range; generating an additional drift rate description for parameter items with obvious time-varying drift; generating a structural health indicator by combining the parameter drift rate, candidate interval coverage, and factor importance weights; and outputting the parameter posterior distribution indexed by parameter item and including source composition, time coverage, and weight summary, which, together with the structural health indicator, serves as the output of this step.

[0022] Furthermore, the durability modeling process includes:

[0023] The durability modeling process includes: hierarchical modeling preparation based on location and environmental labels; dividing the tank body, lining layer, and support neighborhood into different modeling subdomains; establishing parameter slots related to material temperature coupling and contact hysteresis within each subdomain; using strength reduction features, elastic modulus decay features, and crack threshold features as parallel prediction targets; enabling a multi-task learning process; organizing the training phase according to task-shared representations and task-specific branches; incorporating similarity scores and alignment counts into the sample weight mapping to assign different participation weights; and constructing long-term window slow-change slices from the durability feature summary table in time-series order. The training process involves receiving parameter terms from the posterior distribution of parameters as conditional inputs to adjust the parameter slot state. During the inference phase, a material performance evolution curve is generated for each part and environment combination, consisting of the trajectory of strength reduction, elastic modulus decay, and crack threshold change over time. Confidence descriptions from sample weight mapping and conditional parameter inputs are added. Based on the service life and environmental loading regime, key periods and warning periods are marked for each trajectory and mapped to the upper and lower bounds of the lifespan interval. Two sets of views are output at the part level and the structural interface level, respectively, generating the material performance evolution curve and lifespan interval.

[0024] Furthermore, the automated optimization process includes:

[0025] The automated optimization process includes: constructing a constraint pair of reliability and maintainability constraints. The reliability constraints include a three-axis joint constraint of bearing axis, deformation axis, and contact axis for the load-bearing and deformation requirements at the part level and the contact behavior requirements at the structural interface level. The maintainability constraints include a three-dimensional schedulable model for water outage periods, spare parts availability, and construction accessibility. A candidate measure library containing categories such as support replacement, limit reinforcement, local reinforcement, temperature control insulation, and monitoring encryption is read. An automatic scheme generation process is executed. First, at the part level, the material performance evolution curve and lifespan range are scanned to mark the time periods and parts where triggering measures are needed. Then, at the structural interface level, drift information related to support hysteresis is checked to form a part-interface joint trigger list. After filtering out infeasible items according to maintainability constraints, the candidate measure library is called to match and generate an initial scheme. To verify the feasibility of the initial scheme, a secondary simulation verification process is entered. The working condition settings and observation channel mapping copy of the fluid-structure response prediction package are called. The measure parameters in the initial scheme are written into the model configuration items to generate a revised model configuration list. Numerical solutions are run under representative working conditions and output. The observation channel sequence is consistent with that of the monitoring side. The key time period response in the material performance evolution curve is compared according to the location label and structural interface level. A multi-scale morphological matching algorithm is used to compare extreme values, waveform similarity and key inflection point deviations at the macro, meso and micro scales. Passed and failed items are marked in the scheme entries and a rollback prompt is generated. Rollback paths are triggered for failed schemes. The first rollback branch is based on the expansion range of the parameter posterior distribution to shrink or expand the measure parameters or perform Monte Carlo sampling re-verification. The second rollback branch is based on the drift information of the structural health indicator to adjust the trigger time of the measure or change the combination of measures. Rollback stops when the scheme passes or reaches the upper limit. Multi-objective screening is performed on the verified schemes that pass. Under the premise of meeting the reliability and maintainability constraints, the impact of water outage, cost and schedule indicators are ranked and the weight settings and the difference summary between schemes are recorded to form a candidate scheme set. The candidate scheme set is structured and encapsulated. The encapsulation content includes scheme parameters, applicable location, trigger time, verification evidence and rollback record, and is accompanied by the evidence chain hash value containing upstream evidence and simulation evidence. A scheme set and safety assessment report are generated.

[0026] The following are its main beneficial effects:

[0027] (1) By constructing a fusion input set, the monitoring data and model output are accurately matched in the time and space dimensions, so that subsequent parameter identification can be based on reliable data in the same position and window, avoiding evaluation bias caused by mismatch between data and model.

[0028] (2) By binding the hierarchical inversion task list with uncertainty weights, the decoupled identification of the overall and local responses of the structure is realized. The introduction of parameter drift constraints and robust suppression strategies ensures that the parameter update process conforms to physical laws, so that the identification results can truly reflect the actual state of the structure.

[0029] (3) By combining automated optimization processing with secondary simulation verification processing, the verifiability of the optimized scheme after generation is realized. The joint loading of reliability constraints and maintainability constraints ensures that the output scheme has engineering feasibility and avoids the limitation that the optimized scheme is difficult to implement in traditional methods.

[0030] (4) By preserving evidence of bypass reference segments and constructing a knowledge base of corroborating evidence during the parameter inversion process, the observation gap caused by data quality issues is filled, and the evaluation conclusions are made to maintain continuity and traceability on the time axis.

[0031] (5) By encapsulating the evidence chain hash value in the scheme generation process, each output scheme can be traced back to the upstream monitoring data and simulation verification records, providing an auditable basis for periodic review and further optimization in the operation and maintenance phase. Attached Figure Description

[0032] Figure 1 A flowchart illustrating a method for safety assessment and optimization design of a high-flow-rate aqueduct structure provided in this application embodiment;

[0033] Figure 2 A flowchart illustrating step S100 provided in an embodiment of this application;

[0034] Figure 3 A flowchart illustrating step S200 provided in an embodiment of this application;

[0035] Figure 4 A flowchart illustrating step S300 provided in an embodiment of this application;

[0036] Figure 5 This is a flowchart illustrating step S400 provided in an embodiment of this application. Detailed Implementation

[0037] Example 1: Refer to Figure 1 This is a flowchart illustrating a method for safety assessment and optimization design of a high-flow-rate aqueduct structure provided in an embodiment of the present invention. The process may include at least steps S100-S400:

[0038] S100: Based on the triple probe deployment scheme, perform probe group configuration, data access and raw monitoring sequence processing to generate monitoring baseline data packets;

[0039] S200. Based on the monitoring baseline data packet, perform observation point mapping, model configuration and solution control processing to generate a fluid-solid response prediction packet;

[0040] S300. Based on the fluid-solid response prediction package, perform time location alignment, parameter inversion and multi-factor evaluation to generate parameter posterior distribution and structural health indicator.

[0041] S400. Based on the posterior distribution of the parameters and the structural health indicator, perform durability feature construction, durability modeling and automated optimization processing to generate a set of solutions and a safety assessment report.

[0042] like Figure 2 As shown, Figure 2 This is a flowchart illustrating step S100 provided in an embodiment of this application. Step S100 includes at least steps S110-S130:

[0043] S110. Obtain the triple probe layout scheme, perform grouping processing of the mid-span section and support neighborhood and temperature gradient sensitive area, and obtain the probe group configuration.

[0044] The triple probe deployment scheme includes three sets of locations: candidate locations at the mid-span section, candidate locations in the support neighborhood, and candidate locations in the temperature gradient sensitive area. It also provides a description of the mounting surface, accessible channels, wiring restrictions, environmental interference sources, and safety isolation zone for each candidate location.

[0045] Specifically, using the aforementioned triple probe deployment scheme as input, the construction and monitoring sides align their positions according to the same coordinate origin, longitudinal baseline, and transverse baseline. The aqueduct centerline, net width of the aqueduct body, lining thickness, and support centerline are used as positioning anchor points to complete the unified projection of candidate mid-span section positions onto the structural reference coordinate system. For each candidate position, a position identifier, coordinate ternary field, installation posture field, and accessory interface field are recorded. Furthermore, for candidate positions in the support neighborhood, based on the force transmission path of the support structure and the constraint direction of the limiting components, the deformation sensitivity in each direction is determined. Under the condition of meeting the minimum installation spacing and avoidance boundary requirements, a priority field is marked for subsequent grouping and sorting.

[0046] Understandably, for candidate locations of temperature gradient sensitive areas, the first step is to search for locations that are prone to forming temperature differences and temperature gradients based on the radiation exposure, sunshine duration, ventilation conditions, and water temperature difference of the external environment of the aqueduct. Then, the candidate locations are cross-checked with the distance constraints of the measuring points along the channel and the distance constraints of the structural variable cross-sections. Locations that are inaccessible or of high risk during the operation and maintenance phase are eliminated, and the remaining locations are marked as valid locations of temperature gradient sensitive areas.

[0047] For three types of effective points—mid-span section, support neighborhood, and temperature gradient sensitive area—parallel combinations within the same section and sequential combinations between adjacent sections are established according to the three-way combination rule of displacement gauge—strain gauge—temperature probe. The relative spacing, included angle, and installation surface matching degree of each probe within the combination are calculated. When the combination meets the three constraints of installation surface matching degree, cable path accessibility, and interface redundancy, the initial grouping entries of the three-way combination are generated, and the combination number, role within the group, and relative position vector are recorded for each entry. Based on the initial groupings described above, for the mid-span section type, the direction of maximum deflection at mid-span is selected as the preferred direction for displacement gauges, and the direction consistent with the direction of the principal tensile strain is selected as the axial direction of the strain gauges. Temperature probes are then placed at heat-sensitive points near the displacement gauges on the sidewalls or top plates of the same section. For the support neighborhood type, the direction crossing the support centerline is selected as the preferred direction for displacement gauges, and the direction along the support sliding and limiting directions is selected as the axial direction of the strain gauges. Temperature probes are then placed at the thermal coupling points of the transition zone between the support steel and the concrete. For the temperature gradient sensitive area type, the thickness direction between the outer and inner surfaces is selected as the preferred direction for temperature probes, and displacement gauges and strain gauges are placed near the same location to synchronously record deformation and stress states.

