Coastal environmental protection mixed tower light weight concrete corrosion data cloud processing system

Abstract

The present application relates to the technical field of concrete structure durability monitoring and intelligent operation and maintenance, in particular to a cloud processing system for corrosion data of coastal environment-friendly light-weight concrete towers, comprising: a data preprocessing module, which receives coastal environment data, dynamic load data and light-weight concrete body data through an edge gateway, and performs interpolation and smoothing processing on a preset time microelement to generate standard coupled data flow; a nonlinear conversion calculation module, which inputs the standard coupled data flow into a physical information neural network model to calculate structural micro-strain, local porosity change value and dynamic equivalent diffusion coefficient; a space-time evolution deduction module, which deduces three-dimensional permeation field data based on a preset reduced-order model of intrinsic orthogonal decomposition, and calculates the dynamic loss rate of the steel structure section; a risk quantification and decision module, which calculates the structure safety index and the remaining health life prediction value, and generates a dynamic operation and maintenance strategy; the present application realizes cloud quantification of the real corrosion risk inside the coastal light-weight concrete tower.

Description

Technical Field

[0001] This invention relates to the field of concrete structure durability monitoring and intelligent operation and maintenance technology, specifically a cloud processing system for corrosion protection data of lightweight concrete composite towers used in coastal environmental protection projects. Background Technology

[0002] In the current mixed tower operation and maintenance environment of coastal wind farms, the tower body is under high salt spray, high humidity and high wind load conditions for a long time. Various sensors will continuously generate multi-source heterogeneous monitoring data including surface environment, dynamic load and the state of lightweight concrete body. To assess structural health, existing corrosion monitoring solutions primarily focus on superficial assessments and alarms of surface environmental indicators, failing to integrate scattered cross-domain data into a unified computational framework. Furthermore, due to inconsistent sampling frequencies and susceptibility to network latency along the coast, direct analysis often leads to time-series misalignment and misjudgment of features. Simultaneously, existing mechanisms generally treat lightweight concrete as a static homogeneous medium, neglecting the nonlinear amplification effect of alternating loads and extreme wind conditions on the expansion of micropores in the material, making it difficult to accurately quantify the enhancing effect of dynamic response on ion diffusion capacity. Moreover, the lack of quantitative methods to extrapolate damage consequences from internal three-dimensional penetration to the steel reinforcement cross-section results in corrosion maintenance decisions heavily reliant on subjective human assessment, leading to severe response delays and an inability to accurately locate high-risk repair areas. Therefore, how to effectively integrate multi-source heterogeneous sensor data, accurately quantify the actual corrosion evolution risk inside the mixed tower under complex stress environment coupling, and realize the transformation of anti-corrosion strategy into predictive maintenance has become an urgent technical problem to be solved. Summary of the Invention

[0003] The purpose of this invention is to provide a cloud-based data processing system for corrosion protection data of lightweight concrete composite towers used in coastal environmental protection projects, thereby solving the following technical problems: This approach avoids relying solely on surface concentration for single-feature analysis in existing technologies, or neglecting the nonlinear impact of dynamic loads on the microporous structure and diffusion capacity of materials. It enables the cloud-based spatiotemporal quantification of the actual corrosion risk inside the mixing tower, thereby accurately generating dynamic operation and maintenance strategies and guiding external terminals to perform local anti-corrosion repairs.

[0004] The objective of this invention can be achieved through the following technical solutions: A cloud-based data processing system for corrosion protection data of lightweight concrete in coastal environmental protection hybrid towers is provided. The system includes: a data preprocessing module, which receives coastal environmental data including surface chloride ion concentration and salt spray deposition rate, dynamic load data, and lightweight concrete body data including initial porosity of the coastal environmental protection hybrid tower through an edge gateway, performs interpolation and smoothing on the dynamic load data and the lightweight concrete body data to generate a standard coupled data stream; and constructs a spatial voxel mesh of the lightweight concrete of the coastal environmental protection hybrid tower based on a preset three-dimensional geometric model. The nonlinear conversion calculation module is used to input the standard coupled data stream into the physical information neural network model, calculate the structural micro-strain caused by the dynamic load data, and calculate the local porosity change value based on the structural micro-strain and the initial porosity in the lightweight concrete body data, so as to generate a dynamic equivalent diffusion coefficient. The spatiotemporal evolution simulation module is used to input the dynamic equivalent diffusion coefficient into a preset reduced-order model based on intrinsic orthogonal decomposition, to deduce the three-dimensional permeability field data, and to calculate the cross-sectional dynamic loss rate of the steel reinforcement structure inside the lightweight concrete of the coastal environmental protection hybrid tower based on the three-dimensional permeability field data; the risk quantification and decision-making module is used to calculate the structural safety index and the predicted value of remaining healthy life based on the cross-sectional dynamic loss rate and the dynamic load data, and to generate a dynamic operation and maintenance strategy based on the predicted value of remaining healthy life.

[0005] Optionally, the data preprocessing module includes: a heterogeneous data receiving unit, used to receive the coastal environmental data and the lightweight concrete body data through a preset asynchronous channel, and to receive the dynamic load data through a preset real-time channel; A time alignment unit is used to interpolate the coastal environmental data collected at a first sampling rate based on the preset time micro-element, and to smooth downsample the dynamic load data collected at a second sampling rate, thereby aligning the coastal environmental data and the dynamic load data in the time dimension, wherein the second sampling rate is greater than the first sampling rate; a noise filtering unit is used to identify and filter out abnormal noise points in the aligned data to output the standard coupled data stream.

[0006] Optionally, the nonlinear transformation calculation module includes: a strain extraction unit, used to extract dynamic vibration characteristics from the dynamic load data through the physical information neural network model, and calculate the micro-strain of the structure; A porosity update unit is used to calculate the nonlinear disturbance of the initial porosity in the lightweight concrete body data by the micro-strain of the structure based on a preset fatigue model, so as to obtain the local porosity change value; a dynamic equivalent diffusion coefficient generation unit is used to calculate and output the dynamic equivalent diffusion coefficient based on the local porosity change value and a preset diffusion law model.

[0007] Optionally, the spatiotemporal evolution deduction module includes: an infiltration field calculation unit, used to take the dynamic equivalent diffusion coefficient as the input of the preset reduced-order model based on intrinsic orthogonal decomposition, calculate the infiltration path of chloride ions corresponding to the surface chloride ion concentration in the coastal environmental data inside the lightweight concrete, so as to generate the three-dimensional infiltration field data. The arrival time prediction unit is used to calculate the penetration time of chloride ions to the surface of the steel structure based on the three-dimensional permeation field data; the loss rate quantification unit is used to calculate the dynamic loss rate of the cross section of the steel structure within a time period consisting of the sum of multiple preset time micro-elements after the penetration time.

[0008] Optionally, the risk quantification and decision-making module includes: a safety assessment unit, used to convert the dynamic loss rate of the cross section into the residual bearing capacity based on a preset mechanical mapping model, extract the frequency shift characteristics in the dynamic load data to determine the stiffness correction coefficient, and perform fatigue damage assessment on the dynamic load data based on the rainflow counting method to determine the residual strength; and calculate the structural safety index by combining the residual bearing capacity and the maximum load peak value, and combining the stiffness correction coefficient and the residual strength. The lifespan prediction unit is used to calculate the predicted value of the remaining healthy lifespan based on the gradient of the structural safety index over time. The strategy generation unit is used to generate a first dynamic operation and maintenance strategy containing local anti-corrosion repair instructions when the remaining health life prediction value is lower than a preset life threshold; and to generate a second dynamic operation and maintenance strategy containing regular monitoring instructions when the remaining health life prediction value is equal to or higher than the preset life threshold.

[0009] Optionally, coastal environmental data may also include temperature and humidity data; The dynamic load data includes wind speed and direction data, multi-level vibration frequency data, and strain gauge data; the lightweight concrete body data also includes moisture content.

[0010] Optionally, the three-dimensional permeation field data is the three-dimensional spatiotemporal distribution matrix of chloride ions inside the lightweight concrete.

[0011] Optionally, the strategy generation unit is also used to: generate a three-dimensional visualized risk heat map based on the three-dimensional permeation field data and the cross-sectional dynamic loss rate; The three-dimensional visualized risk heat map is encapsulated into control commands, and the control commands are sent to an external drone terminal to guide the external drone terminal to execute the local anti-corrosion repair commands.

[0012] Optionally, before the heterogeneous data receiving unit receives the data, the coastal environmental data, the dynamic load data, and the lightweight concrete body data are preprocessed by the edge gateway, wherein the edge gateway is used to: compress the collected coastal environmental data, the dynamic load data, and the lightweight concrete body data based on the Huffman coding algorithm; Based on the block transmission strategy, the compressed data is transmitted to the heterogeneous data receiving unit through the asynchronous channel and the real-time channel respectively.

[0013] Optionally, the system further includes a model update module, which is used to: receive real cross-sectional loss data fed back during the actual anti-corrosion maintenance process from an external terminal, and compare the real cross-sectional loss data with the cross-sectional dynamic loss rate to calculate the comparison error; If the absolute value of the comparison error is greater than a preset error threshold, the parameters of the physical information neural network model are fine-tuned using the real cross-sectional loss data; if the absolute value of the comparison error is equal to or less than the preset error threshold, the current parameters of the physical information neural network model are maintained.

