A method and system for detecting external defects of a covered liquefied hydrocarbon storage tank

Abstract

The application relates to the technical field of intelligent sensors, and discloses a method and system for detecting appearance defects of a covered soil type liquefied hydrocarbon storage tank. The method comprises the following steps: acquiring ground surface temperature distribution data above a soil layer, generating an initial temperature sequence and separating a medium-high frequency disturbance component, determining a soil heat capacity delay effect parameter and correcting the temperature sequence; a super-temperature difference threshold value is used to strip the soil thermal physical influence to obtain a pure thermal anomaly signal; a thermal distribution law is restored and mapped to a three-dimensional coordinate system of the storage tank to determine a preliminary defect boundary; a gradient is calculated to accurately locate a corrosion layer damage position, heat insulation attribute data and a fire risk assessment model are fused to generate an early warning report containing a defect position, a risk level and a processing suggestion. The method can realize accurate and efficient detection of appearance defects of the storage tank and meet the high-precision and continuity requirements of safety monitoring.

Description

Technical Field

[0001] This invention relates to the field of intelligent sensor technology, and in particular to a system for detecting and managing appearance defects in soil-covered liquefied hydrocarbon storage tanks. Background Technology

[0002] Currently, in the field of intelligent sensors, with the continuous expansion of energy storage scale and the increasing demand for safe operation and maintenance of storage tanks, the detection of appearance defects in soil-covered liquefied hydrocarbon storage tanks is a core link in ensuring the safe operation of equipment, and is directly related to the safety of energy storage and the stability of the surrounding environment.

[0003] Existing methods for detecting external defects in soil-covered liquefied hydrocarbon storage tanks in the industry mainly rely on manual excavation or single infrared scanning. For example, this involves periodically excavating to expose the tank for flaw detection, or using ordinary infrared equipment to collect surface temperature data without depth processing, or lacking effective removal of soil thermal inertia interference. However, these methods are clearly insufficient in complex operating environments. Manual excavation is costly and time-consuming, making continuous monitoring impossible; single infrared scanning is easily affected by soil thermal capacity delay and thermal resistance smoothing, masking weak defect signals; and without signal correction for soil thermophysical properties, it is difficult to distinguish between environmental interference and genuine defect signals, especially in scenarios with thick soil layers or changing soil moisture, leading to misjudgments and missed detections, and making it impossible to accurately locate defects.

[0004] In summary, existing technologies are insufficient for the accurate and efficient detection of external defects in earth-covered liquefied hydrocarbon storage tanks, and cannot meet the high-precision and continuous requirements for safety monitoring of energy storage equipment. Summary of the Invention

[0005] This invention provides a method and system for detecting appearance defects in earth-covered liquefied hydrocarbon storage tanks, so as to achieve accurate and efficient detection of appearance defects in earth-covered liquefied hydrocarbon storage tanks and meet the high precision and continuous requirements for safety monitoring of energy storage equipment.

[0006] Firstly, in order to solve the above-mentioned technical problems, the present invention provides a method for detecting appearance defects in soil-covered liquefied hydrocarbon storage tanks, comprising:

[0007] Obtain temperature distribution data of the surface area above the overburden layer;

[0008] The temperature distribution data is converted into a two-dimensional temperature field matrix, and transient temperature disturbance information is extracted to generate an initial temperature sequence.

[0009] The initial temperature sequence is subjected to Fourier transform to extract the corresponding spectrum and separate the mid-to-high frequency perturbation components. Based on the amplitude attenuation characteristics and phase lag law of the mid-to-high frequency perturbation components, the delay effect parameters caused by soil heat capacity are determined.

[0010] The initial temperature sequence is time-domain corrected based on the delay effect parameter to obtain the corrected temperature sequence;

[0011] The temperature difference between the local area and the surrounding area in the corrected temperature sequence is calculated. If the temperature difference exceeds the preset temperature difference judgment threshold, the corresponding temperature distribution data is aligned with the real-time collected surface temperature sequence. The soil thermophysical influence is removed by establishing a heat exchange model to obtain a pure thermal anomaly signal.

[0012] The pure thermal anomaly signal is input into a preset thermal distribution restoration model to restore the thermal distribution pattern caused by the damage to the anti-corrosion layer. The thermal distribution pattern is mapped to a preset three-dimensional coordinate system of the storage tank, and the coordinates of the abnormal area are extracted to determine the preliminary boundary of the defect detection.

[0013] Gradient calculation is applied to the pure thermal anomaly signal within the preliminary boundary-defined analysis range, and the correspondence between the gradient peak value and the thermal signal attenuation mode is determined. Based on the correspondence, the precise location information of the anti-corrosion layer damage is determined.

[0014] The system integrates the location information with the pre-acquired natural thermal insulation attribute data of the storage tank, inputs it into a preset fire risk assessment model, outputs the corresponding risk probability value, and generates a defect warning report based on the extent to which the risk probability value exceeds the preset probability judgment threshold.

[0015] Secondly, the present invention provides a system for detecting appearance defects in soil-covered liquefied hydrocarbon storage tanks, comprising:

[0016] The temperature distribution acquisition module is used to acquire temperature distribution data of the surface area above the soil cover layer;

[0017] The temperature sequence generation module is used to convert the temperature distribution data into a two-dimensional temperature field matrix, extract transient temperature disturbance information, and generate an initial temperature sequence.

[0018] The delay parameter determination module is used to perform Fourier transform on the initial temperature sequence, extract the corresponding spectrum, separate the mid-to-high frequency disturbance components, and determine the delay effect parameters caused by soil heat capacity based on the amplitude attenuation characteristics and phase lag law of the mid-to-high frequency disturbance components.

[0019] A time-domain correction module is used to perform time-domain correction on the initial temperature sequence based on the delay effect parameter to obtain a corrected temperature sequence.

[0020] An anomaly signal extraction module is used to calculate the temperature difference between a local area and the surrounding area in the corrected temperature sequence. If the temperature difference exceeds a preset temperature difference judgment threshold, the corresponding temperature distribution data is aligned with the real-time collected surface temperature sequence, and a heat exchange model is established to remove the soil thermophysical influence and obtain a pure thermal anomaly signal.

[0021] The preliminary boundary determination module is used to input the pure thermal anomaly signal into a preset thermal distribution restoration model, restore the thermal distribution pattern caused by the damage to the anti-corrosion layer, map the thermal distribution pattern to a preset three-dimensional coordinate system of the storage tank, extract the coordinates of the abnormal area, and determine the preliminary boundary of the defect detection.

[0022] The precise positioning module is used to apply gradient calculation to the pure thermal anomaly signal within the preliminary boundary-defined analysis range, determine the correspondence between the gradient peak value and the thermal signal attenuation mode, and determine the precise positioning information of the anti-corrosion layer damage based on the correspondence.

[0023] The early warning report generation module is used to integrate the location information with the pre-acquired natural thermal insulation attribute data of the storage tank, input a preset fire risk assessment model, output the corresponding risk probability value, and generate a defect early warning report based on the extent to which the risk probability value exceeds the preset probability judgment threshold.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] (1) This invention acquires surface temperature distribution data above the soil cover layer through a multispectral infrared sensor array, generates an initial temperature sequence, separates the mid-to-high frequency disturbance components and extracts amplitude attenuation and phase lag characteristics, determines soil thermal capacity delay effect parameters, breaks through the limitation of traditional single infrared scanning being unable to cope with soil thermal inertial interference, explores the characteristics of weak thermal anomaly signals caused by defects, eliminates the interference of soil thermal capacity delay and thermal resistance smoothing, provides high-precision basic data support for defect detection, effectively improves the signal capture rate of minor damage to the anti-corrosion layer, and solves the problem of weak thermal anomalies being masked by soil thermal background.

[0026] (2) The present invention performs time-domain correction on the initial temperature sequence based on the delay effect parameter, aligns the surface temperature sequence when the temperature difference threshold is exceeded, and obtains a pure thermal anomaly signal by stripping away the soil thermophysical influence, restores the thermal distribution law and maps it to the three-dimensional coordinate system of the storage tank. It breaks through the limitation of traditional simple signal processing that it is difficult to distinguish between environmental interference and real defects, accurately captures the exclusive thermal distribution characteristics of the anti-corrosion layer damage, provides multi-dimensional basis for defect location, significantly improves the accuracy of defect identification in complex soil environment, and makes up for the high misjudgment and omission rate of the existing technology.

[0027] (3) Based on the preliminary boundary definition of the analysis range, this invention achieves accurate defect location through gradient calculation and thermal signal attenuation mode matching. It integrates thermal insulation attribute data and fire risk assessment model to generate early warning reports, breaking through the limitations of long detection cycle and ambiguous positioning of traditional manual excavation. It provides operation and maintenance personnel with accurate defect location and risk level basis, solves the problems of low detection efficiency and poor safety of soil-covered storage tank defects, takes into account both detection accuracy and operation and maintenance timeliness, and meets the high precision and continuous requirements of safety monitoring of energy storage equipment. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of a method for detecting appearance defects in a soil-covered liquefied hydrocarbon storage tank according to the first embodiment of the present invention;

[0029] Figure 2 This is a schematic diagram of the appearance defect detection system for a soil-covered liquefied hydrocarbon storage tank provided in the second embodiment of the present invention. Detailed Implementation

[0030] 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.

[0031] Reference Figure 1 The first embodiment of the present invention provides a method for detecting appearance defects in a soil-covered liquefied hydrocarbon storage tank, comprising the following steps:

[0032] S101, Obtain temperature distribution data of the surface area above the overburden layer;

[0033] S102, the temperature distribution data is converted into a two-dimensional temperature field matrix, and transient temperature disturbance information is extracted to generate an initial temperature sequence;

[0034] S103, Perform Fourier transform on the initial temperature sequence, extract the corresponding spectrum, separate the mid-to-high frequency perturbation components, and determine the delay effect parameters caused by soil heat capacity based on the amplitude attenuation characteristics and phase lag law of the mid-to-high frequency perturbation components.

