A crack prediction method and system for a high-pile wharf

Abstract

This invention discloses a method and system for predicting cracks in high-pile wharves, relating to the field of civil engineering technology. The method includes: acquiring sample crack image data and sample sonar data of the high-pile wharf; performing crack identification and parameter measurement on the sample crack image data to obtain sample crack identification results; extracting features from the sample sonar data to obtain sample sonar data identification results; determining the sample level label of the crack based on the width and depth of the cracks in the imported three-dimensional digital model and preset judgment rules; training an initial Long Short-Term Memory (LSTM) network to obtain an LSM network, and using the LSM network to predict the crack image data and sonar data to obtain the extended prediction results for all cracks in the high-pile wharf. This method improves the accuracy of crack prediction in high-pile wharves.

Description

Technical Field

[0001] This application relates to the field of civil engineering technology, and in particular to a method and system for predicting cracks in high-pile wharves. Background Technology

[0002] The area where the wharf is located, where water and land meet, is a region where the interactions between the hydrosphere, lithosphere, atmosphere, and biosphere are most frequent, active, and intense. It is a dynamic zone with constantly changing biological, chemical, and geological characteristics, possessing environmental features of both marine and terrestrial environments, and is also the area most frequently and severely affected by river and marine natural disasters. The wharf is subjected to strong dynamic factors over a long period, inevitably causing cracks in concrete components such as piles, beams, and panels. This is especially true during operation, when long-term or instantaneous loads such as cargo stacking and ship collisions exceed the bearing capacity of the concrete components, significantly accelerating crack propagation. Furthermore, in marine or saline-alkali environments, the salt from seawater and groundwater penetrates the piles, causing electrochemical corrosion with the reinforcing steel. The expansion of the steel and its subsequent compression of the concrete is also a major cause of pile cracking. This crack propagation can lead to severe structural damage, adversely affecting the safe operation of the high-pile wharf.

[0003] Current crack detection methods for high-pile wharves largely rely on manual on-site inspections. However, manual on-site inspections are prone to errors in determining the length, width, and number of cracks, and it is difficult to accurately obtain crack depth data, resulting in low accuracy in crack detection for high-pile wharves. Summary of the Invention

[0004] Therefore, it is necessary to provide a method and system for predicting cracks in high-pile wharves to address the aforementioned technical problems. This method improves the accuracy of crack prediction for high-pile wharves.

[0005] The following technical solution is adopted in this specification: This specification provides a method for predicting cracks in high-pile wharves, including: Obtain a three-dimensional digital model integrating the structural parameters of the high-pile wharf; Sample crack image data and sample sonar data of the high-pile wharf were collected. Crack identification and parameter measurement were performed on the sample crack image data to obtain sample crack identification results. The sample crack identification results include the length, width, number and distribution location of the cracks. Feature extraction is performed on the sample sonar data to obtain the sample sonar data identification results; the sonar data identification results include the maximum depth, average depth and depth distribution curve of each crack; Import the sample crack identification results and sample sonar data identification results into the three-dimensional digital model; Based on the width and depth of cracks in the imported 3D digital model and the preset judgment rules, the sample level labels of cracks are determined; the sample level labels include micro-cracks, small cracks, medium cracks and large cracks. The initial long short-term memory network is trained based on sample crack image data, sample sonar data, sample level labels, environmental parameters, and load parameters to obtain the long short-term memory network. The long short-term memory network is then used to predict the expansion of cracks in the high-pile wharf by using the crack image data and sonar data.

[0006] Preferably, based on the width and depth of the cracks in the imported 3D digital model and preset judgment rules, the sample level label of the cracks is determined, specifically including: When the width of a crack is less than or equal to a first width threshold, or the depth of a crack is less than or equal to a first depth threshold, the crack sample level is determined to be a microcrack. When the width of a crack is greater than the first width threshold and less than or equal to the second width threshold, or when the depth of a crack is less than or equal to the thickness of the protective layer c, the crack sample level is determined to be a small crack. When a crack is located on a slab, if the crack width is greater than the third width threshold and less than or equal to the fourth width threshold, or the crack depth is greater than the protective layer thickness, the crack sample level is determined as a medium crack. If the crack width is greater than the fourth width threshold, or the crack depth is greater than or equal to 1 / 2 of the component thickness or has penetrated through, the crack sample level is determined as a large crack. When a crack is located on a beam or pile, if the crack width is greater than the third width threshold and less than or equal to the fifth width threshold, or if the crack depth is greater than the protective layer thickness, the crack sample level is determined as a medium crack. If the crack width is greater than the fifth width threshold, or if the crack depth is greater than or equal to 1 / 2 of the component thickness or has penetrated through the crack, the crack sample level is determined as a large crack.

[0007] Preferably, the sample crack image data and sample sonar data are double-aligned using timestamps and coordinates to bind the surface parameters and depth parameters of the same crack; the surface parameters include the crack length, crack width, and crack distribution location.

[0008] Preferably, the process of constructing a three-dimensional digital model integrating the structural parameters of a high-pile wharf specifically includes: Obtain the structural parameters of the high-pile wharf; the structural parameters include the pile foundation parameters, beam parameters, panel parameters and additional parameters of the high-pile wharf; the additional parameters include the wharf design load level and crack resistance design index. Based on the structural parameters of the high-pile wharf, a three-dimensional digital model of the high-pile wharf is generated using lightweight modeling technology based on modeling information model.

[0009] Preferably, the method further includes: After collecting sample crack image data, the sample crack image data is purified. The purified sample crack image data were sequentially subjected to Gaussian filtering, histogram equalization, and image stitching to obtain the preprocessed sample crack image data.

[0010] Preferably, the method further includes: After collecting sample sonar data, the interference signals generated by water reflection waves and UAV body vibration are separated and removed from the sample sonar data by filtering algorithm to obtain denoised sample sonar data. Wavelet transform technology was used to separate the reflected waves from the crack interface in the denoised sample sonar data to obtain the effective depth signal; Based on the real-time ambient temperature detected, the deviation of the sound wave propagation speed in the sample sonar data is corrected to improve the calculation accuracy of the effective depth signal, thus obtaining the preprocessed sample sonar data.