[0014] The beneficial effects of this invention are: 1) This invention performs interpolation, smooth downsampling, and alignment processing on multi-source heterogeneous monitoring data at preset time micro-elements, and filters out abnormal noise. This design effectively overcomes the time misalignment caused by different sampling rates between coastal environment, dynamic load, and body data, and provides a time-consistent standard data stream basis for subsequent coupled calculations. 2) This invention utilizes a physical information neural network to extract dynamic vibration features and calculate structural micro-strain, and combines a fatigue and diffusion model to generate a dynamic equivalent diffusion coefficient. This design breaks through the limitations of static models, accurately depicts the nonlinear changes in porosity caused by dynamic loads, and truly reflects the dynamic enhancement process of the internal permeability of concrete under harsh working conditions. 3) This invention inputs the dynamic equivalent diffusion coefficient into a reduced-order model to deduce the three-dimensional infiltration field, and combines the penetration time to estimate the dynamic loss rate of the steel reinforcement section; this design realizes the rapid and reduced-dimensional deduction of the evolution path of chloride ions in concrete and the actual damage consequences of steel reinforcement, avoiding the drawbacks of traditional methods that rely solely on surface environmental monitoring to reveal the true internal risks. 4) This invention combines the dynamic loss rate of the cross section with the peak value of the maximum load to calculate the structural safety index, and predicts the remaining healthy life based on the changing gradient to generate a dynamic operation and maintenance strategy; this design directly transforms the underlying corrosion data into intuitive operation and maintenance decision indicators, realizing the transformation of coastal mixed towers from periodic passive inspection to precise predictive maintenance. 5) This invention can generate a three-dimensional visualized risk heat map based on the three-dimensional permeation field and the dynamic loss rate of the cross section, and encapsulate it as a control command to be sent to an external UAV terminal; this design builds a data interaction channel between cloud-based risk quantification and on-site hardware execution, which can accurately guide the UAV to perform local anti-corrosion repair, improving the spatial positioning accuracy and response efficiency of on-site operations; 6) This invention receives real cross-sectional loss data from actual maintenance feedback and fine-tunes the parameters of the physical information neural network when the prediction comparison error exceeds the limit; this design constructs a data closed loop based on real field feedback, effectively overcoming prediction deviations caused by long-term material aging or sensor drift, and ensuring the long-term accuracy of the cloud-based anti-corrosion treatment system. Attached Figure Description

[0015] The invention will now be further described with reference to the accompanying drawings.

[0016] Figure 1 This is a block diagram of a cloud-based data processing system for corrosion protection of lightweight concrete composite towers in coastal areas, as described in this application. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1 As shown, a cloud-based data processing system for corrosion protection data of lightweight concrete in coastal environmental protection hybrid towers is disclosed. The system includes: a data preprocessing module, used to receive coastal environmental data including surface chloride ion concentration and salt spray deposition rate, dynamic load data, and lightweight concrete body data including initial porosity of the coastal environmental protection hybrid tower through an edge gateway; interpolating and smoothing the dynamic load data and the lightweight concrete body data to generate a standard coupled data stream; and constructing a spatial voxel mesh of the lightweight concrete of the coastal environmental protection hybrid tower based on a preset three-dimensional geometric model. The nonlinear conversion calculation module is used to input the standard coupled data stream into the physical information neural network model, calculate the structural microstrain caused by the dynamic load data, and calculate the local porosity change value corresponding to the spatial voxel grid based on the structural microstrain and the initial porosity in the lightweight concrete body data, so as to generate a dynamic equivalent diffusion coefficient. The spatiotemporal evolution deduction module is used to input the dynamic equivalent diffusion coefficient into a preset reduced-order model based on intrinsic orthogonal decomposition, deduce the three-dimensional permeability field data, and calculate the cross-sectional dynamic loss rate of the steel reinforcement structure inside the lightweight concrete of the coastal environmental protection mixed tower based on the three-dimensional permeability field data. The risk quantification and decision-making module is used to calculate the structural safety index and the predicted value of remaining healthy life based on the dynamic loss rate of the cross section and the dynamic load data, and to generate a dynamic operation and maintenance strategy based on the predicted value of remaining healthy life.

[0019] This embodiment provides a cloud-based processing mechanism for corrosion protection data of lightweight concrete composite towers in coastal areas. Specifically, the No. 7 composite tower of a coastal wind farm is taken as the continuous monitoring object. The composite tower is located under high salt spray, high humidity and high wind load conditions throughout the year. The tower adopts a composite structure of environmentally friendly lightweight concrete outer layer and internal steel reinforcement skeleton. The system's main operating line is set as follows: from the daily inspection stage to the typhoon warning stage, and then to the life review stage after the strong wind impact. The entire process always revolves around the same tower body, the same batch of sensors, and the same cloud model, thus forming a complete engineering closed loop. Specifically, the system receives three types of raw data through a data preprocessing module; the first type is coastal environmental data, such as surface chloride ion concentrations recorded sequentially as 1.8, 2.1, and 2.4 mol / m³ within a certain hour. 3 The salt spray deposition rates were recorded as 0.12, 0.15, and 0.17 mg / (cm³). 2 ·d); The second category is dynamic load data, such as wind-induced vibration acceleration sequences and strain gauge sequences collected within the same minute; the third category is lightweight concrete body data, such as the initial porosity of 0.18 recorded during tower construction acceptance; since the sampling frequencies of the three types of data are not consistent, the system presets 10 seconds as a time element; if the environmental data is collected once per hour, while the vibration data is collected 100 times per second, the environmental data is first linearly interpolated according to adjacent time points, and the vibration data is smoothed and summarized according to a 10-second window; Taking a 10-second window as an example, the system can obtain a set of standard coupled data streams: surface chloride ion concentration 2.25, salt spray deposition rate 0.16, equivalent wind load peak 38kN, micro-vibration main frequency 2.8Hz, and initial porosity 0.18; after this processing, the originally misaligned discrete data are mapped to a unified time. Based on this, the nonlinear transformation calculation module inputs the above-mentioned standard coupled data stream into the physical information neural network model; this model does not only perform statistical fitting, but also embeds diffusion constraints and structural strain constraints simultaneously; the loss function of the physical information neural network is composed of a weighted sum of the data-driven regression loss term and the residual loss term of the partial differential equation of the physical law, ensuring that the output structural micro-strain strictly follows the linear elasticity equilibrium equation and mass conservation constraints in the spatiotemporal distribution; to illustrate how the data flows, the following simplified deduction can be made: assuming the model input vector is [2.25, 0.16, 38, 2.8, 0.18], after the network forward calculation; Specifically, the physical information neural network model contains a mechanical decoupling subnetwork and a diffusion boundary subnetwork. The mechanical decoupling subnetwork only extracts the dynamic load features in the input vector to calculate the micro-strain of the structure, thus avoiding the physical interference of chemical environment parameters on the transient mechanical response. The surface chloride ion concentration and the salt spray deposition rate are transmitted to the diffusion boundary subnet to update the diffusion boundary conditions; the structural micro-strain of a local area of ​​the tower body within the 10-second window is obtained as 120 με. The system calculates the micro-strain transformation as a disturbance to the local porosity based on a preset fatigue model; specifically, the preset fatigue model uses a nonlinear microcrack propagation empirical formula to establish the micro-strain of the structure. With local porosity disturbance The quantitative mapping relationship is calculated using the following formula:

[0020] in, The material reference damage coefficient is determined by the mix proportions of lightweight concrete. This is a strain sensitivity constant; for example, when 120 με corresponds to a porosity increment of 0.006, the local porosity is updated from 0.18 to 0.186; combined with a preset diffusion law model, the porosity change and the environmental chloride ion conditions are jointly mapped to a dynamic equivalent diffusion coefficient; specifically, the preset diffusion law model uses a modified Brugmann equation or Abrams formula to establish the diffusion coefficient. With local porosity The nonlinear mapping relationship is calculated using the following formula:

[0021] in, The initial diffusion coefficient is . The initial porosity, The value is an empirical constant determined by the characteristics of lightweight concrete aggregate, ranging from 1.5 to 2.5. This explicit physical equation ensures that the pore expansion caused by micro-strain can be accurately quantified as an enhancement of ion migration ability. For example, the basic diffusion coefficient under static conditions is When the porosity increases to 0.186 and the vibration breathing effect is significant, the dynamic equivalent diffusion coefficient can be corrected to... ; Therefore, the system no longer stops at a shallow judgment of surface concentration, but forms a continuous calculation chain of surface environment—structural micro-strain—porosity change—enhanced diffusion capacity; the spatiotemporal evolution deduction module, based on the preset order reduction model of intrinsic orthogonal decomposition, quickly deduces the three-dimensional penetration of chloride ions inside concrete. To avoid overly abstract explanations, the local protective layer of the mixed tower can be simplified to eight voxel units of 2×2×2, with each voxel representing a material region of a certain thickness range. Let the initial concentration of the first voxel layer near the outer surface be [0.8, 0.7, 0.9, 0.85], and the initial concentration of the second voxel layer near the steel reinforcement surface be [0.2, 0.25, 0.15, 0.18]. When the dynamic equivalent diffusion coefficient increases, the concentration of the second voxel layer derived by the reduced-order model in the next cycle can be updated to [0.32, 0.38, 0.29, 0.34]. When the chloride ion concentration on the steel rebar surface corresponding to a certain voxel reaches the corrosion initiation threshold, the system records the penetration time and estimates the dynamic loss rate of the cross-section by accumulating multiple time micro-elements; for example, the original cross-sectional area of ​​the steel rebar is 100 mm². 2 It then evolved to 97 mm in the following period. 2 Therefore, the dynamic loss rate of the cross section in this stage can be denoted as 3%; Furthermore, the risk quantification and decision-making module calculates the structural safety index and the predicted value of remaining healthy life based on the cross-sectional dynamic loss rate and dynamic load data. For ease of understanding, a simple engineering example can be used: assuming that the remaining bearing capacity after corrosion is converted to 420kN, and the maximum load peak under extreme wind conditions is 300kN, then the structural safety index can be calculated as 420 / 300=1.40. The structural safety index is not calculated as a single static ratio, but rather as a weighted assessment of strength reserve factor, stiffness reduction factor, and fatigue damage accumulation factor. Specifically, the system extracts frequency shift characteristics from dynamic loads; if the first-order natural frequency of the hybrid tower decreases by more than 5% due to corrosion and cross-sectional losses, this is incorporated into the safety index. Stiffness correction factor; Meanwhile, by combining the rainflow counting method to assess fatigue damage under dynamic loads, the remaining bearing capacity is corrected from the static limit value to the residual strength under alternating stress. This multi-factor coupling model avoids the optimistic estimation of remaining life caused by neglecting fatigue effects. If the index drops from 1.62 to 1.52 and then to 1.40 within three consecutive months, the system can estimate the remaining healthy lifespan based on the decline gradient; for example, the predicted value is 11 months. Based on this predicted value, the system outputs a dynamic operation and maintenance strategy; if the predicted value is lower than the 12-month threshold, local anti-corrosion repair and key retesting are triggered; if it is higher than the threshold, routine monitoring is maintained. Furthermore, if environmental data within a micro-element is missing at a certain time but vibration data is complete, the structural response characteristics of that window are retained, and the most recent valid environmental data is called for short-term compensation; if the loss of vibration data exceeds a set proportion, such as exceeding 30% of the sample size of that window, then that window will not participate in the dynamic diffusion coefficient update, but will only maintain the result of the previous valid cycle and be marked as a low confidence state. If the porosity update result exceeds the allowable range of the project, such as less than 0 or greater than 0.6, the system will perform truncation correction and record the abnormal log to prevent subsequent simulation instability; if the calculated safety index is lower than 1, the strategy will be directly upgraded to high-risk handling, and load limit or shutdown suggestions will be output first, rather than just regular maintenance instructions. For example, 48 hours before Typhoon Hailan made landfall, the chloride ion concentration on the outer surface of the No. 7 mixing tower increased rapidly. However, the system did not directly alarm based on the surface value alone. Instead, it combined the increased vibration frequency caused by the rising wind speed to identify that the micro-strain in the 30-42 meter height area on the windward side of the tower was continuously increasing. It further deduced that the dynamic equivalent diffusion coefficient of the protective layer in this area was about 58% higher than the monthly average. 12 hours after the typhoon, the system updated the three-dimensional permeability field and determined that two areas near the steel bars were close to the corrosion initiation boundary. Finally, it gave an operation and maintenance strategy of implementing local anti-corrosion sealing and key retesting within 14 days. The purpose of this step is to unify the previously scattered environmental monitoring, load monitoring, and material parameter monitoring into a single consequence-driven data link, thereby achieving cloud-based quantification of the actual corrosion risk inside coastal lightweight concrete towers, rather than remaining at the level of surface alarms.

[0022] In a preferred embodiment of the present invention, the data preprocessing module includes: a heterogeneous data receiving unit, used to receive the coastal environmental data and the lightweight concrete body data through an asynchronous channel, and to receive the dynamic load data through a real-time channel; A time alignment unit is used to interpolate the coastal environmental data collected at a first sampling rate based on the preset time micro-element, and to smooth downsample the dynamic load data collected at a second sampling rate, thereby aligning the coastal environmental data and the dynamic load data in the time dimension, wherein the second sampling rate is greater than the first sampling rate; a noise filtering unit is used to identify and filter out abnormal noise points in the aligned data to output the standard coupled data stream.

[0023] This embodiment provides a preprocessing step for heterogeneous monitoring sources. Specifically, after the aforementioned No. 7 hybrid tower continued to enter the strong wind season, the system found that if data with different sampling frequencies and different link delays were directly sent into the subsequent model, it would cause the environmental response to be updated but the vibration response not yet aligned or the load signal spikes to be misjudged as material mutations. Therefore, this embodiment further refines the reception, time alignment and noise filtering. Specifically, the heterogeneous data receiving unit adopts dual-channel access; environmental data and material body data change slowly, making them suitable for batch input via asynchronous channels; dynamic load data changes rapidly, making them suitable for continuous push via real-time channels; taking an actual receiving cycle as an example, the asynchronous channel receives the environmental packet from 9:00 to 10:00 at 10:00, which includes a surface chloride ion concentration of 2.0 and a salt spray deposition rate of 0.14. The real-time channel continuously received one vibration sample every 0.01 seconds from 9:59:50 to 10:00:00, for a total of 1000 samples. Since the arrival times of asynchronous packets and real-time streams are not completely consistent, the system first puts them into the cache queue according to data type, and then performs unified arrangement based on timestamp. The processing of the time alignment unit can be illustrated through simplified data demonstration; assuming that the chloride ion concentrations at 9:00 and 10:00 are 1.8 and 2.0 respectively, and the system needs to obtain the estimated value at 9:30, then 1.9 can be taken in a linear manner; regarding vibration data, if 1000 samples are collected between 9:30 and 9:30:10, the local peak values ​​in the original sequence may be [0.12, 0.15, 0.14, 0.60, 0.13…], where 0.60 may be caused by instantaneous mechanical disturbances or electrical noise; The system first applies a smoothing window to the 1000-point sequence, for example, by taking the mean or median value in groups of 100 points, and finally obtains 10 smoothed points, which are then summarized into the equivalent vibration intensity of the 10-second time element. In this way, the environmental data is obtained from low-frequency interpolation to fine time resolution, and the vibration data is compressed from high frequency to time granularity consistent with the environmental data. After alignment, the noise filtering unit performs abnormal noise identification. To make the processing logic reproducible, the three features within a certain time micro-element can be set as surface chloride ion concentration 2.1, salt spray deposition rate 0.16, and equivalent vibration intensity 0.48. If the equivalent vibration intensity of two consecutive time micro-elements is 0.14 and 0.16, respectively, and the intermediate value suddenly jumps to 0.48, and the strain gauge does not show a corresponding increase, then this point can be identified as an isolated noise point. The system can adopt the neighborhood consistency criterion: when the deviation of a point from its two adjacent points exceeds a set proportion, such as exceeding 200%, and there is a lack of cross-validation from other sensors, the average value of the adjacent points of 0.15 is used as the replacement; similarly, if the chloride ion concentration is negative or unreasonably high for a short period of time, boundary clipping or setting it as missing will be performed directly, and it will be interpolated later. Furthermore, if the asynchronous channel does not receive material ontology data updates for an extended period, the static parameters in the modeling archive will be used by default, but a material parameter not updated flag will be added to the standard coupled data stream for subsequent model confidence adjustments. If the real-time channel delay exceeds the preset limit, for example, more than 3 time micro-element, the current window will not issue an immediate warning, but will only perform offline supplementary calculations to avoid misleading on-site decisions; If the timestamps of environmental data and vibration data conflict, for example, if two versions of the same field appear at the same time, the data packet with the correct verification code and high source credibility will be selected first, and the remaining data will be kept for future reference but will not enter the main calculation chain. For example, during the six consecutive hours before the arrival of the typhoon, the chloride ion concentration on the surface of the No. 7 mixing tower was reported once per hour, while the multi-layer accelerometer continuously uploaded data at 100 Hz; the system first expanded the six environmental points into 2160 10-second micro-element points, and then compressed the high-frequency vibration flow into the same number of window statistical values. In one of the windows, a sudden abnormal increase in vibration amplitude was found, but there was no corresponding change in strain gauge and wind speed. Therefore, it was identified as electromagnetic interference noise and filtered out to ensure that subsequent micro-strain calculations were not amplified by false spikes. The purpose of this step is to provide a time-consistent, noise-controlled, and physically interpretable standard coupled data stream for subsequent nonlinear coupled calculations, thereby achieving stable aggregation of heterogeneous sensor data in the cloud.

[0024] In a preferred embodiment of the present invention, the nonlinear conversion calculation module includes: a strain extraction unit, used to extract dynamic vibration characteristics from the dynamic load data through the physical information neural network model, and calculate the micro-strain of the structure; and a porosity update unit, used to calculate the nonlinear disturbance of the micro-strain of the structure on the initial porosity in the lightweight concrete body data based on a preset fatigue model, so as to obtain the local porosity change value. The dynamic equivalent diffusion coefficient generation unit is used to calculate and output the dynamic equivalent diffusion coefficient based on the local porosity change value and the preset diffusion law model.

