A concrete pole corrosion intelligent evaluation control system based on big data analysis

Abstract

The application discloses a kind of concrete electric pole corrosion intelligent evaluation control systems based on big data analysis, comprising: multi-frequency magnetic flux density acquisition module, for collecting multi-frequency magnetic flux density detection data and preprocessing;Steel flow alignment input construction module, for constructing two-dimensional heat map and generating steel flow coordinate;Align multi-frequency heat map generation module, for performing steel flow unfolding and coordinate alignment;Implicit corrosion field and background field generation module;Cross-frequency reconstruction and observation residual calculation module, for generating cross-frequency reconstruction heat map and calculating residual;Observation consistent closed loop update and evaluation module, for updating steel implicit corrosion field and residual, generate evaluation result;Detection control instruction generation module, for generating detection control instruction to form corrosion intelligent evaluation control closed loop.The application adopts multi-frequency magnetic flux flow alignment and cross-frequency closed loop inversion, evaluates corrosion and self-adapting complex scanning frequency conversion, accurate and efficient.

Description

Technical Field

[0001] This invention relates to the field of pole corrosion assessment technology, and in particular to an intelligent assessment and control system for concrete pole corrosion based on big data analysis. Background Technology

[0002] Current methods for detecting steel reinforcement corrosion on concrete utility poles primarily employ non-destructive testing techniques such as magnetic leakage and magnetic flux density, combined with manual inspections or single-frequency scanning to acquire surface response signals. The sampled values ​​are then mapped to two-dimensional heatmaps or curves for threshold discrimination and empirical grading. Some solutions incorporate data analysis models to extract features from the heatmaps, enabling the location and severity assessment of corrosion areas. While these techniques are easy to deploy in engineering projects, they typically rely on fixed frequencies and sampling trajectories, and the detection results are significantly affected by slowly varying factors such as probe fit, protective layer thickness, and background magnetization differences.

[0003] The aforementioned existing technologies generally lack "manifold alignment" processing for the direction and spacing of rebar, making it difficult to compare heat maps from different pole sections and different scanning batches in a unified coordinate system. Furthermore, the lack of cross-frequency differentiable observation modeling and a closed-loop update mechanism for consistent observations makes it difficult to decouple multi-frequency response differences from background field changes, leading to false alarms, missed alarms, or unstable detection levels. More importantly, most systems fail to form an "assessment-control" closed loop, unable to adaptively trigger rescanning, frequency switching, and encrypted sampling based on cross-frequency residuals and risk heat maps. This results in insufficient evidence in high-risk areas and a difficulty in balancing detection costs and efficiency.

[0004] Therefore, how to provide an intelligent assessment and control system for the corrosion of concrete poles based on big data analysis is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose an intelligent assessment and control system for corrosion of concrete poles based on big data analysis. This invention employs multi-frequency flux manifold alignment and cross-frequency closed-loop inversion to assess corrosion and adaptively rescan and change frequencies, which is accurate and efficient.

[0006] According to an embodiment of the present invention, a smart assessment and control system for corrosion of concrete utility poles based on big data analysis includes: The multi-frequency magnetic flux density acquisition module is used to acquire and preprocess multi-frequency magnetic flux density detection data on the surface of concrete poles. The reinforcement manifold alignment input construction module is used to construct a set of two-dimensional heat maps of multi-frequency magnetic flux density and generate the corresponding reinforcement manifold coordinates; The aligned multi-frequency heat map generation module is used to generate the rebar direction field, rebar spacing field and coordinate confidence field. It performs rebar manifold expansion and coordinate alignment on the multi-frequency magnetic flux density two-dimensional heat map set to obtain the aligned multi-frequency heat map set. The implicit corrosion field and background field generation module is used to generate the initial value of the implicit corrosion field of the steel bars and the set of proxy quantities of the background field; The cross-frequency reconstruction and observation residual calculation module is used to generate a set of cross-frequency reconstruction heatmaps and calculate a set of cross-frequency observation residuals. The observation consistency closed-loop update and evaluation module is used to update the initial value of the implicit corrosion field of steel bars and the cross-frequency observation residual set, and generate corrosion risk heat map and corrosion level assessment results; The detection control command generation module is used to generate detection control commands, trigger rescanning, frequency switching and encrypted sampling, forming a closed loop of intelligent corrosion assessment and control.

[0007] Optionally, modules can be integrated using the following methods: Multi-frequency magnetic flux density detection data were collected from the surface of concrete poles, and preprocessing was performed to obtain a standardized dataset of multi-frequency magnetic flux density. A two-dimensional heat map set of multi-frequency magnetic flux density is constructed based on the standardized multi-frequency magnetic flux density dataset, and the corresponding steel bar manifold coordinates are generated to obtain the steel bar manifold aligned input set; Input the reinforcement manifold alignment input set into the reinforcement manifold coordinate generator, output the reinforcement direction field, reinforcement spacing field and coordinate confidence field, perform reinforcement manifold unfolding and coordinate alignment on the multi-frequency magnetic flux density two-dimensional heat map set to obtain the aligned multi-frequency heat map set; Input the aligned multi-frequency heat map set into the implicit corrosion field representation device to generate the initial value of the implicit corrosion field of the steel bar and the set of background field proxy quantities; The initial value of the implicit corrosion field of steel bars and the set of background field proxy quantities are input into the cross-frequency differentiable observation operator layer to generate a set of cross-frequency reconstruction heatmaps and calculate the set of cross-frequency observation residuals. The cross-frequency observation residual set is input into the observation consistency closed-loop solver to update the initial value of the implicit corrosion field of steel bars and the cross-frequency observation residual set. Based on the updated implicit corrosion field of steel bars, a corrosion risk heat map and corrosion level assessment results are generated. Based on the corrosion risk heat map, corrosion level assessment results and cross-frequency observation residual set, detection and control commands are generated to trigger rescanning, frequency switching and encrypted sampling, forming a closed loop of intelligent corrosion assessment and control.

[0008] Optionally, the generation of the multi-frequency flux density normalized dataset specifically includes: Multi-frequency excitation scanning was performed on the surface of a concrete pole to obtain a set of raw multi-frequency magnetic flux density data. Sensor calibration conversion and zero-point drift correction are performed on the original multi-frequency magnetic flux density data set to obtain the multi-frequency magnetic flux density correction data set; Perform sampling timing alignment and spatial sampling point alignment on the multi-frequency magnetic flux density correction dataset to obtain a multi-frequency magnetic flux density aligned dataset; Anomaly removal is performed on the multi-frequency flux density aligned dataset to obtain a cleaned multi-frequency flux density dataset and anomaly mask dataset. Noise suppression and amplitude uniformity processing are performed on the multi-frequency magnetic flux density cleanup dataset to obtain a multi-frequency magnetic flux density denoised dataset. Perform normalization processing on the multi-frequency magnetic flux density denoised dataset and output a normalized multi-frequency magnetic flux density dataset.

[0009] Optionally, the generation of the reinforcement manifold alignment input set specifically includes: A set of two-dimensional heat map raster mapping rules is obtained based on a multi-frequency magnetic flux density standardized dataset. Based on the set of two-dimensional heat map grid mapping rules, the multi-frequency magnetic flux density standardized data set is written into the two-dimensional heat map grid according to frequency to obtain the multi-frequency magnetic flux density two-dimensional heat map set; The abnormal mask set is propagated to the multi-frequency magnetic flux density two-dimensional heat map set according to the two-dimensional heat map grid mapping rule set to generate a two-dimensional heat map mask set. Hole filling and connectivity repair are performed to obtain a consistent heat map set of steel reinforcement manifold mask. Based on the set of two-dimensional heat map grid mapping rules, the steel manifold coordinates corresponding to each point of the two-dimensional heat map set of multi-frequency magnetic flux density are generated for each grid position; The set of two-dimensional heatmaps of multi-frequency magnetic flux density, the coordinates of the reinforcing bar manifold, and the set of heatmaps of the reinforcing bar manifold mask are encapsulated according to frequency and grid position to form a set of reinforcing bar manifold aligned inputs.

