Intelligent image analysis system and method for safety detection of special equipment

Abstract

The invention relates to the technical field of safety detection, and discloses an intelligent image analysis system and method for safety detection of special equipment. The data acquisition module is used for acquiring three-dimensional image data of the surface of the special equipment and generating point cloud data; the curved surface reconstruction module is used for generating a three-dimensional curved surface model of the surface of the special equipment based on the point cloud data; and the crack detection module is used for analyzing the three-dimensional curved surface model and identifying a crack form and an initial position. By adopting the technical scheme of the three-dimensional convolutional neural network, the three-dimensional curved surface model is analyzed through the deep learning model, and efficient and accurate crack detection is realized. Compared with a traditional method based on two-dimensional image processing in the prior art, the three-dimensional convolutional neural network can process complex three-dimensional space data and automatically recognize and classify crack areas, and the problem that the traditional method is insufficient in precision in complex equipment surface detection is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of safety detection, and particularly to an image intelligent analysis system and method for special equipment safety detection. Background Art

[0002] In the field of traditional special equipment safety detection, the detection and prediction of crack propagation have always been key tasks to ensure equipment safety. Currently, the crack detection of many equipment relies on manual visual inspection and traditional image processing techniques. Although these methods can identify cracks on the equipment surface to a certain extent, they have obvious deficiencies in accuracy and efficiency on complex equipment surfaces.

[0003] In the prior art, crack detection usually relies on two-dimensional image analysis methods. Traditional two-dimensional image methods are difficult to handle complex three-dimensional equipment surfaces. Especially for equipment with curved surfaces or complex structures, two-dimensional images often cannot provide sufficient spatial information. This method has large defects in crack detection accuracy, especially when there are tiny cracks or cracks with complex shapes on the equipment surface, it is easy to have problems of missed detection and false detection.

[0004] In the existing crack propagation path prediction methods, many technologies rely on simple prediction models based on stress concentration or local curvature. These models fail to consider the actual complexity of crack propagation, such as multiple factors like material properties, environmental changes, stress field distribution, etc. Due to the rough processing of path optimization methods, traditional technologies often cannot accurately predict the future propagation path and speed of cracks, which leads to the inability to adjust equipment maintenance and management strategies in a timely and effective manner, thus increasing the risk of equipment failure.

[0005] The prior art lacks the function of real-time prediction of crack propagation trends. Traditional methods usually rely on static analysis of crack historical data, but cannot fully consider the real-time state during equipment operation, such as temperature, stress changes, etc. The propagation of cracks is often affected by complex dynamic factors, and existing methods cannot adjust the prediction results of crack propagation in a timely manner, which makes the crack propagation prediction lose real-time nature and causes potential risks. Especially when dealing with sudden crack propagation, there is a lack of effective emergency response mechanisms. Summary of the Invention

[0006] Aiming at the deficiencies of the prior art, the present invention provides an image intelligent analysis system and method for special equipment safety detection, which solves the problems of insufficient crack detection accuracy, inaccurate crack propagation path prediction, lack of real-time crack propagation prediction, and lag in maintenance decision support in the technology.

[0007] To achieve the above objectives, the present invention is realized through the following technical solutions: An image intelligent analysis system for special equipment safety detection, including; A data acquisition module, which is used to obtain three-dimensional image data of the surface of special equipment and generate point cloud data; A surface reconstruction module, which is used to generate a three-dimensional surface model of the surface of special equipment based on the point cloud data; A crack detection module, which is used to analyze the three-dimensional surface model, identify the crack morphology and starting position; A curvature analysis module, which is used to calculate the curvature information of each point on the surface of special equipment and determine the crack propagation direction based on the curvature information; A crack propagation path detection module, which is used to determine the crack propagation path based on the curvature information and crack morphology, and optimize the possible crack propagation trajectory by combining geodesic calculation methods; A crack propagation prediction module, which is used to predict the future crack propagation trend based on the historical data of crack propagation, combine the optimal control theory, and calculate the crack propagation speed; A real-time feedback and decision support module, which is used to provide equipment maintenance suggestions or control strategies for controlling the crack propagation speed based on the crack propagation prediction results.

[0008] Preferably, the data acquisition module includes a laser scanning device and a stereo vision camera device. The laser scanning device is used to obtain the three-dimensional coordinate information of the surface of special equipment through laser point cloud scanning, and the stereo vision camera device is used to obtain images of multiple perspectives of the surface of special equipment and generate a depth map using a depth estimation algorithm.