[0035] S104, Perform time-domain correction on the initial temperature sequence according to the delay effect parameter to obtain the corrected temperature sequence;

[0036] S105, calculate the temperature difference between the local area and the surrounding area in the corrected temperature sequence. If the temperature difference exceeds the preset temperature difference judgment threshold, align the corresponding temperature distribution data with the real-time collected surface temperature sequence, and obtain a pure thermal anomaly signal by establishing a heat exchange model to remove the soil thermophysical influence.

[0037] S106, The pure thermal anomaly signal is input into a preset thermal distribution restoration model to restore the thermal distribution pattern caused by the damage to the anti-corrosion layer, the thermal distribution pattern is mapped to a preset three-dimensional coordinate system of the storage tank, and the coordinates of the abnormal area are extracted to determine the preliminary boundary of the defect detection.

[0038] S107, Apply gradient calculation to the pure thermal anomaly signal within the preliminary boundary-defined analysis range, determine the correspondence between the gradient peak value and the thermal signal attenuation mode, and determine the precise location information of the anti-corrosion layer damage based on the correspondence.

[0039] S108, integrate the positioning information with the pre-acquired natural thermal insulation attribute data of the storage tank, input the preset fire risk assessment model, output the corresponding risk probability value, and generate a defect warning report based on the extent to which the risk probability value exceeds the preset probability judgment threshold.

[0040] In step S101, acquiring the temperature distribution data of the surface area above the overburden layer includes:

[0041] A multispectral infrared sensor array was used to collect full-coverage data of the surface area above the soil cover layer, obtaining raw temperature data and spatial location information.

[0042] The original temperature data is denoised to obtain denoised temperature data.

[0043] The denoised temperature data is integrated with the spatial location information to form surface temperature distribution data.

[0044] It should be noted that, firstly, when acquiring raw temperature data, an array of uncooled multispectral infrared sensors was constructed. The sensor's detection band covers the mid-far infrared band of 8-14μm. This band can accurately capture the thermal radiation signal of the soil surface and has good sensitivity to identify weak thermal anomalies under the soil-covered storage tank.

[0045] The sensor array is arranged in a planar array, with a spatial resolution of 0.1 meters per pixel. This allows for full coverage acquisition of the area above the soil cover while ensuring spatial detail of the temperature data. When the soil cover area is large, the acquisition range can be expanded by splicing multiple arrays. The sampling frequency is set to 1 frame per second to balance the real-time performance of the data and storage efficiency.

[0046] During the data acquisition process, the geospatial location information corresponding to each pixel is recorded synchronously. The raw temperature data is measured using radiation temperature as the standard and is saved in binary raw data format for easy subsequent processing and analysis.

[0047] Next, the raw temperature data undergoes denoising processing. When removing outliers and duplicate data, median filtering is first used for spatial denoising. A 3×3 filter window is selected, which effectively eliminates isolated noise points caused by sensor noise and environmental stray radiation, while preserving the spatial gradient characteristics of the temperature data. Then, the Laida criterion is used for outlier identification and removal. This criterion is based on the normal distribution characteristics of the temperature data, calculating the mean and standard deviation of all data. Values ​​exceeding the mean ± 3 times the standard deviation are identified as outliers. These values ​​are often caused by momentary sensor malfunctions or strong environmental interference. After removal, linear interpolation is performed at the locations of the outliers to ensure the spatial continuity of the data.

[0048] Subsequently, duplicate data screening was performed on the denoised data. Based on spatial location information and timestamps, completely duplicate pixel data was removed to reduce data redundancy. After screening, the data was subjected to min-max normalization to map the temperature values ​​to the [0,1] interval. The reference range for normalization is the typical surface temperature range of -10℃ to 50℃ in the monitoring scenario of soil-covered storage tanks.

[0049] For example, after applying a 3×3 window mid-range filter to the collected raw temperature data, the mean was calculated to be 18℃ and the standard deviation to be 2℃. Values ​​below 12℃ and above 24℃ were identified as outliers and removed. Interpolation was used to complete the positions of more than 200 outlier points. At the same time, more than 5,000 duplicate data were screened out and removed. After normalization, the data denoising was completed.

[0050] When extracting spatial location information and corresponding temperature values ​​from the denoised data and integrating them to form surface temperature distribution data, the spatial location information in latitude and longitude form and the normalized temperature value are first extracted pixel by pixel from the denoised data. At the same time, the actual value of radiation temperature is restored, forming a three-dimensional data unit containing spatial location, normalized temperature value, and actual radiation temperature value. Then, with the geometric center of the storage tank as the coordinate origin, a plane rectangular coordinate system dedicated to the storage tank is established. The latitude and longitude coordinates of all three-dimensional data units are converted into plane coordinates under this rectangular coordinate system, realizing the unified calibration of spatial location.

[0051] Subsequently, all data units are arranged in a grid according to the horizontal and vertical coordinates of the plane. The size of the grid is consistent with the spatial resolution of the sensor array, which is 0.1 m × 0.1 m. The arithmetic mean of the temperature values ​​in each grid is taken to obtain the representative temperature value of the grid. Finally, the plane coordinates of all grids and the corresponding representative temperature values ​​are integrated to form the surface temperature distribution data. This data is stored in a two-dimensional raster data format, which can intuitively reflect the spatial distribution characteristics of surface temperature above the soil cover layer.

[0052] In step S102, converting the temperature distribution data into a two-dimensional temperature field matrix and extracting transient temperature perturbation information to generate an initial temperature sequence includes:

[0053] The temperature distribution data is stored as a two-dimensional temperature field matrix containing time and spatial information;

[0054] The background temperature change component caused by soil thermal inertia is separated from the two-dimensional temperature field matrix by time difference filtering, and transient temperature disturbance information is extracted.

[0055] The transient temperature disturbance information is integrated in chronological order to generate an initial temperature sequence.

[0056] It should be noted that when collecting temperature field data at various times, the preset time interval is set based on the propagation characteristics of thermal anomaly signals from soil-covered storage tanks and the real-time detection requirements. Soil thermal inertia causes a delay in the propagation of thermal signals. Too short an interval will cause data redundancy, while too long an interval will miss transient thermal anomalies. Therefore, the basic time interval is set to 5 seconds. In areas where the soil cover thickness exceeds 3 meters, the soil thermal delay is more obvious, and the interval can be increased to 10 seconds. In shallow soil-covered areas where the soil cover thickness is less than 1 meter, the thermal signal response is faster, and the interval can be decreased to 2 seconds. Those skilled in the art will know that this interval can be flexibly adjusted within the range of 2 to 10 seconds according to the actual thickness of the soil cover layer of the storage tank.

[0057] In this implementation case, a multispectral infrared sensor array was used during the scanning process to maintain the same spatial resolution as the temperature distribution data acquisition. Temperature field data of the entire area above the soil cover was collected frame by frame, with each frame accompanied by a precise timestamp and spatial coordinate information to ensure the spatiotemporal consistency of the collected data. For example, when scanning a tank area with a 2-meter-thick soil cover, the preset time interval was set to 5 seconds, and the multispectral infrared sensor array continuously collected data for 1 hour, resulting in 720 frames of temperature field data with timestamps and spatial coordinates, completely recording the surface temperature changes during that period.

[0058] Next, when storing the temperature field data as a two-dimensional temperature field matrix containing time and spatial information, a Cartesian coordinate system of the area above the tank's cover layer is used as the basis. The horizontal and vertical coordinates of the spatial dimension are used as the row and column indices of the matrix, and the time dimension is used as the third dimension of the matrix to construct a three-dimensional temperature field data structure. This structure is then decomposed into multiple two-dimensional temperature field matrices according to time slices, with each matrix corresponding to the global temperature distribution at a specific moment. The value of each element in the matrix is ​​the actual radiation temperature value at the corresponding spatial location. Metadata information is also established for the matrix, including key information such as the acquisition timestamp, sensor array parameters, and spatial resolution. The matrix is ​​stored in a floating-point data format, retaining two decimal places for the temperature values ​​to ensure data accuracy. During storage, the data is divided into blocks, with every 100 frames of data forming a data block, facilitating subsequent reading and processing and avoiding operational lag caused by excessively large single-file data volumes.

[0059] For example, 720 frames of temperature field data are decomposed into 720 two-dimensional temperature field matrices according to time. The number of rows and columns of each matrix corresponds to the pixel distribution of the ground area, and the element value is the temperature value of the corresponding location. At the same time, the acquisition time is marked for each matrix. After being divided into blocks of 100 frames, the data is stored as 8 data blocks for easy retrieval later.

[0060] Subsequently, the background temperature change component caused by soil thermal inertia was separated from the two-dimensional temperature field matrix by time-difference filtering. When extracting transient temperature disturbance information, a first-order time-difference algorithm was used to perform difference operations on the two-dimensional temperature field matrix at consecutive time points, calculating the temperature difference at the same spatial location between two adjacent time points to obtain a difference temperature matrix. Then, a low-pass filter was applied to this matrix, using a Butterworth low-pass filter with a cutoff frequency of 0.01Hz. This frequency can effectively filter out the low-frequency background temperature change component caused by soil thermal inertia, while retaining the high-frequency transient temperature disturbance caused by tank defects. After filtering, the validity of the obtained transient temperature data was determined. Values ​​with an absolute temperature difference exceeding 0.01℃ were retained as valid transient temperature disturbance information, while values ​​below this value were considered environmental noise and set to zero. At the same time, the data was subjected to min-max normalization, mapping it to the [0,1] interval. The normalization reference range is the typical transient temperature disturbance range of -0.5℃ to 0.5℃ in the soil-covered storage tank detection scenario.