[0011] Preferably, crack identification and parameter measurement are performed on the sample crack image data to obtain the sample crack identification result, specifically including: The YOLOv8 deep learning algorithm was used to identify cracks and measure parameters in the sample crack image data, and the crack identification results were obtained.

[0012] Preferably, feature extraction is performed on the sample sonar data to obtain the sample sonar data identification result, specifically including: The sonar data sample features are extracted using the sonar's built-in algorithm to obtain the sample sonar data identification results.

[0013] Preferably, the sonar's built-in algorithms include a sound wave reflection time difference ranging algorithm, a peak-valley detection algorithm, and a 3D-based algorithm. σ Outlier removal algorithm based on criteria.

[0014] A crack prediction system for high-pile wharves includes a drawing import and digital modeling module for obtaining structural parameters of the high-pile wharves. Based on the structural parameters, a three-dimensional digital model of the high-pile wharves is generated using BIM (Building Information Modeling) lightweight modeling technology. The structural parameters include pile foundation parameters, beam parameters, panel parameters, and additional parameters of the high-pile wharves. The additional parameters include the wharf design load level and crack resistance design index. The multi-source data acquisition module is used to acquire sample crack image data and sample sonar data of the high-pile wharf; The data processing and grading module is used to identify cracks and measure parameters in sample crack image data, obtaining sample crack identification results. These results include crack length, width, number, and distribution location. Feature extraction is performed on sample sonar data to obtain sample sonar data identification results, including the maximum depth, average depth, and depth distribution curve for each crack. The sample crack identification results and sample sonar data identification results are imported into a 3D digital model. Based on the crack width and depth in the imported 3D digital model and preset judgment rules, the sample crack grade labels are determined. These labels include micro-cracks, small cracks, medium cracks, and large cracks. The predictive analysis module is used to train an initial long short-term memory network based on sample crack image data, sample sonar data, sample level labels, environmental parameters, and load parameters to obtain a long short-term memory network. The long short-term memory network is then used to predict the crack image data and sonar data to obtain the expansion prediction results of all cracks in the high-pile wharf.

[0015] The above-mentioned technical solutions adopted in this specification can achieve the following beneficial effects: The project acquires the structural parameters of the high-pile wharf and generates a 3D digital model of the wharf based on these parameters. This transforms the physical structure of the high-pile wharf into a digital model, providing a unified and accurate platform for subsequent crack identification results, sonar data identification results, and crack analysis. Sample crack image data and sample sonar data of the high-pile wharf are collected. Crack identification and parameter measurement are performed on the sample crack image data to obtain sample crack identification results. These results are then imported into the 3D digital model to achieve accurate identification and quantification of crack surface features. Furthermore, by combining these results with the 3D digital model, the specific location of cracks within the wharf structure is accurately determined. Feature extraction is performed on the sample sonar data to obtain sample sonar data. Based on the identification results, the identification results of the sample sonar data are imported into a three-dimensional digital model to mine the internal features of the cracks and achieve a comprehensive characterization of the cracks from the surface to the interior. Based on the width and depth of the cracks and the judgment rules in the imported three-dimensional digital model, the sample level label of the cracks is determined. Based on the sample crack image data, sample sonar data, sample level label, environmental parameters and load parameters, the initial long short-term memory network is trained to obtain the long short-term memory network. The long short-term memory network is then used to predict the crack image data and sonar data to obtain the propagation prediction results of all cracks in the high-pile wharf. By using multi-dimensional data to train the model, the various factors affecting crack propagation are fully considered, thus improving the accuracy of crack prediction in the high-pile wharf. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0017] Figure 1 This document provides a flowchart illustrating a method for predicting cracks in a high-pile wharf. Figure 2 A general diagram of a crack prediction system for a high-pile wharf provided in this specification; Figure 3 This is a schematic diagram of a crack prediction system for a high-pile wharf provided in this specification. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without creative effort are within the scope of protection of this application.

[0019] As a core hub connecting land and water in ports, high-pile wharves have long been subjected to complex and harsh environments such as marine erosion, dynamic loads, and temperature stress. The initiation and propagation of cracks in concrete components directly threaten structural safety. Currently, the industry generally faces three major technical bottlenecks: First, the acquisition of crack detection data is incomplete and lacks accuracy. Existing technologies mostly rely on manual on-site inspections or single drone aerial photography. Manual inspection is inefficient, crack width measurement is prone to errors, and crack depth data cannot be accurately obtained, requiring destructive sampling for a rough assessment. Second, there is a lack of unified crack classification and grading standards. Data processing relies on manual statistics, making it impossible to quickly determine the distribution proportion and impact of cracks of different grades, resulting in delays in the treatment of key cracks. Third, there is no crack development prediction and scientific treatment guidance system. Existing solutions can only identify crack surfaces and cannot predict subsequent propagation trends, leading to blind maintenance decisions and serious waste of maintenance costs.

[0020] Both operators of high-pile wharves and engineering testing agencies have recognized the core value of integrated intelligent testing: operators hope to reduce safety risks and extend the wharf's service life through efficient testing; testing agencies hope to achieve a closed-loop process for cracks, encompassing "full parameter acquisition - intelligent analysis - prediction and early warning - treatment suggestions." However, current technological applications in this field are mostly limited to single-function levels, lacking an integrated system that combines UAV and sonar collaborative detection, digital modeling, standardized grading, and big data prediction. Furthermore, there are no mature engineering solutions, failing to meet the actual needs of high-pile wharves for accurate testing and efficient assessment.

[0021] Manual on-site inspections lack a unified standard for classifying and grading crack data, making it impossible to quickly determine the proportion of cracks and their subsequent development trends. This results in key crack locations not being addressed in a timely manner, affecting the structural safety and service life of high-pile wharves. Existing technologies have not yet formed an integrated system that combines crack identification, data analysis, statistical classification, and treatment recommendations.

