AI image recognition technology-based main beam deformation detection and lifting positioning compensation system for gantry crane

Abstract

A bridge and portal crane main beam deformation detection and lifting positioning compensation system based on an AI image recognition technology obtains main beam panoramic image data through a camera array, and adopts a multi-frame fusion technology to preprocess collected images for vibration and light changes under complex working conditions to obtain a stable and clear main beam first image; outputting main beam state parameters; judging whether the dip angle change trend and the deformation degree of the main beam exceed safety thresholds or not; generating a main beam deformation feature vector; according to the main beam deformation feature vector, a real-time data stream processing technology is adopted to generate a deviation correction control instruction, the target angle and displacement of an adjusting mechanism are determined, and an accurate deviation correction parameter set is output; generating a new round of deviation rectifying instruction; and driving an adjusting mechanism to perform linkage execution according to the new round of deviation rectification instruction, and determining that the inclination angle change and bending deformation of the main beam are stabilized in a safe range, thereby completing a closed-loop control process. The posture of the main beam can be accurately adjusted, and it is ensured that the dip angle change and bending deformation of the main beam are stabilized within the safety range.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a bridge crane main beam deformation detection and lifting positioning compensation system based on AI image recognition technology. Background Technology

[0002] As core equipment in engineering construction and heavy industry, cranes' safety and stability directly affect operational efficiency and the safety of personnel and property. The main beam, a critical load-bearing component, has a decisive impact on equipment performance under complex working conditions due to its tilt angle changes and bending deformation. In recent years, with the development of cranes towards larger and more intelligent designs, the need for real-time monitoring and adjustment of the main beam's condition has become increasingly urgent, becoming an important research direction for improving equipment reliability. However, traditional monitoring and correction methods have gradually revealed their inadequacies in addressing this need.

[0003] Currently, solutions based on traditional technologies such as tilt sensors have significant limitations in practical applications. Sensors need to be directly mounted on the main beam, making them susceptible to environmental interference such as vibration and temperature changes, thus compromising measurement accuracy and reliability. Furthermore, contact-based measurement methods are difficult to deploy in the complex structures of large cranes and cannot comprehensively reflect the dynamic deformation of the main beam. For main beam correction, traditional methods often rely on manual experience or simple mechanical adjustments, lacking real-time performance and accuracy, and failing to adapt to dynamic requirements under varying load conditions.

[0004] The core challenges in this research area lie in achieving non-contact, precise monitoring of the main beam's tilt angle and deformation, and in realizing automated correction based on real-time data. Key technical factors include the accuracy of dynamic image data acquisition and processing, the ability to automatically identify tilt angles and deformations, and the effective linkage between the monitoring system and the adjustment mechanism. The unresolved issues stem from the difficulty of existing technologies in accurately capturing the spatial attitude changes of the main beam under non-contact conditions, and the lack of closed-loop control capabilities to rapidly convert monitoring results into correction commands, particularly evident under the high-load and high-vibration conditions of large cranes.

[0005] Therefore, how to monitor the tilt angle changes and bending deformation of the crane's main beam in real time using non-contact technology, and achieve automated and precise correction based on this, has become a key issue in improving the safety and service life of cranes. Solving this problem requires breaking through the limitations of traditional methods and exploring new technological paths to meet the high standards required for modern crane operations. Summary of the Invention

[0006] This invention provides a bridge crane main beam deformation detection and lifting positioning compensation system based on AI image recognition technology, mainly comprising: A high-resolution camera array was used to acquire panoramic image data of the main beam. Multi-frame fusion technology was used to preprocess the acquired images to obtain a stable and clear first image of the main beam under complex working conditions of vibration and light changes. Edge feature points are extracted from the first image of the main beam. A deep learning model is used to spatially locate and classify the feature points, determine the current tilt angle change and bending deformation distribution of the main beam, and output the main beam state parameters. After obtaining the main beam state parameters, the spatial attitude deviation of the main beam is calculated through a preset geometric transformation algorithm. Combined with historical data comparison and analysis, it is determined whether the trend of the main beam inclination angle change and the degree of deformation exceed the safety threshold. If the trend of tilt angle change exceeds the threshold, the state parameters are input into the pre-established dynamic recognition network, and the specific deformation type and location are identified based on the image data processing results to generate the main beam deformation feature vector. Based on the deformation characteristic vector of the main beam, real-time data stream processing technology is used to generate correction control commands, determine the target angle and displacement of the adjustment mechanism, and output a precise set of correction parameters. The system obtains the set of correction parameters through a closed-loop control system, drives the hydraulic or servo mechanism to perform the main beam attitude adjustment, and simultaneously collects real-time feedback images during the adjustment process to obtain a second image of the main beam. Based on the extraction and spatial positioning analysis of repeated edge features in the second image of the main beam, it is determined whether the change in tilt angle and deformation state after adjustment have returned to the normal range, and the adjusted state parameters are output. The adjusted state parameters are obtained, and the residual value is calculated by comparing them with the initial state parameters. If the residual value is greater than the preset threshold, the parameters are re-input into the dynamic recognition network to generate a new round of correction instructions. Driven by a new round of correction commands, the adjustment mechanism is executed in conjunction with the data collection and analysis of the final feedback images. This confirms that the main beam's tilt angle and bending deformation have stabilized within a safe range, thus completing the closed-loop control process.

[0007] The process involves acquiring panoramic image data of the main beam using a high-resolution camera array. To address vibration and light variations under complex operating conditions, multi-frame fusion technology is employed to preprocess the acquired images, resulting in a stable and clear first image of the main beam, including: An initial image set was obtained by acquiring panoramic image data of the main beam using a high-resolution camera array; To address vibration and light variations, multiple frames of data are acquired from the initial image set and processed using multi-frame fusion technology to obtain the second image. If residual noise exists in the second image, the second image is smoothed using a mean filtering algorithm to obtain the third image; Based on the pixel distribution of the third image, determine whether the sharpness reaches the preset threshold and obtain the sharpness evaluation result; Based on the sharpness evaluation results, the edge features of the third image are enhanced using the Laplacian operator to obtain the fourth image; The stability effect features of the fourth image are obtained to determine whether they meet the output requirements under complex working conditions, and the final judgment result is obtained. Based on the final judgment result, a stable and clear first image of the main beam is output.