[0048] To avoid excessive local density, the spatial overlap and channel occupancy of each initial group are calculated. When the overlap exceeds the threshold and the channel occupancy reaches the upper limit, groups with high priority and high installation accessibility are retained to form a simplified group set. For subsequent data acquisition, the simplified group set is matched with acquisition channel resources, sampling frequency constraints, time synchronization methods, and power supply methods. Group entries that do not meet channel resource constraints or have inconsistent time synchronization methods are placed in an adjustment list with the reason field recorded. For group entries that meet the constraints, the acquisition channel number, trigger mode, and timestamp source are added and registered. Finally, the simplified group set processed above is summarized to form a probe group configuration. Each probe group includes a description of the relative positions, installation attitude, wiring ports, and acquisition channels of three probes (displacement gauge, strain gauge, and temperature probe) in the same local coordinate system. The output field name is explicitly recorded in the probe group configuration as input for the next subsection, for use by the probe group configuration in S120, thereby completing the grouping processing of the mid-span section, the support neighborhood, and the temperature gradient sensitive area, and outputting the probe group configuration.

[0049] S120. Extract displacement gauges, strain gauges and temperature probes from the probe group configuration, perform unified coordinate binding and time anchor point registration, and generate the original monitoring sequence.

[0050] The probe group configuration includes fields such as group number, displacement gauge entry, strain gauge entry and temperature probe entry, as well as channel mapping, trigger mode and timestamp source.

[0051] Specifically, the probe group configuration is imported into the data access module on the monitoring side. A coordinate mapping table is established through the binding operation from the structural reference coordinate system to the local probe coordinate system. The coordinate mapping table records the correspondence between global coordinates and local coordinates, the rotation relationship of the probe installation posture, and the relative position vector of the probe, ensuring that the observation data of the three probes in the group can be synchronously aligned in the same local coordinate system.

[0052] Furthermore, based on the channel mapping fields, the sampling frequency, range, resolution, and trigger threshold are set for the displacement gauge, strain gauge, and temperature probe, respectively. A unified time anchor registration strategy is adopted, and the reference time, timestamp source, and drift correction window from the time anchor registration strategy are written into the acquisition control parameters. The time anchor registration strategy is used to record a unified time reference for each observation during asynchronous triggering or multi-machine collaboration. To ensure the access process operates within existing constraints, continuity detection and static noise sampling are performed on each probe channel, and the channel status and static noise statistics are written into the channel health registration table. When the channel health registration table shows a channel unavailable or static noise exceeding limits, the corresponding entry is marked as pending maintenance and temporarily not entered into the real-time acquisition process, while retaining data inflow from other available channels in the probe group.

[0053] After the time anchor point registration is completed, multi-source synchronous acquisition is initiated, generating displacement observation sequences for displacement gauges, strain observation sequences for strain gauges, and temperature observation sequences for temperature probes. These three types of observation sequences are then aligned according to probe group numbers and local coordinate indices to generate a triple synchronous observation entry. To adapt to the different response characteristics of mid-span sections and support neighborhoods, the triggering mode switches between passive and periodic triggering based on the role within the group and the structural conditions. Passive triggering is used for rapid recording in response to abrupt changes in conditions, while periodic triggering is used for continuous steady-state recording. For temperature gradient-sensitive areas, periodic triggering is prioritized, combined with a longer time window to record the gradual change process.

[0054] During the generation of synchronous observation entries, all observations carry timestamp source and channel number, sampling frequency and status flag, so as to facilitate subsequent unified coordinate binding and time alignment processing.

[0055] Subsequently, using a unified coordinate binding module, the displacement observation sequence, strain observation sequence, and temperature observation sequence are mapped to the local coordinates of the probe groups. The local coordinates of different probe groups are then mapped to the structural reference coordinates. Mapping error estimation and attitude correction records are written during the mapping process. When the mapping error or attitude correction exceeds the allowable error of the installation attitude, the entry is marked as an attitude anomaly and enters the subsequent anomaly processing branch. After coordinate binding is completed, all synchronous observation entries are merged according to time anchor points to form the original monitoring sequence. The original monitoring sequence is a time series set aggregated by probe group number and observation type, containing displacement, strain, and temperature entries, and includes timestamps, channel status, and attitude correction fields. To clarify subsequent workflows, at the end of this section, the output field name is specified as the original monitoring sequence, serving as input for the next section and used by the original monitoring sequence in S130. The record also indicates that this output serves as one of the data sources for subsequent main steps and as a reference for structural reference coordinate binding, establishing a comparable relationship with the observation point mapping and model observation position registration processing in S200.

[0056] S130. Perform missing detection only annotation and robust noise reduction processing on the original monitoring sequence to generate the monitoring baseline data packet structure;

[0057] The original monitoring sequence consists of displacement, strain, and temperature entries, all of which have been uniformly bound to coordinates and registered with time anchor points.

[0058] Specifically, the original monitoring sequence is imported into the data quality processing module. First, the missing data identification and labeling strategy is executed. The missing data identification identifies the missing observation points within the time window based on the time anchor interval and channel status flag. The missing location, missing time period and corresponding probe group number are recorded in the missing data index. No estimation or imputation is performed on the missing values, only labeling is performed. At the same time, continuous missing time periods caused by the temporary unavailability of the channel are segmented and marked to form a missing data segment description and written into the data quality registration table.

[0059] Furthermore, the original monitoring sequence undergoes anomaly segment identification and robust denoising. Anomaly segment identification is based on the consistency threshold between the status flags in the channel health registration table, the attitude correction records, and the same-site alignment. When three entries from the same probe group exhibit inconsistencies or abrupt changes within the same time window, and these abrupt changes do not occur simultaneously in adjacent structural groups or adjacent time periods, this time window is marked as an anomaly segment candidate. For the anomaly segment candidate, robust denoising is performed without altering the original values. The neighborhood statistics of each observation entry are compared with those of similar entries. Outlier observations are assigned low weights, and denoised reference values ​​are output. The reference values ​​are retained along with the original values ​​and distinguished by a flag. For systematic shifts caused by attitude anomalies, this time period is designated as a correction segment, and the shift field is recorded for interpretation during subsequent same-site alignments.

[0060] After completing the missing measurement annotation and robust denoising of abnormal segments, the displacement, strain and temperature items are uniformly resampled and time-windowed. The observation items with different sampling frequencies are aligned on a common time axis to form an item set arranged with fixed time steps. For each fixed time step, the alignment record of the three items, the missing measurement mark, the abnormal segment mark and the reference value link are saved to form a structured item table.

[0061] To facilitate subsequent model observation point mapping and load loading, the entry table is partitioned and encapsulated according to probe group number, observation type, and location label to generate a monitoring baseline data package. This monitoring baseline data package includes a basic metadata area, an entry index area, and a quality marker area, which can be associated with specific grouping and coordinate mapping relationships for mid-span section classes, support neighborhood classes, and temperature gradient sensitive area classes. To ensure alignment across main steps, observation point mapping pre-loading information and a time axis anchor summary are generated synchronously during encapsulation. The observation point mapping pre-loading information provides a list of probe group positions and observation types in the structural reference coordinates, while the time axis anchor summary provides the time reference required for subsequent site-to-site comparison and uncertainty registration. After encapsulation, the output field name is explicitly stated in the text as "monitoring baseline data package," used as input for the next main step, and called by the monitoring baseline data package in S210. It is also indicated that this output serves as one of the data inputs for subsequent main step parameter inversion and multi-factor weight evaluation, forming a comparable relationship with the fluid-structure response prediction package within the same site and time window.

[0062] In summary, the technical effects of this step are as follows: By performing missing measurement annotation and robust denoising of abnormal segments on the original monitoring data of the mid-span section, the support neighborhood, and the temperature gradient sensitive area, a monitoring baseline data package with quality labels is formed. Time and coordinate constraints are written into the entry structure for subsequent model observation location registration, load condition loading, and same-site fusion processing.

[0063] like Figure 3 As shown, Figure 3 This is a flowchart illustrating step S200 provided in an embodiment of this application. Step S200 includes at least steps S210-S230:

[0064] S210. Obtain the monitoring baseline data packet structure, perform observation point mapping and model observation location registration processing, and obtain the observation mapping table;

[0065] The input source for this section is the monitoring baseline data package, which contains displacement, strain, and temperature entries arranged in a fixed time step, along with missing measurement markers, abnormal segment markers, reference value links, probe group numbers, the correspondence between local coordinates and structural reference coordinates, and a time axis anchor point summary.

[0066] Specifically, the monitoring baseline data package is imported into the observation point mapping module as input. First, three types of labels are parsed under the structural reference coordinates: mid-span section, support neighborhood, and temperature gradient sensitive area. Then, the same type of entries are aggregated according to the probe group number. During the aggregation process, the recorded attitude correction and mapping error estimation are checked. When the error exceeds the limit, the corresponding time window is registered as a reference segment only instead of an alignment segment.

[0067] Furthermore, based on the binding relationship between the structural reference coordinates and local coordinates, a candidate list of observation points is established for each probe group. Displacement, strain, and temperature items are combined into a three-candidate list within the same local coordinate system. At the same time, the time axis anchor point and channel status associated with the candidate item are read, and the alignable and non-alignable time periods are marked. For non-alignable time periods, only the position and time period information are retained in the list, and they are not included in the subsequent model observation position registration.

[0068] Understandably, after the candidate list of observation points is formed, according to the needs of structural modeling for observations, the displacement entries of the mid-span section candidate items are mapped to the deflection observation points of the beam-slab elements, the strain entries of the support neighborhood candidate items are mapped to the stress-strain observation points of the support surrounding members, and the temperature entries of the temperature gradient sensitive area candidate items are mapped to the material temperature field observation points. The three types of observation points are then uniformly included in the observation point index, which records the observation point number, type, three-dimensional location, observation category, and time window availability.