[0022] Current technologies mostly rely on single drones for aerial photography to capture cracks, followed by manual measurement of representative crack lengths and widths, but cannot detect depth. Data processing is largely manual classification and statistics, lacking a standardized grading system and failing to include crack development prediction capabilities. These solutions suffer from fragmented data acquisition, low processing efficiency, and insufficient predictive power. The core reason is the lack of an integrated design for multi-device data fusion, intelligent classification and grading, and development prediction, which fails to meet the demand for comprehensive and accurate crack detection in high-pile wharves.

[0023] Therefore, developing a crack detection system that combines high efficiency, accuracy, and convenience, enabling timely identification and scientific handling of crack problems, has become a crucial and urgent need to ensure the structural safety of port engineering facilities. Developing a crack detection method and system for high-pile wharves has become an urgent industry requirement. The realization of this technology has significant engineering value and practical implications for improving the accuracy and efficiency of crack detection, reducing human safety risks, optimizing maintenance costs, and ensuring the safe and stable operation of high-pile wharf structures.

[0024] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0025] Figure 1 This is a flowchart illustrating a crack prediction method for a high-pile wharf as described in this specification, which specifically includes the following steps: S101: Obtain a three-dimensional digital model of the integrated high-pile wharf structure parameters.

[0026] In an exemplary embodiment, the process of constructing a three-dimensional digital model integrating the structural parameters of a high-pile wharf specifically includes: obtaining the structural parameters of the high-pile wharf; the structural parameters include the pile foundation parameters, beam parameters, panel parameters, and additional parameters of the high-pile wharf; the additional parameters include the wharf design load level and crack resistance design index; and generating a three-dimensional digital model of the high-pile wharf based on the structural parameters of the high-pile wharf using lightweight modeling technology of modeling information model.

[0027] Obtaining the structural parameters of the high-pile wharf specifically includes: obtaining the design drawings of the high-pile wharf; and analyzing the design drawings of the high-pile wharf using a geometric feature recognition algorithm to obtain the structural parameters of the high-pile wharf.

[0028] Specifically, the system automatically analyzes the following structural parameters of the wharf using a geometric feature recognition algorithm: pile foundation parameters: pile diameter, pile spacing, pile length, number of piles, and distribution coordinates; beam parameters: span of crossbeams / longitudinal beams, cross-sectional dimensions, and concrete strength grade; panel parameters: panel thickness, laying range, and expansion joint location; and additional parameters: wharf design load level and crack resistance design index. The analyzed structural parameters are automatically verified. If any data is missing or abnormal, the system will generate a prompt message, allowing for manual correction.

[0029] Specifically, based on the analyzed structural parameters, lightweight BIM modeling technology is used to generate a 1:1 3D digital model of the high-pile wharf. The 3D digital model integrates GPS positioning information. The 3D digital model supports: layered structural display; individual annotation of key components; and preset inspection ranges.

[0030] In actual execution, after the 3D digital model is completed, the system automatically outputs a model verification report, which includes indicators such as parameter matching degree and modeling integrity, to ensure that the model can be directly used for subsequent crack location and data association.

[0031] S102: Collect sample crack image data and sample sonar data of the high-pile wharf, perform crack identification and parameter measurement on the sample crack image data, and obtain sample crack identification results; the sample crack identification results include the length, width, number and distribution location of the cracks.

[0032] Specifically, an industrial-grade multi-rotor drone is used, equipped with a 4K high-definition camera and a GPS positioning module. During actual execution, the system automatically plans the shooting path based on a 3D digital model. For example, it uses the A* path optimization algorithm to ensure coverage of the entire detection area and automatically avoids obstacles such as dock cranes and mooring bollards.

[0033] The specific operational process includes: the drone takes off along a preset path and vertically photographs components such as piles, beams, longitudinal beams, and panels to ensure clear imaging of cracks; the camera acquires images in real time, simultaneously recording the shooting coordinates and timestamps; the image data is transmitted to the ground terminal in real time, and if abnormal images such as blurry or overexposed images occur, for example, if the resolution is lower than 1080P, the system automatically triggers a reshoot command; after shooting, the drone automatically returns to base, and the image data is uploaded in batches to the data processing module. In addition, the drone supports two operating modes: fully automatic mode and manual mode.

[0034] A drone-mounted concrete-penetrating sonar detector is employed. In actual operation, sonar detection and drone aerial photography are integrated and coordinated, requiring no human intervention throughout the process. After the drone's aerial photography module completes full-area image acquisition, the system automatically locates the crack position using image recognition algorithms, marking the crack centerline coordinates and detection points in a 3D digital model. The drone, equipped with a sonar probe, automatically plans the detection path, using dual calibration via GPS and visual positioning. It flies directly above the target point, adjusting its attitude to ensure the probe is perpendicular to the crack surface, guaranteeing an alignment deviation of ≤2mm between the probe center and the crack centerline. The probe emits low-frequency ultrasonic waves to penetrate the concrete and acquire crack depth data, simultaneously recording detection coordinates and timestamps, with a single-point detection time ≤10s. After completing a single detection point, the drone automatically flies to the next target point, storing the detection data in real-time in the drone's storage module. Upon completion of the operation, the data is automatically uploaded in batches to the data processing module, supporting resume capability.

[0035] In an exemplary embodiment, the method further includes: after acquiring sample crack image data, purifying the sample crack image data; and sequentially applying Gaussian filtering, histogram equalization, and image stitching to the purified sample crack image data to obtain preprocessed sample crack image data.

[0036] In an exemplary embodiment, crack identification and parameter measurement are performed on sample crack image data to obtain sample crack identification results. Specifically, this includes: performing crack identification and parameter measurement on sample crack image data using the YOLOv8 deep learning algorithm to obtain sample crack identification results.

[0037] Specifically, an improved YOLOv8 deep learning algorithm is used to identify cracks and measure parameters in the preprocessed images. Independent cracks are automatically identified, and the total number of cracks and the number of cracks in each component are counted. The crack location coordinates are mapped to a 3D digital model, and the component to which the crack belongs is labeled.