[0008] The process involves extracting edge feature points from the first image of the main beam, using a deep learning model to spatially locate and classify these feature points, determining the current tilt angle change and bending deformation distribution of the main beam, and outputting the main beam state parameters, including: Edge feature points are extracted from the main beam image to obtain a set of feature points; A deep learning model is used to spatially locate the feature point set and obtain the location coordinate data; The positioning coordinate data is classified by features to determine the distribution of tilt angle changes; The bending deformation characteristics are calculated from the distribution of tilt angle variation to obtain deformation distribution data; Based on the deformation distribution data and positioning coordinate data, determine whether the state parameters exceed the preset threshold and obtain the parameter evaluation results; Adjust the output of the deep learning model based on the parameter evaluation results to obtain the optimized state parameters; If the optimized state parameters meet the preset conditions, the final state parameters will be output.

[0009] After obtaining the main beam state parameters, the spatial attitude deviation of the main beam is calculated using a preset geometric transformation algorithm. Combined with historical data comparison and analysis, it is determined whether the trend of the main beam's inclination angle change and the degree of deformation exceed the safety threshold, including: After obtaining the main beam state parameters, the spatial attitude deviation is calculated using a geometric transformation algorithm to obtain the deviation data; For the deviation data, historical data are used for comparative analysis to determine the trend of tilt angle changes; By analyzing the trend of tilt angle changes, it can be determined whether the degree of deformation exceeds a preset threshold, and a judgment result can be obtained. Based on the judgment results, relevant change patterns are extracted from historical data to obtain trend distribution data; By using trend distribution data and combining it with spatial attitude deviation, the direction of adjustment of state parameters is determined; By using the adjusted state parameters, the degree of deformation is classified using the support vector machine algorithm to obtain classification data; For categorized data, data processing is performed in conjunction with preset thresholds to determine the final state assessment result.

[0010] If the tilt angle change trend exceeds the threshold, the state parameters are input into a pre-established dynamic recognition network. Based on the image data processing results, the specific deformation type and location are identified, and a main beam deformation feature vector is generated, including: If the tilt angle change exceeds the preset threshold, the state parameters are processed by the dynamic recognition network, and the deformation type and location are identified by combining the image data to obtain the main beam deformation feature vector. The location identification results are analyzed by the deformation feature vector of the main beam, and the deformation type distribution is determined by the clustering algorithm to obtain the type distribution data; Based on the type distribution data and the trend of inclination angle changes, the evolution direction of the main beam's condition is determined, and the condition evolution data is identified. Based on the state evolution data, historical state parameters are obtained, and the current state parameters are adjusted through comparative analysis to obtain the adjusted parameter data; Using the adjusted parameter data, the deformation type is classified using the support vector machine algorithm to obtain the classification result data; Based on the classification results data, the final distribution characteristics of the main beam state are determined by combining image data processing with location recognition information. By analyzing the final distribution characteristics, the dynamic update values ​​of the state parameters are obtained, the stability trend of the main beam is determined, and stability assessment data is obtained.

[0011] The process involves generating correction control commands based on the main beam deformation characteristic vector using real-time data stream processing technology, determining the target angle and displacement of the adjustment mechanism, and outputting a precise set of correction parameters, including: Feature vectors are extracted by the deformation of the main beam, and initial control commands are generated using real-time data stream processing technology to obtain preliminary command data. Based on the preliminary instruction data and the characteristics of the adjustment mechanism, the target angle and displacement are calculated, and the adjustment parameter set is determined. By adjusting the parameter set and processing real-time data through a pre-established mapping model, the direction of the correction control is determined, and the direction correction data is obtained. Based on the direction correction data, obtain the dynamic response information of the adjustment mechanism, and determine the optimized instruction set by comparing and analyzing the updated control commands. Based on the optimized instruction set, the support vector machine algorithm is used to classify the correction effect and obtain the classification result data; By combining the classification results data with the changing trends of real-time data streams, the execution stability of the adjustment mechanism is judged, and stability assessment data is obtained. Based on the stability assessment data, historical response records are obtained, and the distribution characteristics of the adjustment parameters are analyzed using a clustering algorithm to determine the final set of correction parameters.

[0012] The process involves acquiring a set of correction parameters through a closed-loop control system, driving a hydraulic or servo mechanism to adjust the main beam's attitude, and simultaneously acquiring real-time feedback images during the adjustment process to obtain a second image of the main beam, including: The main beam attitude data is acquired through a closed-loop control system, a set of correction parameters is generated, and the hydraulic mechanism is driven to perform adjustments to obtain the adjusted attitude data. Based on the adjusted attitude data, real-time feedback images are acquired, and the edge features of the main beam are extracted using image processing technology to obtain a second image; Using the second image, the image data is processed through a pre-established template matching model to determine the direction of the main beam's attitude deviation and obtain the deviation direction data; Based on the deviation direction data, the response characteristics of the servo mechanism are obtained, and the correction parameter set is updated by comparison and analysis to determine the optimization parameter set; For the optimized parameter set, the servo mechanism is driven to adjust the posture of the main beam, and real-time feedback during the adjustment process is collected to obtain an updated feedback image; By updating the feedback image, a convolutional neural network is used to extract posture change features, determine the posture stability of the adjusted main beam, and obtain stability data. Based on stability data and real-time feedback trends, clustering algorithms are used to analyze the distribution characteristics of attitude adjustments and determine the final adjustment result.