[0010] Optionally, the generation of the aligned multi-frequency heatmap set specifically includes: Input the reinforcement manifold alignment input set into the input terminal of the reinforcement manifold coordinate generator to obtain the coordinate generation input tensor; Input the coordinate generation input tensor into the rebar manifold coordinate generator, and output the rebar direction field, rebar spacing field, and coordinate confidence field; Generate alignment and culling masks based on coordinate confidence fields; Perform reinforcement manifold expansion and coordinate alignment on the multi-frequency magnetic flux density two-dimensional heat map set to obtain the reinforcement manifold expansion coordinate transformation field; Based on the coordinate transformation field of the steel bar manifold expansion, bilinear interpolation resampling and alignment are performed on the multi-frequency magnetic flux density two-dimensional heat map set frequency by frequency to obtain the aligned multi-frequency heat map set, and the alignment mask set is generated simultaneously.

[0011] Optionally, the generation of the initial value of the implicit corrosion field of the reinforcing steel bars and the set of proxy quantities of the background field specifically includes: By inputting the aligned multi-frequency heat map set and the aligned mask set into the input terminal of the implicit corrosion field representation device, the implicit field encoding input set is obtained. A cross-frequency feature coding backbone network with shared parameters is used to perform cross-frequency consistent feature coding on the implicit field coding input set, resulting in a cross-frequency consistent feature map set and a cross-frequency consistent global description vector set; Based on the cross-frequency consistent feature map set and the coordinate transformation field of the rebar manifold expansion, the axial sampling band set and the normal sampling band set of the rebar are generated. Under the constraint of the alignment mask set, the cross-frequency consistent feature map set in the sampling band is aggregated within the band to obtain the axial feature sequence set and the normal feature sequence set of the rebar. The set of axial feature sequences of reinforcing bars and the set of cross-frequency consistent global description vectors are input into the implicit corrosion field initial value generation unit of the implicit corrosion field representation device to generate the initial value of the implicit corrosion field of the reinforcing bars. The set of steel reinforcement normal feature sequences and the set of cross-frequency consistent global description vectors are input into the background field surrogate quantity generation unit of the implicit corrosion field representation device to generate the set of background field surrogate quantities.

[0012] Optionally, the generation of the cross-frequency observation residual set specifically includes: The initial value of the implicit corrosion field of steel bars and the set of background field proxy quantities are input into the cross-frequency differentiable observation operator layer. The set of aligned multi-frequency heat maps and the set of aligned masks are input into the cross-frequency differentiable observation operator layer as observation references and effective region constraints to obtain the cross-frequency observation input set. Within the cross-frequency differentiable observable operator layer, the axial corrosion excitation distribution of steel bars is constructed based on the initial value of the implicit corrosion field of steel bars, and the slowly varying background response distribution is constructed based on the background field surrogate quantity set. Cross-frequency coupling mapping is performed according to the frequency dimension to obtain the cross-frequency predicted magnetic flux response set. The cross-frequency predicted flux response set is mapped to the same grid coordinate system and amplitude domain as the aligned multi-frequency heat map set through the heat map imaging uniformity unit of the cross-frequency differentiable observable operator layer, thus obtaining the cross-frequency reconstructed heat map set. Point-by-point differencing is performed on the cross-frequency reconstructed heatmap set and the aligned multi-frequency heatmap set within the effective area defined by the aligned mask set to generate cross-frequency observation residuals. Cross-frequency consistent residual shaping is then performed to suppress occasional spike residuals in anomalous frequency channels, resulting in a shaped cross-frequency observation residual set.

[0013] Optionally, the generation of the corrosion risk heat map and corrosion level assessment results specifically includes: The set of shaped cross-frequency observation residuals, the set of cross-frequency reconstructed heatmaps, the set of aligned multi-frequency heatmaps, and the set of aligned masks are input into the observation-consistent closed-loop solver. The initial values ​​of the implicit corrosion field of the steel bars and the set of surrogate quantities of the background field are initialized as closed-loop solution states, and the set of initial closed-loop solution states is obtained. Construct observation-consistent target quantities and cross-frequency consistent target quantities based on closed-loop solution of the initial state set; Based on the observation-consistent target quantity and the cross-frequency consistent target quantity, the initial value of the implicit corrosion field of steel bars is updated by differentiable backpropagation through the observation-consistent closed-loop solver; Write the updated set of candidate steel bar implicit corrosion field values ​​back to the initial value of the steel bar implicit corrosion field to form the updated steel bar implicit corrosion field, and regenerate the cross-frequency reconstruction heat map set to calculate the closed-loop residual update state set; Convergence determination is performed based on updating the state set using closed-loop residuals; When the convergence criterion is met, the updated implicit corrosion field of the steel bars is output, and a corrosion risk heat map is generated in the grid coordinate system corresponding to the coordinate transformation field of the steel bar manifold expansion. Corrosion level assessment results are generated based on the corrosion risk heat map.

[0014] Optionally, the generation of the intelligent corrosion assessment and control closed loop specifically includes: The corrosion risk heat map, corrosion level assessment results and shaping cross-frequency observation residual set are aligned and encapsulated to obtain the control trigger input set; Generate a set of control trigger heatmaps based on the set of control trigger inputs; Generate a set of target regions for rescanning and a set of rescanning priorities based on the set of control trigger heatmaps; A set of frequency switching schemes is generated based on the shaped cross-frequency observation residual set; A set of encrypted sampling schemes is generated based on the set of target regions for repeated scanning and the set of frequency switching schemes. The set of rescan trajectory segments, the set of frequency switching schemes, and the set of encrypted sampling schemes are encapsulated into detection control commands and written back to the multi-frequency magnetic flux density detection data acquisition process. The system executes detection and control commands to trigger rescanning, frequency switching, and encrypted sampling, and collects execution feedback to form a closed-loop reflux data set, thereby realizing a closed-loop intelligent assessment and control system for corrosion.

[0015] The beneficial effects of this invention are: This invention unifies multi-frequency thermal images from different pole segments and scanning batches into a manifold coordinate system consistent with the direction / spacing of the reinforcing bars by acquiring and standardizing multi-frequency magnetic flux density and generating and aligning the manifold coordinates of the reinforcing bars. This enables comparable input across positions and frequencies. Furthermore, it introduces a background field proxy using an implicit corrosion field representation and generates a cross-frequency reconstructed thermal image and observation residuals through a cross-frequency differentiable observation operator layer. The implicit corrosion field is then updated and converged in a closed-loop solver that ensures consistent observation. This effectively shields invalid regions and suppresses abnormal frequency spike residuals, thereby improving the stability and accuracy of corrosion risk thermal images and corrosion level assessment results. It also reduces misjudgments and level drift caused by probe fit variations, protective layer thickness fluctuations, and differences in material background magnetization.