[0025] This embodiment provides a nonlinear conversion mechanism that transforms dynamic loads into corrosion propagation capabilities. Specifically, although the aforementioned basic scheme can already align multi-source data, if concrete is still regarded as a static homogeneous medium, the amplification effect of typhoons and long-term alternating vibrations on the microporous structure of lightweight concrete will be underestimated. Therefore, this embodiment further introduces a continuous calculation process for strain extraction, porosity updating, and dynamic equivalent diffusion coefficient generation. Specifically, the strain extraction unit first extracts dynamic vibration characteristics that are more sensitive to the microstructure of the material from the dynamic load data; these characteristics are not limited to a single peak value, but can include the dominant frequency, frequency band energy, strain gauge synchronous response, etc. A simplified deduction can be made: Assume that within a certain 10-second window, the main vibration frequency caused by wind speed is 2.8Hz, the vibration energy index is 0.65, the average value of the strain gauge is 105με, and the peak value is 130με. After receiving these quantities, the physical information neural network does not simply calculate the average, but combines the tower boundary conditions, material stiffness and force constraints to output the structural micro-strain; for example, it outputs 120με in the windward region and 78με in the leeward region; in this way, the system obtains a spatially distinguishable micro-strain distribution, rather than a single average value for the entire tower. The porosity update unit calculates the local porosity change value based on microstrain; the micropores in lightweight concrete will produce microcrack propagation or pore connectivity effect under alternating loads, and the change is usually nonlinear; for the purpose of demonstrating the logic, the initial porosity can be assumed to be 0.180. When the local micro-strain is 50 με, the perturbation value given by the fatigue model is only 0.001; when the micro-strain increases to 120 με, the perturbation value does not increase linearly to 0.0024, but may jump to 0.006; when the micro-strain reaches 180 με, the perturbation value can further increase to 0.012; this shows that once the strain exceeds a certain fatigue-sensitive range, the porosity growth rate will accelerate significantly; thus, the porosity of the windward side region can be updated from 0.180 to 0.186, while that of the leeward side is only updated to 0.182; The dynamic equivalent diffusion coefficient generation unit outputs the diffusion capacity based on the updated porosity and diffusion law model; for ease of understanding, it can be assumed that the static diffusion coefficient corresponds to a porosity of 0.180 as follows: When the porosity is updated to 0.182, the diffusion coefficient increases to ; When the porosity is updated to 0.186, the diffusion coefficient increases to If conditions such as higher moisture content are added, it can be further corrected to... Thus, the system has completed the quantitative mapping from wind-driven vibration to enhanced material permeability. Furthermore, if the deviation between the micro-strain output by the physical information neural network and the measured value of the strain gauge continues to be too large, for example, if the deviation exceeds the set ratio for five consecutive windows, the online update of that region will be suspended, a conservative estimate will be adopted, and the sample will be sent to the subsequent model update queue. If the porosity disturbance calculated by the fatigue model is negative and exceeds the physical allowable range, it is only allowed to be corrected within a small rebound range, for example, not less than 95% of the initial porosity, to avoid unreasonably repairing the material to be denser than the initial state; if the dynamic equivalent diffusion coefficient jumps too fast without corresponding environmental or load basis, the slope limit is implemented to ensure that the change between adjacent windows does not exceed a preset multiple, thereby improving evolution stability. For example, near a height of 35 meters on the windward side of the No. 7 mixing tower, the system continuously detected an increase in the dominant frequency and an expansion in the peak fluctuation of the strain gauges during the outer edge of the typhoon. After strain extraction, the micro-strain in this area increased from the usual 70 με to 125 με; the porosity increased from 0.179 to 0.185 after updating; and the dynamic equivalent diffusion coefficient correspondingly increased from... Upgraded to Even if the surface chloride ion concentration is only about 20% higher than usual, the system can still identify that the internal permeation rate of the area has been significantly amplified. The purpose of this step is to express the vibration-induced micropore changes, which were originally impossible to observe directly, through a calculable intermediate quantity, thereby realizing the stress-diffusion coupling transformation, which is the most critical aspect of corrosion risk in coastal mixed towers.

[0026] In a preferred embodiment of the present invention, the spatiotemporal evolution deduction module includes: a permeation field calculation unit, used to use the dynamic equivalent diffusion coefficient as input to the preset reduced-order model based on intrinsic orthogonal decomposition to calculate the permeation path of chloride ions in the coastal environmental data inside the lightweight concrete, so as to generate the three-dimensional permeation field data; an arrival time prediction unit, used to calculate the penetration time of chloride ions to the surface of the steel structure based on the three-dimensional permeation field data; and a loss rate quantification unit, used to calculate the dynamic loss rate of the cross section of the steel structure within a time period consisting of the sum of multiple preset time micro-elements after the penetration time.

[0027] This embodiment provides a spatiotemporal evolution step from diffusion capacity to the consequences of steel bar damage. Specifically, after the aforementioned conversion is completed, the system already knows that the dynamic equivalent diffusion coefficient of certain areas has increased significantly. However, if we only stop at the intermediate conclusion that diffusion is faster, it is still difficult to arrange maintenance priorities on site. Therefore, this embodiment continues to map diffusion capacity to a three-dimensional penetration field, penetration time, and cross-sectional dynamic loss rate. Specifically, the permeation field calculation unit adopts a reduced-order model based on intrinsic orthogonal decomposition, thereby avoiding the need for a complete high-dimensional finite element solution each time. Considering that the dynamic equivalent diffusion coefficient is updated at a time micro-element frequency of 10 seconds, while the derivation step size of the reduced-order model is usually at the level of days or months, the permeation field calculation unit performs time integration and weighted averaging on all micro-element level dynamic equivalent diffusion coefficients within a preset macro-period before performing three-dimensional permeation derivation, and calculates the equivalent aggregation diffusion coefficient within that macro-period, thereby solving the problem of time scale crossing between high-frequency load response and low-frequency permeation evolution. To illustrate the treatment method, the concrete cover and the location of the reinforcing bars can be simplified into three depth levels: outer layer L1, middle layer L2, and layer adjacent to the reinforcing bars L3; Assuming each layer contains two lateral positions A and B, we obtain six grid points. If the initial chloride ion concentration at a certain moment is L1=[0.90, 0.85], L2=[0.35, 0.30], and L3=[0.10, 0.08], and the dynamic equivalent diffusion coefficient is higher at position A than at position B, then after the next evolution cycle, it can be updated to L1=[0.88, 0.83], L2=[0.46, 0.37], and L3=[0.18, 0.12]. This indicates that chloride ions at position A advance faster along the depth, forming a steeper permeation path. The arrival time prediction unit determines when chloride ions reach the steel bar surface threshold based on the three-dimensional penetration field. For example, if the initial corrosion threshold of the steel bar surface is 0.30, and the concentration at location L3-A continues to increase from 0.18 to 0.24, 0.29, and 0.33 in three cycles, the system determines that location A penetrates to the steel bar surface at the end of the third cycle. If each cycle represents 30 days, the penetration time is predicted to be 90 days; position B may only reach 0.30 after the fifth cycle, thus obtaining a penetration time of 150 days; in this way, the system not only knows that corrosion will occur, but also knows when corrosion will begin. The loss rate quantification unit performs a periodic cumulative estimation of the cross-sectional loss of the reinforcing bar after the penetration time; it can be assumed that the original diameter of a certain reinforcing bar corresponds to a cross-sectional area of ​​100mm². 2 The diameter at position A decreased to 99.2 mm in the first cycle after penetration. 2 The second cycle reduced the diameter to 98.4 mm. 2 The third cycle reduced to 97.1mm. 2 The cumulative cross-sectional dynamic loss rates for the three periods are 0.8%, 1.6%, and 2.9%, respectively. If penetration at location B occurs later, its loss rate curve will lag accordingly; the system's final output is a dynamic loss result associated with spatial location, rather than a single average corrosion value. Furthermore, if the input of the reduced-order model exceeds the training coverage range, for example, if the dynamic equivalent diffusion coefficient is significantly higher than the historical maximum sample, the system switches to conservative extrapolation mode and simultaneously reduces the output confidence level. If the geometric modeling of certain areas is incomplete, making it impossible to accurately locate the surface of the reinforcing bars, then only the penetration risk level will be output for that area, without the precise penetration time. If the penetration time has not yet been reached during the loss rate quantification process, then the dynamic loss rate of the section will be maintained at zero or an extremely low background value, so as not to enter the corrosion accumulation stage in advance and avoid generating false maintenance requirements. For example, within a height range of 35 to 38 meters on the windward side of the No. 7 mixing tower, the system deduced a significantly faster penetration path than the surrounding area based on the updated diffusion coefficient. The results showed that the concentration near the reinforcing steel in this area would reach 0.24, 0.31, and 0.39 in the next three 30-day cycles, thus locking the penetration time near the end of the second cycle. Furthermore, the system estimated that the dynamic cross-sectional loss rate of the corresponding reinforcing steel within 90 days after penetration could reach 2.6%, significantly higher than the 0.7% in the same height area on the leeward side. The purpose of this step is to ultimately translate the changes in front-end sensor data and intermediate material parameters into measurable internal damage consequences, thereby enabling spatiotemporal prediction of the steel corrosion process in coastal lightweight concrete towers.