[0038] S103: Perform feature extraction on the sample sonar data to obtain the sample sonar data identification results; the sonar data identification results include the maximum depth, average depth and depth distribution curve of each crack.

[0039] In an exemplary embodiment, the method further includes: after acquiring sample sonar data, separating and removing interference signals generated by water reflection waves and UAV body vibration in the sample sonar data using a filtering algorithm to obtain denoised sample sonar data; using wavelet transform technology to separate the reflected waves of the crack interface from the denoised sample sonar data to obtain an effective depth signal; and correcting the deviation of the sound wave propagation speed in the sample sonar data according to the detected real-time environmental temperature to improve the calculation accuracy of the effective depth signal, thereby obtaining preprocessed sample sonar data.

[0040] The preprocessed dataset is automatically validated. Invalid data, such as images where cracks cannot be identified or data with distorted depth signals, account for ≤3% of the total data. If the threshold is exceeded, a prompt to re-acquire the data is made.

[0041] In an exemplary embodiment, sample crack image data and sample sonar data are double-aligned using timestamps and coordinates to ensure that surface parameters and depth parameters of the same crack are bound together; surface parameters include crack length, crack width, and crack distribution location.

[0042] In an exemplary embodiment, feature extraction is performed on the sample sonar data to obtain the sample sonar data recognition result. Specifically, this includes: extracting features from the sample sonar data using a built-in sonar algorithm to obtain the sample sonar data recognition result.

[0043] In one exemplary embodiment, the sonar's built-in algorithms include a time-of-flight ranging algorithm, a crest-and-trough detection algorithm, and a 3D-based algorithm. σ Outlier removal algorithm based on criteria.

[0044] Data collected by drone aerial photography and sonar detection are double-aligned using timestamps and coordinates to ensure precise binding of surface and depth parameters for the same crack, providing complete data support for subsequent full-dimensional analysis. Surface parameters include length and distribution.

[0045] S104: Import the sample crack identification results and sample sonar data identification results into the three-dimensional digital model.

[0046] S105: Based on the width and depth of cracks in the imported 3D digital model and the preset judgment rules, determine the sample level label of the cracks; the sample level label includes micro cracks, small cracks, medium cracks and large cracks.

[0047] In an exemplary embodiment, based on the width and depth of cracks in the imported 3D digital model and the judgment rules, the sample level label of the crack is determined, specifically including: when the crack width is less than or equal to a first width threshold, or the crack depth is less than or equal to a first depth threshold, the crack sample level is determined as a micro-crack; when the crack width is greater than the first width threshold and less than or equal to a second width threshold, or the crack depth is less than or equal to the protective layer thickness c, the crack sample level is determined as a small crack; when the crack is located on a plate, when the crack width is greater than a third width threshold and less than or equal to a fourth width threshold... When the crack width is greater than the thickness of the protective layer, or the crack depth is greater than the thickness of the protective layer, the crack sample level is determined as a medium crack. When the crack width is greater than the fourth width threshold, or the crack depth is greater than or equal to 1 / 2 of the component thickness or has penetrated through the structure, the crack sample level is determined as a large crack. When the crack is located on a beam or pile, when the crack width is greater than the third width threshold and less than or equal to the fifth width threshold, or the crack depth is greater than the thickness of the protective layer, the crack sample level is determined as a medium crack. When the crack width is greater than the fifth width threshold, or the crack depth is greater than or equal to 1 / 2 of the component thickness or has penetrated through the structure, the crack sample level is determined as a large crack.

[0048] Specifically, when the crack width is less than or equal to a first width threshold of 0.1, or the crack depth is less than or equal to a first depth threshold of 10, the crack sample level is determined to be a micro-crack; when the crack width is greater than a first width threshold of 0.1 and less than or equal to a second width threshold of 0.2, or the crack depth is less than or equal to the protective layer thickness c, the crack sample level is determined to be a small crack; when the crack is located on the plate, when the crack width is greater than a third width threshold of 0.3 and less than or equal to a fourth width threshold of 1, or the crack depth is greater than the protective layer thickness, the crack sample level is determined to be a micro-crack. The crack grade is determined as medium crack. When the crack width is greater than the fourth width threshold 1, or the crack depth is greater than or equal to 1 / 2 of the component thickness or has penetrated through, the crack sample grade is determined as large crack. When the crack is located on a beam or pile, when the crack width is greater than the third width threshold 0.3 and less than or equal to the fifth width threshold 3, or the crack depth is greater than the protective layer thickness, the crack sample grade is determined as medium crack. When the crack width is greater than the fifth width threshold 3, or the crack depth is greater than or equal to 1 / 2 of the component thickness or has penetrated through, the crack sample grade is determined as large crack.

[0049] Specifically, based on the Technical Specifications for Inspection and Evaluation of Port Hydraulic Structures (JTJ302-2006) and the Technical Specifications for Inspection and Evaluation of Hydraulic Structures in Water Transport Engineering (JTS 304-2019), a four-level crack classification and grading standard was established, and the level of each crack was automatically determined, as shown in Table 1.

[0050] Table 1 After the classification is completed, the system adds a classification label to each crack and highlights it in different colors in the 3D model (micro cracks (Class A) - green, small cracks (Class B) - yellow, medium cracks (Class C) - orange, large cracks (Class D) - red).

[0051] Relationship between depth and protective layer: The thickness of the protective layer (c) is a key benchmark for assessing crack depth. Once the depth exceeds the thickness of the protective layer, it means that the crack has extended to the surface of the reinforcing steel, and its hazard level jumps from Class B to Class C, requiring fundamentally different treatment measures.

[0052] S105: The initial long short-term memory network is trained based on sample crack image data, sample sonar data, sample level labels, environmental parameters and load parameters to obtain the long short-term memory network. The long short-term memory network is then used to predict the crack image data and sonar data to obtain the propagation prediction results of all cracks in the high-pile wharf.

[0053] By combining current crack data, structural characteristics, environmental parameters, and historical data, machine learning models are used to predict crack propagation trends and identify structural safety risks.