[0016] Meanwhile, this invention couples risk heatmaps, level assessments, and cross-frequency residuals as control trigger inputs, automatically generating rescan target areas, frequency switching schemes, and encrypted sampling trajectories, and writing them back to the acquisition process. This forms a closed-loop adaptive detection mechanism of "assessment-control-reacquisition-reassessment," enabling the system to dynamically focus sampling resources on high-risk and cross-frequency inconsistent areas. Without significantly increasing the overall detection time, it improves the evidence density and verification efficiency in key areas, thereby achieving earlier identification and more reliable graded assessment of corrosion evolution, balancing detection efficiency and engineering feasibility. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 The flowchart shows a smart assessment and control system for corrosion of concrete poles based on big data analysis proposed in this invention. Figure 2 This invention presents a flowchart of the multi-frequency magnetic flux density standardization and two-dimensional heat map construction process for an intelligent assessment and control system for corrosion of concrete poles based on big data analysis. Figure 3 This is a flowchart illustrating the inversion update and adaptive detection control closed-loop process of an intelligent assessment and control system for concrete pole corrosion based on big data analysis proposed in this invention. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0019] refer to Figures 1-3 A smart assessment and control system for corrosion of concrete utility poles based on big data analysis includes: The multi-frequency magnetic flux density acquisition module is used to acquire and preprocess multi-frequency magnetic flux density detection data on the surface of concrete poles. The reinforcement manifold alignment input construction module is used to construct a set of two-dimensional heat maps of multi-frequency magnetic flux density and generate the corresponding reinforcement manifold coordinates; The aligned multi-frequency heat map generation module is used to generate the rebar direction field, rebar spacing field and coordinate confidence field. It performs rebar manifold expansion and coordinate alignment on the multi-frequency magnetic flux density two-dimensional heat map set to obtain the aligned multi-frequency heat map set. The implicit corrosion field and background field generation module is used to generate the initial value of the implicit corrosion field of the steel bars and the set of proxy quantities of the background field; The cross-frequency reconstruction and observation residual calculation module is used to generate a set of cross-frequency reconstruction heatmaps and calculate a set of cross-frequency observation residuals. The observation consistency closed-loop update and evaluation module is used to update the initial value of the implicit corrosion field of steel bars and the cross-frequency observation residual set, and generate corrosion risk heat map and corrosion level assessment results; The detection control command generation module is used to generate detection control commands, trigger rescanning, frequency switching and encrypted sampling, forming a closed loop of intelligent corrosion assessment and control.

[0020] In this embodiment, the modules are interconnected using the following method: Multi-frequency magnetic flux density detection data were collected from the surface of concrete poles, and preprocessing was performed to obtain a standardized dataset of multi-frequency magnetic flux density. A two-dimensional heat map set of multi-frequency magnetic flux density is constructed based on the standardized multi-frequency magnetic flux density dataset, and the corresponding steel bar manifold coordinates are generated to obtain the steel bar manifold aligned input set; Input the reinforcement manifold alignment input set into the reinforcement manifold coordinate generator, output the reinforcement direction field, reinforcement spacing field and coordinate confidence field, perform reinforcement manifold unfolding and coordinate alignment on the multi-frequency magnetic flux density two-dimensional heat map set to obtain the aligned multi-frequency heat map set; Input the aligned multi-frequency heat map set into the implicit corrosion field representation device to generate the initial value of the implicit corrosion field of the steel bar and the set of background field proxy quantities; The initial value of the implicit corrosion field of steel bars and the set of background field proxy quantities are input into the cross-frequency differentiable observation operator layer to generate a set of cross-frequency reconstruction heatmaps and calculate the set of cross-frequency observation residuals. The cross-frequency observation residual set is input into the observation consistency closed-loop solver to update the initial value of the implicit corrosion field of steel bars and the cross-frequency observation residual set. Based on the updated implicit corrosion field of steel bars, a corrosion risk heat map and corrosion level assessment results are generated. Based on the corrosion risk heat map, corrosion level assessment results and cross-frequency observation residual set, detection and control commands are generated to trigger rescanning, frequency switching and encrypted sampling, forming a closed loop of intelligent corrosion assessment and control.

[0021] In this embodiment, the generation of the multi-frequency magnetic flux density standardized dataset specifically includes: Multi-frequency excitation scanning was performed on the surface of a concrete pole to obtain a set of raw multi-frequency magnetic flux density data. The multi-frequency excitation scanning acquisition involves deploying magnetization excitation components and magnetic flux density sensing components on the surface of concrete poles, establishing scanning trajectories along the height and circumferential directions, and acquiring a set of original multi-frequency magnetic flux density data including height direction position markers, circumferential direction position markers, sampling time markers, and sensor working status markers. Sensor calibration conversion and zero-point drift correction are performed on the original multi-frequency magnetic flux density data set to obtain the multi-frequency magnetic flux density correction data set; The magnetic flux density correction data is obtained by subtracting the zero-point reference voltage value from the sampled voltage value and then dividing by the sensitivity calibration coefficient; Perform sampling timing alignment and spatial sampling point alignment on the multi-frequency magnetic flux density correction dataset to obtain a multi-frequency magnetic flux density aligned dataset; The spatial sampling point alignment is based on the position marks in the height direction and the position marks in the circumferential direction to perform equidistant resampling of sampling points with unequal spacing. The aligned magnetic flux density value at the target equidistant sampling point is obtained by weighted summation of the magnetic flux density correction values ​​of the two original sampling points adjacent to the target sampling point according to linear interpolation weight. Anomaly removal is performed on the multi-frequency flux density aligned dataset to obtain a cleaned multi-frequency flux density dataset and anomaly mask dataset. The anomaly removal process is based on the sensor's working status marking to remove saturated segments, disconnected segments, and poor contact segments. The remaining data is then subjected to pulse interference point detection and replacement. The pulse interference point detection uses the median and the median of the absolute deviation within a sliding window centered on the sampling point to be judged to construct a robust statistic. When the absolute value of the robust statistic exceeds a preset threshold, the sampling point is written into the anomaly mask set and replaced with the median of the window. Noise suppression and amplitude uniformity processing are performed on the multi-frequency magnetic flux density cleanup dataset to obtain a multi-frequency magnetic flux density denoised dataset. The noise suppression uses wavelet thresholding to denoise the data and propagates the set of anomaly masks to the denoised data to maintain data chain consistency. Wavelet thresholding denoising performs soft thresholding shrinkage on wavelet coefficients. After taking the sign of the wavelet coefficients, the absolute value of the coefficients is subtracted from the threshold parameter and the maximum value is taken with zero to obtain the shrinkage amplitude. Perform normalization processing on the multi-frequency magnetic flux density denoised dataset and output a normalized multi-frequency magnetic flux density dataset. The standardization process calculates the central statistic and scale statistic separately according to the frequency, maps the denoised data into a dimensionless standardized representation, and encapsulates the corresponding height direction position marker, circumferential direction position marker and anomaly mask set together. The dimensionless standardized representation is obtained by subtracting the central statistic from the denoised magnetic flux density value and then dividing by the scale statistic.

[0022] In this embodiment, the generation of the reinforcement manifold alignment input set specifically includes: A set of two-dimensional heat map raster mapping rules is obtained based on a multi-frequency magnetic flux density standardized dataset. The set of two-dimensional heat map raster mapping rules is based on the height direction position markers and circumferential direction position markers in the multi-frequency magnetic flux density standardized data set. The height direction raster division rules and circumferential direction raster division rules are determined, and the height direction raster index mapping rules and circumferential direction raster index mapping rules are generated to obtain the set of two-dimensional heat map raster mapping rules. The height direction grid index is obtained by subtracting the height direction start position value from the position value corresponding to the height direction position marker, dividing by the height direction grid interval, and rounding down. The circumferential direction grid index is obtained by subtracting the circumferential direction start position value from the position value corresponding to the circumferential direction position marker, dividing by the circumferential direction grid interval, and rounding down. Based on the set of two-dimensional heat map grid mapping rules, the multi-frequency magnetic flux density standardized data set is written into the two-dimensional heat map grid according to frequency to obtain the multi-frequency magnetic flux density two-dimensional heat map set; The heatmap values ​​of the corresponding frequencies and grid positions in the multi-frequency magnetic flux density two-dimensional heatmap set are obtained by weighting and summing the standardized magnetic flux density values ​​of the sampling point set mapped to the grid according to the validity markers determined by the anomaly mask set, and then dividing by the sum of the validity markers. The validity marker takes a value of one for valid and a value of zero for invalid. The abnormal mask set is propagated to the multi-frequency magnetic flux density two-dimensional heat map set according to the two-dimensional heat map grid mapping rule set to generate a two-dimensional heat map mask set. Hole filling and connectivity repair are performed to obtain a consistent heat map set of steel reinforcement manifold mask. The two-dimensional heat map mask set corresponds point-by-point with the two-dimensional heat map set of multi-frequency magnetic flux density. The mask value at the corresponding grid position is determined by summing the validity markers of the set of sampling points mapped to that grid. When the sum of the validity markers is greater than zero, the mask value is one; otherwise, it is zero. Based on the set of two-dimensional heat map grid mapping rules, the steel manifold coordinates corresponding to each point of the two-dimensional heat map set of multi-frequency magnetic flux density are generated for each grid position; The reinforcement manifold coordinates include height-normalized coordinates, circumferential-normalized coordinates, and circumferential-periodic embedded coordinates, based on the reinforcement manifold mask consistency heatmap set to shield invalid grid positions. The height-direction normalized coordinates are obtained by subtracting the height-direction start position value from the height-direction position value corresponding to the grid, and then dividing by the difference between the height-direction end position value and the height-direction start position value. The circumferential-direction normalized coordinates are obtained by subtracting the circumferential-direction start position value from the circumferential-direction position value corresponding to the grid, and then dividing by the difference between the circumferential-direction end position value and the circumferential-direction start position value. The circumferential-direction periodic embedded coordinates are obtained by inputting the circumferential-direction normalized coordinates into the sine and cosine functions respectively, and multiplying the circumferential-direction normalized coordinates by twice the constant of pi as the independent variable of the function. The set of two-dimensional heatmaps of multi-frequency magnetic flux density, the coordinates of the reinforcing bar manifold, and the set of heatmaps of the reinforcing bar manifold mask are encapsulated according to frequency and grid position to form a set of reinforcing bar manifold aligned inputs.