[0028] In a preferred embodiment of the present invention, the risk quantification and decision-making module includes: a safety assessment unit, used to convert the dynamic loss rate of the cross section into the residual bearing capacity based on a preset mechanical mapping model, extract the frequency shift characteristics in the dynamic load data to determine the stiffness correction coefficient, and perform fatigue damage assessment on the dynamic load data based on the rainflow counting method to determine the residual strength; and to calculate the structural safety index by combining the residual bearing capacity and the maximum load peak value, and combining the stiffness correction coefficient and the residual strength. The lifespan prediction unit is used to calculate the remaining healthy lifespan prediction value based on the gradient of the structural safety index over time; the strategy generation unit is used to generate a first dynamic operation and maintenance strategy containing local anti-corrosion repair instructions when the remaining healthy lifespan prediction value is lower than a preset lifespan threshold; and to generate a second dynamic operation and maintenance strategy containing routine monitoring instructions when the remaining healthy lifespan prediction value is equal to or higher than the preset lifespan threshold.

[0029] This embodiment provides a risk assessment and operation and maintenance decision-making process with consequence quantification as its core. Specifically, the aforementioned simulation can already give the dynamic loss rate of the steel bar section, but if manual judgment is still required to determine whether to repair, response lag is likely to occur during the high salt spray season. Therefore, this embodiment further converts the loss rate into load-bearing safety and remaining life, and automatically generates differentiated operation and maintenance strategies. Specifically, the safety assessment unit maps the dynamic loss rate of the cross section to the remaining bearing capacity. This can be illustrated by a simplified example: Suppose that the steel bars in a certain monitoring area can provide a bearing capacity of 500kN when they are not corroded. When the cumulative dynamic loss rate of the cross section is 4%, the remaining bearing capacity after conversion by the mechanical mapping model is 470kN. When the loss rate increases to 8%, the remaining bearing capacity may drop to 435kN. At the same time, the system extracts the maximum load peak value within the same assessment period from the dynamic load data, such as the peak value of 320kN during the typhoon stage; the structural safety index can be obtained by the ratio of the remaining bearing capacity to the maximum load peak value; for example, 470 / 320=1.47, 435 / 320=1.36; thus, the system can compare the safety boundaries of different regions and different time periods on a uniform scale. The life prediction unit further observes the gradient of the safety index change; assuming that the index of a certain region for the most recent four periods is 1.62, 1.54, 1.45 and 1.36 respectively, the downward trend is more obvious; The system can estimate the time required to reach the critical safety line based on the descent slope; for example, when the critical line is set to 1.20, the remaining healthy lifespan can be estimated to be about 8 months based on the current gradient; if the exponential sequence at another position is 1.70, 1.68, 1.66, 1.64, then its remaining healthy lifespan can be greater than 24 months. The strategy generation unit generates operation and maintenance results based on the lifespan threshold. If the lifespan threshold is set to 12 months, the system outputs the first dynamic operation and maintenance strategy when the predicted value is 8 months. This strategy may include local anti-corrosion repair instructions, key area retesting plans, and temporary load control suggestions. If the predicted value is 18 months, a second dynamic operation and maintenance strategy will be output, such as maintaining monthly routine monitoring, quarterly drone inspections, and model review for the next cycle. In this way, the system does not trigger alarms based on a single concentration threshold, but allocates maintenance resources according to the differences in lifespan. Furthermore, if the safety index of a certain area drops briefly but recovers, the system will detect whether the change is caused by extreme load transients; only when multiple consecutive cycles show a downward trend will lifetime convergence prediction be performed. If the safety index is close to or below the critical line, for example, ≤1.05, the system will no longer wait for the lifespan threshold judgment, but will directly output the emergency response strategy; if different sensor sources cause the load-bearing capacity mapping results to differ too much, the more conservative side will be used as the decision basis, and on-site verification will be prompted to avoid delays in maintenance due to optimistic estimates. For example, on the 5th day after the typhoon, the dynamic loss rate of the section of the 37-meter area on the windward side of the No. 7 hybrid tower reached 5.2% after the update, which is equivalent to a remaining bearing capacity of 458kN. The peak load recorded during the same period was 330kN, resulting in a safety index of 1.39. Based on the trend of the index dropping from 1.58 to 1.49 and then to 1.39 over the past three months, the system predicts that the remaining healthy lifespan of the area is about 10 months, which is lower than the 12-month threshold. Therefore, the system automatically generates an operation and maintenance strategy of local anti-corrosion repair and key retesting. Furthermore, to maintain consistency in terminology throughout the text, in this embodiment and its related embodiments, the safety index is used as an abbreviation for the structural safety index. Both refer to the same evaluation quantity and represent the ratio of the remaining bearing capacity to the peak value of the maximum load within the same evaluation period. The lifespan prediction value is a shorthand for the remaining healthy lifespan prediction value. Both represent the estimated remaining time before the structure evolves from its current state to the preset critical safety threshold. Accordingly, the threshold judgment object called by the strategy generation unit is always the remaining healthy lifespan prediction value, rather than other independent lifespan parameters. The purpose of this step is to transform the complex diffusion evolution results into operationally executable safety and life indicators, thereby enabling the transition of coastal hybrid towers from periodic maintenance to predictive maintenance.

[0030] In a preferred embodiment of the present invention, the coastal environmental data further includes temperature data and humidity data; the dynamic load data includes wind speed and direction data, multi-level vibration frequency data, and strain gauge acquisition data; the lightweight concrete body data further includes moisture content.

[0031] This embodiment provides a refinement mechanism for expanding input elements. Specifically, if the basic scheme only uses surface chloride ion concentration, salt spray deposition rate, vibration data, and initial porosity, although it can complete the basic deduction, the diffusion rate and microcrack response may deviate under continuous high humidity or large diurnal temperature differences. Therefore, this embodiment incorporates temperature, humidity, wind speed and direction, multi-level vibration frequency, strain gauge data, and material moisture content into a unified processing framework. Specifically, temperature and humidity are used to modify chloride ion migration activity and material wettability; for example, the same surface chloride ion concentration of 2.2 has different effects on promoting internal diffusion at 18°C ​​and 70% humidity versus at 30°C and 95% humidity. The system can assign an environmental correction factor of 0.9 to the former and 1.2 to the latter; wind speed and direction are used to determine the load directionality and the location of the windward area; if the wind direction changes from southeast to due east, the high-risk area of ​​the tower may shift from the original southeast quadrant to the due east quadrant. Multi-level vibration frequencies can reflect the differences in dynamic response at different heights and structural levels, while strain gauge data provides a direct calibration basis for micro-strain calculations; material moisture content reflects the water content in micropores, and under the same porosity, regions with higher moisture content are more likely to accelerate ion migration. For a more intuitive demonstration, the input data for a 30-meter height area of ​​the tower can be set as follows: temperature 28℃, humidity 92%, wind speed 17m / s, wind direction 15° east of south, dominant frequency 2.9Hz, secondary frequency 5.6Hz, strain gauge peak value 128με, and moisture content 11%. The data for the 45-meter height area are: temperature 26℃, humidity 89%, wind speed 19m / s, same wind direction, dominant frequency 3.4Hz, secondary frequency 6.1Hz, strain gauge peak value 142με, and moisture content 9%. After system calculation, the former area may diffuse faster due to higher moisture content, while the latter area may have greater pore disturbance due to stronger load. Although the sources of risk are different, both can be included in a unified model to obtain the final result. Furthermore, if the temperature and humidity sensor malfunctions, data from nearby height sensors or weather stations on the same tower can be used as a substitute, but the substitution range is limited to within a preset height difference; if wind direction data is missing, the system will temporarily maintain the most recent valid wind direction while reducing the weight of positioning on the windward side. If there is a significant conflict between the vibration frequencies of multiple levels, such as abnormal frequencies in higher levels while completely stable frequencies in lower levels, the system will combine strain gauge data for consistency verification to avoid misjudgment caused by single-level equipment failure; if the moisture content has not been updated for a long time, the most recent laboratory measurement value will be maintained and a conservative coefficient will be added to the risk calculation. For example, during the post-typhoon No. 7, during the humidification phase, the surface chloride ion concentration did not continue to rise, but the humidity increased from 85% to 96%, and the material moisture content increased from 9% to 12%. Based on this, the system adjusted the diffusion correction factor for several areas upward; at the same time, the wind direction gradually shifted from southeast to east, causing the windward hotspot to move from the southeast side of the tower to the due east side; the results showed that although the total environmental salt load changed only slightly, the risk of local internal infiltration was still increasing. The purpose of this step is to make the coupled model more complete in representing the actual working conditions of coastal mixed towers by supplementing key data reflecting the degree of wettability, directional load, and material moisture content, thereby improving the accuracy of risk prediction and spatial positioning capabilities.