[0013] The process involves extracting and spatially locating the repeated edge features from the second image of the main beam, determining whether the adjusted tilt angle change and deformation state have returned to the normal range, and outputting the adjusted state parameters, including: Edge feature data is obtained by processing the second image of the main beam using edge detection technology; Based on the edge feature data, spatial positioning technology is used to analyze and determine the spatial location data of the main beam; Based on spatial location data, geometric calculation methods are used to obtain tilt angle variation data; By combining the tilt angle variation data with the pre-established deformation model, the deformation state is determined and deformation distribution data is obtained; For deformation distribution data, if the deformation exceeds a preset threshold, the state parameter adjustment value is determined through parameter mapping technology. Based on the adjusted state parameters, a convolutional neural network is used to extract the change features, determine the stability after adjustment, and obtain stability data. By updating stability data and edge feature data, the final adjustment state parameters are obtained.

[0014] The process involves acquiring the adjusted state parameters, calculating the residual value by comparing it with the initial state parameters, and if the residual value is greater than a preset threshold, then re-inputting the parameters into the dynamic recognition network to generate a new round of correction instructions, including: The residual value is obtained by comparing the state parameters with the initial state. For the residual value, the value is determined by calculating the residual. If the residual value is greater than the preset threshold, it is processed by dynamic identification technology. Acquire dynamic recognition results and generate correction instructions through network processing; Based on the correction command, the state parameters are updated using state adjustment technology to obtain the adjusted data; By comparing the adjusted data with the initial state again, new residual values ​​are obtained. If the new residual value is less than a preset threshold, the final state is determined through instruction generation technology. Based on the final state, the system parameters are updated using a data input method to obtain a stable result.

[0015] The process involves driving the adjustment mechanism in conjunction with the new round of correction commands, collecting and analyzing the final feedback image, determining that the main beam's tilt angle and bending deformation have stabilized within a safe range, and completing the closed-loop control process, including: The adjustment data is obtained through the correction instruction generation module, which drives the adjustment mechanism to operate and obtains the linkage execution status. Based on the feedback images collected during the linkage execution status, image processing technology is used to analyze and determine the trend of the main beam inclination angle. If the tilt angle changes beyond the stable range, the feedback image is processed by a convolutional neural network to determine the degree of bending deformation. Based on the degree of bending deformation, an update command is generated, which drives the adjustment mechanism to execute again and obtain a new feedback image. The deformation stability state is analyzed by the new feedback image. If the stability range is reached, the closed-loop control result is recorded. The closed-loop control results are processed using control process optimization techniques to determine the final stable range data. The system parameters are updated based on the final stable range data, and the control process is adjusted.

[0016] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses an intelligent posture correction method for a main beam. The method acquires panoramic images of the main beam using a high-resolution camera array, employs multi-frame fusion technology to process vibration and lighting changes under complex working conditions, extracts edge features, and utilizes a deep learning model for spatial localization and classification to determine the beam's tilt angle changes and bending deformation distribution. Based on state parameters, the invention judges the deformation trend, uses a dynamic recognition network to identify specific deformation types and locations, and generates deformation feature vectors. Based on this, the invention generates real-time correction control commands to drive the adjustment mechanism to perform posture adjustments. Through closed-loop control and multiple iterations, this invention can precisely adjust the main beam's posture, ensuring that its tilt angle changes and bending deformation remain within a safe range, thus achieving intelligent and precise posture correction control for the main beam. Attached Figure Description

[0017] Figure 1 This is a flowchart of the present invention.

[0018] Figure 2 This is a schematic diagram of the correction method of the present invention.

[0019] Figure 3 This is another schematic diagram of the correction method of the present invention. Detailed Implementation

[0020] To further understand the content of this invention, a detailed description of the invention is provided in conjunction with the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0021] like Figure 1-3 This embodiment describes a bridge gantry crane main beam deformation detection and lifting positioning compensation system based on AI image recognition technology. Specific steps may include: Step S101: Acquire panoramic image data of the main beam using a high-resolution camera array. For vibration and light changes under complex working conditions, preprocess the acquired images using multi-frame fusion technology to obtain a stable and clear first image of the main beam.

[0022] A high-resolution camera array is used to acquire panoramic image data of the main beam, resulting in an initial image set. Multiple frames are then extracted from the initial image set to account for vibration and lighting variations, and multi-frame fusion technology is employed to obtain a second image. If residual noise exists in the second image, it is smoothed using a mean filtering algorithm to obtain a third image. Based on the pixel distribution of the third image, its sharpness is assessed to determine if it meets a preset threshold, yielding a sharpness evaluation result. Using this sharpness evaluation result, the Laplacian operator is applied to enhance the edge features of the third image, resulting in a fourth image. The stability characteristics of the fourth image are then acquired to determine if it meets the output requirements under complex working conditions, resulting in a final judgment result. Based on the final judgment result, a stable and clear first image of the main beam is output.

[0023] For example, a high-resolution camera array was used to acquire panoramic images of the main beam. First, eight 50-megapixel industrial cameras were deployed at key locations on the bridge. Synchronous triggering technology ensured that all cameras simultaneously acquired images within 1 millisecond, with each camera continuously capturing images at 30 frames per second to obtain high-definition images of the main beam at different times. To address vibration and light variations under complex conditions, a multi-frame fusion technique based on feature point matching was used for image preprocessing. First, the SIFT algorithm was used to extract key feature points from each frame. Then, the RANSAC algorithm was used for feature point matching, eliminating mismatches and retaining feature point pairs with a matching degree higher than 95%. Next, a weighted average method was used to fuse the multiple frames, with the weights dynamically adjusted according to image clarity and lighting conditions. For areas with significant lighting variations, a histogram equalization algorithm was used for brightness correction to ensure uniform brightness in the fused image. Through this processing, a stable and clear first image of the main beam was obtained, with an image resolution of 8000×6000 pixels and a signal-to-noise ratio improved to 45dB, providing a high-quality data foundation for subsequent structural health monitoring.

[0024] Step S102: Extract edge feature points from the first image of the main beam, use a deep learning model to spatially locate and classify the feature points, determine the current tilt angle change value and bending deformation distribution of the main beam, and output the main beam state parameters.