[0023] In this embodiment, the generation of the aligned multi-frequency heatmap set specifically includes: Input the reinforcement manifold alignment input set into the input terminal of the reinforcement manifold coordinate generator to obtain the coordinate generation input tensor; The input end of the reinforcement manifold coordinate generator aligns the multi-frequency magnetic flux density two-dimensional heat map set in the reinforcement manifold input set with frequency uniformity amplitude range, arranges the reinforcement manifold coordinates with channel alignment, and broadcasts the reinforcement manifold mask uniform heat map set point by point according to grid position; Input the coordinate generation input tensor into the rebar manifold coordinate generator, and output the rebar direction field, rebar spacing field, and coordinate confidence field; The rebar orientation field represents the main direction of the rebar in the two-dimensional heat map grid, the rebar spacing field represents the equivalent spacing of the rebar in the two-dimensional heat map grid, and the coordinate confidence field represents the confidence level of the rebar orientation field and rebar spacing field at each grid position. The direction vector of the steel reinforcement direction field at the corresponding grid position is obtained by normalizing the gradient vectors obtained from the coordinate generation input tensor in the height and circumferential directions. The normalization is to use the square root of the sum of the squares of the components of the gradient vector in the height and circumferential directions and the sum of the stability terms as the normalization factor to normalize the gradient vector. The rebar spacing field is determined by the autocorrelation curve of the local sampling sequence extracted along the rebar normal at the corresponding grid position. The autocorrelation curve is obtained by summing the local sampling sequence and its displacement version point by point. The rebar spacing field is calculated by converting the displacement corresponding to the autocorrelation peak value except for zero displacement. The confidence value of the coordinate confidence field at the corresponding grid position is obtained by weighted summation of the mask values ​​of the directional consistency index, peak significance index and the heat map set of the rebar manifold mask consistency, and then input into a bounded compression function. The directional consistency index is calculated by the directional consistency of the rebar direction field in the local neighborhood, and the peak significance index is calculated by the ratio of the autocorrelation main peak to the secondary peak. Generate alignment and culling masks based on coordinate confidence fields; The participating alignment mask and the culling mask correspond point by point to the grid position of the two-dimensional heat map set of multi-frequency magnetic flux density. The value of the participating alignment mask at the corresponding grid position is determined by comparing the confidence value of the coordinate confidence field at that grid position with the confidence threshold. When the confidence value is not less than the confidence threshold, the participating alignment mask is set to one; otherwise, it is set to zero. The culling mask is obtained by inverting the participating alignment mask. Hole filling and connectivity repair are performed on the participating alignment mask to obtain a continuous participating alignment region. Perform reinforcement manifold expansion and coordinate alignment on the multi-frequency magnetic flux density two-dimensional heat map set to obtain the reinforcement manifold expansion coordinate transformation field; The reinforcement manifold unfolding and coordinate alignment are based on the reinforcement direction field to construct the reinforcement tangential and reinforcement normal at each grid position, and based on the reinforcement spacing field to construct the equivalent spacing benchmark with unified normal scale across the position, generating a coordinate transformation field from the two-dimensional heat map grid coordinates to the reinforcement manifold unfolding coordinates. In the process of generating the coordinate transformation field, the grid position corresponding to the culling mask is shielded to avoid invalid regions from participating in the construction of the coordinate transformation field, thus obtaining the reinforcement manifold unfolding coordinate transformation field. Based on the coordinate transformation field of the steel bar manifold expansion, bilinear interpolation resampling and alignment are performed on the multi-frequency magnetic flux density two-dimensional heat map set frequency by frequency to obtain the aligned multi-frequency heat map set, and the alignment mask set is generated simultaneously. Bilinear interpolation is the interpolation result obtained by weighting and summing the heatmap values ​​of four source grids adjacent to the target's expanded coordinates according to the bilinear interpolation weights. The alignment mask set is the effective region marker of the alignment multi-frequency heatmap set.

[0024] In this embodiment, the generation of the initial value of the implicit corrosion field of the reinforcing steel bars and the set of proxy quantities of the background field specifically includes: By inputting the aligned multi-frequency heat map set and the aligned mask set into the input terminal of the implicit corrosion field representation device, the implicit field encoding input set is obtained. The input of the implicit corrosion field display device will maintain an independent channel for the aligned multi-frequency heat map set according to the frequency dimension, and will broadcast the aligned mask set point by point according to the frequency and grid position, and then align and splice it with the aligned multi-frequency heat map set to obtain the implicit field code input set. The effective area of ​​the implicit field code input set is limited by the aligned mask set. A cross-frequency feature coding backbone network with shared parameters is used to perform cross-frequency consistent feature coding on the implicit field coding input set, resulting in a cross-frequency consistent feature map set and a cross-frequency consistent global description vector set; The cross-frequency feature coding backbone network encodes each frequency channel and performs feature alignment and fusion in the grid position dimension. The cross-frequency consistent feature map set represents the local corrosion response structure of the aligned multi-frequency heat map set, and the cross-frequency consistent global description vector set represents the overall background response and excitation coupling state of the pole segment. Based on the cross-frequency consistent feature map set and the coordinate transformation field of the rebar manifold expansion, the axial sampling band set and the normal sampling band set of the rebar are generated. Under the constraint of the alignment mask set, the cross-frequency consistent feature map set in the sampling band is aggregated within the band to obtain the axial feature sequence set and the normal feature sequence set of the rebar. The axial sampling band set and the normal sampling band set of the reinforcing bars are generated by aligning them along the tangential and normal directions of the reinforcing bars, respectively. The invalid positions of the axial feature sequence set and the normal feature sequence set of the reinforcing bars are shielded by the alignment mask set. The set of axial feature sequences of reinforcing bars and the set of cross-frequency consistent global description vectors are input into the implicit corrosion field initial value generation unit of the implicit corrosion field representation device to generate the initial value of the implicit corrosion field of the reinforcing bars. The implicit corrosion field initial value generation unit performs weighted normalization aggregation on the feature vectors in the adjacent axial neighborhood according to the validity label determined by the alignment mask set in the axial position dimension to obtain the corrosion candidate description vector set for each axial position. After the global description vector set consistent with the cross-frequency is aligned and spliced ​​in the channel dimension, a linear mapping is applied and a bounded compression function is input to obtain the initial value of corrosion intensity of axial control points at each axial position. According to the alignment mask set, the initial value of corrosion intensity of axial control points is gated and shielded to generate the set of initial values ​​of corrosion intensity of effective axial control points. Piecewise cubic interpolation is performed on the set of initial values ​​of corrosion intensity of effective axial control points with adjacent effective axial control points as segment intervals to generate the initial value of corrosion intensity field continuously parameterized along the steel bar axis and skip the invalid label interval. The set of steel bar normal feature sequences and the set of cross-frequency consistent global description vectors are input into the background field surrogate quantity generation unit of the implicit corrosion field representation device to generate the set of background field surrogate quantities. The background field surrogate quantity generation unit performs weighted normalization aggregation on the feature vectors in the adjacent normal neighborhood according to the validity label determined by the alignment mask set in the normal position dimension to obtain the background candidate description vector set for each normal position. After the global description vector consistent with the cross-frequency is aligned and spliced ​​in the channel dimension, a linear mapping is applied to generate the normal background response sequence. Low-frequency constraint filtering is performed to extract the slowly varying background components to obtain the background field surrogate quantity set. The low-frequency constraint filtering is achieved by combining multi-scale sliding window smoothing along the normal of the steel bar with monotonic variation constraint. The masking is performed at the positions marked as invalid in the alignment mask set and filled with the smooth extension value of the adjacent valid positions. The background field proxy set represents the slowly varying background components in the two-dimensional thermal map set of multi-frequency magnetic flux density caused by changes in the protective layer thickness, changes in probe adhesion, and differences in material background magnetization.