[0069] After indexing, model observation location registration is performed. This registration process is based on a combination of aqueduct geometric segments, support structure types, and hydraulic segments. A bidirectional reference relationship is established between observation points and model nodes / elements segment by segment. Simultaneously, a set of observation channel identifiers and unit lists matching the observation type is set. The observation channel identifiers are used to ensure consistent field naming in subsequent numerical prediction outputs. During registration, if multiple candidate model nodes or elements exist near the same observation point, the optimal option is selected based on a comprehensive score of geometric distance and structural alignment. The second-best option is reserved as a backup mapping, which is used for local replacement in the subsequent numerical robustness verification stage.

[0070] After registration, an observation mapping table is generated. The observation mapping table includes fields such as observation point number, model node or unit identifier, observation category, time window availability, spare mapping list and unit list. At the end of this section, the output field name is explicitly stated as the observation mapping table, which serves as the input for the next subsection and is called by the observation mapping table in S220. It is also indicated that this observation mapping table will be used in subsequent main steps to align the model output with the monitored quantities, forming a consistent entry point with the same site comparison and uncertainty registration in the parameter inversion stage.

[0071] S220. Extract the structural nodes, hydrodynamic boundaries and support locations from the observation mapping table, perform three-dimensional mesh generation and support nonlinear constitutive configuration and fluid-solid boundary setting, and generate a model configuration list.

[0072] The input source for this section is the aforementioned observation mapping table, which has established the correspondence between observation points and model nodes or units and provides a time window availability and backup mapping list.

[0073] Specifically, the observation mapping table is imported into the model configuration module. First, the candidate set of structural nodes is read. Based on the geometric axis of the aqueduct, the net width of the aqueduct body, the lining thickness, and the longitudinal segmentation and transverse zoning information, a three-dimensional geometric skeleton is constructed. On this geometric skeleton, combined with the distribution of observation points in the structural reference coordinates, the mesh density level of the local densification area is determined. Among them, the areas where the mid-span section class and the support neighborhood class are located are set with a higher mesh density, and the areas where the temperature gradient sensitive area is located are set with a layered mesh division strategy in the thickness direction.

[0074] Furthermore, regarding the support location, the spatial distribution of observation points around the support is read from the observation mapping table to infer the force path and limiting direction of the support structure. The nonlinear constitutive parameter set of the support is then configured, covering hysteresis response, slip threshold, friction segment order, and recoverable segment settings. These parameter sets are registered as placeholder parameters that can be updated in subsequent parameter inversion stages, and the registration information is recorded in the model configuration item in key-value form. For the hydrodynamic boundary, the inlet section location, downstream boundary location, and free surface positioning line are read. A hydrodynamic boundary list is established based on the high-flow-rate operating conditions of the aqueduct design. The hydrodynamic boundary list provides a parameterized description of the inflow rate, inflow direction, turbulence intensity, and free surface treatment strategy for each operating condition, and establishes a correlation with the availability of the time window on the structural side, enabling direct comparison of the flow state and observation period within the same time window.

[0075] To improve the consistency between 3D mesh generation, boundary setting, and observation mapping, the shape quality index and geometric fitting error of the cell containing the observation point or its nearest neighbor cells are evaluated in real time during the generation process. When the index does not meet the preset threshold, local re-meshing is automatically triggered, and the corresponding node and cell identifiers are updated. At the same time, the candidate with the closest distance and higher alignment is selected from the backup mapping list in the observation mapping table to replace the original mapping. The replacement action is tracked for backtracking during subsequent robustness checks. After completing the 3D mesh generation, support nonlinear constitutive configuration, and fluid-solid boundary setting, a model configuration list is generated. The model configuration list includes geometric skeleton parameters, mesh generation parameters, support constitutive parameter groups, hydrodynamic boundary list, observation channel mapping copy, and case set placeholders. At the end of this section, the output field name is explicitly stated as "model configuration list" as input for the next subsection, for use by the model configuration list in S230. It is also noted that this model configuration list is used in subsequent main steps for case set loading and numerical robustness checks, as well as as a structural prior in parameter inversion and multi-factor weight evaluation, running through the data-model fusion main line of this invention.

[0076] S230. Load the load case set and verify the numerical robustness of the model configuration list to generate the fluid-solid response prediction package structure.

[0077] The input source for this section is the aforementioned model configuration list, which already includes configuration items such as geometric skeleton parameters, mesh generation parameters, support nonlinear structure parameter groups, and hydrodynamic boundary list.

[0078] Specifically, the model configuration list is imported into the solver control module. First, based on the high-flow usage scenario, a set of working conditions is established and loaded into the solver queue one by one. The set of working conditions covers the operating states such as conventional water flow, fluctuation transition, seismic coupling, and superposition of temperature difference effects. In each working condition item, the structural side time history length, boundary side inflow evolution, free liquid surface treatment strategy and support constitutive activation segment are recorded.

[0079] Subsequently, the solution control step size and nonlinear iteration threshold are set according to the mesh generation parameters and support constitutive parameter set. At the same time, the observation channel mapping copy is activated so that the nodal displacement, element strain and material temperature field corresponding to the observation type can be directly extracted from the numerical field of each time step.

[0080] Furthermore, numerical robustness checks are performed simultaneously with the loading of the load case set. The numerical robustness check is divided into three stages: pre-check, process check, and post-check. The pre-check quickly scans the mesh quality, the consistency of the initial segment of the support constitutive model, and the hydrodynamic boundary. If the element shape quality or boundary setting does not meet the threshold, the corresponding configuration item is read back from the model configuration list and a local reconstruction is triggered. The process check monitors the iteration convergence state and the boundedness of key quantities during the solution process. When the number of iterations in a continuous time step exceeds the set threshold or an unacceptable jump occurs in a key quantity, it automatically reverts to the previous time step and reduces the control step size, while recording the triggering reason and the load case items involved. After completing the solution of the entire load case, the post-check checks the continuity and comparability of the observation channel sequence output by each load case. If a period of time is found that does not match the availability of the time window in the observation mapping table, the period of time is marked as for reference only and a summary of the reason is attached.

[0081] After the above solution and verification process is completed, the nodal displacement sequence, element strain sequence, and material temperature field sequence under each working condition are summarized and aligned with the channels and positions consistent with the monitored quantities according to the observation channel mapping copy. This is then packaged into a fluid-solid response prediction package. The fluid-solid response prediction package includes a working condition index, time axis, observation channel number, numerical sequence body, and verification markers. Solving control parameters and rollback records are retained at the working condition level for subsequent interpretation. To establish alignment relationships across main steps, same-site alignment tags and time window alignment tags are written during packaging, allowing direct access to the monitoring baseline data package under the same probe group number and time window. After packaging, the output field name is explicitly stated as "fluid-solid response prediction package" at the end of this section, serving as input for subsequent main steps and for use by the fluid-solid response prediction package in S310. It is also stated that this output serves as one of the preliminary information sources in durability modeling and dual-constraint optimization, used for secondary simulation verification of candidate schemes and rollback path recording. In summary, the technical effects of this step are as follows: By loading the operating condition set and conducting numerical robustness checks throughout the early, middle, and late stages, a fluid-solid response prediction package is formed that is consistent with the observation channel and can be called at the same location and time window. The solution control and rollback information are also encapsulated for direct connection to subsequent data-model fusion and optimization closed loop.

[0082] like Figure 4 As shown, Figure 4 This is a flowchart illustrating step S300 provided in an embodiment of this application. Step S300 includes at least steps S310-S330:

[0083] S310. Obtain the fluid-solid response prediction package structure, perform time alignment, same-site comparison and uncertainty registration processing to obtain the fusion input set;

[0084] The monitoring baseline data package includes displacement, strain, and temperature entries organized based on fixed time steps, along with missing measurement markers, abnormal segment markers, reference value links, probe group numbers, correspondence between local coordinates and structural reference coordinates, and time axis anchor point summaries; the fluid-solid response prediction package includes nodal displacement sequences, element strain sequences, and material temperature field sequences organized by operating condition index, along with site alignment labels, time window alignment labels, solution control parameters, and rollback records.

[0085] Specifically, the two inputs mentioned above are imported into the time and position alignment module. First, based on the time axis anchor point summary and the time window alignment label, a mapping relationship is established between the fixed time step on the monitoring side and the output time history on the model side. The mapping relationship uses the time window as the basic unit, mapping each fixed time step to the corresponding model output time slice. On this basis, the monitoring entries and model entries with the same probe group number and the same observation channel number are matched at the same location. During the comparison process, the observation point index and the observation channel mapping copy are called to verify the consistency of the three-dimensional position, observation category and unit list on both sides. When there is a unit inconsistency or position deviation exceeding the limit, the alignment attempt is recorded as a correction item and transferred to the alignment correction branch.

[0086] In the alignment correction branch, the backup mapping list provided by the observation mapping table is used first for local replacement. When the backup mapping still cannot meet the position and unit consistency constraints, the time window is registered as a reference window only and removed from the same-site comparison set. At the same time, a summary of the reason for removal and the relevant working condition index is added to the quality label area. Further, after completing the alignment at the time and position levels, the monitoring items and model items are classified into three categories: mid-span section, support neighborhood, and temperature gradient sensitive area. A same-site comparison set is constructed in each category. Consistency judgment is performed on the items in the comparison set. The consistency judgment is carried out based on attitude correction records, reference value links, and statistical results of the same neighborhood. The judgment result generates a flag on the item indicating whether it is included in the statistical fusion.