[0061] Next, the transient temperature disturbance information is integrated in chronological order. When generating the initial temperature sequence, the normalized temperature disturbance value of each spatial location is extracted from the transient temperature disturbance matrix of each frame. Using the spatial location as a unique identifier, the temperature disturbance values ​​of the same spatial location at different times are arranged in chronological order according to the timestamp to form a one-dimensional temperature disturbance sequence for that location.

[0062] Then, the one-dimensional temperature disturbance sequences of all spatial locations are integrated in the row and column order of the Cartesian coordinate system to construct the initial temperature sequence of the entire domain. This sequence is a two-dimensional array structure, with rows and columns corresponding to spatial locations and the third dimension being the time series, which fully characterizes the transient temperature disturbance features of each spatial location above the overburden layer as time changes.

[0063] After integration, the initial temperature sequence is validated to remove abnormal sequences with disordered timestamps or missing data, ensuring the temporal continuity and spatial integrity of the sequence.

[0064] In step S103, a Fourier transform is performed on the initial temperature sequence to extract the corresponding spectrum, and the mid-to-high frequency perturbation components are separated. Based on the amplitude attenuation characteristics and phase lag law of the mid-to-high frequency perturbation components, the delay effect parameters caused by soil heat capacity are determined, including:

[0065] The initial temperature sequence is segmented, and the frequency spectrum distribution of each time period is extracted by Fourier transform.

[0066] The low-frequency background component and the mid-to-high-frequency disturbance component in the frequency spectrum distribution are separated to obtain the mid-to-high-frequency thermal signal;

[0067] The mid-to-high frequency thermal signal is subdivided into frequency bands, and the energy distribution characteristics of sub-bands with an exponential decay trend over time are extracted and integrated in chronological order to form an amplitude decay sequence.

[0068] The amplitude decay sequence is subjected to exponential fitting to calculate the decay rate constant and phase lag time value, forming a set of characteristic parameters to determine the delay effect parameters.

[0069] It should be noted that, firstly, the initial temperature sequence is segmented. When extracting the frequency spectrum distribution for each time period, the initial temperature sequence is divided into equal-length segments according to the time length. The basic segment duration is set to 30 minutes. This duration balances the frequency resolution and time resolution of the spectrum analysis, ensuring that the spectral features are not indistinct due to excessively short segments, nor are transient thermal signal characteristics lost due to excessively long segments. In areas with thicker overburden and slower thermal signal changes, the segment duration can be increased to 60 minutes. In areas with thinner overburden and faster thermal signal response, the duration can be decreased to 15 minutes. Those skilled in the art will understand that the duration can be adjusted within the range of 15 to 60 minutes based on the actual situation of the overburden layer of the storage tank.

[0070] Next, Fourier transform is applied to the subsequences after each segmentation for spectral analysis, converting the time-domain signal into a frequency-domain signal. The frequency spectrum distribution corresponding to each subsequence is extracted, including core information such as frequency, amplitude, and phase. Simultaneously, min-max normalization is performed on the spectral data, mapping the amplitude to the [0,1] interval. The normalization reference range is the amplitude extreme range of the initial temperature sequence spectrum, ensuring the comparability of the spectral data of each segment. For example, the initial temperature sequence of 2 consecutive hours is divided into 30-minute segments, resulting in 4 time subsequences. After Fourier transform, the frequency spectrum distribution of each subsequence is extracted. After normalization, the amplitude of each segment's spectrum is distributed between 0 and 1, facilitating subsequent component separation.

[0071] Subsequently, when separating the low-frequency background component and the mid-to-high-frequency disturbance component in the frequency spectrum distribution, the threshold for dividing the frequency interval was set based on the frequency characteristics of soil thermal inertia. These frequency characteristics were derived from statistical analysis of historical monitoring data of the storage tank and comprehensive calibration using multiphysics simulation. Specifically, background temperature changes caused by soil thermal inertia are mostly concentrated in the low-frequency range below 0.001Hz, while thermal anomalies caused by tank defects are mostly distributed in the mid-to-high-frequency range of 0.001Hz-0.1Hz. Therefore, 0.001Hz was set as the critical threshold for high-to-low frequency separation. The low-frequency characteristics of thermal inertia in sandy soils are narrower, so the critical threshold can be lowered to 0.0005Hz. The frequency band affected by thermal inertia in clay soils is slightly wider, so the threshold can be raised to 0.002Hz. These values ​​can be flexibly adjusted according to the soil medium type. Based on this critical threshold, components in the frequency spectrum distribution that are above the threshold are defined as mid-to-high frequency disturbance components, and those that are below the threshold are defined as low-frequency background components. All mid-to-high frequency disturbance components are directly extracted and integrated in frequency order to obtain mid-to-high frequency thermal signals. At the same time, the amplitude and phase information of the mid-to-high frequency thermal signals are retained to provide data support for subsequent extraction of attenuation features.

[0072] For example, using 0.001Hz as the critical threshold, the mid-to-high frequency disturbance components of 0.001Hz-0.1Hz are extracted from the frequency spectrum distribution of each segment. The integrated mid-to-high frequency thermal signal clearly retains the frequency domain characteristics of thermal disturbances that may be caused by tank defects.

[0073] In this implementation case, when performing frequency band subdivision on the mid-to-high frequency thermal signal, extracting the energy distribution characteristics of sub-bands including the soil thermal capacity delay effect, and integrating them in time sequence to form an amplitude attenuation sequence, wavelet packet decomposition technology is used to perform frequency band subdivision on the mid-to-high frequency thermal signal. The db4 wavelet is selected as the base wavelet, and the number of decomposition layers is set to 4. This parameter can subdivide the mid-to-high frequency band into 16 equal-width sub-bands, which can accurately capture the energy change characteristics of different sub-bands.

[0074] After obtaining the signals of each sub-band through wavelet packet decomposition, the energy value of each sub-band is calculated. Specifically, the absolute energy is calculated by summing the squares of the amplitudes of the sampling points within the sub-band, and the target sub-bands with a clear and continuous energy decay trend over time are selected using the correlation coefficient. The energy distribution characteristics of these sub-bands are extracted. For example, the energy change sequence of each sub-band over time is calculated, and a first-order exponential decay model is fitted to the sequence. If the goodness of fit R² > 0.8 and the decay rate constant is greater than zero, the sub-band is determined to contain a delay effect, and its energy distribution characteristics are extracted. Next, the amplitude peaks of the corresponding sub-bands for each segment are arranged sequentially according to time. During extraction, the maximum value of the signal envelope is obtained using Hilbert transform to eliminate local noise interference, forming a one-dimensional amplitude decay sequence. Each element of the sequence corresponds to the amplitude peak of a time segment. Simultaneously, the sequence is subjected to min-max normalization, mapped to the [0,1] interval. The normalization reference range is the extreme value of the sequence's amplitude, namely the maximum initial peak value of the first time period and the minimum residual peak value of the last time period, eliminating the influence of numerical magnitude differences and converting the amplitude with physical dimensions into a standardized relative decay ratio.

[0075] Next, an exponential fitting was performed on the amplitude decay sequence. When calculating the decay rate constant and phase lag time, a first-order exponential decay model was used to curve-fit the amplitude decay sequence. This model accurately characterizes the amplitude decay law of the thermal signal caused by soil heat capacity. During the fitting process, the least squares method was used to solve the model parameters, with the mean square error as the loss function. Iterative optimization was stopped when the loss value was stably less than 0.001, ensuring the accuracy of the fitting. The decay rate constant in the model was obtained through fitting calculations. This constant reflects the rate of decay of the thermal signal amplitude. Simultaneously, the phase difference between different frequency components and the reference signal was calculated from the phase information of the original mid-to-high frequency thermal signal. The average value was taken to obtain the phase lag time value, which characterizes the thermal signal propagation delay caused by soil heat capacity. The reference signal is the background temperature signal obtained by low-pass filtering (cutoff frequency 0.001Hz) of the initial temperature sequence. By integrating the decay rate constant and the phase lag time value, a set of characteristic parameters is formed. This set of parameters is the delayed effect parameter characterizing the influence of soil heat capacity. The value of the parameter set will change with the physical properties of soil such as heat capacity and thermal conductivity, and can be directly used for subsequent time-domain correction of temperature series.

[0076] For example, after fitting the amplitude decay sequence with a first-order exponential decay model, the decay rate constant is obtained as 0.02 / min, and the phase lag time value of 3 seconds is calculated. The two parameters are integrated into a feature parameter set, which is used as the delay effect parameter for this detection in subsequent time-domain correction processing.

[0077] In step S104, the step of performing time-domain correction on the initial temperature sequence based on the delay effect parameter to obtain the corrected temperature sequence includes:

[0078] The initial temperature sequence is segmented in the time domain to obtain multiple time windows;

[0079] The delay effect parameter is applied to the data within the multiple time windows for compensation, generating an intermediate temperature sequence with preliminary time-domain correction;

[0080] Fourier transform and frequency domain analysis were performed on the intermediate temperature sequence to extract the main frequency components within each time window, and the low-frequency smooth components related to thermal resistance characteristics were separated to obtain the frequency characteristic distribution.

[0081] The frequency characteristic distribution is weighted and calculated, using the negative correlation value of the proportion of low-frequency energy as the weight to generate a weighted characteristic distribution;

[0082] The intermediate temperature sequence is then corrected by applying the weighted feature distribution to obtain the corrected temperature sequence.