[0009] Preferably, the surface reconstruction module uses a grid interpolation method to convert the point cloud data into a continuous and smooth three-dimensional surface model.

[0010] Preferably, the crack detection module analyzes the three-dimensional surface model based on a three-dimensional convolutional neural network, extracts feature information through a deep learning model, and classifies cracks.

[0011] Preferably, the curvature analysis module calculates the local curvature information of each point on the surface of special equipment and determines the crack propagation trend along the geodesic direction of the equipment surface according to Riemannian geometry theory.

[0012] Preferably, the crack propagation path detection module optimizes the crack propagation path based on the variational method, combines the curvature information and crack morphology of the surface of special equipment, calculates the optimal crack propagation trajectory along the geodesic direction, and determines the minimum energy path of crack propagation.

[0013] Preferably, the crack propagation prediction module uses an optimal control method, predicts the future crack propagation path by establishing a dynamic model of crack propagation, and calculates the crack propagation speed.

[0014] Preferably, the real-time feedback and decision support module combines with an edge computing platform to analyze the crack propagation situation in real time and provide decision-making suggestions for equipment shutdown maintenance, preventive maintenance, or adjustment of operating parameters based on the prediction results.

[0015] Preferably, the real-time feedback and decision support module includes a graphical user interface for displaying the crack propagation state, prediction results, and maintenance suggestions of the equipment to the operator and providing interactive control options.

[0016] An image intelligent analysis method for special equipment safety detection includes the following steps; S1. Use a laser scanning device and a stereo vision camera device to obtain three-dimensional image data of the surface of special equipment and generate point cloud data; S2. Generate a three-dimensional surface model of the surface of special equipment by using a grid interpolation method based on the point cloud data; S3. Use a three-dimensional convolutional neural network to analyze the three-dimensional surface model and extract the starting position of the crack; S4. Curvature analysis, calculate the local curvature information of each point on the surface of special equipment, and determine the crack propagation trend according to Riemannian geometry theory; S5. Optimize the crack propagation path based on the variational method, combine the local curvature information, stress distribution, and material properties of the surface of special equipment, and calculate the optimal propagation trajectory of the crack along the geodesic direction; S6. Establish a dynamic model of crack propagation by using an optimal control method, predict the future propagation path of the crack, and calculate the crack propagation speed; S7. Provide equipment maintenance suggestions based on the crack propagation prediction results and perform real-time adjustment in combination with edge computing.

[0017] The present invention provides an image intelligent analysis system and method for special equipment safety detection. It has the following beneficial effects: 1. By adopting the technical solution of a three-dimensional convolutional neural network, the present invention analyzes the three-dimensional surface model through a deep learning model, realizing efficient and accurate crack detection. Compared with the traditional method based on two-dimensional image processing in the prior art, the three-dimensional convolutional neural network can process complex three-dimensional space data, automatically identify and classify crack regions, and solve the problem of insufficient accuracy in the detection of complex equipment surfaces by traditional methods.

[0018] 2. By adopting the variational method and geodesic calculation method to optimize the crack propagation path, the present invention obtains the optimal trajectory of crack propagation by comprehensively considering curvature, stress distribution, and material properties. Different from the simple path calculation method in the prior art, the optimization method of the present invention can more accurately simulate the crack propagation process, avoiding problems such as equipment damage or untimely maintenance caused by inaccurate path prediction.

[0019] 3. The present invention predicts crack propagation through optimal control theory. By establishing a kinetic model, it can predict the propagation trend and speed of cracks in real time. Compared with the existing solutions that fail to predict crack propagation in real time, the present invention can adjust the prediction results according to the real-time operating state of the equipment, reduce the uncontrollability of crack propagation and the occurrence of accidents, and significantly improve the safety of the equipment.