[0087] Understandably, for items determined to enter statistical fusion, the timestamp source and channel status of the monitoring-side items, the solution control parameters and backoff records of the model-side items are recorded, and a corresponding site comparison fragment is generated. For items determined not to enter statistical fusion but need to be retained, they are recorded as bypass reference fragments. The bypass reference fragments do not participate in subsequent weight allocation and parameter updates, but are only used for corroboration and tracing. Subsequently, the uncertainty registration and processing stage begins. Uncertainty registration revolves around three types of sources: one is uncertainty from the monitoring side, which generates observation-side uncertainty identifiers based on the channel health registration form, missing measurement markers, and abnormal segment markers; another is uncertainty from the model side, which generates prediction-side uncertainty identifiers based on solution backoff records, time step reduction records, and local repartition records; and the third is uncertainty from the alignment process, which generates alignment-side uncertainty identifiers based on the number of spare mapping replacements, the number of unit conversions, and position deviation correction records. Three types of uncertainty identifiers are bound to the same-site alignment segments, and a comprehensive uncertainty description is generated for each time window. The comprehensive uncertainty description records the source of influence, frequency of occurrence, and severity level in the form of semantic fields. Finally, the same-site alignment segments entering statistical fusion are concatenated on the time axis at fixed time steps to form a fused segment set aggregated by probe group number and observation type; the bypass reference segments are encapsulated separately by category and a reference relationship is established with the fused segment set.

[0088] After completing the above processing, a fusion input set is output. The fusion input set includes a fusion fragment set, a bypass reference fragment, three types of uncertainty identifiers, a comprehensive uncertainty description, and alignment metadata. The output field name is explicitly stated in the text as the fusion input set, which serves as the input for the next subsection and is called by the fusion input set in S320. It is also pointed out that this fusion input set has a continuous field inheritance relationship with the parameter posterior candidate set, the parameter posterior distribution, and the structural health indicator in the subsequent main steps, and forms a closed-loop co-position comparison entry with the fluid-solid response prediction package and the monitoring baseline data package.

[0089] S320. Extract displacement, strain and temperature response from the fused input set, perform hierarchical Bayesian-filter parameter inversion processing, and generate a parameter posterior candidate set.

[0090] The fusion input set provides a collection of fusion segments organized at fixed time steps, along with three types of uncertainty identifiers and a comprehensive uncertainty description. Specifically, the fusion input set is imported into the parameter inversion control module. First, based on the response characteristics of the mid-span section class, the support neighborhood class, and the temperature gradient sensitive area class, a hierarchical inversion task list is constructed. The hierarchical inversion task list focuses on the slow variables of the overall structural response and the support system response at the upper level, and on the fast variables of the local component and material temperature field at the lower level. Subsequently, the displacement response corresponding to the upper-level task and the strain response corresponding to the support neighborhood are extracted from the fusion input set, and the temperature response corresponding to the lower-level task is extracted from the fusion input set. Different confidence levels are assigned to segments from different sources based on the comprehensive uncertainty description.

[0091] Furthermore, before the inversion is executed, the placeholder parameters of the support nonlinear constitutive parameter group and material temperature coupling parameters in the model configuration list are read and registered as updatable parameter slots in the inversion control module. For each task, a binding relationship is established between the observation sequence, the prediction sequence, and the uncertainty weight. The observation sequence comes from the fusion fragment set of the fusion input set, the prediction sequence comes from the aligned channel output of the fluid-structure response prediction package, and the uncertainty weight comes from the level mapping of the three types of uncertainty identifiers. Subsequently, a hierarchical filtering update process is entered. The filtering update process is executed sequentially at each fixed time step. For the upper-level task, the displacement response and the strain response in the support neighborhood are used to further update the parameter slots related to the overall structure and the support, and the update record for this time step is generated based on the update residual and the uncertainty weight. For the lower-level task, the temperature response and the local strain response are used to update the parameter slots related to material temperature coupling in one step, and when there is a bypass reference fragment, only the circumstantial evidence is recorded without generating an update action.

[0092] To constrain the drift magnitude and rate of change of parameters over time, parameter drift constraints and robust suppression strategies are introduced during the filtering update process. Parameter drift constraints control the parameter change range by reading the prior change boundary of the hierarchical task, while robust suppression strategies suppress parameters by performing neighborhood statistics on the update residuals to prevent abnormal segments from causing unacceptable jumps to parameter slots. Understandably, after the cyclic update across time steps, a time-series update trajectory is formed for each parameter slot, and the trajectory is segmented and aggregated. The aggregation rules segment the data according to the working condition index and time window, grouping consecutive time periods within the same working condition into the same segment, and calculating the center value and expansion range within the segment to form candidate intervals. For time periods with multiple backup mapping replacements or multiple unit conversions, the expansion range of the candidate interval is appropriately increased; for time periods surrounded by a large number of bypass reference segments, the candidate interval is only registered and not used for downstream health determination. Finally, the candidate intervals of the upper-level and lower-level tasks are summarized to form a parameter posterior candidate set. This parameter posterior candidate set covers parameter items related to the coupling of concrete equivalent elastic modulus and temperature, as well as parameter items related to the hysteresis of friction pendulum supports. It also retains the source composition, time coverage, and uncertainty notes for each candidate interval. After completing the above processing, the output field name is explicitly stated in the main text as "parameter posterior candidate set," serving as input for the next subsection and for use by the parameter posterior candidate set in S330. Simultaneously, it is indicated that the candidate intervals in this parameter posterior candidate set will be further filtered and delimited in the multi-factor influence weight assessment and health indicator generation processing, and a parameter inheritance chain will be established with the subsequent main steps of durability feature construction and automated scheme generation.

[0093] In one embodiment, the input source for S320 is the aforementioned fusion input set, which provides a set of fusion segments organized at fixed time steps and bound with three types of uncertainty identifiers (monitoring end, model end, and alignment end) and a comprehensive uncertainty description. Specifically, the fusion input set is imported into the parameter inversion control module. First, based on the response characteristics of the mid-span section class, support neighborhood class, and temperature gradient sensitive area class, a hierarchical inversion task list is constructed. The hierarchical inversion task list focuses on slow variables (such as the equivalent elastic modulus of concrete) in the overall structural response and support system response at the upper level, and on fast variables (such as temperature coupling parameters) in the local component and material temperature field at the lower level. Subsequently, the displacement response corresponding to the upper-level task and the strain response corresponding to the support neighborhood are extracted from the fusion input set, and the temperature response corresponding to the lower-level task is extracted from the fusion input set. Different confidence levels are assigned to segments from different sources based on the comprehensive uncertainty description. Furthermore, before the inversion is executed, the placeholder parameters of the support nonlinear constitutive parameter group and material temperature coupling parameters in the model configuration list are read and registered as updatable parameter slots in the inversion control module. For each task, a binding relationship is established between the observation sequence, the prediction sequence, and the uncertainty weight. The observation sequence comes from the fusion segment set of the fusion input set, the prediction sequence comes from the aligned channel output of the fluid-solid response prediction package, and the uncertainty weight comes from the level mapping of the three types of uncertainty identifiers. To align the observation sequence and the prediction sequence at the time step level and quantify the residuals, a state-space model is introduced, where the state vector represents the parameter slot value and the observation vector represents the physical quantity in the fusion segment. Formula ① is used to calculate the weighted residual. This tool eliminates time series bias through cross-correlation analysis to ensure the consistency of data at the same point and time window. Formula ①:

[0094]

[0095] in: Indicates a fixed time step The weighted residual vector; This represents the uncertainty weight matrix, derived from the level mapping of the three types of uncertainty indicators, with a value range of [0,1]. The observation sequence vector is extracted from displacement, strain, and temperature segments in the fused input set. Represents the predicted sequence vector, extracted from the aligned channel output of the fluid-solid response prediction package; symbol This indicates element-wise multiplication.

[0096] Data source → Indicator → Variable mapping: Extract displacement observations from the fused input set and denot them as... Strain observations and denoted as Temperature observation value and recorded as Together they form Displacement prediction values ​​are extracted from the fluid-solid response prediction package and denoted as... The strain prediction value is denoted as Temperature prediction value and recorded as Together they form Weights are extracted from the uncertainty identifier and formed Formula ① The S320 intermediate index is used for residual quantization. Subsequently, Equation ② is used to construct the state transition in the Bayesian filter update. This tool describes the evolution of parameter slots over time through state-space equations, avoiding the non-convergence problem of traditional single-layer inversion. Equation ②:

[0097] in: Indicates at time step The parameter state vector includes upper-level slow-variable parameters (such as the equivalent elastic modulus of concrete) and lower-level fast-variable parameters (such as temperature coupling parameters). The state transition matrix is ​​derived from the prior change boundary of the parameter slots. : at time step The parameter state vector; Represents the control input matrix; This represents the control input vector, which originates from the initial values ​​of placeholder parameters in the model configuration list, and has the same dimensions as the parameters. This represents the process noise vector, and its range is constrained by the prior boundary.

[0098] Data source → Indicator → Variable mapping: Extract the support nonlinear constitutive parameter set from the model configuration list and denot it as... Material temperature coupling parameters are denoted as Together they form State transition characteristics are extracted from prior changing boundaries and formed. and Formula ② directly uses the residual from Formula ①. The state update is correlated through the observation equation. This section outputs the weighted residual index and the predicted state value, which serve as inputs for hierarchical filtering updates, generating an intermediate state vector for subsequent update processing; the output fields are named intermediate state and residual set, and are called by subsequent processing in S320.