[0032] In a preferred embodiment of the present invention, the three-dimensional permeation field data is the three-dimensional spatiotemporal distribution matrix of chloride ions inside the lightweight concrete.

[0033] This embodiment provides a method for organizing three-dimensional permeability field data. Specifically, in the aforementioned spatiotemporal evolution process, if only textual descriptions or single-point concentration values ​​are saved, it is difficult to support subsequent risk visualization, regional comparison, and UAV path planning. Therefore, this embodiment explicitly organizes the three-dimensional permeability field as a three-dimensional spatiotemporal distribution matrix of chloride ions inside lightweight concrete. Specifically, the so-called three-dimensional spatiotemporal distribution matrix can be understood as a combination of spatial three-dimensional grid and time series data structure; for ease of explanation, a local region can be divided into eight voxels of 2×2×2, which are represented by position numbers P1 to P8 respectively. For each time slice t1, t2, and t3, the corresponding concentration values ​​are recorded. At time t1, the matrix can be represented as [P1=0.20, P2=0.24, P3=0.18, P4=0.22, P5=0.08, P6=0.10, P7=0.07, P8=0.09]; and at time t2, it is updated to [P1=0.27, P2=0.31, P3=0.23, P4=0.28, P5=0.12, P6=0.16, P7=0.10, P8=0.14]. If P2 corresponds to the area closer to the surface of the reinforcing bar, it can be directly used to determine its penetration progress; the system does not require a 2×2×2 grid, and in engineering, it can be expanded to a higher resolution grid according to the tower size and computing resources; Furthermore, to avoid ambiguity in the representation of three-dimensional matrices and time series, the system can organize the three-dimensional infiltration field data into a set of three-dimensional concentration matrices arranged in chronological order during implementation. That is, each time slice corresponds to a three-dimensional concentration matrix, and multiple time slices are connected end to end to form a complete spatiotemporal distribution dataset. In other words, at any single moment, the main data is a three-dimensional spatial concentration matrix; when called across multiple moments, it is represented as a matrix sequence with a time index; this maintains the expression of the three-dimensional spatiotemporal distribution matrix in the embodiment, and keeps the storage, retrieval and subsequent calculation objects consistent, avoiding the misunderstanding of it as a single static matrix. Specifically, the reduced-order model pre-extracts a snapshot matrix of the chloride ion diffusion process through high-dimensional finite element simulation, and then uses intrinsic orthogonal decomposition to extract the preceding... basis functions of the first space During real-time cloud computing, the system uses the dynamic equivalent diffusion coefficient as a parameter input and obtains the modal coefficients by solving a system of low-dimensional ordinary differential equations. Ultimately, through the formula

[0034] This method rapidly reconstructs a three-dimensional permeation field matrix encompassing height, circumference, and depth. Compared to traditional layer-by-layer iteration, this approach reduces computation time by over 90% while maintaining three-dimensional spatial resolution. The advantage of this data structure is that it can directly support the calculation of spatial adjacency relationships. For example, if P2, P4, and P6 increase continuously, it indicates that the infiltration path may advance along a certain oblique channel. If only P1 increases while the adjacent voxels remain stable, it is more likely to be local noise or model instability, which needs to be verified. The system can also slice according to three dimensions: height, orientation, and thickness to extract the concentration evolution curve of a specific region. Furthermore, to facilitate integration with the subsequent cross-sectional loss rate quantification module, the system can also add a spatial positioning index to each voxel. The spatial positioning index includes at least a height range, a circumferential angle range, and a protective layer depth range. In this way, when a voxel approaches the corrosion initiation threshold within a certain number of consecutive time slices, the system can directly trace its corresponding physical location in the tower without having to repeatedly convert between the permeation field and the geometric model. Furthermore, if some voxels cannot be reliably updated due to missing local data, the voxel inherits the concentration of the previous valid time step and is marked with an interpolation tag. If some points in the matrix have physically unreasonable negative or excessively large values, range correction is performed, and the upstream diffusion coefficient input is checked for abnormalities. If the spatial grid resolution is too high, causing the cloud simulation to time out, it can automatically degrade to a coarser grid version to prioritize the continuity of risk identification. For example, in a local cubic model of a 35-meter area on the windward side of the No. 7 mixed tower, the system divides the protective layer into eight voxels and continuously records the chloride ion distribution matrix for the next six cycles. By comparison, it can be seen that the two voxels near the outer wall grow faster, while the voxel in the upper right corner near the steel bar suddenly approaches the threshold in the fourth cycle, suggesting that this path may be a key repair zone in the future. Furthermore, to ensure that the three-dimensional spatiotemporal distribution matrix of the three-dimensional infiltration field data and a set of three-dimensional concentration matrices arranged in chronological order refer to the same information throughout the text, the system adopts a unified indexing rule: the main body of data within a single time slice is recorded as a three-dimensional concentration matrix, whose three spatial dimensions correspond to the height interval, the circumferential angle interval, and the depth interval of the protective layer, respectively; when calling data across multiple time slices, the matrix sequence is formed according to the time index order. In a single-time context, the three-dimensional infiltration field data is equivalent to a single three-dimensional concentration matrix; in a multi-time context, it is equivalent to a sequence of three-dimensional concentration matrices with time indexes. Neither of these changes the fact that the object they represent is the spatial distribution of chloride ions within lightweight concrete. To avoid unit confusion, the values ​​of each voxel in the matrix uniformly represent the chloride ion concentration at the corresponding voxel location, and the concentration units used are consistent with those in the upstream infiltration extrapolation module. Spatial location indexes are only used for location and are not mixed with concentration values. The purpose of this step is to standardize the internal penetration results into a storable, comparable, and retrievable spatiotemporal matrix, thereby achieving a unified data foundation for subsequent loss rate quantification, risk heatmap generation, and external device linkage.

[0035] In a preferred embodiment of the present invention, the strategy generation unit is further configured to: generate a three-dimensional visualized risk heat map based on the three-dimensional permeability field data and the cross-sectional dynamic loss rate; encapsulate the three-dimensional visualized risk heat map into control instructions, and send the control instructions to an external drone terminal to guide the external drone terminal to execute the local anti-corrosion repair instructions.

[0036] This embodiment provides a visualized linkage step for external execution devices. Specifically, although the aforementioned strategy can output the judgment that local repair is needed, if it still relies on manual search of the target area layer by layer on the entire hybrid tower, it will increase the operation time and reduce the response efficiency. Therefore, this embodiment integrates the three-dimensional permeation field and the cross-sectional dynamic loss rate into a risk heat map and converts it into control commands that can be executed by UAVs. Specifically, the system first calculates the comprehensive risk value of each spatial unit based on the three-dimensional permeability field data and the dynamic loss rate of the cross section. For ease of explanation, the permeability risk scores of four units can be set to 0.6, 0.8, 0.4, and 0.7, respectively, and the cross section loss weight scores can be set to 0.3, 0.5, 0.2, and 0.6, respectively. The comprehensive risk values ​​can then be weighted to obtain 0.9, 1.3, 0.6, and 1.3. The system then maps this value to a color level, for example, 0 to 0.7 is green, 0.7 to 1.1 is yellow, and greater than 1.1 is red. As a result, a continuous thermal distribution is generated on the surface of the three-dimensional model of the tower, with the red area representing the priority repair point. Furthermore, considering that high-risk elements in the three-dimensional permeation field may be located inside the concrete, while the actual target of the UAV is located on the outer surface of the tower, the system performs surface projection or neighborhood mapping on the internal risk voxels before encapsulating the control commands. Specifically, this can be understood as follows: when an internal voxel is determined to be high-risk, the system projects it onto the nearest external working surface according to its corresponding height, circumferential position and protective layer thickness direction, and expands it to the surrounding area to form a workable area according to a preset safety margin. For example, if a high-risk voxel is located at a height of 36m, an azimuth of 95°, and a thickness of 40mm from the outer surface, the system can map it as an outer surface work zone with a height of 35.5m to 36.5m and an azimuth of 90° to 100°, so that the UAV receives a spatial target that can be directly reached and sprayed, rather than an abstract internal node number. The strategy generation unit extracts high-risk areas from the heatmap as external execution tasks; when encapsulating these tasks, it can include the area's height range, circumferential angle range, estimated dwell time, repair material type, and job priority. For example, a high-risk area can be defined as follows: height 35m to 38m, azimuth 85° to 110°, priority 1, recommended spraying of anti-corrosion sealant, and estimated operation time 12 minutes. After receiving the information, the drone terminal can automatically plan the flight path according to the preset route protocol, first arrive at the high-risk area, and then perform image verification, local spraying or marking operations. Furthermore, to avoid mishandling caused by direct spraying when the cloud prediction results are inconsistent with the location of surface defects on site, the drone can first complete a close-range image verification or point re-measurement according to the task package requirements before executing the local anti-corrosion repair command. When the visual recognition of surface cracks, erosion, water seepage marks, etc. are consistent with the risk area on the heat map, it can then switch to spraying or sealing operation mode. If the on-site verification results are significantly inconsistent with the heat map, the task is sent back to the cloud and transferred to manual confirmation. This additional step does not change the main link of sending control commands and guiding the execution of local anti-corrosion repair commands, but rather makes the external execution process and prediction results have a more stable engineering connection. Furthermore, if the heat map shows that the risk area is too scattered, making it impossible to cover in a single flight, the system will split it into multiple sub-task packages according to priority and spatial proximity; if the drone's current battery is insufficient to complete the highest priority operation, only the inspection and verification task will be issued, and the painting task will not be issued. If there is a discrepancy between the heat map and the drone's real-time visual positioning, such as if the tower's attitude recognition error exceeds the limit, the system will automatically switch to manual confirmation mode to prevent accidental or missed spraying. If external terminal communication is interrupted, the control commands will be retained in the cloud task queue and resent after the link is restored. For example, during the special response to Typhoon No. 7, the three-dimensional heat map generated by the system showed a continuous red risk zone at a depth of 35 to 38 meters on the east side and about 10° east of due south. The cloud encapsulates the area into a high-priority task package and sends it to the inspection drone; after the drone arrives at the site, it locks the target area based on the heat map, first takes high-definition images for verification, and then performs local anti-corrosion spraying, thereby avoiding indiscriminate treatment on the entire tower. Furthermore, to maintain terminology consistency, the control instructions are not independent risk objects generated separately from the heatmap, but rather task data packets that are protocol-encapsulated by taking the spatial coordinates, color levels, job priorities, and job types of high-risk areas in the three-dimensional visualized risk heatmap. Therefore, the risk is represented by a three-dimensional visual heatmap at the cloud display level and by control commands at the external terminal communication level. The two correspond to different carriers of the same risk result and do not constitute a new evaluation caliber. The coordinate parameters contained in the control commands maintain a one-to-one mapping relationship with the voxel indexes in the three-dimensional spatiotemporal distribution matrix, ensuring that the UAV positioning accuracy and cloud prediction accuracy are aligned in physical space. After receiving the data, the external drone terminal interprets the height range, circumferential angle range, and work zone boundary according to the same spatial positioning index, avoiding ambiguity with the internal voxel coordinates in the three-dimensional penetration field data; The purpose of this step is to directly transform cloud computing results into spatial tasks that can be executed by external devices, thereby achieving precise guidance and efficiency improvement in the local anti-corrosion repair of mixed towers along the coast.