[0054] It receives multi-dimensional input parameters, divided into core parameters and auxiliary parameters: Core parameters: current crack level, size (length, width, depth), distribution location, and material properties of the component to which it belongs (concrete strength, reinforcement ratio); auxiliary parameters: environmental parameters (seawater salinity, annual average temperature, tidal frequency, rainfall), load parameters (wharf cargo load, ship collision frequency), historical data (crack change records in the past 1-3 years; if unavailable, import historical data of the same type of wharf by default).

[0055] The parameters support automatic import and manual supplementation. For example, environmental parameters can be connected to local weather station and port operation system data, and the system will prompt correction for abnormal parameters.

[0056] A Long Short-Term Memory (LSTM) machine learning model is employed, initially trained based on a training sample library. During actual implementation, as the system continues to run, newly collected crack data and maintenance feedback data are automatically added to the sample library. The LSTM is automatically updated monthly: adjusting LSTM parameters; optimizing prediction accuracy; and outputting an LSTM update report, including metrics such as sample library increments and changes in prediction error.

[0057] Based on the trained LSTM model, the crack propagation prediction results are output: Short-term forecast (within 1 year): Crack width / depth propagation rate (mm / month), and changes in grade after propagation.

[0058] Long-term forecast (1-5 years): final crack size, whether it will cause secondary structural damage (such as steel corrosion, reduced load-bearing capacity of components).

[0059] Risk identification: Based on the prediction results, identify high-risk cracks (i.e. cracks that may reach the level of "large cracks" or affect key stress-bearing parts of the structure after expansion) and mark them as key targets for attention.

[0060] The prediction results are presented as trend curves (width / depth changes over time) and text descriptions, and can be exported to an evaluation report.

[0061] When applying the crack prediction method for high-pile wharves provided in this manual, it is not necessary to consider... Figure 1 The steps shown are executed in sequence. The specific execution order of each step can be determined as needed, and this manual does not impose any restrictions on it.

[0062] A crack prediction system for high-pile wharves, characterized by comprising: a drawing import and digital modeling module for acquiring structural parameters of the high-pile wharf and generating a three-dimensional digital model of the high-pile wharf based on these parameters using lightweight BIM (Building Information Modeling) technology; the structural parameters include pile foundation parameters, beam parameters, panel parameters, and additional parameters; the additional parameters include the wharf's design load level and crack resistance design index; a multi-source data acquisition module for acquiring sample crack image data and sample sonar data of the high-pile wharf; and a data processing and grading module for crack identification and parameter measurement of the sample crack image data to obtain sample crack identification results; the sample crack identification results include the length, width, number, and distribution location of the cracks; and a special processing module for the sample sonar data. The process involves extracting features to obtain sample sonar data identification results. These results include the maximum depth, average depth, and depth distribution curve for each crack. The sample crack identification results and the sample sonar data identification results are then imported into a three-dimensional digital model. Based on the crack width and depth in the imported three-dimensional digital model and preset judgment rules, sample level labels for the cracks are determined. These labels include micro-cracks, small cracks, medium cracks, and large cracks. A prediction and analysis module is used to train an initial long short-term memory network based on the sample crack image data, the sample sonar data, the sample level labels, environmental parameters, and load parameters to obtain a long short-term memory network. This network is then used to predict the crack image data and sonar data to obtain the expansion prediction results for all cracks in the high-pile wharf.

[0063] Specifically, Figure 2 This is a schematic diagram of the high-pile wharf crack prediction system provided by the present invention, as shown below. Figure 2As shown, the high-pile wharf crack prediction system 10 includes: drawing import and digital modeling module 100, multi-source data acquisition module 200, data processing and classification module 300, prediction analysis module 400, and suggestion output and visualization module 500.

[0064] Each module achieves real-time data interaction via 5G wireless communication or industrial Ethernet, with a data transmission latency of ≤100ms. It supports simultaneous operation of multiple devices and fully automated operation of the entire process, ensuring the efficiency and accuracy of testing and evaluation.

[0065] The Drawing Import and Digital Modeling Module 100 is used to import high-pile wharf design drawings and building information model (BIM) models, automatically extract structural parameters and generate a 1:1 three-dimensional digital model, providing a precise spatial reference for crack location and detection range planning.

[0066] The drawing import and digital modeling module 100 includes: drawing parsing module 101 and 3D modeling module 102.

[0067] The drawing parsing module 101 supports direct import of AutoCAD 2016 and above (.dwg / .dxf format) and BIM model files (.rvt format). Through geometric feature recognition algorithms, it automatically parses the core structural parameters of the wharf: pile foundation parameters: pile diameter, pile spacing, pile length, number of piles and distribution coordinates; beam parameters: span of horizontal / longitudinal beams, cross-sectional dimensions, concrete strength grade; panel parameters: panel thickness, laying range, expansion joint location; additional parameters: wharf design load level, crack resistance design index.

[0068] The parsed parameters are automatically validated. If there is missing or abnormal data (such as parameters exceeding the industry standard range), the system will generate a prompt message and support manual supplementation and correction.

[0069] Based on the analyzed structural parameters, the 3D modeling module 102 uses lightweight BIM modeling technology to generate a 1:1 3D digital model of the high-pile wharf, integrating GPS positioning information. The model supports: layered structural display (pile base, beam layer, panel layer); individual annotation of key components (e.g., number, dimensions, material properties); and preset detection ranges (users can manually select key detection areas, or the system can automatically identify crack-prone areas such as pile tops, beam ends, and panel corners). During actual execution, the system automatically outputs a model verification report after modeling is completed, including indicators such as parameter matching degree and modeling integrity, ensuring the model can be directly used for subsequent crack location and data association.

[0070] The multi-source data acquisition module 200 is used to collect crack parameters (length, width, number, depth, and distribution location) of concrete components of high-pile wharves in all dimensions through collaborative operation of drone aerial photography and sonar detection.

[0071] The multi-source data acquisition module 200 includes a drone aerial photography module 201 and a sonar detection module 202.