[0025] Edge feature points are extracted from the main beam image to obtain a feature point set. A deep learning model is used to spatially locate the feature point set, obtaining location coordinate data. Feature classification is performed on the location coordinate data to determine the tilt angle variation distribution. Bending deformation features are calculated from the tilt angle variation distribution to obtain deformation distribution data. Based on the deformation distribution data and the location coordinate data, it is determined whether the state parameters exceed a preset threshold, obtaining parameter evaluation results. The output of the deep learning model is adjusted based on the parameter evaluation results to obtain optimized state parameters. If the optimized state parameters meet the preset conditions, the final state parameters are output.

[0026] For example, firstly, images of the main beam surface are acquired at a rate of 30 frames per second using a high-resolution industrial camera. The Canny edge detection algorithm is used to extract feature points, with high and low thresholds set to 120 and 80 respectively, and a Gaussian filter kernel size of 5×5, ultimately obtaining approximately 1500 edge feature point coordinates. These feature points are then input into an improved ResNet50 network. The network adds a spatial attention module at the input, adjusts the convolution kernel size to 3×3, and sets the stride to 1. A 128-dimensional feature vector is extracted through five convolutional layers. Transfer learning is applied based on ImageNet pre-trained weights, and fine-tuning is performed using a dataset containing 100,000 steel structure samples, achieving a classification accuracy of 97%. For the spatial localization task, the DBSCAN algorithm based on feature point clustering is used, with a neighborhood radius ε = 5 pixels and a minimum sample size MinPts = 15. The feature points are divided into 20 spatial clusters, and the localization accuracy of the center coordinates of each cluster reaches ±3mm. The tilt angle calculation module fits the spatial curve formed by the centers of each cluster using the least squares method. When the vertical offset of the characteristic point measured at the mid-span of the main beam reaches 15mm, the tilt angle change value is calculated to be 02° based on the beam length of 35m. The bending deformation analysis adopts the cubic spline interpolation method, and a monitoring section is set every 5m in the longitudinal direction of the beam. The maximum bending strain is measured to occur at 14m from the support, with a strain of 325με and a corresponding deflection value of 2mm. The system finally outputs a 12-dimensional state parameter vector containing tilt angle, deflection, strain distribution matrix, etc., with a data update frequency of 10Hz.

[0027] Step S103: After obtaining the main beam state parameters, calculate the spatial attitude deviation of the main beam through a preset geometric transformation algorithm, and compare and analyze historical data to determine whether the trend of the main beam inclination angle change and the degree of deformation exceed the safety threshold.

[0028] After obtaining the main beam's state parameters, spatial attitude deviations are calculated using a geometric transformation algorithm to obtain deviation data. Historical data is then compared and analyzed to determine the inclination angle change trend. This trend is used to determine whether the deformation exceeds a preset threshold, yielding a judgment result. Based on the judgment result, relevant change patterns are extracted from historical data to obtain trend distribution data. This trend distribution data, combined with the spatial attitude deviation, determines the adjustment direction of the state parameters. The adjusted state parameters are then used to classify the deformation degree using a support vector machine algorithm, obtaining classification data. Finally, this classification data is processed in conjunction with a preset threshold to determine the final state assessment result.

[0029] For example, after obtaining the main beam's state parameters, a geometric transformation algorithm based on Euler angles is used to calculate the spatial attitude deviation of the main beam. The initial attitude is set as a reference benchmark. The current attitude is compared with the benchmark attitude using a rotation matrix, and the deviation angles of the main beam in the X, Y, and Z directions are calculated to be 3°, 1°, and 0.5°, respectively. Combining historical data, a time series analysis method is used to predict the trend of the main beam's inclination angle change. An ARIMA model is used to fit 1000 sets of inclination angle data from the past, predicting that the inclination angle change will increase by 0.2° within the next 10 minutes, and determining whether the current inclination angle change exceeds the preset safety threshold of 5°. Regarding the degree of deformation of the main beam, a deformation assessment method based on finite element analysis is used. The main beam is divided into 100 element nodes, and the overall deformation distribution is calculated through node displacement. When the maximum node displacement reaches 12mm, combined with the material's elastic modulus of 210GPa and the section moment of inertia of 0.2m... 4 The system calculates the overall deformation of the main beam to be 15%, comparing it to the safety threshold of 2% to determine whether it is within a safe range. Through real-time data updates, the system generates a spatial attitude deviation report every 5 seconds, including the deviation angle, deformation degree, and safety status assessment results, providing data support for subsequent decision-making.

[0030] Step S104: If the trend of tilt angle change exceeds the threshold, the state parameters are input into the pre-established dynamic recognition network to identify the specific deformation type and location based on the image data processing results, and generate the main beam deformation feature vector.

[0031] If the tilt angle change exceeds a preset threshold, the state parameters are processed through a dynamic recognition network, and the deformation type and location are identified by combining image data to obtain the main beam deformation feature vector. The location identification results are analyzed using the main beam deformation feature vector, and a clustering algorithm is used to determine the deformation type distribution, obtaining type distribution data. Based on the type distribution data and the tilt angle change trend, the evolution direction of the main beam state is determined, and the state evolution data is determined. Based on the state evolution data, historical state parameters are obtained, and the current state parameters are adjusted through comparative analysis to obtain adjusted parameter data. Using the adjusted parameter data, a support vector machine algorithm is used to classify the deformation types, obtaining classification result data. Based on the classification result data, the location identification information is processed by image data to determine the final distribution characteristics of the main beam state. Based on the final distribution characteristics, the dynamically updated values ​​of the state parameters are obtained, the stability trend of the main beam state is judged, and stability assessment data is obtained.