[0037] In a preferred embodiment of the present invention, before the heterogeneous data receiving unit receives data, the coastal environmental data, the dynamic load data, and the lightweight concrete body data are preprocessed by the edge gateway. The edge gateway is used to: compress the collected coastal environmental data, the dynamic load data, and the lightweight concrete body data based on the Huffman coding algorithm; and transmit the compressed data to the heterogeneous data receiving unit through the asynchronous channel and the real-time channel respectively based on the block transmission strategy.

[0038] This embodiment provides an edge-side compression and chunked transmission mechanism. Specifically, the aforementioned preprocessing scheme assumes that the cloud can stably receive raw or quasi-raw data. However, in coastal wind farms, the connection between the tower base and the cloud is often affected by network jitter, especially during typhoons when high-frequency vibration traffic increases dramatically. If all data is transmitted as is, it can easily cause real-time link congestion. Therefore, this embodiment adds compression and chunked transmission steps on the edge gateway side. Specifically, the edge gateway first performs Huffman coding compression on the original signal; the focus here is not on the complex coding details, but on compressing the high-repetition monitoring sequence before transmission. A simplified demonstration can be performed: the vibration state label sequence within a 10-second window is AAAABBAACC, where A represents low-amplitude stability, B represents medium-amplitude fluctuation, and C represents high-amplitude fluctuation. If direct transmission is required, 10 status codes need to be sent. After using Huffman coding, the shortest codeword can be assigned to the most frequently occurring A, and the longer codewords can be assigned to B and C, thereby reducing the total number of bits. For continuous numerical data, similar compression processing can also be performed after quantization and binning. Environmental data and material data have lower update frequencies and higher repetition rates, so the compression benefits are more obvious. Edge gateways perform chunked transmission based on data timeliness; dynamic payload data is divided into smaller real-time blocks, such as one block every 10 seconds, and uploaded quickly through the real-time channel; environmental data and material body data can be uploaded in larger blocks, such as one block every 30 minutes or one block per hour, through the asynchronous channel. For example, the vibration flow from 9:00 to 9:10 can be divided into 60 real-time blocks, which are sent to the cloud sequentially; environmental data within the same hour is divided into only 1 asynchronous block; this ensures the real-time performance of high-frequency load data while avoiding low-frequency data from occupying real-time link resources. Furthermore, if the verification of the data block after Huffman compression fails, the edge gateway will trigger re-encoding and retransmission of the block; if the real-time channel is temporarily congested, the latest blocks will be retained first and some low-priority historical blocks will be discarded, while the missing interval will be marked in the block header for the cloud to fill in according to the rules. If the asynchronous channel is interrupted, the environment block and material block will be cached and retransmitted in batches after the link is restored. If a block is out of order during transmission, the cloud will reorder it according to the block number. Blocks that are not assembled after the waiting time limit will be marked as incomplete blocks and go through the degradation process. For example, on the eve of Typhoon No. 7 making landfall, the sampling frequency of the tower vibration sensor was increased to 100Hz, and the edge gateway compressed 10 consecutive seconds of raw vibration data into a real-time transmission block, with the average size compressed from the original 100 units to about 55 units; the temperature, humidity, salt spray and chloride ion data of the environmental sensor were aggregated into an asynchronous block per hour; even if the wind farm backhaul link was interfered with, the real-time channel could prioritize the transmission of load data and support timely updates of risk status in the cloud. The purpose of this step is to first complete data slimming and transmission offloading at the edge, so as to achieve real-time delivery of high-frequency payload data and stable supplementation of low-frequency environmental data, thereby reducing the cloud receiving pressure under the complex network conditions along the coast.

[0039] In a preferred embodiment of the present invention, the system further includes a model update module, which is configured to: receive real cross-sectional loss data fed back during the actual anti-corrosion maintenance process input by an external terminal, and compare the real cross-sectional loss data with the cross-sectional dynamic loss rate to calculate the comparison error; if the comparison error exceeds a preset error threshold, fine-tune the parameters of the physical information neural network model using the real cross-sectional loss data; and maintain the current parameters of the physical information neural network model if the comparison error is equal to or less than the preset error threshold.

[0040] This embodiment provides a model update step based on on-site feedback; Specifically, after the aforementioned full-link operation, the system is able to continuously output the dynamic loss rate of steel bar sections and life prediction results. However, if the actual maintenance results are not used for back-calibration for a long time, the model may gradually deviate from the true state due to material aging, construction differences, or sensor drift. Therefore, this embodiment introduces a maintenance feedback-driven closed-loop update mechanism. Specifically, the model update module receives actual cross-sectional loss data after actual anti-corrosion maintenance or grooving verification; for example, the system previously predicted a dynamic loss rate of 4.8% for the steel reinforcement cross-section in a certain area, but the actual loss found during on-site dismantling and testing was 5.5%. The comparison error can be calculated as the difference or relative error, for example, the difference is 0.7 percentage points; if the preset error threshold is 0.5 percentage points, then the sample is judged as an over-threshold sample and needs to participate in parameter fine-tuning; if the prediction at another position is 2.1% and the actual measurement is 2.3%, the difference is only 0.2 percentage points, then the model is considered to maintain acceptable accuracy at that position; During the fine-tuning process, the system is not fully retrained. Instead, it prioritizes using real samples related to the tower, the season, and the material state to make small-step parameter corrections to the physical information neural network. For ease of understanding, a batch of maintenance feedback can be assumed to contain three location samples: predicted values ​​[4.8, 3.2, 2.1] and measured values ​​[5.5, 3.0, 2.3]. If the system identifies the first group as exceeding the threshold and the latter two groups as not exceeding the threshold, the first group is used as the key fine-tuning sample, while retaining its corresponding environmental and load context, so as to improve the model's sensitivity to pore disturbance under similar high humidity and high vibration conditions; after fine-tuning, the retained samples are used to verify whether the error before and after the update has decreased. Furthermore, considering that the real cross-sectional loss data is located at the end of the entire simulation link, and that the error source may come from the physical information neural network, or from the subsequent reduced-order penetration simulation, cross-sectional loss mapping, or field measurement deviation, the system first performs an error attribution screening before performing parameter fine-tuning. If the samples exceeding the threshold simultaneously exhibit consistent characteristics of low upstream micro-strain prediction, slow porosity update, and downstream permeability and bearing capacity mapping parameters within the normal range, then the error is mainly attributed to insufficient physical information neural network representation, and parameter fine-tuning is initiated. Specifically, in order to solve the gradient breakage problem of physical constraints in backpropagation, considering that the physical mapping link from micro-strain calculation to final cross-sectional loss quantification contains non-differentiable operations, the parameter fine-tuning adopts a gradient-free optimization algorithm, such as Bayesian optimization or genetic algorithm. The parameter fine-tuning uses a gradient-based optimization algorithm to fine-tune the weight parameters of the physical information neural network, and combines a gradient-free optimization algorithm to inversely optimize the empirical constants in the preset fatigue model, thereby completing the end-to-end parameter fine-tuning. If the error is more likely to originate from the offset of the on-site slotting position, inaccurate positioning of the rebar, or the lack of necessary correction in the downstream quantification process, the main parameters of the physical information neural network will not be adjusted immediately. Instead, the sample will be marked as a sample to be reviewed. In this way, errors not caused by the front-end strain extraction can be avoided from being incorrectly fed back into the physical information neural network. Furthermore, before being used for fine-tuning, the actual cross-sectional loss data can be registered with the time window corresponding to when the predicted value was generated; The system prioritizes extracting environmental data, dynamic load data, porosity update trajectory, and three-dimensional permeability field trajectory of the maintenance point within a preset traceability period before trenching as supporting context, rather than using only a single final value for updating; in this way, the model learns not only the result of the final loss being too large or too small, but also the working condition evolution path that led to the result, thus making the fine-tuning closer to the actual engineering situation. Furthermore, if the number of on-site feedback samples is too small, for example, only a single sample and the sampling conditions are incomplete, the system will not immediately update the main model, but will first put it into the confirmation queue. If the measured data returned by different maintenance teams conflict with each other, the set of data with the more complete detection process and the time closer to the slot opening time shall be used as the standard, and the remaining data shall be archived as abnormal samples; if the verification error increases after fine-tuning, the parameters before the update shall be rolled back to avoid erroneous samples from contaminating the model; if the comparison error is always within the threshold, the current parameters shall be kept unchanged, and only the accumulated samples shall be used for subsequent periodic evaluation. For example, during the partial anti-corrosion repair of the No. 7 hybrid tower, the maintenance personnel cut a groove 36 meters on the east side for inspection and measured the actual loss rate of the steel reinforcement section to be 5.4%, while the system's previous prediction was 4.6%. Since the error exceeded the set threshold, the model update module extracted environmental, vibration and material moisture content data for the past two months at that location and fine-tuned the physical information neural network. In the verification of adjacent areas, the updated model reduced the prediction error from 0.8 percentage points to 0.3 percentage points. The purpose of this step is to back-inject the actual damage results from maintenance into the cloud model, thereby enabling the coastal mixed tower corrosion prediction system to maintain continuous calibration and accuracy during long-term operation.