[0072] The 201 drone aerial photography module utilizes an industrial-grade multi-rotor drone equipped with a 4K high-definition camera and GPS positioning module. During actual operation, the system automatically plans the shooting path based on a 3D digital model, ensuring coverage of the entire detection area and automatically avoiding obstacles such as dock cranes and mooring bollards. The specific operation process is as follows: The drone takes off according to a preset path and vertically photographs components such as pile foundations, crossbeams, longitudinal beams, and panels; the camera acquires images in real time, simultaneously recording the shooting coordinates and timestamps; the image data is transmitted to the ground terminal in real time, and if abnormal images such as blurriness or overexposure occur, the system automatically triggers a reshoot command; after shooting is completed, the drone automatically returns to base, and the image data is uploaded in batches to the data processing module.

[0073] In addition, the 201 drone aerial photography module supports two operating modes: fully automatic mode and manual mode.

[0074] Sonar detection module 202 adopts a UAV-mounted concrete-penetrating sonar detector.

[0075] In actual operation, the sonar detection module 202 and the UAV aerial photography module 201 work together as a unified whole, requiring no human intervention. After the UAV aerial photography module completes the full-domain image acquisition, the system automatically locates the crack position through image recognition algorithms and marks the crack centerline coordinates and detection points in the 3D digital model. The UAV, equipped with a sonar probe, automatically plans the detection path and, through dual calibration of GPS and visual positioning, flies directly above the target point. It adjusts its body attitude to ensure that the probe is perpendicular to the crack surface, ensuring that the alignment deviation between the probe center and the crack centerline is ≤2mm. The probe automatically starts emitting low-frequency ultrasonic waves to penetrate the concrete and obtain crack depth data, simultaneously recording the detection coordinates and timestamp (precisely aligned with the aerial photography data using millisecond-level timestamps). The detection time for a single point is ≤10s. After completing a single detection point, the UAV automatically flies to the next target point. The detection data is stored in real time in the onboard storage module and automatically uploaded in batches to the data processing module after the operation is completed, supporting breakpoint resume.

[0076] The data collected by the UAV aerial photography module 201 and the sonar detection module 202 are double-aligned through timestamps and coordinates to ensure that the surface parameters (length, width, distribution) and depth parameters of the same crack are accurately bound, providing complete data support for subsequent full-dimensional analysis.

[0077] The data processing and grading module 300 is used for preprocessing, feature extraction, standardized classification and grading, and statistical analysis of crack data collected from multiple sources.

[0078] The data processing and classification module 300 includes a data preprocessing module 301, a crack feature extraction module 302, a classification and grading module 303, and a data statistics module 304.

[0079] The data preprocessing module 301 is used to clean up the acquired image data and sonar data: Image preprocessing: Gaussian filtering is used to remove noise, histogram equalization is used to enhance crack contrast, and image stitching is used to generate a complete surface image of the component; Sonar data preprocessing: Environmental interference signals are removed, effective depth signals are extracted through wavelet transform, and temperature effects are corrected.

[0080] The preprocessed dataset is automatically validated, with invalid data accounting for ≤3%. If the percentage exceeds the threshold, a prompt to re-collect the data is displayed.

[0081] The crack feature extraction module 302 is used to identify cracks and measure parameters in preprocessed images using an improved YOLOv8 deep learning algorithm. It automatically identifies individual cracks, counts the total number of cracks and the number of cracks in each component, maps the crack location coordinates to a 3D digital model, and labels the component to which the crack belongs. After feature extraction from sonar data, it directly outputs the maximum depth, average depth, and depth distribution curve of a single crack.

[0082] The classification and grading module 303 is used to establish a four-level crack classification and grading standard based on the Technical Specification for Inspection and Evaluation of Port Hydraulic Structures (JTJ302-2006) and the Technical Specification for Inspection and Evaluation of Hydraulic Structures of Waterway Engineering (JTS 304-2019), and automatically determine the grade of each crack.

[0083] Data statistics module 304 is used to automatically generate crack data statistics reports, including: overall statistics: total number of cracks, percentage of cracks of each level, total crack length, average width / depth; component statistics: crack density (cracks / m²) of each component (pile foundation, crossbeam, longitudinal beam, panel), distribution of the highest level cracks; trend statistics: comparison of crack distribution density in different areas of the same component, percentage of crack levels in key areas.

[0084] The report supports exporting to Excel and PDF formats, and the data can be linked with the 3D model. Clicking on a crack record in the report will automatically locate and highlight the crack in the model.

[0085] The predictive analysis module 400 is used to combine current crack data, structural characteristics, environmental parameters and historical data, and predict crack propagation trends through machine learning models to identify structural safety risk points.

[0086] The predictive analysis module 400 includes: a parameter input module 401, a model training and update module 402, and an extended prediction module 403.

[0087] The parameter input module 401 receives multi-dimensional input parameters, which are divided into core parameters and auxiliary parameters: Core parameters: current crack level, size (length, width and depth), distribution location, and material properties of the component to which it belongs (concrete strength, reinforcement ratio); Auxiliary parameters: environmental parameters (seawater salinity, annual average temperature, tidal frequency, rainfall), load parameters (wharf cargo load, ship collision frequency), and historical data (crack change records in the past 1-3 years; if no data is available, historical data of the same type of wharf will be imported by default).

[0088] The system supports automatic import of parameters (such as environmental parameters, which can be connected to local weather stations and port operation system data) and manual supplementation. The system will prompt corrections for abnormal parameters.

[0089] The model training and update module 402 uses an LSTM machine learning model, initially trained based on a training sample library. During actual execution, as the system continues to run, newly collected crack data and maintenance effect feedback data are automatically added to the sample library. The model is automatically updated monthly: correcting model parameters; optimizing prediction accuracy; and outputting a model update report, including indicators such as sample library increments and changes in prediction error.

[0090] Extended prediction module 403, based on a trained LSTM model, outputs crack propagation prediction results: Short-term forecast (within 1 year): Crack width / depth propagation rate (mm / month), and changes in grade after propagation.