[0032] For example, when the inclination angle of the main beam exceeds a preset safety threshold, the system inputs the real-time collected state parameters into a pre-trained convolutional neural network (CNN) model. This model, based on the ResNet-50 architecture, extracts features and classifies the deformation images of the main beam through transfer learning. The input image resolution is 1920×1080. After preprocessing, it is trained using the Adam optimizer with a learning rate of 0.01. The training set contains 5000 labeled images, covering deformation types such as bending, twisting, and local indentation. The model output is a probability distribution of deformation types. When the bending deformation probability reaches 85%, the system further locates the deformation area using a region growing algorithm, generating a deformation feature vector containing the deformation center coordinates (X=1200, Y=450), deformation area (S=3500mm²), and maximum deformation depth (D=8mm). Combining the material properties of the main beam, a deformation assessment method based on strain energy density is used to calculate the strain energy density of the deformation area as 15J / m³, which is compared with the safety threshold of 1J / m³ to determine whether the deformation is within a controllable range. The system updates the deformation feature report every 10 seconds to provide data support for subsequent maintenance decisions.

[0033] Step S105: Based on the deformation characteristic vector of the main beam, real-time data stream processing technology is used to generate correction control commands, determine the target angle and displacement of the adjustment mechanism, and output a precise set of correction parameters.

[0034] Feature vectors are extracted from the main beam deformation, and initial control commands are generated using real-time data stream processing technology to obtain preliminary command data. Based on the preliminary command data and the characteristics of the adjustment mechanism, the target angle and displacement are calculated to determine the adjustment parameter set. Using the adjustment parameter set, real-time data is processed through a pre-established mapping model to determine the direction of the correction control, resulting in direction correction data. For the direction correction data, the dynamic response information of the adjustment mechanism is acquired, and the control commands are updated through comparative analysis to determine the optimized command set. Based on the optimized command set, the correction effect is classified using a support vector machine algorithm to obtain classification result data. Using the classification result data and the changing trend of the real-time data stream, the execution stability of the adjustment mechanism is assessed, resulting in stability evaluation data. For the stability evaluation data, historical response records are acquired, and the distribution characteristics of the adjustment parameters are analyzed using a clustering algorithm to determine the final correction parameter set.

[0035] For example, in the analysis of the main beam deformation feature vector, displacement data of the main beam is first collected in real time by sensors. For instance, if the displacement value collected at a certain moment is 5 mm, the Fast Fourier Transform (FFT) algorithm is then used to perform frequency domain analysis on the data, extracting the vibration frequency of the main beam as 8 Hz. Based on this data, a Kalman filter algorithm is used to predict the deformation state of the main beam. The prediction result shows that the displacement deviation of the main beam at the next moment is 2 mm. To generate a correction control command, the system calculates the required adjustment angle of 5 degrees and the displacement of 2 mm based on the prediction result and the target displacement value (set to 0 mm). Next, the system generates a precise correction parameter set through a PID control algorithm, where the proportional coefficient is 8, the integral time is 5 seconds, and the derivative time is 2 seconds. Finally, the system sends the correction parameter set to the adjustment mechanism, which makes real-time adjustments based on the target angle and displacement to ensure that the deformation of the main beam is effectively controlled. The entire process is implemented through real-time data stream processing technology to ensure rapid response and accurate execution of the correction command.

[0036] Step S106: Obtain the correction parameter set through the closed-loop control system, drive the hydraulic or servo mechanism to perform main beam attitude adjustment, and simultaneously collect real-time feedback images during the adjustment process to obtain the second image of the main beam.

[0037] The main beam's attitude data is acquired through a closed-loop control system, generating a correction parameter set. This parameter set drives the hydraulic mechanism to perform adjustments, resulting in adjusted attitude data. Real-time feedback images are acquired based on the adjusted attitude data, and image processing techniques are used to extract the main beam's edge features, yielding a second image. Using this second image, a pre-established template matching model is applied to process the image data, determining the direction of the main beam's attitude deviation and obtaining deviation direction data. Based on the deviation direction data, the response characteristics of the servo mechanism are acquired, and the correction parameter set is updated through comparative analysis to determine an optimized parameter set. For the optimized parameter set, the servo mechanism is driven to adjust the main beam's attitude, while simultaneously acquiring real-time feedback during the adjustment process, resulting in an updated feedback image. Using the updated feedback image, a convolutional neural network is employed to extract attitude change features, assessing the stability of the adjusted main beam's attitude and obtaining stability data. Based on the stability data and the changing trends in real-time feedback, a clustering algorithm is used to analyze the distribution characteristics of the attitude adjustment, determining the final adjustment result.

[0038] For example, in a closed-loop control system, the system collects real-time attitude data of the main beam using high-precision sensors, such as detecting a tilt angle of 5 degrees and a horizontal displacement deviation of 2 millimeters. Based on this data, the system uses the least squares method to fit and analyze the attitude of the main beam, calculating the deviation matrix between the current attitude and the target attitude. Subsequently, the system generates a set of correction parameters using a fuzzy control algorithm, where the fuzzy rule base contains 15 rules. The input variables are the angle deviation and displacement deviation, and the output variables are the pressure value of the hydraulic mechanism and the speed of the servo motor. Based on the calculation results, the system sets the pressure value of the hydraulic mechanism to 12 MPa and the speed of the servo motor to 1200 rpm. During the adjustment process, the system acquires real-time images of the main beam using a high-speed industrial camera, and uses image processing algorithms to perform edge detection and feature extraction on the images to obtain a second image of the main beam. By comparing the differences between the second image and the target image, the system further optimizes the correction parameters, such as adjusting the pressure value of the hydraulic mechanism to 15 MPa and the speed of the servo motor to 1180 rpm. Throughout the process, the system adopts an adaptive control strategy to ensure the accuracy and stability of the main beam attitude adjustment.

[0039] Step S107: Extract and analyze the repeated edge features of the second image of the main beam, determine whether the change in tilt angle and deformation state after adjustment have returned to the normal range, and output the adjusted state parameters.

[0040] Edge feature data is obtained by processing the second image of the main beam using edge detection technology. Based on this edge feature data, spatial positioning technology is used to determine the spatial location data of the main beam. According to the spatial location data, geometric calculation methods are used to obtain the tilt angle change data. By combining the tilt angle change data with a pre-established deformation model, the deformation state is determined, resulting in deformation distribution data. If the deformation exceeds a preset threshold, parameter mapping technology is used to determine the adjustment values ​​for the state parameters. Based on these adjustment values, a convolutional neural network is used to extract change features and determine the stability after adjustment, obtaining stability data. Finally, the adjusted state parameters are obtained by updating the stability data and edge feature data.