[0099] Following the aforementioned intermediate state vector and weighted residual, the hierarchical filtering update process begins. This process is executed sequentially at each fixed time step. For the upper-level task, the displacement response and the strain response in the support neighborhood are used to update the parameter slots related to the overall structure and the supports in one step. For the lower-level task, the temperature response and the local strain response are used to update the parameter slots related to material temperature coupling in one step. Equation ③ is used for the Bayesian filtering update step. This tool minimizes the mean square error through Kalman gain, achieving parameter estimation with uncertainty awareness. Equation ③:

[0100]

[0101] in: Indicates at time step The Kalman gain matrix; The prediction error covariance matrix is ​​derived from the state transition process and its dimension is the unit of parameter variance (e.g., Pascal²). The observation matrix represents the mapping of the state vector to the observation space; The observation noise covariance matrix is ​​derived from the weighted residuals. Uncertainty; symbol This indicates the matrix transpose.

[0102] Data source → Indicator → Variable mapping: Obtained from state prediction in formula ② Extracted from the binding relationship between observed and predicted sequences The weighted residuals from formula ① calculate Formula ③ directly uses the state prediction error from Formula ②. The residual information from Formula ①. Formula ④ is used for parameter state updates. This tool fuses observations and predictions through a linear combination and implements robust suppression based on uncertainty weights. Formula ④:

[0103] in: Indicates at time step Updated parameter state vector; This represents the state vector predicted by formula ②; The observation vector is derived from the fused input set; other symbols are the same as in formula ③. Time step index.

[0104] Data source → Indicator → Variable mapping: obtained from formula ② From formula ①, we get From formula ③, we get Formula ④ directly uses the Kalman gain from Formula ③. The observed values ​​are given by Formula ①. To constrain parameter drift, parameter drift constraints and robust suppression strategies are introduced during the update process, and the feasible region is determined using a Lyapunov function. This section outputs the updated parameter state sequence, generating the parameter update trajectory; the output field name is Parameter Update Trajectory Set, which is called by the trajectory aggregation processing of S320.

[0105] Following the aforementioned parameter update trajectory, after the cyclic update across time steps, a time-series update trajectory is formed for each parameter slot. This trajectory is then segmented and aggregated. The aggregation rules are based on the working condition index and time window, with consecutive time periods within the same working condition grouped into the same segment. The center value and extension range within each segment are calculated to form candidate intervals. Formula ⑤ is used for trajectory segmentation and aggregation. This tool handles the differences between slow and fast changes through time window segmentation and weighted averaging. Formula ⑤:

[0106]

[0107] in: Indicates the first A vector of aggregated center values ​​of parameters for each time window; Time window index; Time step index; Indicates the first The set of time steps for each time window is derived from the working condition index; The weighting coefficient is derived from the uncertainty weight and the evidence retention of the bypass reference segment, and its value range is [0,1]. This indicates the parameter status after being updated by formula ④.

[0108] Data source → Indicator → Variable mapping: Extracted from parameter update trajectory set Extracted from the comprehensive uncertainty description of the fused input set. Divided by working condition index Formula ⑤ directly uses the state update from Formula ④. Formula 6 is used to calculate the expanded range of the candidate interval. This tool avoids the influence of outliers through neighborhood statistics. Formula 6:

[0109] in: Indicates the first The parameter extension range vector for each time window.

[0110] Data source → Indicator → Variable mapping: Obtained from formula ⑤ ; obtained from formula ④ Formula ⑥ directly uses the aggregation center of Formula ⑤. For time periods with multiple backup mapping replacements or unit conversions, the expanded range is appropriately increased; for bypass reference segments, candidate intervals are only registered. Finally, the candidate intervals from the upper-level and lower-level tasks are summarized to form a parameter posterior candidate set. This parameter posterior candidate set covers parameter items related to the coupling of concrete's equivalent elastic modulus with temperature and parameter items related to the hysteresis of friction pendulum supports, and retains the source composition, time coverage, and uncertainty notes for each candidate interval. The output field is named "Parameter Posterior Candidate Set," serving as input for subsequent main steps and used by S330 to access the parameter posterior candidate set.

[0111] This section summarizes the technical effects: parameter inversion is achieved through hierarchical state space and Bayesian filtering, ensuring the fusion of data and model under spatiotemporal constraints; trajectory aggregation and uncertainty weighting enhance the robustness of the results; and the output candidate set provides a verifiable range for downstream health assessment.

[0112] S330. Perform multi-factor influence weight evaluation and health indicator generation on the posterior candidate set of parameters to generate the posterior distribution of parameters and the structure of structural health indicator.

[0113] The input source for this section is the aforementioned posterior candidate set of parameters. This set provides candidate intervals related to the coupling of the equivalent elastic modulus of concrete with temperature and candidate intervals related to the hysteresis of friction pendulum supports, and records the source composition, time coverage, and uncertainty notes of the candidate intervals. Specifically, the posterior candidate set of parameters is imported into the multi-factor evaluation module. First, a set of factors such as water pressure, temperature, support state, and constraint conditions is obtained. The factor set originates from the factor space definition required for the weighted evaluation results and is compared item by item with the working condition index in the fluid-solid response prediction package to establish a mapping from factors to working conditions. Subsequently, within each working condition, candidate intervals overlapping with the time period of that working condition are collected to construct a three-dimensional evaluation unit of working condition—factor—candidate interval. Further, the weights of factor importance and factor uncertainty are calculated for each evaluation unit. Factor importance is generated based on the dominance and participation of the factor under that working condition, and factor uncertainty is generated based on the comprehensive uncertainty description from the monitoring side and the model side. These two types of weights are bound to the candidate intervals to form a weighted candidate set. Interval screening and delimitation are performed on the weighted candidate set. The screening strategy follows two principles: first, data from different sources within the weighted candidate set should be comparable at the same location and within the same time window; if not comparable, they are considered circumstantial evidence; second, for candidate intervals that have undergone multiple alternative mapping replacements or unit conversions, their weights are reduced and extended towards the safety side during delimitation. After screening, the weighted candidate intervals for each parameter item are aggregated. The aggregation process involves grouping by operating condition and performing weighted segmentation and merging within each group to obtain the posterior center and extended range of the parameter item. For parameter items with significant time-varying drift, an additional drift rate description is generated, which serves as one of the inputs to the health indicator. Understandably, after obtaining the posterior center and extended range of each parameter item, a structural health indicator is generated. This indicator comprehensively uses information such as parameter drift rate, candidate interval coverage, and factor importance weights to form a multi-component description, and establishes a correspondence with the observation category, allowing the health descriptions of the mid-span section, support neighborhood, and temperature gradient sensitive area categories to be retrieved and accessed separately. To establish a connection with subsequent main steps, while generating the structural health indicator, a posterior distribution of parameters is output. This posterior distribution is indexed according to parameter items and includes source composition, time coverage, and weight summaries, forming a parameter-side input that can be invoked for subsequent durability feature construction and automated optimization. Finally, the posterior distribution of parameters and the structural health indicator are encapsulated into a unified data structure, and the output field name is explicitly stated in the text as "posterior distribution of parameters and structural health indicator," serving as input for subsequent main steps and invoked by the posterior distribution of parameters and structural health indicator in S410. It is also indicated that this output, together with the monitoring baseline data package and the fluid-structure response prediction package, constitutes a data fusion closed-loop entry point across main steps and serves as the parameter basis for secondary simulation verification and backtracking path selection.In summary, the technical effects of this step are as follows: by evaluating the weights of multiple factors and filtering and delimiting intervals, the posterior candidate set of parameters is transformed into a searchable posterior distribution of parameters, and a callable structural health indicator is generated within the same data structure, enabling continuous connection and consistent calling between the parameter side, the monitoring side, and the model side in subsequent steps.

[0114] like Figure 5 As shown, Figure 5 This is a flowchart illustrating step S400 provided in an embodiment of this application. Step S400 includes at least steps S410-S430:

[0115] S410. Obtain the posterior distribution of parameters and the structure health indicator structure, perform durability feature construction, in-service monitoring and accelerated degradation feature docking processing to obtain durability modeling input;

[0116] The input sources for this section are the posterior distribution of the parameters, the structural health indicator, and the monitoring baseline data package. The posterior distribution of the parameters has been indexed by parameter item and includes source composition, time coverage, and weight summary. The structural health indicator has been established with independently searchable descriptions according to mid-span section class, support neighborhood class, and temperature gradient sensitive area class. The monitoring baseline data package contains displacement, strain, and temperature entries registered with unified coordinates and time anchors, and includes missing measurement markers and abnormal segment markers. Specifically, the aforementioned three types of inputs are imported into the durability feature construction module. First, based on the partitions formed by the structural health indicator on the time and location axes, the monitoring entries are divided into two data segments according to the structural location and service environment: material response segment and construction interface segment. The material response segment extracts features for the long-term slow-change behavior of the trench concrete and lining layer, while the construction interface segment extracts features for the contact damage and thermal cycling coupling behavior of the support neighborhood and the limiting part. For each data segment, the corresponding parameter item in the parameter posterior distribution is read, and the observation items in the segment are paired with the parameter posterior center and extension range, and the binding relationship between the features and parameter sources in the segment is registered. Furthermore, after completing the data segmentation, feature dictionaries are constructed for carbonization-related, freeze-thaw-related, wet-dry cycle-related, and temperature gradient-related data. These feature dictionaries enumerate the required observation windows, statistical operators, and anomaly marker combinations for each scenario, and incorporate missing measurement markers and anomaly segment markers from the monitoring side into the operator masking rules. During feature extraction, displacement and strain entries within the same data segment are first divided into time windows and aligned with temperature entries. Then, trend quantities, fluctuation quantities, and extreme value quantities are output according to the requirements of the feature dictionaries, with segment indexes, location labels, and environmental labels appended to the output records. For locations susceptible to sudden environmental changes, drift descriptions are preferentially read from the structural health indicator and used as weight adjustment terms for durability features, forming an end-to-end traceable weight trajectory. Understandably, in order to access accelerated degradation information from the test side, the corresponding scenario entries in the accelerated degradation test library are imported into the homogenization interface. The test side entries include fields such as test group, loading regime, medium properties, and observation channel. Through the homogenization interface, the test channel name and the monitoring channel name are mapped to a unified observation channel vocabulary, and the test time axis is mapped to the equivalent service time. After homogenization, the test side entries are paired with the monitoring side features by comparing environmental characteristics such as ambient temperature, water content, and external medium concentration. During the pairing process, the similarity score and the summary of the source of deviation are recorded. When the similarity score is lower than the threshold, the test entry is only registered as an illustrative entry and does not participate in parameter fitting.Subsequently, a durability feature summary table is established, consisting of three parts: material response segment features, structural interface segment features, and experimental side isomorphic entries. The table records location labels, environmental labels, parameter sources, and similarity scores at the row level. For monitoring entries involving multiple backup mapping replacements or unit conversions, the table simultaneously records the alignment and conversion count fields for weight suppression in the downstream modeling stage. Finally, based on the index order of the parameter posterior distribution, parameters related to material temperature coupling and support hysteresis are bound to the durability feature summary table at the field level, forming a data view that can be directly read by the modeler. Within this section, the output field name is explicitly stated as "durability modeling input," serving as the input for the next subsection, for use by the durability modeling input in S420. It is also indicated that this durability modeling input forms a sequential relationship with the material performance evolution curve and lifespan interval in subsequent main steps, and together with the fluid-solid response prediction package, constitutes the pre-processing data for the secondary simulation verification of the scheme.