[0083] It should be noted that when the initial temperature sequence is segmented in the time domain to obtain multiple time windows, the duration of the time domain segment is set according to the time variation law of soil thermal resistance characteristics. The signal smoothing effect caused by soil thermal resistance has stability in a short period of time. The basic segment duration is set to 20 minutes. This duration can ensure that the thermal resistance influence characteristics within each time window are relatively uniform, and will not lose local transient thermal anomaly information due to excessively long segments. In areas where the moisture content of the cover soil changes rapidly and the thermal resistance characteristics fluctuate greatly, the segment duration can be reduced to 10 minutes. In areas where the properties of the cover soil are stable and the thermal resistance changes slowly, the duration can be increased to 30 minutes. Those skilled in the art will know that the duration can be flexibly adjusted within the range of 10 to 30 minutes according to the actual environmental characteristics of the tank cover soil.

[0084] Subsequently, the initial temperature sequence is divided into equal-length segments according to the fixed duration. During the segmentation process, the timestamps and spatial location information of the sequence are preserved. Each time window forms an independent sub-temperature sequence, and finally, multiple consecutive time window datasets are obtained, which provide a basis for subsequent window-by-window correction.

[0085] Next, a delay effect parameter is applied to the data within multiple time windows for compensation. When generating the intermediate temperature sequence for preliminary time-domain correction, a time-domain compensation algorithm is constructed based on the previously determined soil heat capacity delay effect parameters, namely the decay rate constant and the phase lag time value. The algorithm performs phase shifting and amplitude correction on the temperature data point by point within each time window. The duration of the phase shift is consistent with the phase lag time value, and the amplitude correction compensates for the amplitude decay caused by soil heat capacity according to the decay rate constant. During the compensation process, the first sampling point of each time window is used as the time reference, and the temperature values ​​of all subsequent sampling points are synchronously corrected to ensure the temporal consistency of the data within the window. After the correction is completed, the corrected data of each time window are spliced ​​together in the original time order to form the intermediate temperature sequence for preliminary time-domain correction. At the same time, the sequence is subjected to min-max normalization to map the temperature values ​​to the [0,1] interval. The normalization reference range is the temperature extreme value of the initial temperature sequence to eliminate numerical deviations during the compensation process.

[0086] For example, by substituting the phase lag time of 3 seconds and the decay rate constant of 0.02 / min into the compensation algorithm, the temperature data of the 6 time windows were corrected point by point and then spliced ​​together. After normalization, the intermediate temperature sequence was obtained, which initially eliminated the signal delay and attenuation caused by soil heat capacity.

[0087] Subsequently, frequency domain analysis was performed on the intermediate temperature sequence to extract the main frequency components within each time window. Low-frequency smoothing components related to thermal resistance characteristics were separated to obtain the frequency characteristic distribution. Fourier transforms were then performed on the subsequences of each time window in the intermediate temperature sequence to convert the time-domain signal into a frequency-domain signal, extracting the frequency spectrum distribution of each window, including core information such as frequency and amplitude. The signal smoothing effect caused by soil thermal resistance is mainly reflected in the low-frequency band. 0.0005Hz was used as the critical threshold for dividing the low-frequency smoothing component. This threshold was set based on the frequency characteristics of historical soil thermal resistance monitoring data. For sandy soils, the low-frequency characteristics of thermal resistance are narrower, so the threshold can be lowered to 0.0003Hz. For clay soils, the low-frequency influence of thermal resistance is slightly wider, so the threshold can be raised to 0.0008Hz. Adjustments can be made according to the soil medium type.

[0088] Based on this threshold, the low-frequency smoothed component and the mid-to-high-frequency effective component of each window are separated. The energy proportion and dominant frequency of the low-frequency smoothed component within each window are statistically analyzed. These characteristics are then integrated sequentially according to the time window to form a frequency characteristic distribution characterizing the effect of thermal resistance smoothing. This distribution intuitively reflects the degree of influence of thermal resistance on the signal within different time periods. For example, after performing Fourier transforms on the subsequences of six time windows, the low-frequency smoothed component is separated using a threshold of 0.0005Hz. The statistical results show that the low-frequency energy proportion in each window is between 20% and 70%. After sequential integration, a frequency characteristic distribution is formed, clearly presenting the temporal variation of the thermal resistance effect.

[0089] In this implementation case, a weighted calculation is performed on the frequency characteristic distribution. When generating the weighted characteristic distribution, the energy proportion of the low-frequency smoothing component within each time window is used as the core basis for weighting. The higher the energy proportion, the more significant the thermal resistance smoothing effect within that time period, and the larger the corresponding suppression weight value should be. After normalizing the low-frequency energy proportion of each window, 1 minus the normalized value is taken as the suppression weight, realizing a positive correlation between the degree of thermal resistance influence and the suppression weight, and achieving a direct correlation between the degree of thermal resistance influence and the weight value. For abnormal windows with excessively high or low low-frequency energy proportions, the neighborhood averaging method is used to smooth and correct the weight values, avoiding the impact of abnormal weights in a single window on the overall correction effect. After correction, the weight values ​​are arranged in the order of the time windows to generate a weighted characteristic distribution. This distribution corresponds one-to-one with the time windows of the intermediate temperature series, providing a weight basis for subsequent secondary correction. For example, after normalizing the proportion of low-frequency energy in each time window, six weight values ​​are obtained. The weight value of one window with an abnormally high proportion is corrected by neighborhood averaging. After arranging them in order, a weighted feature distribution is generated, which is accurately matched with the six time windows of the intermediate temperature sequence.

[0090] Next, a secondary correction is performed on the intermediate temperature sequence using a weighted feature distribution. The resulting corrected temperature sequence is obtained through a weighted fusion method. Specifically, a Fourier transform is performed on the intermediate temperature sequence to obtain its spectrum. The amplitudes of low-frequency components below 0.0005Hz in this spectrum are multiplied by their corresponding weights in the weighted feature distribution, then combined with other frequency components and subjected to an inverse Fourier transform to obtain the secondary-corrected temperature sequence. After completing the above secondary correction for each time window, the corrected data from each window are seamlessly spliced ​​together in their original chronological order to form a complete secondary-corrected data sequence. This sequence is then subjected to inverse normalization to restore the actual physical dimensions of the temperature values. Simultaneously, data smoothing is performed, using a 5-point moving average to eliminate numerical jumps during the correction process. Finally, a corrected temperature sequence is obtained that eliminates the effects of soil heat capacity delay and thermal resistance smoothing.

[0091] In step S105, the calculation of the temperature difference between the local area and the surrounding area in the corrected temperature sequence, if the temperature difference exceeds a preset temperature difference judgment threshold, aligns the corresponding temperature distribution data with the real-time collected surface temperature sequence, and obtains a pure thermal anomaly signal by establishing a heat exchange model to remove the soil thermophysical influence. This includes:

[0092] The corrected temperature sequence is divided into local regions to obtain multiple sets of local temperature difference points;

[0093] Based on the set of local temperature difference points, the temperature difference between each local area and the surrounding area is calculated. If the temperature difference exceeds the preset temperature difference judgment threshold, it is marked as an abnormal area, forming an abnormal area set.

[0094] The set of abnormal regions is time-aligned with the real-time collected surface temperature sequence to generate a surface temperature reference sequence.

[0095] A heat exchange model is established based on the surface temperature reference sequence, and the temperature data corresponding to the set of anomalous areas is input into the heat exchange model to separate the soil thermophysical components and obtain a preliminary pure thermal anomaly signal.

[0096] The preliminary pure thermal anomaly signal is subjected to spectral filtering to remove residual low-frequency components, resulting in the final pure thermal anomaly signal.

[0097] It should be noted that when dividing the calibration temperature sequence into local regions to obtain multiple sets of local temperature difference points, the monitoring area above the tank's cover layer is spatially divided using a grid of equal size. The basic grid size is set to 5 meters × 5 meters. This size is determined by considering the diffusion range of thermal anomalies in the cover-type tank and the required detection accuracy. This avoids the grid being too large and obscuring minor thermal anomalies, while also preventing data redundancy due to a grid that is too small. In areas with thinner cover layers and more concentrated thermal anomaly signals, the grid size can be reduced to 3 meters × 3 meters. In areas with thicker cover layers and a larger thermal anomaly diffusion range, the grid can be expanded to 8 meters × 8 meters. Those skilled in the art will know that the grid size can be adjusted within the range of 3 meters to 8 meters based on the actual situation of the tank's cover layer. After division, all temperature sampling points within each grid are grouped into a set of local temperature difference points. The temperature values ​​and spatial coordinate information within the set are extracted to ensure the spatial independence and data integrity of each set, providing a foundation for subsequent local temperature difference calculations.

[0098] For example, the area above the soil cover of a 100-meter-diameter storage tank is divided into 5-meter × 5-meter sections to obtain 400 independent sets of local temperature difference points. Each set contains all temperature sampling data within the corresponding grid.

[0099] Next, based on the set of local temperature difference points, the temperature difference between each local area and the surrounding area is calculated. If the temperature difference exceeds the preset temperature difference judgment threshold, it is marked as an abnormal area. When forming an abnormal area set, the average temperature value within each set of local temperature difference points is calculated first. Then, taking the set as the center, the average temperature value of the eight adjacent grids is selected to calculate the average temperature of the surrounding area. The difference between the two is the relative temperature difference of the local area.

[0100] It is worth noting that the temperature difference judgment threshold is set based on historical thermal anomaly data from tank defect detection. After stripping away the soil's thermophysical effects, the effective thermal anomaly temperature difference caused by corrosion layer damage is mostly above 0.8℃, therefore the basic threshold is set at 0.8℃. Sandy soils have fast thermal conductivity and relatively smaller thermal anomaly temperature differences, so the threshold can be lowered to 0.6℃. Clay soils have slow thermal conductivity and more significant thermal anomaly temperature differences, so the threshold can be raised to 1.0℃. This can be flexibly adjusted according to the soil medium type. Local areas with relative temperature differences exceeding this threshold are marked as anomalous areas. If multiple adjacent anomalous areas form a connected region, they are merged into a whole. Finally, all independent anomalous areas and connected anomalous areas are integrated to form an anomalous area set.