[0020] 4. The present invention introduces an edge computing platform for real-time feedback and decision support. By analyzing crack propagation data in real time and providing operation suggestions, it helps equipment managers make maintenance decisions quickly. Different from the existing systems that require remote computing and analysis, the edge computing solution ensures higher real-time performance and data processing efficiency, greatly reduces the system response time, and improves the accuracy of decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is the system framework diagram of the present invention; Figure 2 is the method flow chart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0023] Please refer to the attached Figure 1 , the embodiment of the present invention provides an image intelligent analysis system for special equipment safety detection, including; A data acquisition module for acquiring three-dimensional image data of the surface of special equipment and generating point cloud data; Specifically, the data acquisition module uses two methods, laser scanning and stereo vision, for data acquisition to meet the requirements of different application scenarios. Among them, laser scanning can provide high-density point cloud data and is suitable for the surface detection of equipment with complex shapes; stereo vision is based on the principle of multi-view imaging and can achieve efficient three-dimensional data reconstruction under specific working conditions.

[0024] In this embodiment, the laser scanning device mainly uses coherent laser scanning technology to measure the distance of the surface of special equipment with high precision by using a laser beam to obtain spatial point cloud data. The basic principle of laser scanning is to measure the time of flight of light, that is, the time difference between the laser emission and the reflection back.

[0025] Point cloud data can be expressed as; ; Wherein: is the spatial coordinate of the -th point in the point cloud; is the laser reflection intensity of this point; is the total number of point clouds collected.

[0026] In some embodiments, the laser scanning device adopts multi-line laser scanning, and multiple laser emission arrays work simultaneously to improve the scanning speed and reduce the scanning dead angle.

[0027] In a possible implementation manner, the stereo vision camera device uses a binocular camera for three-dimensional data acquisition. The basic principle of binocular imaging is to calculate the depth information of the target object using parallax, and the specific calculation formula is as follows; ; Wherein: is the depth coordinate of the target point; is the baseline distance between the binocular cameras; is the focal length of the camera lens; is the parallax of the corresponding points in the left and right images.

[0028] As an option, the system can combine structured light or phase-encoded light field technology to further improve the accuracy of three-dimensional reconstruction. Especially when detecting low-contrast surfaces, these technologies can effectively improve the stability of data acquisition.

[0029] The data acquisition module of the present invention can dynamically adjust parameters according to the different materials, geometric features, and optical properties of special equipment. On the surface of highly reflective materials or transparent materials, ordinary laser scanning may result in data loss. In this embodiment, the data acquisition module can adopt polarization light filtering or adaptive exposure adjustment strategies to ensure the integrity of the collected data.

[0030] In order to improve the detection ability of micro-cracks, the present invention supports super-resolution reconstruction methods, and improves the data acquisition accuracy through multi-frame image fusion technology. The mathematical model of this method can be described as follows: ; Wherein: is the image after super-resolution enhancement; is the -th low-resolution image; It is a super-resolution reconstruction function.

[0031] A surface reconstruction module for generating a three-dimensional surface model of the special equipment surface based on point cloud data; Specifically, in this embodiment, the surface reconstruction module converts the three-dimensional point cloud data obtained by the data acquisition module into a continuous and smooth three-dimensional surface model. Generally, due to noise interference or equipment precision limitations, the point cloud data may be incomplete or have errors, so it must be processed by appropriate algorithms. The accuracy of surface reconstruction directly affects the accuracy of subsequent modules such as crack detection, curvature analysis, and crack propagation path optimization.

[0032] The invented surface reconstruction module uses the mesh interpolation method to complete the generation of the three-dimensional surface. It can effectively eliminate the noise in the point cloud data and generate a smooth and seamless three-dimensional surface. Specifically, the surface reconstruction module optimizes the discrete representation of the point cloud to ensure that the reconstructed three-dimensional model has good continuity and geometric accuracy.

[0033] In this embodiment, the surface reconstruction module uses the mesh interpolation method. This method constructs a mesh by triangulating (Delaunay Triangulation) the point cloud data and uses the relationships between these mesh points for interpolation to obtain the surface model of the equipment surface. The mesh interpolation method is carried out through the following steps: Triangulation: Perform Delaunay triangulation on the point cloud data to generate a set of triangular meshes.

[0034] Local interpolation: Calculate the points within each triangular patch through an interpolation algorithm to generate a smooth surface.

[0035] The mesh interpolation method can quickly and effectively generate a surface model when the data acquisition is relatively dense and has high computational efficiency. Specifically, the point cloud data After triangulation, the local interpolation calculation for each triangle can be carried out through the following formula: ; Where: is the coordinate on the triangle; represents the function is the height value at this coordinate point, that is, the spatial position of the surface at this point.