[0117] S420: Extract strength reduction features, elastic modulus decay features and crack threshold features from the durability modeling input, perform multi-task learning training and inference, and generate material performance evolution curves and life ranges.

[0118] The input source for this section is the aforementioned durability modeling input. This durability modeling input has been recorded in fields such as row-level record location labels, environment labels, parameter sources, similarity scores, and alignment counts, and is bound at the field level to the parameter items of the parameter posterior distribution. Specifically, the durability modeling input is imported into the modeler. First, hierarchical modeling preparation is performed according to location labels and environment labels. This hierarchical modeling preparation divides the tank body, lining layer, and support neighborhood into different modeling subdomains, and establishes parameter slots related to material temperature coupling and contact hysteresis within each subdomain. Simultaneously, strength reduction features, elastic modulus attenuation features, and crack threshold features are used as parallel prediction targets. A multi-task learning training and inference process is enabled in the modeler. Multi-task learning is labeled as Multi-Task Learning (MLL) upon its first appearance, and this process is organized during the training phase in the form of shared representations between tasks and task-specific branches. During the training preparation phase, similarity scores and alignment counts are weighted suppression terms and written into the sample weight map. The sample weight map assigns different participation weights to entries included in the batch when constructing the training batch. For samples containing supporting evidence entries, they are only used for stabilizing the shared representation and do not affect parameter updates for task-specific branches. Further, during the training execution phase, the overall durability feature table is sliced ​​according to time series order. For the material response segment, a slowly varying segment with a long time window is constructed; for the construction interface segment, a cyclic segment with a medium time window is constructed; and a transition segment with a short time window is generated at the intersection of the two segment sequences. The training process receives parameter terms from the parameter posterior distribution as conditional inputs. These conditional inputs are aligned with the segments in both the time and location dimensions and are used to adjust the state of the material temperature coupling groove and the contact hysteresis groove. During training, the participating location labels and environment labels are recorded for each batch so that they can be replayed according to the labels during subsequent curve generation. In the inference phase, without introducing new monitoring items, material performance evolution curves are generated for each part and its environmental combination. The material performance evolution curves consist of three parts: the trajectory of strength reduction over time, the trajectory of elastic modulus decay over time, and the trajectory of crack threshold over time. Confidence descriptions from sample weight mapping and conditional parameter inputs are added to each trajectory. While generating the curves, critical periods and warning periods are marked for each trajectory according to the service life setting and environmental loading regime. The critical periods and warning periods are mapped to the upper and lower bounds of the lifespan interval. The width of the upper and lower bounds varies due to the influence of the sample weight suppression term and the proportion of supporting evidence items.Understandably, to establish a consistent calling relationship between the curve and subsequent simulation verification, two sets of views are output for the curve and lifespan interval at the part level and the structural interface level, respectively. The part level view is used for the deduction of reinforcement strategies for the trench body and lining, while the structural interface level view is used for the deduction of replacement and reinforcement strategies for the support and limiting parts. Both sets of views retain the summary of the conditional parameter inputs when the curve was generated and register the index correspondence with the posterior distribution of the parameters, which facilitates subsequent sensitivity replay and scheme comparison by parameter item. After completing the above process, the material performance evolution curve and lifespan interval are output, and the output field name is explicitly stated in the text as "material performance evolution curve and lifespan interval," serving as the input for the next subsection for S430 to call the material performance evolution curve and lifespan interval. It is also indicated that this output, together with the posterior distribution of the parameters and the structural health indicator, enters the automated optimization and secondary simulation verification process, forming a direct data channel from durability prediction to scheme selection.

[0119] S430: Based on the material property evolution curve and life range, perform automated optimization and secondary simulation verification based on reliability-maintainability dual constraints to generate a scheme set and safety assessment report structure.

[0120] The input sources for this section are the aforementioned material performance evolution curves and lifespan ranges, the aforementioned posterior distribution of parameters, and the aforementioned structural health indicator. The aforementioned material performance evolution curves and lifespan ranges provide replayable views at both the component level and the structural interface level. The aforementioned posterior distribution of parameters provides parameter item indexes and source summaries. The aforementioned structural health indicator provides health descriptions and drift information for component classifications. Specifically, the three types of inputs are imported into the automated optimization module. First, constraint pairs of reliability constraints and maintainability constraints are constructed. The reliability constraints provide threshold ranges for the load-bearing and deformation requirements at the component level and the contact behavior requirements at the structural interface level. The maintainability constraints provide restrictions on construction accessibility, water outage periods, spare parts availability, and lifecycle maintenance windows. After the constraint pairs are established, the candidate measure library is read. The candidate measure library includes categories such as support replacement, limit reinforcement, local reinforcement, temperature control insulation, and monitoring encryption. Each category entry has a parameterized description and a list of applicable components. Subsequently, the automatic generation process is executed. The automatic generation process first scans the curve and lifespan range at the part level to mark the time period and part set that need to trigger measures. Then, at the interface construction level, it checks the drift information related to support hysteresis to form a joint trigger list of parts and interfaces. For the joint trigger list, items that cannot be implemented are screened out according to maintainability constraints. For the remaining items, the candidate measure library is called to match them according to part adaptation and interface adaptation to generate initial plans. In each initial plan, the parameters, trigger time period, part label and environment label involved are recorded. To verify the feasibility of the initial scheme, a secondary simulation verification process is initiated. This process calls the working condition settings and observation channel mapping copy of the fluid-structure response prediction package, writes the measure parameters from the initial scheme into the model configuration items, and generates a revised model configuration list. Numerical solutions are then run under representative working conditions, outputting an observation channel sequence consistent with the monitoring side. On the output sequence side, the responses in key time periods of the material performance evolution curves are compared according to location labels and structural interface levels. The sensitivity of relevant parameter items from the posterior distribution of the parameters is checked. After comparison and checking, items that pass and fail are marked in the scheme entries, and a rollback prompt is generated. Understandably, for scheme entries that fail, a rollback path is triggered. The rollback path executes from two branches: one branch shrinks or expands the measure parameters based on the extended range of the posterior distribution of the parameters and re-runs the secondary simulation verification; the other branch adjusts the measure triggering time or changes the measure combination based on the drift information of the structural health indicator. Both stop when the rollback count reaches the upper limit or the verification passes. For the verified schemes, a multi-objective screening process is initiated. Under the premise that the reliability and maintainability constraints are met, the multi-objective screening process reads and sorts indicators such as water outage impact, cost and construction period. The sorting process records the weight settings and a summary of the differences between schemes to form a candidate scheme set.Finally, the candidate solution set is structured and encapsulated, including solution parameters, applicable locations, triggering time periods, verification evidence, and rollback records. The document explicitly states that the output fields are the solution set and the safety assessment report, which serve as inputs for subsequent main steps and cross-main steps, for periodic review and further optimization during the operation and maintenance phase. It also indicates that this output is fed back into the probe group configuration and operating condition settings on the monitoring side for retesting and re-inversion in subsequent cycles, forming a closed loop from monitoring to modeling, from inversion to durability, from durability to optimization, and back to monitoring.

[0121] In summary, the technical effects of this step are as follows: Through automated optimization and secondary simulation verification driven by the dual constraints of reliability and maintainability, the durability prediction results and parameter-side uncertainties are transformed into evidence-supported feasible solutions. Records and rollback traces are preserved in a unified solution set and safety assessment report structure, forming a closed-loop data chain for long-term operation and maintenance.