[0101] For example, after calculating the relative temperature difference of each local area, areas with a difference exceeding 0.8℃ are marked as abnormal, and 12 adjacent abnormal areas are merged to form 3 connected areas, which together with 5 independent abnormal areas form an abnormal area set.

[0102] Next, the set of abnormal areas is time-aligned with the real-time collected surface temperature sequence. When generating the surface temperature reference sequence, the timestamp of the corrected temperature sequence is used as the benchmark to match the timestamp of the real-time collected surface temperature sequence. For sampling points with inconsistent timestamps, linear interpolation is used to complete the sequence, ensuring that the time resolution of the two sequences is completely consistent.

[0103] The spatial coordinates of the anomalous region set are correlated with the aligned surface temperature sequence. The temperature value of each anomalous region at each time point is extracted. At the same time, normal regions without thermal anomalies around the anomalous regions are selected as references, and their corresponding temperature values ​​at the time points are extracted. The temperature data of the anomalous regions and the reference regions are integrated in chronological order to generate a surface temperature reference sequence. This sequence contains temperature information that varies in both spatial location and time, providing a spatiotemporally synchronized data source for the subsequent establishment of heat exchange models.

[0104] It is worth noting that when establishing the heat exchange model based on the surface temperature reference sequence, the model comprehensively considers the three core heat exchange mechanisms of soil: conduction, convection, and radiation. Built upon the fundamental laws of heat transfer, the model incorporates over 800 sets of measured heat exchange data for its training set. These data not only originate from long-term monitoring records at the storage tank site but also incorporate simulation tests in a controlled sandbox in the laboratory, comprehensively covering soil state changes under different seasonal transitions and various complex meteorological conditions, ensuring the model's strong robustness to external environmental disturbances. The dataset is divided into training and validation sets in a 7:3 ratio, and a BP neural network is used to train the model with a three-layer network structure. The model's input layer is precisely set to six core physical and environmental features that directly affect the heat exchange process, specifically: surface temperature, soil thermal conductivity, soil volumetric moisture content (humidity), ambient wind speed (affecting convective heat transfer), solar radiation intensity (affecting radiative heat transfer), and cover layer thickness. Soil volumetric moisture content is obtained in real-time by pre-installing multiple soil temperature and humidity sensors in the tank's cover layer. Soil thermal conductivity is calculated using empirical formulas based on soil type and moisture content, or obtained in real-time by thermal pulse probes. The hidden layer has 12 nodes, and the output layer contains predicted values ​​of soil thermophysical components. The activation function is ReLU, the loss function is mean squared error, and the learning rate is set to 0.01. Training stops when the loss value is consistently less than 0.005. By inputting the surface temperature reference sequence into the trained heat exchange model, the model fits the change curve of the soil thermophysical components. By subtracting the predicted value of the curve from the temperature data corresponding to the set of anomalous areas, the thermophysical effects of soil heat conduction, convection, and radiation can be removed, and a preliminary pure thermal anomaly signal can be obtained. At the same time, the signal is subjected to min-max normalization and mapped to the [0,1] interval. The normalization reference range is the extreme value range of the preliminary thermal anomaly signal.

[0105] For example, by inputting the surface temperature reference sequence into the trained heat exchange model, the soil thermophysical component curve is fitted, and a preliminary pure thermal anomaly signal is obtained after difference calculation. After normalization, the thermal anomaly characteristics caused by tank defects are effectively highlighted.

[0106] Next, the preliminary pure thermal anomaly signal is subjected to spectral filtering to remove residual low-frequency components. A Butterworth high-pass filter is used for frequency domain filtering of the preliminary pure thermal anomaly signal, with a cutoff frequency set to 0.01Hz. This frequency is set based on the low-frequency characteristics of soil thermophysical components. Residual interference caused by soil thermal capacity and thermal resistance is mostly concentrated in the low-frequency band below 0.01Hz. High-pass filtering can effectively remove these residual low-frequency components, retaining the mid-to-high-frequency thermal anomaly signal caused by tank defects. First, a Fourier transform is performed on the preliminary pure thermal anomaly signal to convert the time-domain signal into a frequency-domain signal. Components below the cutoff frequency are set to zero in the frequency domain. Then, an inverse Fourier transform is used to restore the signal to the time domain, obtaining the filtered thermal anomaly signal. The amplitude of the filtered signal is corrected to restore the signal amplitude to the actual physical dimensions. At the same time, a three-point moving average is used for smoothing to eliminate numerical fluctuations during the filtering process, finally obtaining a pure thermal anomaly signal without soil thermophysical influence.

[0107] For example, the initial pure thermal anomaly signal is subjected to Butterworth high-pass filtering with a cutoff frequency of 0.01Hz. After frequency domain transformation, component zeroing, inverse transformation and smoothing, the final pure thermal anomaly signal is obtained. This signal completely removes the influence of soil thermophysical effects and accurately reflects the thermal anomaly caused by the damage to the tank's anti-corrosion layer.

[0108] In step S106, the pure thermal anomaly signal is input into a preset thermal distribution restoration model to restore the thermal distribution pattern caused by the damage to the anti-corrosion layer. The thermal distribution pattern is then mapped to a preset three-dimensional coordinate system of the storage tank, and the coordinates of the abnormal area are extracted to determine the preliminary boundary of the defect detection.

[0109] It should be noted that when inputting the pure thermal anomaly signal into the preset thermal distribution reconstruction model to reconstruct the thermal distribution pattern caused by the damage to the anti-corrosion layer, the thermal distribution reconstruction model is constructed based on the three-dimensional steady-state heat conduction equation of heat transfer. It is trained and optimized by combining the thermophysical parameters of the tank's anti-corrosion layer and the covering soil medium. The training set contains over 900 sets of measured and simulated thermal distribution data under different anti-corrosion layer damage sizes, locations, and covering soil thicknesses. The measured data is obtained through a thermocouple array embedded in the tank surface. The simulated data uses finite element software (such as COMSOL) to establish a tank-soil heat conduction model, setting different damage parameters to calculate the correspondence between surface temperature anomalies and the thermal distribution on the tank surface. After the training set is constructed, it is divided into a training set and a validation set in an 8:2 ratio. The model adopts a U-Net network structure. The input layer is a two-dimensional feature map of the pure thermal anomaly signal, and the output layer is a thermal distribution pattern map of the tank surface. The activation function is Sigmoid, the loss function is mean squared error, and the learning rate is set to 0.005. Training stops when the loss value fluctuation is less than 0.001 for 20 consecutive rounds to ensure reconstruction accuracy.

[0110] Subsequently, the pure thermal anomaly signal normalized to the [0,1] interval is input into the trained model. The model, through heat conduction inversion calculations, reconstructs the actual heat distribution pattern transmitted from the damaged area of ​​the tank's anti-corrosion layer to the ground surface, including core features such as the intensity, range, and diffusion direction of the thermal anomaly. The output heat distribution pattern is presented in the form of a two-dimensional heat map, intuitively reflecting the thermal field changes caused by the damage. For example, inputting the extracted pure thermal anomaly signal from the eastern side of the storage tank into the model reconstructs an elliptical heat distribution pattern in that area, with the highest thermal anomaly intensity in the core area, gradually decreasing towards the periphery, highly consistent with the actual thermal diffusion characteristics of the damaged anti-corrosion layer.

[0111] Next, the heat distribution pattern is mapped onto a preset 3D coordinate system of the storage tank at a preset resolution. When forming a set of spatial anomalies, the 3D coordinate system is established with the center of the tank bottom as the origin, the tank axis as the X-axis, the radial direction as the Y-axis, and the height direction as the Z-axis. The parameters of the coordinate system match the actual size of the storage tank 1:1 to ensure the spatial accuracy of the mapping. The heat distribution pattern map is mapped in the form of a grid, and the grid size is the mapping resolution. For normal areas, a uniform grid of 0.1 meters / pixel is used; for critical areas such as the tank head and welds, the resolution is increased to 0.05 meters / pixel by secondary interpolation and resampling to enhance local details.

[0112] The thermal distribution pattern map is mapped pixel by pixel to the three-dimensional coordinate system of the storage tank at this resolution. Pixels in the map whose thermal anomaly intensity values ​​exceed a preset threshold are extracted. This threshold is set based on the mean and standard deviation of the thermal distribution pattern. The mean + 1 standard deviation is taken as the benchmark. Pixels exceeding this value are judged as anomalies. The three-dimensional coordinate information of all anomalies is extracted and integrated to form a spatial anomaly point set. Each point in the set contains X, Y, Z three-dimensional coordinates and the corresponding thermal anomaly intensity value.

[0113] For example, the restored elliptical heat distribution pattern is mapped to the three-dimensional coordinate system of the storage tank at a ratio of 0.1 meters / pixel. 86 pixels with heat anomaly intensity exceeding the threshold are extracted and integrated to obtain a set of spatial anomaly points containing three-dimensional coordinates and intensity values, which accurately correspond to the heat anomaly area on the surface of the tank.

[0114] Finally, cluster analysis was performed on the spatial anomaly set to extract the geometric contours of continuous anomaly regions and obtain the preliminary boundaries for defect detection. The DBSCAN density clustering algorithm was then used to cluster the spatial anomaly set. This algorithm can effectively identify continuous anomaly regions of arbitrary shapes and can remove isolated noise anomalies. The algorithm's neighborhood radius was set to 0.5 meters, and the minimum number of contained points was set to 5. This parameter combination was not empirically determined but strictly based on the physical thermal anomaly diffusion law of the damaged anti-corrosion layer of the soil-covered storage tank and the initial spatial mapping resolution. The neighborhood radius represents the connectivity distance of the anomaly points, and the minimum number of contained points represents the minimum number of points required to constitute a continuous anomaly region. For areas with thicker soil cover and a larger thermal anomaly diffusion range, the neighborhood radius can be increased to 0.8 meters, while the minimum number of contained points remains unchanged at 5.