[0036] In the processing of point cloud data, the surface reconstruction module of the present invention fully combines advanced algorithms such as Poisson surface reconstruction and mesh interpolation, which not only improves the accuracy of surface generation but also enhances the robustness of the system. These technologies can effectively handle complex geometric shapes, noisy data, and data sparsity problems that may occur in actual engineering, providing a reliable basis for subsequent crack detection and propagation analysis.

[0037] A crack detection module for analyzing a three-dimensional surface model to identify the starting position of a crack; Specifically, in this embodiment, the crack detection module is used to analyze the three-dimensional surface data generated by the surface reconstruction module, extract the morphological information of the crack, and determine the starting position of the crack. Generally, the forms of cracks are complex and may manifest as different types such as microcracks, cross cracks, fatigue cracks, stress corrosion cracks, etc. Therefore, it is difficult to achieve high-precision detection only relying on traditional image processing methods.

[0038] The present invention uses a three-dimensional convolutional neural network (3D-CNN) for crack detection. This method can make full use of point cloud data and surface information, extract the depth features of the crack area, and improve the recognition ability. Specifically, the crack detection module is based on the input three-dimensional surface model, combines information such as surface normal vectors, curvature distributions, and topological structures, judges the geometric features of the crack, and classifies it.

[0039] In this embodiment, the crack detection module uses a 3D-CNN deep learning network, and the input data is a three-dimensional surface model, which is input into the neural network in the form of voxels. A voxel is a three-dimensional data representation method that can effectively retain the spatial structure information of the crack. For the three-dimensional surface model , first convert it into a voxel representation ; ; Where: represents the spatial coordinates of the voxel; is the surface feature of the voxel, including information such as curvature, gradient, and normal vector Generally, the curvature of the crack area changes greatly, and preliminary screening can be carried out through Gaussian curvature. The definition of Gaussian curvature is as follows: ; Where: represents the local curvature feature of the surface; is the principal curvature of the surface; As an option, the crack detection module uses the surface gradient analysis method to determine whether cracks may exist in a local area. For a specific point , calculate the surface gradient; ; Where: represents a specific point in space, usually a point in a three-dimensional coordinate system, ; represents a scalar function, usually a function of the space coordinates ; represents the gradient of the function at the point , which is a vector pointing in the direction of the fastest growth of the function, and its magnitude represents the magnitude of the rate of change; represents with respect to the partial derivative in the direction, that is, the rate of change in the direction; represents with respect to the partial derivative in the direction, that is, the rate of change in the direction; represents with respect to the partial derivative in the direction, that is, the rate of change in the direction.

[0040] In a possible implementation, the crack detection module combines principal component analysis (PCA) for crack morphology modeling. PCA can be used to analyze the main direction of the crack area and establish a local feature space. For a given point cloud subset , calculate the covariance matrix.

[0041] ; Where: represents the covariance matrix; represents each point in the dataset, which is a point in three-dimensional space; represents the centroid of the point cloud, that is, the average value of the coordinates of all points; represents the number of points in the point cloud; ( ) represents each point The offset relative to the centroid of the point cloud reflects the position of each point relative to the centroid in three-dimensional space; ( represents the transpose of the offset and is used to calculate the matrix product to obtain the covariance matrix.

[0042] The eigenvectors obtained from PCA analysis correspond to the main directions of the cracks and can effectively distinguish the cracks from the background area.

[0043] As can be seen from the above description, the crack detection module of the present invention combines three-dimensional deep learning, surface gradient analysis, and PCA morphological modeling methods, which can effectively improve the accuracy of crack recognition and provide high-quality data support for subsequent crack propagation prediction.

[0044] The curvature analysis module is used to calculate the curvature information of each point on the surface of special equipment and determine the propagation direction of cracks based on the curvature information; Specifically, in this embodiment, the curvature analysis module is used to analyze the geometric features of the equipment surface. Curvature plays an important role in the prediction of crack propagation. Cracks usually propagate along areas with larger curvature. Therefore, accurately calculating the local curvature of each point on the equipment surface is decisive for subsequent crack propagation path optimization. The work content of the curvature analysis module directly affects the accuracy of subsequent crack detection, crack propagation prediction, and decision support.