[0041] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A cloud-based data processing system for corrosion-resistant lightweight concrete composite towers used in coastal environmental protection projects, characterized in that: The system includes: a data preprocessing module, used to receive coastal environmental data including surface chloride ion concentration and salt spray deposition rate, dynamic load data, and lightweight concrete body data of the coastal environmental protection hybrid tower including initial porosity, through an edge gateway, and perform interpolation and smoothing processing on the dynamic load data and the lightweight concrete body data to generate a standard coupled data stream; and to construct a spatial voxel mesh of the lightweight concrete of the coastal environmental protection hybrid tower based on a preset three-dimensional geometric model; The nonlinear conversion calculation module is used to input the standard coupled data stream into the physical information neural network model, calculate the structural microstrain caused by the dynamic load data, and calculate the local porosity change value corresponding to the spatial voxel grid based on the structural microstrain and the initial porosity in the lightweight concrete body data, so as to generate a dynamic equivalent diffusion coefficient. The spatiotemporal evolution deduction module is used to input the dynamic equivalent diffusion coefficient into a preset reduced-order model based on intrinsic orthogonal decomposition, deduce the three-dimensional permeation field data, and calculate the cross-sectional dynamic loss rate of the steel reinforcement structure inside the lightweight concrete of the coastal environmental protection mixed tower based on the three-dimensional permeation field data. The risk quantification and decision-making module is used to calculate the structural safety index and the predicted value of remaining healthy life based on the dynamic loss rate of the cross section and the dynamic load data, and to generate a dynamic operation and maintenance strategy based on the predicted value of remaining healthy life.

2. The cloud processing system for corrosion protection data of lightweight concrete composite towers for coastal environmental protection as described in claim 1, characterized in that, The data preprocessing module includes: a heterogeneous data receiving unit, used to receive the coastal environmental data and the lightweight concrete body data through a preset asynchronous channel, and to receive the dynamic load data through a preset real-time channel; A time alignment unit is used to interpolate the coastal environmental data collected at a first sampling rate based on the preset time micro-element, and to smooth downsample the dynamic load data collected at a second sampling rate, thereby aligning the coastal environmental data and the dynamic load data in the time dimension, wherein the second sampling rate is greater than the first sampling rate; a noise filtering unit is used to identify and filter out abnormal noise points in the aligned data to output the standard coupled data stream.

3. The cloud processing system for corrosion protection data of lightweight concrete composite towers for coastal environmental protection as described in claim 1, characterized in that, The nonlinear conversion calculation module includes: a strain extraction unit, used to extract dynamic vibration characteristics from the dynamic load data through the physical information neural network model, and calculate the micro-strain of the structure; A porosity update unit is used to calculate the nonlinear disturbance of the initial porosity in the lightweight concrete body data by the micro-strain of the structure based on a preset fatigue model, so as to obtain the local porosity change value; a dynamic equivalent diffusion coefficient generation unit is used to calculate and output the dynamic equivalent diffusion coefficient based on the local porosity change value and a preset diffusion law model.

4. The cloud processing system for corrosion protection data of lightweight concrete composite towers for coastal environmental protection as described in claim 1, characterized in that, The spatiotemporal evolution deduction module includes: a permeation field calculation unit, which uses the dynamic equivalent diffusion coefficient as the input of the preset order reduction model based on intrinsic orthogonal decomposition to calculate the permeation path of chloride ions corresponding to the surface chloride ion concentration in the coastal environmental data inside the lightweight concrete, so as to generate the three-dimensional permeation field data. The arrival time prediction unit is used to calculate the penetration time of chloride ions to the surface of the steel structure based on the three-dimensional permeation field data; the loss rate quantification unit is used to calculate the dynamic loss rate of the cross section of the steel structure within a time period consisting of the sum of multiple preset time micro-elements after the penetration time.

5. The cloud processing system for corrosion protection data of lightweight concrete composite towers for coastal environmental protection as described in claim 1, characterized in that, The risk quantification and decision-making module includes: a safety assessment unit, used to convert the dynamic loss rate of the cross section into the remaining bearing capacity based on a preset mechanical mapping model, extract the frequency shift characteristics in the dynamic load data to determine the stiffness correction coefficient, and perform fatigue damage assessment on the dynamic load data based on the rainflow counting method to determine the residual strength; and to calculate the structural safety index by combining the remaining bearing capacity and the maximum load peak value, and combining the stiffness correction coefficient and the residual strength. The lifespan prediction unit is used to calculate the predicted value of the remaining healthy lifespan based on the gradient of the structural safety index over time. The strategy generation unit is used to generate a first dynamic operation and maintenance strategy containing local anti-corrosion repair instructions when the remaining health life prediction value is lower than a preset life threshold; and to generate a second dynamic operation and maintenance strategy containing regular monitoring instructions when the remaining health life prediction value is equal to or higher than the preset life threshold.

6. The cloud processing system for corrosion protection data of lightweight concrete composite towers for coastal environmental protection as described in claim 1, characterized in that, The coastal environmental data also includes temperature and humidity data; The dynamic load data includes wind speed and direction data, multi-level vibration frequency data, and strain gauge data; the lightweight concrete body data also includes moisture content.

7. The cloud processing system for corrosion protection data of lightweight concrete composite towers for coastal environmental protection as described in claim 1, characterized in that, The three-dimensional permeation field data is the three-dimensional spatiotemporal distribution matrix of chloride ions inside the lightweight concrete.

8. The cloud processing system for corrosion protection data of lightweight concrete composite towers for coastal environmental protection as described in claim 5, characterized in that, The strategy generation unit is also used to: generate a three-dimensional visualized risk heat map based on the three-dimensional permeation field data and the cross-sectional dynamic loss rate; The three-dimensional visualized risk heat map is encapsulated into control commands, and the control commands are sent to an external drone terminal to guide the external drone terminal to execute the local anti-corrosion repair commands.

9. The cloud processing system for corrosion protection data of lightweight concrete composite towers for coastal environmental protection as described in claim 2, characterized in that, Before receiving data, the heterogeneous data receiving unit preprocesses the coastal environmental data, the dynamic load data, and the lightweight concrete body data by the edge gateway. The edge gateway is used to compress the collected coastal environmental data, the dynamic load data, and the lightweight concrete body data based on the Huffman coding algorithm. Based on the block transmission strategy, the compressed data is transmitted to the heterogeneous data receiving unit through the asynchronous channel and the real-time channel respectively.

10. The cloud processing system for corrosion protection data of lightweight concrete composite towers for coastal environmental protection as described in claim 1, characterized in that, The system also includes a model update module, which is used to: receive real cross-sectional loss data fed back during the actual anti-corrosion maintenance process from an external terminal, and compare the real cross-sectional loss data with the cross-sectional dynamic loss rate to calculate the comparison error; If the absolute value of the comparison error is greater than a preset error threshold, the parameters of the physical information neural network model are fine-tuned using the real cross-sectional loss data. If the absolute value of the comparison error is equal to or less than the preset error threshold, the current parameters of the physical information neural network model are maintained.

Images