[0091] Long-term forecast (1-5 years): final crack size, whether it will cause secondary structural damage (such as steel corrosion, reduced load-bearing capacity of components).

[0092] Risk identification: Based on the prediction results, identify high-risk cracks (i.e. cracks that may reach the level of "large cracks" or affect key stress-bearing parts of the structure after expansion) and mark them as key targets for attention.

[0093] The prediction results are presented as trend curves (width / depth changes over time) and text descriptions, and can be exported to an evaluation report.

[0094] The visualization module 500 is used to display the classification and grading results and prediction conclusions in a three-dimensional visualization manner.

[0095] The visualization module 500 includes: a key location annotation module 501 and a visualization display module 502.

[0096] The key location marking module 501 is used to automatically determine key locations for treatment based on crack level, predicted risk, and structural stress characteristics: Level 1 key locations: Locations of large cracks, medium cracks predicted to upgrade to large cracks within one year, and crack locations at critical structural stress points (such as the connection between pile top and beam). Level 2 key locations: Locations of medium cracks, and areas with concentrated distribution of small cracks (density ≥ 3 cracks / m²). Level 3 key locations: Locations of small cracks and micro-cracks.

[0097] Key locations are marked with different identifiers in the 3D digital model (Level 1 - flashing red marker, Level 2 - orange marker, Level 3 - yellow marker), and the location coordinates (GPS coordinates and model relative coordinates) are output to facilitate the positioning of construction personnel.

[0098] The visualization module 502 provides multi-dimensional display functions: 3D model interaction: supports rotation, zoom, and layered viewing; clicking on a crack displays its detailed parameters (size, grade, prediction results, and processing suggestions).

[0099] Data visualization charts: displaying pie charts of crack distribution at various levels, bar charts of crack density in components, and crack propagation trend lines.

[0100] Report generation: Automatically generates the "High Pile Wharf Crack Detection and Assessment Report", which includes modules such as detection overview, data statistics, classification and grading results, predictive analysis, treatment suggestions, and a list of key locations. It supports custom editing and export (Word and PDF formats).

[0101] The intelligent system for crack detection and assessment of high-pile wharves, as described in this invention, achieves a comprehensive breakthrough in detection accuracy, efficiency, and intelligence compared to existing technologies. Its core beneficial effects are as follows: (1) Multi-source collaborative detection enables accurate acquisition of crack parameters in all dimensions. This invention innovatively employs a collaborative detection mode combining UAV aerial photography and penetrating sonar, integrating an improved YOLOv8 image recognition algorithm with sonar depth detection technology. This overcomes the limitations of traditional manual inspection or single-device detection, which can only acquire surface parameters. On one hand, UAV aerial photography enables non-contact, precise measurement of crack length, width, number, and distribution location, with a width detection accuracy of 0.05mm, a measurement error ≤ ±0.02mm, and a length error ≤ ±0.03mm. On the other hand, sonar detection can acquire crack depth data without destructive sampling, achieving a closed loop of parameters across all dimensions (length, width, and depth). Simultaneously, a dual alignment mechanism using timestamps and coordinates ensures precise binding of surface parameters and depth data, providing a complete and reliable data source for subsequent analysis, thus comprehensively improving the judgment criteria compared to traditional technologies.

[0102] (2) Standardized grading and full-process integration significantly improve the efficiency of testing and evaluation and the scientific nature of decision-making. This invention establishes a four-level crack classification and grading standard that conforms to industry norms. Through an automated data processing module, it achieves intelligent crack level determination, completely solving the drawbacks of traditional technologies that lack unified standards and rely on manual statistical analysis. The system provided by this invention shortens the inspection cycle of high-pile wharves, improves inspection efficiency, and shortens the maintenance decision-making cycle, avoiding the cost waste caused by traditional blind maintenance. Actual engineering verification shows that it saves the average annual maintenance cost per wharf, reduces the lag rate in treating key cracks, and significantly improves the scientific and economical aspects of wharf operation and management.

[0103] (3) Deep learning prediction enables accurate prediction of structural safety risks. This invention, based on an LSTM machine learning model, integrates current crack data, structural characteristics, environmental parameters, and historical data to construct a dynamically updated crack propagation prediction model. This overcomes the limitations of traditional technologies, which can only detect cracks after they occur and cannot predict risks. The model automatically iterates and optimizes through real-time supplementary engineering data, and can roughly predict crack propagation trends and structural safety risks over the next 1-5 years, issuing early warnings 3-6 months in advance.

[0104] Advantages of this invention: Comprehensive and accurate inspection: Multi-source equipment collaboratively covers all dimensions of crack surface and internal parameters, quantifying errors within the engineering allowable range, and significantly improving data accuracy compared to traditional manual inspection, solving the pain points of traditional technology that only measures surface and not depth or coarseness; Integration of intelligence and standardization: The four-level grading standard conforms to industry norms, the whole process is automated to reduce manual intervention, and data processing efficiency is effectively improved. At the same time, the output inspection reports and maintenance plans are highly universal and can be directly connected to the port engineering management system; Balance of safety and economy: The non-contact operation of drones completely avoids the risks of falls and drowning from manual high-altitude and water-side operations. The inspection process does not require interruption of the core terminal operation, reducing economic losses by tens of thousands of yuan per inspection; Strong adaptability and scalability: Supports the inspection of high-pile wharves of different scales and service years, is compatible with mainstream design file formats such as CAD / BIM, and the model supports data supplementation for multiple scenarios (such as weather and operational loads). It can be further expanded to the inspection of other concrete port structures (such as caisson wharves and approach bridges).

[0105] In actual implementation, the visualization module supports integration with the port operation and management platform, pushing detection results and early warning information to the management personnel's terminals in real time, enabling efficient collaboration in maintenance decision-making.