[0045] The curvature analysis module adopts the curvature theory in Riemannian geometry to accurately model the three-dimensional surface. In the three-dimensional surface model, curvature is not only an important feature of the surface morphology but also a key factor affecting the crack propagation direction. By calculating the local curvature of each point on the equipment surface, the possible paths of crack propagation can be obtained, and the evolution of cracks can be predicted.

[0046] In this embodiment, the curvature analysis module first evaluates the local curvature of each point on the equipment surface through normal vector calculation. Specifically, for any point on the three-dimensional surface, the normal vector of this point is first calculated through gradient operation; ; where: and respectively represent the derivatives of this point in the and directions.

[0047] Through this normal vector information, the principal curvature can be further calculated.

[0048] In general, the calculation of curvature uses Gaussian curvature and mean curvature to characterize the degree of curvature of a surface. Gaussian curvature and mean curvature are respectively defined as: ; ; where: and are respectively the two principal curvatures at this point; is the Gaussian curvature, representing the overall degree of curvature of the surface at this point; is the mean curvature, describing the average curvature trend of the surface.

[0049] The curvature analysis module further infers the crack propagation direction through the principal curvature direction. Crack propagation usually occurs along the direction of the minimum curvature. Therefore, the system determines the possible crack propagation path based on the calculated principal curvature direction.

[0050] In a possible implementation, the curvature analysis module of the present invention combines multi-scale analysis, that is, by calculating the local curvature at different scales, the stability of crack propagation prediction is enhanced. Specifically, the system first calculates the overall curvature of the device surface at the global scale, and then calculates the curvature of the area near the crack at the local scale. This multi-scale analysis method can effectively capture the microscopic details of crack propagation and improve the prediction accuracy of the crack propagation path.

[0051] The crack propagation path detection module is used to determine the crack propagation path based on curvature information, stress distribution, and crack morphology, and optimize the possible crack propagation trajectory in combination with the geodesic calculation method; Specifically, in this embodiment, after the surface reconstruction and crack detection module completes data processing, the crack propagation path detection module is responsible for analyzing the crack propagation path and determining the future development direction of the crack. The task of this module is to infer the crack propagation trajectory and optimize the crack propagation path based on data such as curvature information, crack morphology, and stress distribution calculated by the foregoing modules. Therefore, the crack propagation path detection module is closely connected with the previously mentioned curvature analysis module and crack detection module, providing further analysis results and providing key support for crack propagation prediction and equipment safety monitoring.

[0052] Specifically, in this embodiment, the crack propagation path detection module first determines the preferred direction of crack propagation by judging the crack position and combining the local curvature information of the device surface. Generally, cracks tend to propagate along regions of stress concentration or smaller curvature, which are usually the "propagation channels" of cracks. Therefore, curvature information and stress distribution are the keys to predicting the crack propagation path.

[0053] In a possible implementation, the crack propagation path detection module calculates the optimal propagation path by the variational method. The variational method obtains the minimum energy propagation path of the crack by optimizing the objective function. Specifically, the objective function can be expressed as; ; where: is the crack propagation path; is the local curvature; is the stress distribution; represents the rate of change of the path; is the weight coefficient, controlling the influence of curvature, stress, and path change.

[0054] Specifically, the three terms in the objective function respectively represent: The influence of curvature on the crack propagation path: Cracks tend to propagate along regions of smaller curvature; The influence of stress distribution on crack propagation: Cracks usually propagate along regions of higher stress, so stress information is taken as part of the optimization; The penalty term for the path rate of change: This term is used to prevent excessive bending of the crack path and ensure the physical reasonableness of the propagation path.

[0055] By solving this variational problem, the optimal path of crack propagation, i.e., the minimum energy path, can be obtained. This process uses numerical optimization methods, such as the gradient descent method or the Newton method, to solve the optimal solution of the crack propagation path.

[0056] The crack propagation path detection module can also combine the geodesic calculation method and use the geodesics on the surface to determine the crack propagation direction. Geodesics are the shortest paths on the surface, and crack propagation often follows the direction of the minimum surface curvature. Therefore, by calculating the geodesics on the surface, more accurate guidance for the crack propagation path can be provided. Specifically, the calculation of geodesics can be achieved by solving the following equation; ; where: represents the time or path parameter The derivative of; is the metric of the geodesic tangent vector; is the tangent vector of the geodesic.

[0057] Through the solution of this equation, the propagation path of the crack can be obtained.