[0025] In this embodiment, the generation of the cross-frequency observation residual set specifically includes: The initial value of the implicit corrosion field of steel bars and the set of background field proxy quantities are input into the cross-frequency differentiable observation operator layer. The set of aligned multi-frequency heat maps and the set of aligned masks are input into the cross-frequency differentiable observation operator layer as observation references and effective region constraints to obtain the cross-frequency observation input set. Within the cross-frequency differentiable observable operator layer, the axial corrosion excitation distribution of steel bars is constructed based on the initial value of the implicit corrosion field of steel bars, and the slowly varying background response distribution is constructed based on the background field surrogate quantity set. Cross-frequency coupling mapping is performed according to the frequency dimension to obtain the cross-frequency predicted magnetic flux response set. The axial corrosion excitation distribution of the reinforcing bars is obtained by sampling the initial value of the implicit corrosion field of the reinforcing bars at equidistant intervals at the axial discrete positions corresponding to the coordinate transformation field of the reinforcing bar manifold. The corrosion intensity sequence of the axial control points is obtained. Strip copying and normalization expansion are performed along the reinforcing bar normal within a preset normal expansion width to form a corrosion excitation distribution map. Invalid grid positions are masked based on the alignment mask set. The excitation value of the corrosion excitation distribution map at any grid position is obtained by multiplying the corrosion intensity value of the axial sampling position corresponding to the grid position by the normal expansion weight. The normal expansion weight is calculated by the distance from the normal distance to the center line of the strip through a single-peak attenuation function. The slow-varying background response distribution is formed by creating a normal background response sequence in the normal position dimension of the reinforcement by setting the background field proxy quantity set, performing broadcast replication in the axial dimension of the reinforcement to cover all axial positions of the same reinforcement, forming a slow-varying background response distribution map, and performing masking at the positions marked as invalid by the alignment mask set and filling them with the smooth extension values ​​of the adjacent valid positions. The cross-frequency coupling mapping is achieved by applying a frequency-dependent corrosion response diffusion kernel to the axial corrosion excitation distribution of the steel bars and a frequency-dependent background amplitude modulation kernel to the slowly varying background response distribution. The predicted heatmap of the corresponding frequency in the cross-frequency predicted flux response set is obtained by adding the result of performing a two-dimensional convolution between the corrosion response diffusion kernel of the corresponding frequency and the corrosion excitation distribution map and the result of performing a two-dimensional convolution between the background amplitude modulation kernel of the corresponding frequency and the slowly varying background response distribution map. The cross-frequency predicted flux response set is mapped to the same grid coordinate system and amplitude domain as the aligned multi-frequency heat map set through the heat map imaging uniformity unit of the cross-frequency differentiable observable operator layer, thus obtaining the cross-frequency reconstructed heat map set. The thermal imaging uniformity unit performs amplitude domain calibration, grid alignment resampling, and boundary uniformity under alignment mask set constraints on the cross-frequency predicted flux response set at each frequency, ensuring that the cross-frequency reconstructed thermal image set and the aligned multi-frequency thermal image set can be compared point by point. Point-by-point differencing is performed on the cross-frequency reconstructed heatmap set and the aligned multi-frequency heatmap set within the effective area defined by the aligned mask set to generate cross-frequency observation residuals. Cross-frequency consistent residual shaping is then performed to suppress occasional spike residuals in anomalous frequency channels, resulting in a shaped cross-frequency observation residual set. The residual heatmap of the corresponding frequency in the cross-frequency observation residual is obtained by performing point-by-point masking on the difference between the reconstructed heatmap of the frequency and the observation heatmap using the effective region mask, and then performing point-by-point multiplication between the effective region mask and the difference result.

[0026] In this embodiment, the generation of the corrosion risk heat map and corrosion level assessment results specifically includes: The set of shaped cross-frequency observation residuals, the set of cross-frequency reconstructed heatmaps, the set of aligned multi-frequency heatmaps, and the set of aligned masks are input into the observation-consistent closed-loop solver. The initial values ​​of the implicit corrosion field of the steel bars and the set of surrogate quantities of the background field are initialized as closed-loop solution states, and the set of initial closed-loop solution states is obtained. Construct observation-consistent target quantities and cross-frequency consistent target quantities based on closed-loop solution of the initial state set; The observation consistency target quantity is obtained by summing the absolute values ​​of the residual heatmaps of each frequency channel point by point within the effective area defined by the shaped cross-frequency observation residual set and aggregating them over the entire grid range. The cross-frequency consistency target quantity is obtained by calculating the residual dispersion of each grid position along the frequency channel dimension within the effective area of ​​the shaped cross-frequency observation residual set and aggregating them over the entire grid range. Based on the observation-consistent target quantity and the cross-frequency consistent target quantity, the initial value of the implicit corrosion field of steel bars is updated by differentiable backpropagation through the observation-consistent closed-loop solver; The observation-consistent closed-loop solver calculates the update direction through the differentiable mapping relationship of the initial value of the implicit corrosion field of the steel bars by the cross-frequency differentiable observation operator layer. During the update process, axial continuity constraints and amplitude non-negativity constraints are introduced to obtain the set of candidate steel bar implicit corrosion field update quantities. The axial continuity constraint is constructed by the differential penalty of adjacent positions along the steel bar axis, and the amplitude non-negativity constraint is achieved by performing non-negative truncation on the candidate updated corrosion intensity field. The set of candidate steel bar implicit corrosion field update quantities is obtained by recursively updating the implicit corrosion field of the steel bars along the update direction according to the preset iteration step size. Write the updated set of candidate steel bar implicit corrosion field values ​​back to the initial value of the steel bar implicit corrosion field to form the updated steel bar implicit corrosion field, and regenerate the cross-frequency reconstruction heat map set to calculate the closed-loop residual update state set; Convergence determination is performed based on updating the state set using closed-loop residuals; The convergence determination is based on the observation consistency target quantity satisfying the descent condition and the cross-frequency consistency target quantity satisfying the consistency condition. If the termination condition is not met, the iterative closed loop of updating the implicit corrosion field of the steel bars and the cross-frequency observation residual set is returned. The descent condition is that the relative descent rate of the observation consistency target quantity in two adjacent iterations is not less than the preset descent threshold. The relative descent rate is obtained by dividing the difference of the observation consistency target quantity in two adjacent iterations by the sum of the observation consistency target quantity and the stability term in the previous iteration. The consistency condition is that the cross-frequency consistency target quantity in the current iteration is not greater than the preset consistency threshold. When the convergence criterion is met, the updated implicit corrosion field of the steel bars is output, and a corrosion risk heat map is generated in the grid coordinate system corresponding to the coordinate transformation field of the steel bar manifold expansion. The heatmap value of the corrosion risk heatmap is obtained by fusing the corrosion intensity of the updated steel bar implicit corrosion field at the corresponding grid position with its local axial gradient amplitude, and masking is performed at the positions marked as invalid by the alignment mask set; Corrosion level assessment results are generated based on corrosion risk heat maps; The corrosion level assessment results consist of a level distribution map obtained by applying a grading threshold to the corrosion risk heat map within the effective area, and a statistical analysis of the area proportion of each level area.