[0122] In one embodiment, the input sources for S430 are the aforementioned material performance evolution curve and lifespan range, the posterior distribution of parameters, and the structural health indicator. The aforementioned material performance evolution curve and lifespan range provide replayable views at both the location level and the structural interface level. The posterior distribution of parameters provides a parameter item index and source summary. The structural health indicator provides a health description and drift information for location classification. Specifically, the three types of inputs are imported into the automated optimization module. First, a constraint pair of reliability constraints and maintainability constraints is constructed. The reliability constraints provide threshold ranges for the load-bearing and deformation requirements at the location level and the contact behavior requirements at the structural interface level. The maintainability constraints provide restrictions on construction accessibility, water outage periods, spare parts availability, and lifecycle maintenance windows. After the constraint pair is established, a candidate measure library is read. The candidate measure library includes categories such as support replacement, limit reinforcement, local reinforcement, temperature control insulation, and monitoring encryption. Each category entry has a parameterized description and a list of applicable locations. Subsequently, the automatic generation process is executed. This process first scans the curves and lifespan intervals at the component level, identifying the time periods and component sets requiring triggering measures. Then, at the interface level, it checks drift information related to support hysteresis, forming a joint trigger list for components and interfaces. For this joint trigger list, unimplementable items are filtered out based on maintainability constraints. The remaining items are then matched against a candidate measure library based on component and interface adaptation to generate an initial plan. To quantify the degree to which reliability constraints are met, a linear programming model is introduced. This tool evaluates the feasibility of the plan under load and deformation requirements by constructing a cost function. Formula ⑦ defines the reliability cost function, which incorporates the critical response thresholds in the material performance evolution curve. Formula ⑦:

[0123]

[0124] in: Indicates the first The cost of each solution under reliability constraints; Indicates the first The weighting coefficients of each reliability index are derived from the location priority of the structural health indicator, and the value range is [0,1]. Indicates the first The threshold for each reliability requirement is derived from the load-bearing or deformation requirements in the reliability constraint pair; Indicates the first The first scheme is in the The response values ​​for each indicator are derived from the responses at key time points in the material property evolution curve; The number of reliability indicators is derived from the definition of constraint pairs; Indicates the summation symbol; Reliability index.

[0125] Data source → Indicator → Variable mapping: Extract the carrying capacity threshold from the reliability constraint and denot it as... Deformation threshold and denoted as Together they form The hierarchical response of the affected area is extracted from the material property evolution curve and denoted as follows: Weights are extracted from the structural health indicator and denoted as... Formula ⑦ As an intermediate metric for S430, it is used for reliability assessment. Subsequently, Equation ⑧ is used to define the maintainability cost function, which is modeled through linear constraints and construction and resource limitations. Equation ⑧:

[0126] in: Indicates the first The cost-effectiveness of each solution under maintainability constraints; Indicates the first The penalty coefficient for each maintainability factor is derived from the constraints in the maintainability constraint pair (such as water outage periods), and its value range is [0,1]. This indicates an indicator function that takes the value 1 when the condition is true and 0 otherwise. Indicates the first The time window required for each measure is derived from the parameterized description (e.g., days) of the candidate measure library. Indicates the first The available time window for each measure is derived from the maintenance window in the maintainability constraint pair; Indicates the number of maintainability factors; This represents the function that takes the maximum value.

[0127] Data source → Indicator → Variable mapping: Extracting water outage periods based on maintainability constraints and recording them as... Construction can be done at the window and recorded as Together they form Extract the time window of the measure from the candidate measure library and record it as . The penalty coefficient is extracted from the constraint pair and denoted as... Formula ⑧ directly uses the output of Formula ⑦ to evaluate the overall feasibility of the scheme. This section outputs reliability and maintainability cost indicators as input for scheme selection; the output field name is double-constraint cost set, which is called by subsequent processing of S430.

[0128] Following the aforementioned dual-constraint cost indicators, the process moves to the automatic solution generation and initial verification phase. For the joint trigger list, unimplementable items are filtered out based on maintainability constraints. The remaining items are then matched against a candidate measure library based on location and interface adaptation to generate an initial solution. To integrate reliability and maintainability costs, formula ⑨ is used to construct a comprehensive scoring function. This tool achieves multi-objective optimization through weighted summation. Formula ⑨:

[0129]

[0130] in: Indicates the first A comprehensive score for each option; The weight representing the reliability cost is derived from the parameter sensitivity in the posterior distribution of the parameters, and its value range is [0,1]. The weight representing the maintainability cost is derived from the drift information of the structural health indicator, and its value range is [0,1]. This represents the reliability cost of formula ⑦; This represents the maintainability cost of formula ⑧.

[0131] Data source → Indicator → Variable mapping: obtained from formula ⑦ ; obtained from formula ⑧ Weights are extracted from the posterior distribution of the parameters. Weights are extracted from the structural health indicator. Formula 9 directly uses the values ​​of Formulas 7 and 8. Subsequently, to verify the feasibility of the initial scheme, a secondary simulation verification process is initiated, calling the working condition settings and observation channel mapping copy of the fluid-solid response prediction package, and writing the measure parameters from the initial scheme into the model configuration items. Formula 10 is used to define the verification comparison index; this tool ensures the consistency between the simulation output and the material property curve through cross-correlation analysis. Formula 10:

[0132] in: Indicates the first Validation difference indicators for each scheme; Indicates the first The first scheme in the second simulation The output values ​​of each observation channel are derived from the simulation results of the corrected model configuration list; The first part of the material property evolution curve represents the... The response values ​​for each key time period are derived from curve data; Indicates the number of observation channels; Indicates the summation symbol; : Observation channel index.

[0133] Data source → Indicator → Variable mapping: Extracting observations from the secondary simulation output and recording them as... The response value is extracted from the material property evolution curve and denoted as... Formula 10 directly uses the score from Formula 9 to prioritize the validation of high-scoring solutions. This section outputs the comprehensive score and validation difference index, generating an initial solution set; the output field name is "initial solution set," which is called by the rollback validation process of S430.

[0134] Following the aforementioned comprehensive scoring and verification difference indicators, for failed scheme items, a fallback path is triggered. This fallback path executes along two branches: one branch adjusts or expands the measure parameters based on the extended range of the parameter posterior distribution; the other branch adjusts the triggering period of the measures or changes the combination of measures based on the drift information of the structural health indicator. Formula ⑪ is used for the parameter adjustment branch; this tool optimizes the measure parameters using a linear programming model. Formula ⑪:

[0135]

[0136]

[0137] in: Indicates the first The adjustment amount of the measures parameters for each plan; This represents the L2 norm, used to minimize the adjustment range; This indicates the adjusted validation difference indicator. The function; This represents the validation difference threshold, derived from the extended range of the posterior distribution of the parameters; This represents the parameters used to solve the minimization problem.

[0138] Data source → Indicator → Variable mapping: Obtained from formula 10 Thresholds are extracted from the posterior distribution of the parameters. Formula 11 directly uses the verification difference from Formula 10. Formula 12 is used for multi-objective ranking; this tool integrates indicators such as water outage impact, cost, and construction period through weighted summation. Formula 12:

[0139] in: Indicates the first The final score for each option; , , This represents the weighting coefficient, which is derived from the weighting settings in the multi-objective screening process, and its value range is [0,1]. Indicates the first Cost indicators for each scheme; Indicates the first The project schedule targets for each plan; This represents the overall score of formula ⑨.

[0140] Data source → Indicator → Variable mapping: Obtained from formula ⑨ The cost is extracted from the candidate measure library and recorded as follows: The construction period is recorded as Weights are extracted by setting multi-objective filtering. , , Formula 12 directly uses the scoring from Formula 9. Finally, the candidate solution set is structurally encapsulated to generate a solution set and a security assessment report. The output fields are named "Solution Set" and "Security Assessment Report," serving as input for subsequent main steps and cross-main steps, for periodic review and further optimization during the later operation and maintenance phases.

[0141] This section summarizes the technical effects: automatic scheme generation is achieved through a dual-constraint cost function and comprehensive scoring; secondary simulation verification and backtracking paths ensure the feasibility of the scheme; and multi-objective ranking outputs a set of feasible optimization schemes.

Claims

1. A method for safety assessment and optimization design of a large-flow-rate aqueduct structure, characterized in that, include: S100: Based on the triple probe deployment scheme, perform probe group configuration, data access and raw monitoring sequence processing to generate monitoring baseline data packets; S200. Based on the monitoring baseline data packet, perform observation point mapping, model configuration and solution control processing to generate a fluid-solid response prediction packet; S300. Based on the fluid-solid response prediction package, perform time location alignment, parameter inversion and multi-factor evaluation to generate parameter posterior distribution and structural health indicator. S400. Based on the posterior distribution of the parameters and the structural health indicator, perform durability feature construction, durability modeling and automated optimization processing to generate a set of solutions and a safety assessment report.

2. The method according to claim 1, characterized in that, The process of configuring probe groups and accessing data in S100 specifically includes: The probe group configuration process includes: generating initial groups based on three candidate locations: mid-span section, support neighborhood, and temperature gradient sensitive area, according to the three-way combination rule of displacement gauge-strain gauge-temperature probe; calculating the spatial overlap and channel occupancy of each initial group to form a simplified group set; and recording the combination number, group role, relative position vector, installation attitude, wiring port and acquisition channel mapping for each simplified group to form the probe group configuration. The data access processing includes: establishing a coordinate mapping table from the structural reference coordinate system to the local probe coordinate system according to the probe group configuration; setting the sampling frequency, range, resolution and trigger threshold of the docking acquisition hardware; recording the reference time and timestamp source using the same time anchor point registration strategy; and performing continuity detection and static noise sampling on each probe channel to generate a channel health registration table.

3. The method according to claim 1, characterized in that, The process of processing the original monitoring sequence described in S100 specifically includes: The three types of observation sequences acquired are identified and labeled for missing data, and the missing data index is recorded without interpolation. Robust denoising processing is performed on candidate anomaly segments, and reference values ​​are output and retained along with the original values. Displacement, strain and temperature entries are uniformly resampled and divided into time windows to form a structured entry table arranged with fixed time steps and accompanied by missing data markers, anomaly segment markers, reference value links, probe group numbers, correspondence between local coordinates and structural reference coordinates, and time axis anchor point summaries. This table is then encapsulated into a monitoring baseline data package.