[0041] For example, the system performs Canny edge detection on the second image of the main beam, setting high and low thresholds of 100 and 200 respectively, extracting the feature point set of the main beam contour, detecting straight line segments through Hough transform, fitting the straight line equations of the upper and lower edges of the main beam with slopes of 0.87 and 0.92 respectively, and calculating the current tilt angle as 2 degrees. The SIFT algorithm is used to match the 128-dimensional feature descriptors of the second image with the standard template image, achieving a matching success rate of 92%. After removing mismatched points using the RANSAC algorithm, the spatial displacement deviation of the key corner points of the main beam is calculated to be 3 mm. The tilt angle and displacement data are input into a pre-trained BP neural network model, which contains three hidden layers with 64, 32, and 16 nodes respectively, using ReLU as the activation function and the Sigmoid function as the output layer. The probability of predicting the main beam deformation state returning to the normal range is 97%. Based on the prediction results, the system outputs adjusted state parameters: tilt angle error of 5 degrees, displacement error of 8 mm, and structural stress distribution standard deviation of 16 MPa, meeting the preset threshold conditions. Kalman filtering was used to suppress noise in the image measurement data during the process. The process noise covariance was set to 0.1 and the observation noise covariance was set to 0.5 to ensure data stability.

[0042] Step S108: Obtain the adjusted state parameters, calculate the residual value by comparing it with the initial state parameters, and if the residual value is greater than the preset threshold, re-input the parameters into the dynamic recognition network to generate a new round of correction instructions.

[0043] The residual value is obtained by comparing the state parameters with the initial state. For the residual value, a residual calculation method is used to determine its value. If the residual value is greater than a preset threshold, dynamic recognition technology is applied. The dynamic recognition result is obtained and a correction instruction is generated through network processing. Based on the correction instruction, the state parameters are updated using state adjustment technology to obtain adjusted data. The adjusted data is compared with the initial state again to obtain a new residual value. If the new residual value is less than the preset threshold, the final state is determined using instruction generation technology. Based on the final state, the system parameters are updated using a data input method to obtain a stable result.

[0044] For example, 3D point cloud data of the main beam after adjustment is acquired using a laser scanner. The ICP algorithm is used for registration with the initial design model, and the iteration termination condition is set to a point cloud mean square error of less than 5 mm. After registration, the coordinate residuals of key measurement points are calculated, with the maximum residual occurring at the mid-span position, reaching 8 mm. The residual data is input into a dynamic recognition network based on LSTM. This network's input layer contains 12 temporal feature nodes, the hidden layer uses a bidirectional LSTM structure with 128 units, and the output layer is mapped to a 6-dimensional correction vector through a fully connected layer. The network is trained using the Adam optimizer with an initial learning rate of 0.01 and a batch size of 32, achieving a prediction accuracy of 93% on the validation set. When the residual at the mid-span position exceeds a preset 2 mm threshold, the network outputs a correction instruction set containing parameters such as the hydraulic jacking device stroke adjustment (+2 mm) and the stay cable tension adjustment value (-5 kN). Meanwhile, the Monte Carlo method was used to perform 500 random simulations of the correction process, and the maximum equivalent stress after stress redistribution was calculated to be 185 MPa, which is lower than 85% of the material's yield strength. During the process, a sliding window mechanism was set to smooth the time-series residual data. The window width was set to 5 sampling periods, and the weighting coefficients were configured according to a Gaussian distribution to ensure the stability of dynamic identification.

[0045] Step S109: Drive the adjustment mechanism to perform linkage execution according to the new round of correction command, collect and analyze the final feedback image, and determine that the main beam tilt angle change and bending deformation have stabilized within the safe range, thus completing the closed-loop control process.

[0046] The adjustment data is acquired through the correction command generation module, driving the adjustment mechanism to operate and obtaining the linkage execution status. Feedback images are collected based on the linkage execution status and analyzed using image processing technology to determine the trend of the main beam's inclination angle. If the inclination angle change exceeds the stable range, the feedback image is processed using a convolutional neural network to determine the degree of bending deformation. An update command is generated based on the degree of bending deformation, driving the adjustment mechanism to execute again and acquire new feedback images. The deformation stability state is analyzed using the new feedback images; if a stable range is reached, the closed-loop control result is recorded. Control flow optimization technology is used to process the closed-loop control result to determine the final stable range data. System parameters are updated based on the final stable range data, completing the control flow adjustment.

[0047] For example, driven by a new round of correction commands, the adjustment mechanism, through coordinated execution, first adjusts the main beam in real time based on a preset PID control algorithm, setting the proportional gain Kp to 8, the integral time Ti to 5 seconds, and the derivative time Td to 2 seconds. The system collects the tilt angle data of the main beam through high-precision sensors at a sampling frequency of 100Hz to ensure the real-time performance and accuracy of the data. The collected data is processed by a Kalman filter algorithm to filter out noise interference, resulting in a smooth tilt angle change curve. Subsequently, the system uses finite element analysis software to simulate the bending deformation of the main beam, setting the material elastic modulus to 210GPa, Poisson's ratio to 3, and the mesh generation accuracy to 1mm, calculating the deformation of the main beam under different loads. By comparing the actual collected tilt angle data with the simulation results, the system determines whether the tilt angle change and bending deformation of the main beam are stable within a safe range. If the tilt angle change exceeds a preset threshold of ±5°, the system will automatically trigger the adjustment mechanism for secondary correction until the tilt angle stabilizes within ±2°. Finally, the system acquires the final feedback image of the main beam through image acquisition equipment, uses image processing algorithms for edge detection and feature extraction, and analyzes whether the geometry of the main beam meets the design requirements. If all parameters meet safety standards, the system completes the closed-loop control process, ensuring the stability and safety of the main beam.