[0027] In this embodiment, the generation of the intelligent corrosion assessment and control closed loop specifically includes: The corrosion risk heat map, corrosion level assessment results and shaping cross-frequency observation residual set are aligned and encapsulated to obtain the control trigger input set; The alignment encapsulation aligns the corrosion risk heatmap and the shaping cross-frequency observation residual set point by point according to the grid coordinate system consistent with the alignment multi-frequency heatmap set, and writes the level distribution map and area proportion statistics of the corrosion level assessment results into the metadata field of the control trigger input set; Generate a set of control trigger heatmaps based on the set of control trigger inputs; The control trigger heatmap set is obtained by fusing the corrosion risk heatmap and the shaping cross-frequency observation residual set in the grid position dimension. The fusion method is to perform risk enhancement on the corrosion risk heatmap, perform cross-frequency inconsistency enhancement on the shaping cross-frequency observation residual set, and perform masking at the positions marked as invalid by the alignment mask set. Generate a set of target regions for rescanning and a set of rescanning priorities based on the set of control trigger heatmaps; The set of target areas for rescanning is obtained by performing threshold filtering and connected component clustering on the set of control trigger heatmaps. The set of rescanning priorities is determined by the weight mapping of the aggregated results of control trigger heatmap values ​​in each connected component and the corresponding levels of corrosion level assessment results. The set of target areas for rescanning is mapped back to the scanning trajectory coordinates along the height and circumferential directions on the surface of the concrete pole to obtain the set of rescanning trajectory segments. A set of frequency switching schemes is generated based on the shaped cross-frequency observation residual set; The frequency replacement scheme set is obtained by statistically analyzing the residual ratio and residual dispersion of each frequency channel in the frequency channel dimension of the shaped cross-frequency observation residual set. The frequency channels with high residual ratio and high residual dispersion are identified as abnormal frequency channels that need to be replaced. Among the remaining frequency channels, the target frequency channel with the largest difference from the corrosion risk heat map response is selected as the replacement frequency, resulting in a frequency replacement scheme set including the retained frequency set, the replacement frequency set, and the frequency switching sequence. A set of encrypted sampling schemes is generated based on the set of target regions for repeated scanning and the set of frequency switching schemes. The encrypted sampling scheme set increases the sampling density along the height and circumferential directions on the set of rescan trajectory segments covered by the set of rescan target areas. Based on the local peak distribution of the control trigger heatmap set within the set of rescan target areas, a set of encrypted sampling points is generated. The set of encrypted sampling points performs higher density circumferential and height direction supplementary sampling in the neighborhood of local peaks. Positions marked as invalid by the alignment mask set are removed from the set of encrypted sampling points to obtain an executable encrypted sampling trajectory set. The set of rescan trajectory segments, the set of frequency switching schemes, and the set of encrypted sampling schemes are encapsulated into detection control commands and written back to the multi-frequency magnetic flux density detection data acquisition process. The detection control commands include a set of target areas for rescanning, a set of rescanning trajectory segments, a set of executable encrypted sampling trajectories, a set of frequency switching schemes, and an update item for acquisition process parameters. The update item for acquisition process parameters drives the magnetization excitation component and the magnetic flux density sensing component to perform corresponding excitation configuration switching and sampling configuration switching during the rescanning, frequency switching, and encrypted sampling stages. The system executes detection and control commands to trigger rescanning, frequency switching, and encrypted sampling, and collects execution feedback to form a closed-loop reflux data set, thereby realizing a closed-loop intelligent assessment and control system for corrosion. The closed-loop reflux data set includes the original multi-frequency magnetic flux density data set obtained by rescanning, the original multi-frequency magnetic flux density data set after frequency conversion, the original multi-frequency magnetic flux density data set obtained by encrypted sampling, and the corresponding height direction position mark, circumferential direction position mark, sampling time mark and sensor working status mark, and is refluxed to the corrosion intelligent assessment to realize the control closed loop.

[0028] Example 1: To verify the feasibility of this invention in practice, it was applied to the operation and maintenance scenario of power distribution lines in a coastal area with high salt spray. The concrete poles in this area have a long service life, and the salt spray and humid environment brought by sea breezes significantly increase the risk of steel reinforcement corrosion. However, the poles often only show minor cracks or no obvious signs. Traditional methods based on single-frequency magnetic flux density or manual experience are easily affected by changes in protective layer thickness, probe fit differences, and uneven material background magnetization, leading to the superposition of thermal artifacts and corrosion signals. This makes it difficult to stably locate the direction and spacing of the steel reinforcement, and the evaluation results fluctuate with the inspectors and inspection batches. Furthermore, after discovering suspected anomalies, confirmation usually relies on repeated full-pole scanning, which is time-consuming and lacks a targeted rescanning strategy, easily causing missed or over-detection and increasing the pressure of power outage coordination. This invention addresses these pain points by constructing an aligned multi-frequency thermal image set using multi-frequency magnetic flux density detection data. It achieves stable evaluation and on-demand rescanning through implicit corrosion field inversion updates and adaptive detection control closed loops, thereby solving the problems of "inconsistent cross-frequency behavior, strong background interference, and inefficient rescanning."

[0029] In this scenario, inspectors, equipped with magnetization excitation and magnetic flux density sensing components, perform multi-frequency excitation scanning and acquisition along the height and circumferential directions of the pole. During acquisition, they simultaneously record height and circumferential position markers, sampling time markers, and sensor operating status markers. The acquired raw multi-frequency magnetic flux density data undergoes sensor calibration and zero-point drift correction within the system, along with sampling time alignment and spatial sampling point alignment, ensuring comparability of data from different frequencies and trajectory segments within a unified grid coordinate system. Subsequently, saturated segments, disconnected segments, and poorly connected segments are removed using anomaly mask sets, and pulse interference points are replaced. Combined with noise suppression and amplitude consistency processing, a standardized multi-frequency magnetic flux density data set is output. The system writes the standardized data into a two-dimensional heat map grid based on the position markers, forming a two-dimensional heat map set of multi-frequency magnetic flux density. The anomaly mask is then propagated into a two-dimensional heat map mask set to fill holes and repair connectivity, ensuring stable subsequent alignment. Based on this, the manifold coordinates of the reinforcing bars are generated and encapsulated together with the set of two-dimensional heatmaps of multi-frequency magnetic flux density into a manifold aligned input set.

[0030] After the rebar manifold alignment input set enters the rebar manifold coordinate generator, the system outputs the rebar orientation field, rebar spacing field, and coordinate confidence field. Based on the coordinate confidence field, it forms an alignment mask and a rejection mask to avoid low-confidence regions from biasing subsequent inferences. Subsequently, the system constructs a rebar manifold unfolded coordinate transformation field and performs rebar manifold unfolding and coordinate alignment on the multi-frequency magnetic flux density two-dimensional heatmap set, obtaining an aligned multi-frequency heatmap set and an alignment mask set, making the rebar-related responses exhibit a more consistent structured morphology along the unfolded coordinates. Next, the aligned multi-frequency heatmap set is input into the implicit corrosion field representation device to generate a set of initial values ​​for the implicit corrosion field and a background field surrogate quantity. The background field surrogate quantity is used to characterize the slowly varying background components caused by changes in protective layer thickness, probe fit, and material background magnetization, reducing the coupling between the corrosion signal and the background term from the source. The initial values ​​for the implicit corrosion field and the background field surrogate quantity set are then input into the cross-frequency differentiable observation operator layer to generate a cross-frequency reconstructed heatmap set, which is compared with the aligned multi-frequency heatmap set within the effective region to obtain a shaped cross-frequency observation residual set. The shaped cross-frequency observation residual set, together with the cross-frequency reconstruction heat map set, the aligned multi-frequency heat map set, and the aligned mask set, enters the observation consistency closed-loop solver. The system uses the observation consistency target quantity and the cross-frequency consistency target quantity as the driving force to perform differentiable backpropagation update and constraint correction, iteratively updates the implicit corrosion field of steel bars and simultaneously suppresses the cross-frequency inconsistency residuals, and outputs the corrosion risk heat map and corrosion level assessment results.