[0106] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0107] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A crack prediction method of a high-pile wharf, characterized by, include: Obtain a three-dimensional digital model integrating the structural parameters of the high-pile wharf; Sample crack image data and sample sonar data of the high-pile wharf are collected. Crack identification and parameter measurement are performed on the sample crack image data to obtain sample crack identification results. The sample crack identification results include the length, width, number and distribution location of the cracks. Feature extraction is performed on the sample sonar data to obtain the sample sonar data identification results; the sonar data identification results include the maximum depth, average depth, and depth distribution curve of each crack. The sample crack identification results and the sample sonar data identification results are imported into the three-dimensional digital model; Based on the width and depth of cracks in the imported 3D digital model and the preset judgment rules, the sample level labels of cracks are determined; the sample level labels include micro-cracks, small cracks, medium cracks and large cracks. The initial long short-term memory network is trained based on the sample crack image data, the sample sonar data, the sample level labels, environmental parameters, and load parameters to obtain the long short-term memory network. The long short-term memory network is then used to predict the expansion of all cracks in the high-pile wharf by using the crack image data and sonar data.

2. The crack prediction method of a high-pile wharf according to claim 1, wherein, The process of determining the sample grade label of cracks based on the width and depth of cracks in the imported 3D digital model and preset judgment rules specifically includes: When the width of a crack is less than or equal to a first width threshold, or the depth of a crack is less than or equal to a first depth threshold, the crack sample level is determined to be a microcrack. When the width of a crack is greater than the first width threshold and less than or equal to the second width threshold, or when the depth of a crack is less than or equal to the thickness of the protective layer c, the crack sample level is determined to be a small crack. When a crack is located on a slab, if the crack width is greater than the third width threshold and less than or equal to the fourth width threshold, or the crack depth is greater than the protective layer thickness, the crack sample level is determined as a medium crack. If the crack width is greater than the fourth width threshold, or the crack depth is greater than or equal to 1 / 2 of the component thickness or has penetrated through, the crack sample level is determined as a large crack. When a crack is located on a beam or pile, if the crack width is greater than the third width threshold and less than or equal to the fifth width threshold, or if the crack depth is greater than the protective layer thickness, the crack sample level is determined as a medium crack. If the crack width is greater than the fifth width threshold, or if the crack depth is greater than or equal to 1 / 2 of the component thickness or has penetrated through the crack, the crack sample level is determined as a large crack.

3. The crack prediction method of a high-pile wharf according to claim 1, wherein, The sample crack image data and the sample sonar data are double-aligned using timestamps and coordinates to bind the surface parameters and depth parameters of the same crack; the surface parameters include the crack length, crack width, and crack distribution location.

4. The crack prediction method of a high-pile wharf according to Claim 1, wherein The process of constructing a three-dimensional digital model integrating the structural parameters of a high-pile wharf specifically includes: Obtain the structural parameters of the high-pile wharf; the structural parameters include the pile foundation parameters, beam parameters, panel parameters and additional parameters of the high-pile wharf; the additional parameters include the wharf design load level and crack resistance design index. Based on the structural parameters of the high-pile wharf, a three-dimensional digital model of the high-pile wharf is generated using lightweight modeling technology based on modeling information model.

5. The crack prediction method of a high-pile wharf according to Claim 1, wherein The method further includes: After collecting sample crack image data, the sample crack image data is purified. The purified sample crack image data were sequentially subjected to Gaussian filtering, histogram equalization, and image stitching to obtain the preprocessed sample crack image data.

6. The crack prediction method of a piled wharf according to claim 1, wherein, The method further includes: After collecting sample sonar data, the interference signals generated by water reflection waves and UAV body vibration are separated and removed from the sample sonar data by filtering algorithm to obtain denoised sample sonar data. Wavelet transform technology was used to separate the reflected waves from the crack interface in the denoised sample sonar data to obtain the effective depth signal; Based on the real-time ambient temperature detected, the deviation of the sound wave propagation speed in the sample sonar data is corrected to improve the calculation accuracy of the effective depth signal, thus obtaining the preprocessed sample sonar data.

7. The crack prediction method for high-pile wharves as described in claim 1, characterized in that, The process of identifying cracks and measuring parameters in the sample crack image data to obtain sample crack identification results specifically includes: The sample crack image data was subjected to crack identification and parameter measurement using the YOLOv8 deep learning algorithm to obtain the sample crack identification results.

8. The crack prediction method for high-pile wharves as described in claim 1, characterized in that, The step of extracting features from the sample sonar data to obtain the sample sonar data identification result specifically includes: The sample sonar data is used to extract features from the sonar data using a built-in sonar algorithm to obtain the sample sonar data identification result.

9. The crack prediction method for high-pile wharves as described in claim 8, characterized in that, The sonar's built-in algorithms include a sound wave reflection time difference ranging algorithm, a peak-valley detection algorithm, and a 3D-based algorithm. σ Outlier removal algorithm based on criteria.

10. A crack prediction system for high-pile wharves, characterized in that, include: The drawing import and digital modeling module is used to obtain the structural parameters of the high-pile wharf, and based on the structural parameters, it uses BIM lightweight modeling technology to generate a three-dimensional digital model of the high-pile wharf. The structural parameters include the pile foundation parameters, beam parameters, panel parameters and additional parameters of the high-pile wharf. The additional parameters include the wharf design load level and crack resistance design index. The multi-source data acquisition module is used to acquire sample crack image data and sample sonar data of the high-pile wharf; The data processing and grading module is used to perform crack identification and parameter measurement on the sample crack image data to obtain sample crack identification results; the sample crack identification results include the length, width, number, and distribution location of the cracks; and to perform feature extraction on the sample sonar data to obtain sample sonar data identification results; the sonar data identification results include the maximum depth, average depth, and depth distribution curve of each crack. The sample crack identification results and the sample sonar data identification results are imported into the three-dimensional digital model; Based on the width and depth of cracks in the imported 3D digital model and the preset judgment rules, the sample level labels of cracks are determined; the sample level labels include micro-cracks, small cracks, medium cracks and large cracks. The predictive analysis module is used to train an initial long short-term memory network based on the sample crack image data, the sample sonar data, the sample level labels, environmental parameters, and load parameters to obtain a long short-term memory network. The long short-term memory network is then used to predict the crack image data and sonar data to obtain the expansion prediction results of all cracks in the high-pile wharf.

Images