[0115] This algorithm clusters spatial anomaly points, removes isolated noise points, aggregates connected anomalies into one or more continuous anomaly regions, extracts the outermost boundary points of each continuous anomaly region, and connects the boundary points in spatial order to form a closed geometric contour. This contour is the preliminary boundary for defect detection. The preliminary boundary is stored in the form of a three-dimensional coordinate sequence, which clearly defines the approximate spatial range of the anti-corrosion layer damage.

[0116] For example, DBSCAN density clustering was performed on 86 spatial anomalies, 3 isolated noise points were removed, and the remaining 83 connected anomalies were aggregated into a continuous anomaly region. The outer 22 boundary points were extracted and connected to form a closed elliptical contour, which is the preliminary boundary for detecting defects in the corrosion protection layer on the east side of the storage tank.

[0117] In step S107, the step of applying gradient calculation to the pure thermal anomaly signal within the preliminary boundary-defined analysis range, determining the correspondence between the gradient peak value and the thermal signal attenuation mode, and determining the precise location information of the anti-corrosion layer damage based on the correspondence includes:

[0118] Gradient calculation is performed on the pure thermal anomaly signal within the preliminary boundary-defined analysis range to extract the spatial gradient intensity distribution and obtain a gradient amplitude value sequence;

[0119] Multi-scale sampling analysis is performed on the gradient amplitude value sequence along the radial direction to obtain a set of correspondences between the gradient peak position and the thermal signal intensity attenuation gradient;

[0120] Candidate damage boundary points at the radial gradient peak positions where the signal strength attenuation rate exceeds a preset attenuation threshold and the peak amplitude is greater than a preset amplitude threshold are retained to form the positioning contour of the anti-corrosion layer damage.

[0121] The positioning contour is superimposed with the attenuation characteristics of the pure thermal anomaly signal for verification, thereby obtaining accurate positioning information of the anti-corrosion layer damage.

[0122] It should be noted that, firstly, based on the preliminary boundary defining the analysis range, when performing gradient calculations on the pure thermal anomaly signals within the range, the analysis range is delineated using the preliminary boundary of defect detection as the boundary. Invalid data outside the boundary is eliminated, and gradient calculations are performed only on the pure thermal anomaly signals within the boundary. The Sobel gradient operator is selected to perform two-dimensional convolution processing on the signals. This operator can simultaneously extract gradient components in both horizontal and vertical directions, accurately capturing the spatial edge features of the thermal anomaly signals. The convolution window is set to 3×3 to balance the accuracy and efficiency of gradient calculation. After calculating the gradient values ​​in the horizontal and vertical directions respectively, the gradient amplitude value of each spatial sampling point is obtained through vector synthesis. All gradient amplitude values ​​are integrated according to the spatial position order of the three-dimensional coordinate system of the storage tank to form a gradient amplitude value sequence. At the same time, the sequence is subjected to min-max normalization processing, mapping it to the [0,1] interval. The normalization reference range is the extreme value of the gradient amplitude of the sequence, eliminating the influence of numerical magnitude differences on subsequent analysis.

[0123] Next, multi-scale sampling analysis was performed on the gradient amplitude value sequence along the radial direction to obtain the correspondence between the gradient peak position and the thermal signal intensity attenuation gradient. The geometric center of the thermal anomaly region enclosed by the initial boundary was used as the radial sampling origin. Three sampling scales of 0.2 meters, 0.5 meters, and 1.0 meters were set along the radial direction. This multi-scale setting was combined with the diffusion characteristics of the thermal anomaly caused by the corrosion layer damage. The small scale captured fine gradient changes, the medium scale located the core attenuation boundary, and the large scale verified the overall attenuation trend. In areas with thicker overburden layers and a larger thermal anomaly diffusion range, the maximum sampling scale could be increased to 1.5 meters. Subsequences of the gradient amplitude value sequence were extracted along each radial direction at different scales. The gradient peak position in each subsequence was identified, and the thermal signal intensity attenuation rate at the corresponding position was calculated, i.e., the decrease in thermal anomaly signal intensity per unit radial distance. Each gradient peak position was correlated with its corresponding thermal signal intensity attenuation gradient. All radial and all-scale correlated data were integrated to form a correspondence set between the gradient peak position and the thermal signal intensity attenuation gradient. Each data unit in the set contains four core pieces of information: sampling scale, radial direction, peak position, and attenuation gradient.

[0124] Candidate damage boundary points where the signal intensity attenuation rate at the radial gradient peak location exceeds a preset attenuation threshold and the peak amplitude is greater than a preset amplitude threshold are retained to form the positioning contour of the anti-corrosion layer damage. It should be noted that the attenuation threshold is set based on the difference in thermal signal attenuation characteristics between anti-corrosion layer damage and natural soil thermal diffusion. At the actual damage boundary, a sudden change in the medium causes a surge in the thermal signal intensity attenuation rate, typically exceeding 0.15℃ / m, while the attenuation rate of natural thermal diffusion in normal soil is mostly lower than this value. Therefore, the basic threshold is set at 0.15℃ / m. Sandy soils have fast thermal conductivity and a high natural attenuation rate, so the threshold can be increased to 0.2℃ / m; clay soils have slow thermal conductivity and a low natural attenuation rate, so the threshold can be decreased to 0.1℃ / m. The amplitude threshold is taken as 2.5 times the mean of the normalized gradient amplitude value sequence. This multiple is set based on the statistical distribution characteristics of historical background thermal noise. Taking 2.5 times the mean effectively filters out random noise peaks caused by minor environmental disturbances, accurately distinguishing the true damage gradient peaks. For example, when the normalized mean is 0.2, the amplitude determination threshold is 0.5.

[0125] Subsequently, gradient peak positions that simultaneously satisfy both the attenuation rate exceeding the attenuation threshold and the peak amplitude exceeding the amplitude threshold are selected from the corresponding relationship set. These positions are used as candidate damage boundary points. All candidate damage boundary points are connected sequentially according to their spatial positions in the three-dimensional coordinate system of the storage tank to form a closed anti-corrosion layer damage location contour. If there are discontinuities among the candidate points, linear interpolation is used to complete the continuity of the contour. For example, 18 candidate points that simultaneously satisfy the dual threshold conditions are selected from 24 gradient peak positions. After connecting them according to their spatial positions and interpolating to complete the discontinuities, an approximately elliptical anti-corrosion layer damage location contour is formed.

[0126] Finally, the attenuation characteristics of the positioning contour and the pure thermal anomaly signal are superimposed for verification to obtain accurate positioning information of the anti-corrosion layer damage. First, the average intensity of the pure thermal anomaly signal inside the positioning contour is calculated, with a basic acceptable threshold set at 0.08℃. This threshold setting comprehensively considers the noise equivalent temperature difference baseline of the infrared sensor and the effective thermal anomaly minimum value after the heat from minor damage penetrates the overburden layer to reach the surface. For critical areas of storage tanks with higher detection sensitivity requirements, the threshold can be lowered to 0.06℃. Then, a 0.3-meter buffer zone is set outside the positioning contour, and the average gradient of the thermal signal intensity within the buffer zone is calculated, with a basic acceptable threshold set at 0.04℃ / meter. This threshold characterizes the rapid attenuation characteristics of the thermal signal outside the contour, verifying that the contour is the true damage boundary. The width of the buffer zone can be adjusted within the range of 0.2-0.4 meters according to the thickness of the overburden layer.

[0127] If the positioning contour meets the dual verification conditions of the internal average anomaly intensity exceeding the threshold and the average gradient in the outer buffer zone being lower than the threshold, then the contour is confirmed as the true boundary of the anti-corrosion layer damage. The three-dimensional coordinate sequence, the area enclosed by the contour, the internal average thermal anomaly intensity, and other information are extracted and integrated to form accurate anti-corrosion layer damage positioning information. If the verification conditions are not met, the sampling scale and judgment threshold are readjusted, and the gradient analysis and boundary point screening process is repeated.

[0128] For example, when verifying the aforementioned elliptical positioning profile, the measured internal average abnormal intensity was 0.11℃ and the average gradient within the 0.3-meter buffer zone on the outside was 0.02℃ / meter, both of which met the verification conditions. The three-dimensional coordinates of the profile and the 1.2-square-meter enclosed area were extracted to form the final accurate positioning information of the anti-corrosion layer damage.

[0129] In step S108, the location information is fused with pre-acquired natural thermal insulation attribute data of the storage tank, and input into a preset fire risk assessment model. A corresponding risk probability value is output, and a defect warning report is generated based on the extent to which the risk probability value exceeds a preset probability judgment threshold. This includes:

[0130] Based on the positioning information, the distribution data of the natural thermal insulation properties of the tank wall are obtained, and the spatial distribution feature set of the thermal insulation weakening area is obtained through spatial superposition calculation.

[0131] The set of spatial distribution features is matched with the heat conduction parameters in the preset fire risk assessment model, and the set of probability values ​​for thermal runaway propagation at each defect location is obtained through multi-layer thermal risk propagation simulation.

[0132] Extract the corresponding defect locations from the probability value set that exceed the preset probability judgment threshold, and determine the comprehensive safety impact level value;

[0133] By associating the comprehensive safety impact level value with the overall structural stability parameters of the storage tank, a defect warning report is generated that includes the defect location, risk level, and handling suggestions.