[0031] In actual inspections, the system goes beyond simply outputting assessment results once. It aligns and encapsulates the corrosion risk heatmap, corrosion level assessment results, and the reshaping cross-frequency observation residual set into a control trigger input set, generating a control trigger heatmap set. It automatically assigns the rescan target area and rescan priority, forming a rescan trajectory segment set. Simultaneously, it generates a frequency switching scheme set based on the channel performance of the cross-frequency observation residuals and creates a set of encrypted sampling schemes on the rescan target area. Finally, it encapsulates these into detection control commands, writing them back to the acquisition process to trigger rescanning, frequency switching, and encrypted sampling, forming a closed-loop backflow data set. Through this closed loop, on-site personnel no longer need to repeatedly and blindly scan the entire pole. Instead, they can perform supplementary sampling according to the area and frequency strategies provided by the system, significantly reducing invalid rescanning and waiting time. This also makes the corrosion risk heatmaps from multiple inspections more stable in terms of spatial location and cross-frequency consistency, reducing misjudgments caused by background interference or unstable coordinate alignment. This embodiment was carried out during continuous operation and maintenance periods before and after the flood season and during the humid and high salt spray season of a certain year. It covered multiple pole sections and multiple retest records. The system output rescan range converged and formed a consistent evaluation conclusion more quickly. It also demonstrated better executability and stability in operation and maintenance scheduling and on-site operation coordination, thus verifying the feasibility and closed-loop control value of the invention under real line conditions.

[0032] Table 1 Comparison of Overall Performance in On-site Retests

[0033] As shown in Table 1, the system of this invention achieves simultaneous improvements in four dimensions: identification accuracy, spatial positioning, on-site efficiency, and cross-frequency stability. Taking the corrosion level identification F1 as an example, the system of this invention achieves an accuracy of 0.82, which is 0.13 higher than the 0.69 of single-frequency manual interpretation (an improvement of 18.8%), and 0.07 higher than the 0.75 of multi-frequency fixed threshold (an improvement of 9.3%). The consistency Kappa increases from 0.61 to 0.76, indicating a significant enhancement in interpretation consistency among different personnel and batches. The risk heat map positioning error decreases from 18.7 cm to 9.8 cm (a decrease of 47.6%), which is directly related to the positioning workload of subsequent rescanning and excavation verification. At the same time, the false alarm rate decreases from 12.4% to 7.6%, and the false negative rate decreases from 15.1% to 9.4%, demonstrating that the system of this invention achieves a better balance between reducing unnecessary handling and reducing the risk of missed detection.

[0034] In terms of efficiency and closed-loop control, the rescan trigger rate of this invention is 19.7%, a decrease of 23.1 percentage points (54.0%) compared to 42.8% for single-frequency manual interpretation, and a decrease of 11.9 percentage points (37.7%) compared to 31.6% for multi-frequency fixed threshold; the average detection time per rod is reduced from 21.3 min to 13.6 min (36.2%). The reasons for these improvements are strongly related to the key links in the claims: the rebar manifold alignment unifies the spatial reference of the multi-frequency magnetic flux density two-dimensional heat map to the rebar manifold unfolded coordinates, reducing mismatches caused by rebar orientation, spacing changes, and sampling geometry; the implicit corrosion field representation elevates corrosion from "direct threshold determination" to "continuously invertible field", and then the cross-frequency differentiable observation operator layer generates a cross-frequency reconstructed heat map and forms observation residuals, and finally the observation consistency closed-loop solver iteratively reduces the dispersion of cross-frequency observation residuals (from 0.47 to 0.27, a decrease of 42.6%). With improved cross-frequency consistency, the system becomes less sensitive to occasional noise and single-band anomalies, thus reducing false alarms, rescanning, and duration. The detection and control command generation module uses risk heatmaps, level results, and residuals to jointly drive rescanning, frequency switching, and encrypted sampling, transforming "where to resample and which frequency to switch" from manual experience into data-driven closed-loop decision-making. This allows time to be spent on areas with greater information gain, forming a more stable intelligent corrosion assessment and control closed loop.

[0035] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A smart assessment and control system for corrosion of concrete utility poles based on big data analysis, characterized in that, include: The multi-frequency magnetic flux density acquisition module is used to acquire and preprocess multi-frequency magnetic flux density detection data on the surface of concrete poles. The reinforcement manifold alignment input construction module is used to construct a set of two-dimensional heat maps of multi-frequency magnetic flux density and generate the corresponding reinforcement manifold coordinates; The aligned multi-frequency heat map generation module is used to generate the rebar direction field, rebar spacing field and coordinate confidence field. It performs rebar manifold expansion and coordinate alignment on the multi-frequency magnetic flux density two-dimensional heat map set to obtain the aligned multi-frequency heat map set. The implicit corrosion field and background field generation module is used to generate the initial value of the implicit corrosion field of the steel bars and the set of proxy quantities of the background field; The cross-frequency reconstruction and observation residual calculation module is used to generate a set of cross-frequency reconstruction heatmaps and calculate a set of cross-frequency observation residuals. The observation consistency closed-loop update and evaluation module is used to update the initial value of the implicit corrosion field of steel bars and the cross-frequency observation residual set, and generate corrosion risk heat map and corrosion level assessment results; The detection control command generation module is used to generate detection control commands, trigger rescanning, frequency switching and encrypted sampling, forming a closed loop of intelligent corrosion assessment and control.

2. The intelligent assessment and control system for corrosion of concrete utility poles based on big data analysis according to claim 1, characterized in that, The modules are connected in the following way: Multi-frequency magnetic flux density detection data were collected from the surface of concrete poles, and preprocessing was performed to obtain a standardized dataset of multi-frequency magnetic flux density. A two-dimensional heat map set of multi-frequency magnetic flux density is constructed based on the standardized multi-frequency magnetic flux density dataset, and the corresponding steel bar manifold coordinates are generated to obtain the steel bar manifold aligned input set; Input the reinforcement manifold alignment input set into the reinforcement manifold coordinate generator, output the reinforcement direction field, reinforcement spacing field and coordinate confidence field, perform reinforcement manifold unfolding and coordinate alignment on the multi-frequency magnetic flux density two-dimensional heat map set to obtain the aligned multi-frequency heat map set; Input the aligned multi-frequency heat map set into the implicit corrosion field representation device to generate the initial value of the implicit corrosion field of the steel bar and the set of background field proxy quantities; The initial value of the implicit corrosion field of steel bars and the set of background field proxy quantities are input into the cross-frequency differentiable observation operator layer to generate a set of cross-frequency reconstruction heatmaps and calculate the set of cross-frequency observation residuals. The cross-frequency observation residual set is input into the observation consistency closed-loop solver to update the initial value of the implicit corrosion field of steel bars and the cross-frequency observation residual set. Based on the updated implicit corrosion field of steel bars, a corrosion risk heat map and corrosion level assessment results are generated. Based on the corrosion risk heat map, corrosion level assessment results and cross-frequency observation residual set, detection and control commands are generated to trigger rescanning, frequency switching and encrypted sampling, forming a closed loop of intelligent corrosion assessment and control.

3. The intelligent assessment and control system for corrosion of concrete utility poles based on big data analysis according to claim 2, characterized in that, The generation of the multi-frequency magnetic flux density normalized data set specifically includes: Multi-frequency excitation scanning was performed on the surface of a concrete pole to obtain a set of raw multi-frequency magnetic flux density data. Sensor calibration conversion and zero-point drift correction are performed on the original multi-frequency magnetic flux density data set to obtain the multi-frequency magnetic flux density correction data set; Perform sampling timing alignment and spatial sampling point alignment on the multi-frequency magnetic flux density correction dataset to obtain a multi-frequency magnetic flux density aligned dataset; Anomaly removal is performed on the multi-frequency flux density aligned dataset to obtain a cleaned multi-frequency flux density dataset and anomaly mask dataset. Noise suppression and amplitude uniformity processing are performed on the multi-frequency magnetic flux density cleanup dataset to obtain a multi-frequency magnetic flux density denoised dataset. Perform normalization processing on the multi-frequency magnetic flux density denoised dataset and output a normalized multi-frequency magnetic flux density dataset.