4. The method according to claim 1, characterized in that, The process of mapping observation points and configuring the model as described in S200 specifically includes: The observation point mapping process includes: resolving three types of labels under the structural reference coordinates: mid-span section, support neighborhood, and temperature gradient sensitive area; aggregating similar entries by probe group number; verifying attitude correction and mapping error estimation; establishing a candidate list of observation points for each probe group; mapping displacement entries to beam and slab element deflection observation points; mapping strain entries to stress and strain observation points of support perimeter components; and mapping temperature entries to material temperature field observation points. This generates an observation mapping table containing observation point number, model node or element identifier, observation category, time window availability, backup mapping list, and unit list. The model configuration process includes: constructing a three-dimensional geometric skeleton based on the aqueduct's geometric information; determining the mesh density of the local densification area based on the distribution of observation points; configuring placeholder parameters for the support nonlinear constitutive parameter group and material temperature coupling parameters; establishing a hydrodynamic boundary list associated with the working conditions; and generating a model configuration list that includes geometric skeleton parameters, mesh generation parameters, support constitutive parameter group, hydrodynamic boundary list, observation channel mapping copy, and working condition set placeholders.

5. The method according to claim 1, characterized in that, The solution control process described in S200 specifically includes: A set of working conditions covering conventional water flow, fluctuation transition, seismic coupling, and temperature difference superposition is established and loaded. The solution control step size and nonlinear iteration threshold are set, the observation channel mapping copy is activated, and numerical robustness verification including pre-verification, process verification, and post-verification is performed. When encountering iterative convergence problems or key quantity transitions during process verification, the step size is automatically backed up and reduced. The nodal displacement sequence, element strain sequence, and material temperature field sequence under each working condition are summarized and aligned to the channel and position consistent with the monitored quantities according to the observation channel mapping copy. The result is encapsulated into a fluid-structure response prediction package containing a working condition index, time axis, observation channel number, numerical sequence body, and verification mark. At the time of encapsulation, the same-site alignment label and time window alignment label are written.

6. The method according to claim 1, characterized in that, The time position alignment process described in S300 specifically includes: Based on the time axis anchor point summary and time window alignment labels, a mapping relationship is established between the fixed time step on the monitoring side and the output time history on the model side. Monitoring entries and model entries with the same probe group number and the same observation channel number are paired at the same site. The consistency of the three-dimensional position, observation category and unit list is verified by calling the observation point index and the observation channel mapping copy. If the consistency is still not met after the backup mapping is replaced, the time window is registered as a reference window only and removed from the same site comparison set. The same site comparison set is constructed by partitioning according to the three types of labels. Consistency judgment is performed to generate the same site comparison fragments that enter the statistical fusion and the bypass reference fragments that do not enter the statistical fusion. Uncertainty registration processing is performed. Three types of uncertainty identifiers and comprehensive uncertainty descriptions are generated according to the source of the monitoring side, the source of the model side and the source of the alignment process. The fragments that enter the statistical fusion are concatenated into a fusion fragment set, which is encapsulated together with the bypass reference fragments, the three types of uncertainty identifiers, the comprehensive uncertainty description and the alignment metadata into a fusion input set.

7. The method according to claim 1, characterized in that, The parameter inversion process described in S300 specifically includes: Based on the response characteristics of the mid-span section class, the support neighborhood class, and the temperature gradient sensitive area class, a hierarchical inversion task list is constructed, in which the upper level is oriented towards the slow variables of the overall structure and the support system, and the lower level is oriented towards the fast variables of the local component and material temperature field. The displacement response and support neighborhood strain response corresponding to the upper-level task and the temperature response corresponding to the lower-level task are extracted from the fused input set, and the confidence level is assigned to different segments based on the comprehensive uncertainty description. The placeholder parameters of the support nonlinear constitutive parameter group and the material temperature coupling parameter in the model configuration list are read as updatable parameter slots, and the binding relationship between the observation sequence, the prediction sequence and the uncertainty weight is established. The hierarchical filtering update process is performed. In the upper layer, the displacement response and the strain response of the support neighborhood are used to update the slots of the overall structure and the support-related parameters. In the lower layer, the temperature response and the local strain response are used to update the slots of the material temperature coupling-related parameters. When there is a bypass reference segment, only the circumstantial evidence is recorded and no update action is generated. In the filtering update process, parameter drift constraints and robust suppression strategies are introduced. The parameter drift constraints include parameter change range control based on the prior change boundary of the hierarchical task. The robust suppression strategies include suppression based on the statistics of the update residual neighborhood and a physical consistency penalty function applied when the parameter change direction is contrary to the temperature gradient and water pressure change direction. After the cross-time step cyclic update is completed, a time-series update trajectory is formed for each parameter slot. The time slots are segmented and aggregated according to the working condition index and time window. For time periods with multiple backup mapping replacements or multiple unit conversions, the expansion range of the candidate interval is appropriately increased. For time periods surrounded by a large number of bypass reference segments, only the intervals are registered and not used for downstream health determination. The candidate intervals of the upper and lower level tasks are summarized to form a parameter posterior candidate set that covers the parameters related to the coupling of concrete equivalent elastic modulus and temperature, as well as the parameters related to the hysteresis of friction pendulum support, and retains the source composition, time coverage and uncertainty annotations of each candidate interval.

8. The method according to claim 1, characterized in that, The multi-factor assessment process described in S300 specifically includes: The system acquires a set of factors including water pressure, temperature, support status, and constraints, and compares them item by item with the operating condition index. Within each operating condition, candidate intervals overlapping with the time period of that operating condition are collected to construct a three-dimensional evaluation unit of operating condition-factor-candidate interval. Factor importance and factor uncertainty weights are calculated and bound to candidate intervals to form a weighted candidate set. The weighted candidate set is then filtered and delimited, adhering to the principle that data from different sources within the weighted candidate set should be comparable at the same location and within the same time window. Candidate intervals that have undergone multiple backup mapping replacements or unit conversions are weighted and expanded towards the safety side. For each parameter item, the weighted candidate intervals are grouped by operating condition, and weighted segments are merged to obtain the posterior center and extended range. For parameter items with significant time-varying drift, an additional drift rate description is generated. The structural health indicator is generated by combining the parameter drift rate, candidate interval coverage, and factor importance weights, and the posterior distribution of parameters is output, indexed by parameter item and accompanied by source composition, time coverage, and weight summary.

9. The method according to claim 1, characterized in that, The durability modeling process described in S400 specifically includes: Layered modeling preparation is carried out according to part labels and environment labels. The main body of the tank, the lining layer and the support neighborhood are divided into different modeling subdomains. In each subdomain, parameter tanks related to material temperature coupling and parameter tanks related to contact hysteresis are established. The strength reduction feature, elastic modulus attenuation feature and crack threshold feature are used as parallel prediction targets. A multi-task learning process is enabled. During the training phase, the shared representation between tasks and task-specific branches are organized. The similarity score and alignment number are written into the sample weight mapping to assign different participation weights. The durability feature table is sliced ​​according to time series order to construct long time window slow-varying segments, medium time window cyclic segments and short time window transition segments. During the training process, parameter terms in the posterior distribution of parameters are received as conditional inputs to adjust the state of the parameter slots. During the inference phase, a material performance evolution curve is generated for each part and environment combination, consisting of the trajectory of strength reduction, elastic modulus decay, and crack threshold change over time. A confidence description from sample weight mapping and condition parameter input is added. Based on the service life and environmental loading regime, key periods and warning periods are marked for each trajectory and mapped to the upper and lower bounds of the life interval. Two sets of views are output at the part level and the structural interface level, respectively, generating the material performance evolution curve and life interval.

10. The method according to claim 1, characterized in that, The automated optimization process described in S400 specifically includes: A constraint pair of reliability constraints and maintainability constraints is constructed. The reliability constraints include a three-axis joint constraint of bearing axis, deformation axis and contact axis for the load-bearing and deformation requirements at the part level and the contact behavior requirements at the structural interface level. The maintainability constraints include a three-dimensional schedulable model for water outage periods, spare parts availability and construction reachability. The system reads the candidate measures library, which includes categories such as support replacement, limit reinforcement, local reinforcement, temperature control insulation, and monitoring encryption. It then executes an automatic scheme generation process. First, at the part level, it scans and marks the time period and part set that need to trigger measures by scanning the material performance evolution curve and life range. Then, at the construction interface level, it checks the drift information related to support hysteresis to form a part-interface joint trigger list. After filtering out unimplementable items according to maintainability constraints, it calls the candidate measures library to match and generate an initial scheme. To verify the feasibility of the initial scheme, a secondary simulation verification process is initiated. The working condition settings and observation channel mapping copy of the fluid-solid response prediction package are called. The measures parameters in the initial scheme are written into the model configuration items to generate a revised model configuration list. Numerical solutions are run under representative working conditions and the observation channel sequence consistent with the monitoring side is output. The responses in key time periods of the material performance evolution curve are compared according to the location label and the structural interface level. A multi-scale morphology matching algorithm is used to compare extreme values, waveform similarity and key inflection point deviations from macroscopic, mesoscopic and microscopic scales. Items that pass and fail are marked in the scheme entries and a rollback prompt is generated. For schemes that fail to pass, a fallback path is triggered. The first fallback branch shrinks or expands the measure parameters or performs Monte Carlo sampling and retesting based on the extended range of the parameter posterior distribution. The second fallback branch adjusts the measure triggering period or changes the measure combination based on the drift information of the structural health indicator, and falls back to pass or stops when the upper limit is reached. The verified schemes are screened for multiple objectives. Under the premise of meeting the constraints of reliability and maintainability, the impact of water outage, cost and schedule indicators are ranked and the weight settings and the difference summary between schemes are recorded to form a candidate scheme set. The candidate scheme set is encapsulated in a structured way. The encapsulation content includes scheme parameters, applicable parts, trigger time period, verification evidence and rollback records, and is attached with the evidence chain hash value containing upstream evidence and simulation evidence. The scheme set and safety assessment report are generated.

Images