[0048] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A bridge crane main beam deformation detection and lifting positioning compensation system based on AI image recognition technology, characterized in that, The method includes: A high-resolution camera array was used to acquire panoramic image data of the main beam. Multi-frame fusion technology was used to preprocess the acquired images to obtain a stable and clear first image of the main beam under complex working conditions of vibration and light changes. Edge feature points are extracted from the first image of the main beam. A deep learning model is used to spatially locate and classify the feature points, determine the current tilt angle change and bending deformation distribution of the main beam, and output the main beam state parameters. After obtaining the main beam state parameters, the spatial attitude deviation of the main beam is calculated through a preset geometric transformation algorithm. Combined with historical data comparison and analysis, it is determined whether the trend of the main beam inclination angle change and the degree of deformation exceed the safety threshold. If the trend of tilt angle change exceeds the threshold, the state parameters are input into the pre-established dynamic recognition network, and the specific deformation type and location are identified based on the image data processing results to generate the main beam deformation feature vector. Based on the deformation characteristic vector of the main beam, real-time data stream processing technology is used to generate correction control commands, determine the target angle and displacement of the adjustment mechanism, and output a precise set of correction parameters. The closed-loop control system acquires the set of correction parameters, drives the hydraulic or servo mechanism to perform the main beam attitude adjustment, and simultaneously collects real-time feedback images during the adjustment process to obtain the second image of the main beam. Based on the extraction and spatial positioning analysis of repeated edge features in the second image of the main beam, it is determined whether the change in tilt angle and deformation state after adjustment have returned to the normal range, and the adjusted state parameters are output. The adjusted state parameters are obtained, and the residual value is calculated by comparing them with the initial state parameters. If the residual value is greater than the preset threshold, the parameters are re-input into the dynamic recognition network to generate a new round of correction instructions. Driven by a new round of correction commands, the adjustment mechanism is executed in conjunction with the data collection and analysis of the final feedback images. This confirms that the main beam's tilt angle and bending deformation have stabilized within a safe range, thus completing the closed-loop control process.

2. The bridge gantry crane main beam deformation detection and lifting positioning compensation system based on AI image recognition technology according to claim 1, characterized in that, The process involves acquiring panoramic image data of the main beam using a high-resolution camera array. To address vibration and light variations under complex operating conditions, multi-frame fusion technology is employed to preprocess the acquired images, resulting in a stable and clear first image of the main beam, including: An initial image set was obtained by acquiring panoramic image data of the main beam using a high-resolution camera array; To address vibration and light variations, multiple frames of data are acquired from the initial image set and processed using multi-frame fusion technology to obtain the second image. If residual noise exists in the second image, the second image is smoothed using a mean filtering algorithm to obtain the third image; Based on the pixel distribution of the third image, determine whether the sharpness reaches the preset threshold and obtain the sharpness evaluation result; Based on the sharpness evaluation results, the edge features of the third image are enhanced using the Laplacian operator to obtain the fourth image; The stability effect features of the fourth image are obtained to determine whether they meet the output requirements under complex working conditions, and the final judgment result is obtained. Based on the final judgment result, a stable and clear first image of the main beam is output.

3. The bridge gantry crane main beam deformation detection and lifting positioning compensation system based on AI image recognition technology according to claim 1, characterized in that, The process involves extracting edge feature points from the first image of the main beam, using a deep learning model to spatially locate and classify these feature points, determining the current tilt angle change and bending deformation distribution of the main beam, and outputting the main beam state parameters, including: Edge feature points are extracted from the main beam image to obtain a set of feature points; A deep learning model is used to spatially locate the feature point set and obtain the location coordinate data; The positioning coordinate data is classified by features to determine the distribution of tilt angle changes; The bending deformation characteristics are calculated from the distribution of tilt angle variation to obtain deformation distribution data; Based on the deformation distribution data and positioning coordinate data, determine whether the state parameters exceed the preset threshold and obtain the parameter evaluation results; Adjust the output of the deep learning model based on the parameter evaluation results to obtain the optimized state parameters; If the optimized state parameters meet the preset conditions, the final state parameters will be output.

4. The bridge gantry crane main beam deformation detection and lifting positioning compensation system based on AI image recognition technology according to claim 1, characterized in that, After obtaining the main beam state parameters, the spatial attitude deviation of the main beam is calculated using a preset geometric transformation algorithm. Combined with historical data comparison and analysis, it is determined whether the trend of the main beam's inclination angle change and the degree of deformation exceed the safety threshold, including: After obtaining the main beam state parameters, the spatial attitude deviation is calculated using a geometric transformation algorithm to obtain the deviation data; For the deviation data, historical data are used for comparative analysis to determine the trend of tilt angle changes; By analyzing the trend of tilt angle changes, it can be determined whether the degree of deformation exceeds a preset threshold, and a judgment result can be obtained. Based on the judgment results, relevant change patterns are extracted from historical data to obtain trend distribution data; By using trend distribution data and combining it with spatial attitude deviation, the direction of adjustment of state parameters is determined; By using the adjusted state parameters, the degree of deformation is classified using the support vector machine algorithm to obtain classification data; For categorized data, data processing is performed in conjunction with preset thresholds to determine the final state assessment result.

5. The bridge gantry crane main beam deformation detection and lifting positioning compensation system based on AI image recognition technology according to claim 1, characterized in that, If the tilt angle change trend exceeds the threshold, the state parameters are input into a pre-established dynamic recognition network. Based on the image data processing results, the specific deformation type and location are identified, and a main beam deformation feature vector is generated, including: If the tilt angle change exceeds the preset threshold, the state parameters are processed by the dynamic recognition network, and the deformation type and location are identified by combining the image data to obtain the main beam deformation feature vector. The location identification results are analyzed by the deformation feature vector of the main beam, and the deformation type distribution is determined by the clustering algorithm to obtain the type distribution data; Based on the type distribution data and the trend of inclination angle changes, the evolution direction of the main beam's condition is determined, and the condition evolution data is identified. Based on the state evolution data, historical state parameters are obtained, and the current state parameters are adjusted through comparative analysis to obtain the adjusted parameter data; Using the adjusted parameter data, the deformation type is classified using the support vector machine algorithm to obtain the classification result data; Based on the classification results data, the final distribution characteristics of the main beam state are determined by combining image data processing with location recognition information. By analyzing the final distribution characteristics, the dynamic update values ​​of the state parameters are obtained, the stability trend of the main beam is determined, and stability assessment data is obtained.