4. The intelligent assessment and control system for corrosion of concrete utility poles based on big data analysis according to claim 2, characterized in that, The generation of the reinforcement manifold alignment input set specifically includes: A set of two-dimensional heat map raster mapping rules is obtained based on a multi-frequency magnetic flux density standardized dataset. Based on the set of two-dimensional heat map grid mapping rules, the multi-frequency magnetic flux density standardized data set is written into the two-dimensional heat map grid according to frequency to obtain the multi-frequency magnetic flux density two-dimensional heat map set; The abnormal mask set is propagated to the multi-frequency magnetic flux density two-dimensional heat map set according to the two-dimensional heat map grid mapping rule set to generate a two-dimensional heat map mask set. Hole filling and connectivity repair are performed to obtain a consistent heat map set of steel reinforcement manifold mask. Based on the set of two-dimensional heat map grid mapping rules, the steel manifold coordinates corresponding to each point of the two-dimensional heat map set of multi-frequency magnetic flux density are generated for each grid position; The set of two-dimensional heatmaps of multi-frequency magnetic flux density, the coordinates of the reinforcing bar manifold, and the set of heatmaps of the reinforcing bar manifold mask are encapsulated according to frequency and grid position to form a set of reinforcing bar manifold aligned inputs.

5. The intelligent assessment and control system for corrosion of concrete utility poles based on big data analysis according to claim 2, characterized in that, The generation of the aligned multi-frequency heatmap set specifically includes: Input the reinforcement manifold alignment input set into the input terminal of the reinforcement manifold coordinate generator to obtain the coordinate generation input tensor; Input the coordinate generation input tensor into the rebar manifold coordinate generator, and output the rebar direction field, rebar spacing field, and coordinate confidence field; Generate alignment and culling masks based on coordinate confidence fields; Perform reinforcement manifold expansion and coordinate alignment on the multi-frequency magnetic flux density two-dimensional heat map set to obtain the reinforcement manifold expansion coordinate transformation field; Based on the coordinate transformation field of the steel bar manifold expansion, bilinear interpolation resampling and alignment are performed on the multi-frequency magnetic flux density two-dimensional heat map set frequency by frequency to obtain the aligned multi-frequency heat map set, and the alignment mask set is generated simultaneously.

6. The intelligent assessment and control system for corrosion of concrete utility poles based on big data analysis according to claim 2, characterized in that, The generation of the initial value of the implicit corrosion field of the reinforcing steel and the set of background field proxy quantities specifically includes: By inputting the aligned multi-frequency heat map set and the aligned mask set into the input terminal of the implicit corrosion field representation device, the implicit field encoding input set is obtained. A cross-frequency feature coding backbone network with shared parameters is used to perform cross-frequency consistent feature coding on the implicit field coding input set, resulting in a cross-frequency consistent feature map set and a cross-frequency consistent global description vector set; Based on the cross-frequency consistent feature map set and the coordinate transformation field of the rebar manifold expansion, the axial sampling band set and the normal sampling band set of the rebar are generated. Under the constraint of the alignment mask set, the cross-frequency consistent feature map set in the sampling band is aggregated within the band to obtain the axial feature sequence set and the normal feature sequence set of the rebar. The set of axial feature sequences of reinforcing bars and the set of cross-frequency consistent global description vectors are input into the implicit corrosion field initial value generation unit of the implicit corrosion field representation device to generate the initial value of the implicit corrosion field of the reinforcing bars. The set of steel reinforcement normal feature sequences and the set of cross-frequency consistent global description vectors are input into the background field surrogate quantity generation unit of the implicit corrosion field representation device to generate the set of background field surrogate quantities.

7. The intelligent assessment and control system for corrosion of concrete utility poles based on big data analysis according to claim 2, characterized in that, The generation of the cross-frequency observation residual set specifically includes: The initial value of the implicit corrosion field of steel bars and the set of background field proxy quantities are input into the cross-frequency differentiable observation operator layer. The set of aligned multi-frequency heat maps and the set of aligned masks are input into the cross-frequency differentiable observation operator layer as observation references and effective region constraints to obtain the cross-frequency observation input set. Within the cross-frequency differentiable observable operator layer, the axial corrosion excitation distribution of steel bars is constructed based on the initial value of the implicit corrosion field of steel bars, and the slowly varying background response distribution is constructed based on the background field surrogate quantity set. Cross-frequency coupling mapping is performed according to the frequency dimension to obtain the cross-frequency predicted magnetic flux response set. The cross-frequency predicted flux response set is mapped to the same grid coordinate system and amplitude domain as the aligned multi-frequency heat map set through the heat map imaging uniformity unit of the cross-frequency differentiable observable operator layer, thus obtaining the cross-frequency reconstructed heat map set. Point-by-point differencing is performed on the cross-frequency reconstructed heatmap set and the aligned multi-frequency heatmap set within the effective area defined by the aligned mask set to generate cross-frequency observation residuals. Cross-frequency consistent residual shaping is then performed to suppress occasional spike residuals in anomalous frequency channels, resulting in a shaped cross-frequency observation residual set.

8. The intelligent assessment and control system for corrosion of concrete utility poles based on big data analysis according to claim 2, characterized in that, The generation of the corrosion risk heat map and corrosion level assessment results specifically includes: The set of shaped cross-frequency observation residuals, the set of cross-frequency reconstructed heatmaps, the set of aligned multi-frequency heatmaps, and the set of aligned masks are input into the observation-consistent closed-loop solver. The initial values ​​of the implicit corrosion field of the steel bars and the set of surrogate quantities of the background field are initialized as closed-loop solution states, and the set of initial closed-loop solution states is obtained. Construct observation-consistent target quantities and cross-frequency consistent target quantities based on closed-loop solution of the initial state set; Based on the observation-consistent target quantity and the cross-frequency consistent target quantity, the initial value of the implicit corrosion field of steel bars is updated by differentiable backpropagation through the observation-consistent closed-loop solver; Write the updated set of candidate steel bar implicit corrosion field values ​​back to the initial value of the steel bar implicit corrosion field to form the updated steel bar implicit corrosion field, and regenerate the cross-frequency reconstruction heat map set to calculate the closed-loop residual update state set; Convergence determination is performed based on updating the state set using closed-loop residuals; When the convergence criterion is met, the updated implicit corrosion field of the steel bars is output, and a corrosion risk heat map is generated in the grid coordinate system corresponding to the coordinate transformation field of the steel bar manifold expansion. Corrosion level assessment results are generated based on the corrosion risk heat map.

9. The intelligent assessment and control system for corrosion of concrete utility poles based on big data analysis according to claim 2, characterized in that, The generation of the intelligent corrosion assessment and control closed loop specifically includes: The corrosion risk heat map, corrosion level assessment results and shaping cross-frequency observation residual set are aligned and encapsulated to obtain the control trigger input set; Generate a set of control trigger heatmaps based on the set of control trigger inputs; Generate a set of target regions for rescanning and a set of rescanning priorities based on the set of control trigger heatmaps; A set of frequency switching schemes is generated based on the shaped cross-frequency observation residual set; A set of encrypted sampling schemes is generated based on the set of target regions for repeated scanning and the set of frequency switching schemes. The set of rescan trajectory segments, the set of frequency switching schemes, and the set of encrypted sampling schemes are encapsulated into detection control commands and written back to the multi-frequency magnetic flux density detection data acquisition process. The system executes detection and control commands to trigger rescanning, frequency switching, and encrypted sampling, and collects execution feedback to form a closed-loop reflux data set, thereby realizing a closed-loop intelligent assessment and control system for corrosion.

Images