[0134] It should be noted that when obtaining the distribution data of the natural thermal insulation properties of the tank wall based on the positioning information, and obtaining the spatial distribution feature set of the thermal insulation weakening area through spatial superposition calculation, the basic database of the wall thermal insulation properties in the tank design stage is first retrieved. This database contains basic parameters such as the material, thickness, and thermal conductivity of the thermal insulation layer in each area of ​​the tank. Then, the precise positioning information of the anti-corrosion layer damage is matched with the spatial coordinates of the tank wall to extract the original data of the natural thermal insulation properties of the damage location and the surrounding area.

[0135] Subsequently, a spatial overlay algorithm was used to overlay the damaged location contour with the thermal insulation attribute distribution layer to calculate the change in thermal insulation coefficient caused by the absence of the anti-corrosion layer in the damaged area. The basic value of the thermal insulation coefficient of the normal tank wall was set to 0.045 W / (m·K). The thermal insulation coefficient of the damaged area of ​​the anti-corrosion layer will increase with the degree of damage. All areas with thermal insulation coefficients exceeding the basic value were extracted, and their spatial coordinates, thermal insulation coefficient changes, and damaged areas were integrated to form a spatial distribution feature set of the thermal insulation weakening area. At the same time, the thermal insulation coefficient data in the set were subjected to min-max normalization and mapped to the [0,1] interval. The normalization reference range is the extreme value range of the thermal insulation coefficient of the tank wall.

[0136] It is worth noting that when matching the spatial distribution feature set with the heat conduction parameters in the pre-set fire risk assessment model, the fire risk assessment model is constructed based on heat transfer and fire dynamics. The training set contains over 1000 sets of measured and simulated thermal risk data for different degrees of corrosion protection layer damage, insulation degradation levels, and tank medium temperatures. These data are divided into training and validation sets in an 8:2 ratio. The model employs a deep neural network structure, with the input layer precisely including eight core physical features such as insulation change, damaged area, medium temperature, and soil cover thickness. To prevent overfitting, a Dropout layer is introduced, with two hidden layers and 16 nodes per layer. The ReLU activation function is used, and the Sigmoid output layer outputs probabilities. During training, the Adam optimizer is used, with cross-entropy as the loss function and a learning rate of 0.008. Training stops when the loss value fluctuation is less than 0.001 for 30 consecutive rounds.

[0137] Normalized data from the spatial distribution feature set is input into the model and matched with preset heat conduction parameters. Different degrees of insulation degradation correspond to different correction coefficients for heat conduction parameters; the higher the normalized value of the insulation coefficient, the larger the correction coefficient. After matching, the model performs a three-layer heat risk propagation simulation, simulating the heat propagation process in the insulation layer, tank wall, and soil, respectively, and calculating the probability value of heat runaway propagation at each defect location. The probability value ranges from 0 to 1, and the probability values ​​of all defect locations are integrated to form a probability value set. For example, inputting the insulation weakening feature data of the damaged area on the east side into the model, after heat conduction parameter matching and three-layer propagation simulation, the probability value of heat runaway propagation at this location is 0.78. At the same time, the probability values ​​of two other minor damages in the tank are calculated to be 0.45 and 0.52, forming a probability value set.

[0138] Subsequently, the locations of defects exceeding a preset probability threshold are extracted from the probability value set. When determining the comprehensive safety impact level, the probability threshold is set based on industry standards for tank fire risk prevention and control and historical accident data. When the probability value of thermal runaway propagation exceeds 0.6, the risk of fire increases significantly; therefore, the basic threshold is set to 0.6. For tanks storing highly volatile liquefied hydrocarbon media, fire prevention and control requirements are higher, and the threshold can be lowered to 0.55. For tanks with thick overburden layers and high thermal diffusion difficulty, the threshold can be raised to 0.65. Those skilled in the art will understand that the threshold can be adjusted within the range of 0.55 to 0.65 depending on the type of tank medium and the overburden conditions.

[0139] Defect locations exceeding the threshold are extracted from the probability value set. Based on these probability values, a comprehensive safety impact level is assigned: Level I (0.6-0.75 probability value, corresponding to medium risk); Level II (0.76-0.90 probability value, corresponding to high risk); and Level III (over 0.90 probability value, corresponding to extremely high risk). Each level is assigned a corresponding level value: 1 for medium risk, 2 for high risk, and 3 for extremely high risk, forming the comprehensive safety impact level value for each defect location. For example, if three defect locations exceeding the 0.6 threshold are extracted from the probability value set, two of them (0.78 and 0.85 probability values, corresponding to high risk, level 2) and one (0.62 probability value, corresponding to medium risk, level 1) are identified.

[0140] When generating a defect warning report that includes defect location, risk level, and handling suggestions by linking the comprehensive safety impact level value with the overall structural stability parameters of the storage tank, the overall structural stability parameters of the storage tank are first retrieved. These parameters include key parameters such as the remaining margin of tank wall thickness, weld stress concentration factor, foundation settlement, and tank material fatigue. Each parameter is assigned a weighted weight according to its impact on structural stability. The weight of the remaining margin of tank wall thickness is set to 0.3, the weight of the weld stress concentration factor is set to 0.3, the weight of the foundation settlement is set to 0.2, and the weight of the tank material fatigue is set to 0.2, with a total weight of 1. This weight allocation is based on the structural mechanical characteristics of the storage tank and the statistical laws of accidents, and can be appropriately adjusted according to the service life of the storage tank.

[0141] Next, the comprehensive safety impact level of each defect location is weighted and integrated with the structural stability parameters of the surrounding area to obtain the comprehensive risk value of each defect location. Then, combined with the spatial location of the defect, it is determined whether the defect is located in critical structural areas such as tank welds, heads, and supports. The comprehensive risk value of defects in critical areas needs to be increased by 20%. Finally, the three-dimensional coordinates, thermal runaway propagation probability values, comprehensive safety impact levels, comprehensive risk values, and structural stability correlation analysis results of all defect locations are integrated to formulate targeted handling recommendations for defects of different risk levels. For medium-risk defects, it is recommended to strengthen high-frequency monitoring and formulate maintenance plans; for high-risk defects, it is recommended to immediately shut down and implement emergency reinforcement and repair; for extremely high-risk defects, it is recommended to immediately isolate and replace the damaged tank area. All information is organized in a standardized format to generate a defect early warning report. High-risk and extremely high-risk defects are highlighted in the report to facilitate rapid handling by maintenance personnel.

[0142] In summary, this invention discloses a method for detecting external defects in soil-covered liquefied hydrocarbon storage tanks. The method includes acquiring surface temperature distribution data above the soil cover layer and continuously scanning to generate an initial temperature sequence; separating mid-to-high frequency disturbance components, determining soil heat capacity delay effect parameters, and correcting the temperature sequence; removing soil thermophysical influences when the temperature difference threshold is exceeded to obtain a pure thermal anomaly signal; restoring the heat distribution pattern and mapping it to the tank's three-dimensional coordinate system to determine the preliminary defect boundary; accurately locating the damage site of the anti-corrosion layer through gradient calculation; and integrating insulation property data with a fire risk assessment model to generate an early warning report containing the defect location, risk level, and treatment recommendations. This method achieves accurate and efficient detection of external defects in storage tanks, meeting the requirements for high precision and continuity in safety monitoring.

[0143] Reference Figure 2 The second embodiment of the present invention provides a surface defect detection system for a soil-covered liquefied hydrocarbon storage tank, comprising:

[0144] The temperature distribution acquisition module is used to acquire temperature distribution data of the surface area above the soil cover layer;

[0145] The temperature sequence generation module is used to convert the temperature distribution data into a two-dimensional temperature field matrix, extract transient temperature disturbance information, and generate an initial temperature sequence.

[0146] The delay parameter determination module is used to perform Fourier transform on the initial temperature sequence, extract the corresponding spectrum, separate the mid-to-high frequency disturbance components, and determine the delay effect parameters caused by soil heat capacity based on the amplitude attenuation characteristics and phase lag law of the mid-to-high frequency disturbance components.

[0147] A time-domain correction module is used to perform time-domain correction on the initial temperature sequence based on the delay effect parameter to obtain a corrected temperature sequence.

[0148] An anomaly signal extraction module is used to calculate the temperature difference between a local area and the surrounding area in the corrected temperature sequence. If the temperature difference exceeds a preset temperature difference judgment threshold, the corresponding temperature distribution data is aligned with the real-time collected surface temperature sequence, and a heat exchange model is established to remove the soil thermophysical influence and obtain a pure thermal anomaly signal.

[0149] The preliminary boundary determination module is used to input the pure thermal anomaly signal into a preset thermal distribution restoration model, restore the thermal distribution pattern caused by the damage to the anti-corrosion layer, map the thermal distribution pattern to a preset three-dimensional coordinate system of the storage tank, extract the coordinates of the abnormal area, and determine the preliminary boundary of the defect detection.

[0150] The precise positioning module is used to apply gradient calculation to the pure thermal anomaly signal within the preliminary boundary-defined analysis range, determine the correspondence between the gradient peak value and the thermal signal attenuation mode, and determine the precise positioning information of the anti-corrosion layer damage based on the correspondence.

[0151] The early warning report generation module is used to integrate the location information with the pre-acquired natural thermal insulation attribute data of the storage tank, input a preset fire risk assessment model, output the corresponding risk probability value, and generate a defect early warning report based on the extent to which the risk probability value exceeds the preset probability judgment threshold.

[0152] It should be noted that the soil-covered liquefied hydrocarbon storage tank appearance defect detection system provided in this embodiment of the invention is used to execute all the process steps of the soil-covered liquefied hydrocarbon storage tank appearance defect detection method in the above embodiment. The working principle and beneficial effect of the two are one-to-one, so they will not be described again.