6. The bridge gantry crane main beam deformation detection and lifting positioning compensation system based on AI image recognition technology according to claim 1, characterized in that, The process involves generating correction control commands based on the main beam deformation characteristic vector using real-time data stream processing technology, determining the target angle and displacement of the adjustment mechanism, and outputting a precise set of correction parameters, including: Feature vectors are extracted by the deformation of the main beam, and initial control commands are generated using real-time data stream processing technology to obtain preliminary command data. Based on the preliminary instruction data and the characteristics of the adjustment mechanism, the target angle and displacement are calculated, and the adjustment parameter set is determined. By adjusting the parameter set and processing real-time data through a pre-established mapping model, the direction of the correction control is determined, and the direction correction data is obtained. Based on the direction correction data, obtain the dynamic response information of the adjustment mechanism, and determine the optimized instruction set by comparing and analyzing the updated control commands. Based on the optimized instruction set, the support vector machine algorithm is used to classify the correction effect and obtain the classification result data; By combining the classification results data with the changing trends of real-time data streams, the execution stability of the adjustment mechanism is judged, and stability assessment data is obtained. Based on the stability assessment data, historical response records are obtained, and the distribution characteristics of the adjustment parameters are analyzed using a clustering algorithm to determine the final set of correction parameters.

7. The bridge gantry crane main beam deformation detection and lifting positioning compensation system based on AI image recognition technology according to claim 1, characterized in that, The process involves acquiring a set of correction parameters through a closed-loop control system, driving a hydraulic or servo mechanism to adjust the main beam's attitude, and simultaneously acquiring real-time feedback images during the adjustment process to obtain a second image of the main beam, including: The main beam attitude data is acquired through a closed-loop control system, a set of correction parameters is generated, and the hydraulic mechanism is driven to perform adjustments to obtain the adjusted attitude data. Based on the adjusted attitude data, real-time feedback images are acquired, and the edge features of the main beam are extracted using image processing technology to obtain a second image; Using the second image, the image data is processed through a pre-established template matching model to determine the direction of the main beam's attitude deviation and obtain the deviation direction data; Based on the deviation direction data, the response characteristics of the servo mechanism are obtained, and the correction parameter set is updated by comparison and analysis to determine the optimization parameter set; For the optimized parameter set, the servo mechanism is driven to adjust the posture of the main beam, and real-time feedback during the adjustment process is collected to obtain an updated feedback image; By updating the feedback image, a convolutional neural network is used to extract posture change features, determine the posture stability of the adjusted main beam, and obtain stability data. Based on stability data and real-time feedback trends, clustering algorithms are used to analyze the distribution characteristics of attitude adjustments and determine the final adjustment result.

8. The bridge gantry crane main beam deformation detection and lifting positioning compensation system based on AI image recognition technology according to claim 1, characterized in that, The process involves extracting and spatially locating the repeated edge features from the second image of the main beam, determining whether the adjusted tilt angle change and deformation state have returned to the normal range, and outputting the adjusted state parameters, including: Edge feature data is obtained by processing the second image of the main beam using edge detection technology; Based on the edge feature data, spatial positioning technology is used to analyze and determine the spatial location data of the main beam; Based on spatial location data, geometric calculation methods are used to obtain tilt angle variation data; By combining the tilt angle variation data with the pre-established deformation model, the deformation state is determined and deformation distribution data is obtained; For deformation distribution data, if the deformation exceeds a preset threshold, the state parameter adjustment value is determined through parameter mapping technology. Based on the adjusted state parameters, a convolutional neural network is used to extract the change features, determine the stability after adjustment, and obtain stability data. By updating stability data and edge feature data, the final adjustment state parameters are obtained.

9. The bridge gantry crane main beam deformation detection and lifting positioning compensation system based on AI image recognition technology according to claim 1, characterized in that, The process involves acquiring the adjusted state parameters, calculating the residual value by comparing it with the initial state parameters, and if the residual value is greater than a preset threshold, then re-inputting the parameters into the dynamic recognition network to generate a new round of correction instructions, including: The residual value is obtained by comparing the state parameters with the initial state. For the residual value, the value is determined by calculating the residual. If the residual value is greater than the preset threshold, it is processed by dynamic identification technology. Acquire dynamic recognition results and generate correction instructions through network processing; Based on the correction command, the state parameters are updated using state adjustment technology to obtain the adjusted data; By comparing the adjusted data with the initial state again, new residual values ​​are obtained. If the new residual value is less than a preset threshold, the final state is determined through instruction generation technology. Based on the final state, the system parameters are updated using a data input method to obtain a stable result.

10. The bridge gantry crane main beam deformation detection and lifting positioning compensation system based on AI image recognition technology according to claim 1, characterized in that, The process involves driving the adjustment mechanism in conjunction with the new round of correction commands, collecting and analyzing the final feedback image, determining that the main beam's tilt angle and bending deformation have stabilized within a safe range, and completing the closed-loop control process, including: The adjustment data is obtained through the correction instruction generation module, which drives the adjustment mechanism to operate and obtains the linkage execution status. Based on the feedback images collected during the linkage execution status, image processing technology is used to analyze and determine the trend of the main beam inclination angle. If the tilt angle changes beyond the stable range, the feedback image is processed by a convolutional neural network to determine the degree of bending deformation. Based on the degree of bending deformation, an update command is generated, which drives the adjustment mechanism to execute again and obtain a new feedback image. The deformation stability state is analyzed by the new feedback image. If the stability range is reached, the closed-loop control result is recorded. The closed-loop control results are processed using control process optimization techniques to determine the final stable range data. The system parameters are updated based on the final stable range data, and the control process is adjusted.

Images