[0153] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0154] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for detecting appearance defects in a soil-covered liquefied hydrocarbon storage tank, characterized in that, include: Obtain temperature distribution data of the surface area above the overburden layer; The temperature distribution data is converted into a two-dimensional temperature field matrix, and transient temperature perturbation information is extracted to generate an initial temperature sequence. The initial temperature sequence is subjected to Fourier transform to extract the corresponding spectrum and separate the mid-to-high frequency perturbation components. Based on the amplitude attenuation characteristics and phase lag law of the mid-to-high frequency perturbation components, the delay effect parameters caused by soil heat capacity are determined. The initial temperature sequence is time-domain corrected based on the delay effect parameter to obtain the corrected temperature sequence; The temperature difference between the local area and the surrounding area in the corrected temperature sequence is calculated. If the temperature difference exceeds the preset temperature difference judgment threshold, the corresponding temperature distribution data is aligned with the real-time collected surface temperature sequence. The soil thermophysical influence is removed by establishing a heat exchange model to obtain a pure thermal anomaly signal. The pure thermal anomaly signal is input into a preset thermal distribution restoration model to restore the thermal distribution pattern caused by the damage to the anti-corrosion layer. The thermal distribution pattern is mapped to a preset three-dimensional coordinate system of the storage tank, and the coordinates of the abnormal area are extracted to determine the preliminary boundary of the defect detection. Gradient calculation is applied to the pure thermal anomaly signal within the preliminary boundary-defined analysis range, and the correspondence between the gradient peak value and the thermal signal attenuation mode is determined. Based on the correspondence, the precise location information of the anti-corrosion layer damage is determined. The system integrates the location information with the pre-acquired natural thermal insulation attribute data of the storage tank, inputs it into a preset fire risk assessment model, outputs the corresponding risk probability value, and generates a defect warning report based on the extent to which the risk probability value exceeds the preset probability judgment threshold. The process of performing a Fourier transform on the initial temperature sequence, extracting the corresponding spectrum, separating the mid-to-high frequency perturbation components, and determining the delay effect parameters caused by soil heat capacity based on the amplitude attenuation characteristics and phase lag laws of the mid-to-high frequency perturbation components includes: The initial temperature sequence is segmented, and the frequency spectrum distribution of each time period is extracted by Fourier transform. The low-frequency background component and the mid-to-high-frequency disturbance component in the frequency spectrum distribution are separated to obtain the mid-to-high-frequency thermal signal; The mid-to-high frequency thermal signal is subdivided into frequency bands, and the energy distribution characteristics of sub-bands with an exponential decay trend over time are extracted and integrated in chronological order to form an amplitude decay sequence. The amplitude decay sequence is subjected to exponential fitting to calculate the decay rate constant and phase lag time value, forming a set of characteristic parameters to determine the delay effect parameters. The process of applying gradient calculations to the pure thermal anomaly signal within the preliminary boundary-defined analysis range, determining the correspondence between the gradient peak value and the thermal signal attenuation mode, and determining the precise location information of the anti-corrosion layer damage based on the correspondence includes: Gradient calculation is performed on the pure thermal anomaly signal within the preliminary boundary-defined analysis range to extract the spatial gradient intensity distribution and obtain a gradient amplitude value sequence; Multi-scale sampling analysis is performed on the gradient amplitude value sequence along the radial direction to obtain a set of correspondences between the gradient peak position and the thermal signal intensity attenuation gradient; Candidate damage boundary points at the radial gradient peak positions where the signal strength attenuation rate exceeds a preset attenuation threshold and the peak amplitude is greater than a preset amplitude threshold are retained to form the positioning contour of the anti-corrosion layer damage. The positioning contour is superimposed with the attenuation characteristics of the pure thermal anomaly signal for verification, thereby obtaining accurate positioning information of the anti-corrosion layer damage.

2. The method for detecting appearance defects in a soil-covered liquefied hydrocarbon storage tank according to claim 1, characterized in that, The acquisition of temperature distribution data of the surface area above the overburden layer includes: A multispectral infrared sensor array was used to collect full-coverage data of the surface area above the soil cover layer, obtaining raw temperature data and spatial location information. The original temperature data is denoised to obtain denoised temperature data. The denoised temperature data is integrated with the spatial location information to form surface temperature distribution data.

3. The method for detecting appearance defects in a soil-covered liquefied hydrocarbon storage tank according to claim 1, characterized in that, The step of converting the temperature distribution data into a two-dimensional temperature field matrix and extracting transient temperature perturbation information to generate an initial temperature sequence includes: The temperature distribution data is stored as a two-dimensional temperature field matrix containing time and spatial information; The background temperature change component caused by soil thermal inertia is separated from the two-dimensional temperature field matrix by time difference filtering, and transient temperature disturbance information is extracted. The transient temperature disturbance information is integrated in chronological order to generate an initial temperature sequence.

4. The method for detecting appearance defects in a soil-covered liquefied hydrocarbon storage tank according to claim 1, characterized in that, The step of performing time-domain correction on the initial temperature sequence based on the delay effect parameter to obtain the corrected temperature sequence includes: The initial temperature sequence is segmented in the time domain to obtain multiple time windows; The delay effect parameter is applied to the data within the multiple time windows for compensation, generating an intermediate temperature sequence with preliminary time-domain correction; Fourier transform and frequency domain analysis were performed on the intermediate temperature sequence to extract the main frequency components within each time window, and the low-frequency smooth components related to thermal resistance characteristics were separated to obtain the frequency characteristic distribution. The frequency characteristic distribution is weighted and calculated, and the negative correlation value of the proportion of low-frequency energy is used as the weight to generate a weighted characteristic distribution. The intermediate temperature sequence is then corrected by applying the weighted feature distribution to obtain the corrected temperature sequence.

5. The method for detecting appearance defects in a soil-covered liquefied hydrocarbon storage tank according to claim 1, characterized in that, The calculation of the temperature difference between the local area and the surrounding area in the corrected temperature sequence, if the temperature difference exceeds a preset temperature difference judgment threshold, aligns the corresponding temperature distribution data with the real-time acquired surface temperature sequence, and obtains a pure thermal anomaly signal by establishing a heat exchange model to remove the soil thermophysical influence, including: The corrected temperature sequence is divided into local regions to obtain multiple sets of local temperature difference points; Based on the set of local temperature difference points, the temperature difference between each local area and the surrounding area is calculated. If the temperature difference exceeds the preset temperature difference judgment threshold, it is marked as an abnormal area, forming an abnormal area set. The set of abnormal regions is time-aligned with the real-time collected surface temperature sequence to generate a surface temperature reference sequence. A heat exchange model is established based on the surface temperature reference sequence, and the temperature data corresponding to the set of anomalous areas is input into the heat exchange model to separate the soil thermophysical components and obtain a preliminary pure thermal anomaly signal. The preliminary pure thermal anomaly signal is subjected to spectral filtering to remove residual low-frequency components, resulting in the final pure thermal anomaly signal.

6. The method for detecting appearance defects in a soil-covered liquefied hydrocarbon storage tank according to claim 1, characterized in that, The system integrates the location information with pre-acquired natural thermal insulation attribute data of the storage tank, inputs it into a preset fire risk assessment model, outputs a corresponding risk probability value, and generates a defect early warning report based on the extent to which the risk probability value exceeds a preset probability judgment threshold, including: Based on the positioning information, the distribution data of the natural thermal insulation properties of the tank wall are obtained, and the spatial distribution feature set of the thermal insulation weakening area is obtained through spatial superposition calculation. The set of spatial distribution features is matched with the heat conduction parameters in the preset fire risk assessment model, and the set of probability values ​​for thermal runaway propagation at each defect location is obtained through multi-layer thermal risk propagation simulation. Extract the corresponding defect locations from the probability value set that exceed the preset probability judgment threshold, and determine the comprehensive safety impact level value; By associating the comprehensive safety impact level value with the overall structural stability parameters of the storage tank, a defect early warning report is generated, which includes the defect location, risk level, and handling suggestions.

7. A system for detecting external defects in soil-covered liquefied hydrocarbon storage tanks, characterized in that, For implementing the method as described in any one of claims 1-6, comprising: The temperature distribution acquisition module is used to acquire temperature distribution data of the surface area above the soil cover layer; The temperature sequence generation module is used to convert the temperature distribution data into a two-dimensional temperature field matrix, extract transient temperature disturbance information, and generate an initial temperature sequence. The delay parameter determination module is used to perform Fourier transform on the initial temperature sequence, extract the corresponding spectrum, separate the mid-to-high frequency disturbance components, and determine the delay effect parameters caused by soil heat capacity based on the amplitude attenuation characteristics and phase lag law of the mid-to-high frequency disturbance components. A time-domain correction module is used to perform time-domain correction on the initial temperature sequence based on the delay effect parameter to obtain a corrected temperature sequence. An anomaly signal extraction module is used to calculate the temperature difference between a local area and the surrounding area in the corrected temperature sequence. If the temperature difference exceeds a preset temperature difference judgment threshold, the corresponding temperature distribution data is aligned with the real-time collected surface temperature sequence, and a heat exchange model is established to remove the soil thermophysical influence and obtain a pure thermal anomaly signal. The preliminary boundary determination module is used to input the pure thermal anomaly signal into a preset thermal distribution restoration model, restore the thermal distribution pattern caused by the damage to the anti-corrosion layer, map the thermal distribution pattern to a preset three-dimensional coordinate system of the storage tank, extract the coordinates of the abnormal area, and determine the preliminary boundary of the defect detection. The precise positioning module is used to apply gradient calculation to the pure thermal anomaly signal within the preliminary boundary-defined analysis range, determine the correspondence between the gradient peak value and the thermal signal attenuation mode, and determine the precise positioning information of the anti-corrosion layer damage based on the correspondence. The early warning report generation module is used to integrate the location information with the pre-acquired natural thermal insulation attribute data of the storage tank, input a preset fire risk assessment model, output the corresponding risk probability value, and generate a defect early warning report based on the extent to which the risk probability value exceeds the preset probability judgment threshold.

Images