[0058] The present invention also introduces stress field analysis for further optimizing the crack propagation path. Crack propagation is not only affected by surface curvature but also by the stress field. Especially in high-stress regions, the crack propagation speed may increase. Therefore, stress field information is used as an important input in crack propagation path optimization. By analyzing the stress field, it is possible to guide the crack to propagate along the most dangerous path, thus providing a more accurate prediction basis for subsequent crack propagation prediction.

[0059] The crack propagation path detection module supports dynamic adjustment of the path optimization process. When there are new changes in stress concentration on the device surface, the system can update the prediction results of the crack propagation path in real time to ensure the timeliness of the propagation path. By combining real-time stress monitoring data, the crack propagation path can be dynamically corrected, making the system more real-time and adaptable.

[0060] A crack propagation prediction module, which is used to predict the future propagation trend of the crack based on historical data of crack propagation and calculate the crack propagation speed in combination with the optimal control theory; Specifically, in this embodiment, the crack propagation prediction module is located in the system after crack detection and path optimization. Based on the crack information provided by the aforementioned crack detection module and the path prediction generated by the crack propagation path optimization module, it predicts the future propagation trend of the crack and calculates the crack propagation speed based on stress, curvature, and historical data in combination with the optimal control theory. Through the prediction results of this module, equipment maintenance and safety management can be more precise and targeted, thus avoiding possible equipment failures or accidents.

[0061] In this embodiment, the crack propagation prediction module simulates the crack propagation behavior through the optimal control theory. Specifically, the crack propagation process can be regarded as a dynamic process affected by various factors such as the stress on the device surface, temperature changes, and material properties. By establishing a kinetic model of crack propagation, the crack propagation prediction module can accurately predict the future behavior of the crack.

[0062] Generally, the crack propagation process can be described by a state variable. Let represent the state of the crack, usually the crack length or the crack opening width, and the dynamic equation of the system is as follows; ; Wherein: is the state variable of the crack, representing the size or shape of the crack at time moment; is the crack growth rate, representing the change in the crack state; is the control input, representing the external factors affecting crack growth, such as temperature, stress : represents the kinetic equation of crack growth, which is a function describing the relationship between the crack state and the control input This equation represents the kinetic behavior of crack growth and may involve physical, mechanical, or material correlations.

[0063] As an option, the system adopts an optimal control method to design a control strategy to minimize the crack growth rate and predict the crack growth. The goal is to control the crack growth rate by optimizing the control input and delay the damage of the equipment. The optimal control problem can be expressed as: ; Wherein: is the loss function, usually related to the crack growth rate and external control factors; represents the end point of the time interval of the optimization problem, usually the duration of simulating crack growth; represents the total cost from time 0 to time , aiming to minimize the loss during the crack growth process in the entire time interval. By optimizing to control the crack growth rate and thus reduce the loss.

[0064] The form of the loss function can be defined according to different application scenarios, and common forms include the weighted sum of crack length, crack growth rate, etc.

[0065] Specifically, the loss function may include the following items: ; Wherein: and are the weight coefficients, controlling the contributions of the crack growth rate and the external control input to the objective function; is the crack growth rate, representing the change of the crack; For controlling inputs, such as external adjustment factors like temperature and stress.

[0066] The system uses historical crack growth data for regression analysis to obtain a mathematical model of crack growth. Let be the crack growth state, and the model can be obtained by fitting through historical data ; ; Where: are regression coefficients, representing relevant parameters in the crack growth process.

[0067] The crack growth prediction module combines optimal control theory, regression analysis, and historical data prediction methods, effectively improving the prediction accuracy of crack growth and enhancing the real-time performance and adaptability of the system. This module can not only predict the crack growth trend but also provide a scientific basis for equipment maintenance and optimization, providing strong technical support for the safe operation of special equipment.

[0068] The real-time feedback and decision support module is used to provide equipment maintenance suggestions or control strategies for controlling the crack growth rate based on the crack growth prediction results.

[0069] Specifically, the real-time feedback and decision support module in this embodiment plays a crucial role. Its main task is to provide real-time feedback on the operating state of the equipment and generate decision support information for equipment maintenance based on crack detection, crack growth path optimization, and growth prediction. This module directly affects the formulation of equipment maintenance strategies. By providing timely and accurate data analysis results, it helps operators make quick responses according to the crack growth trend, thereby extending the equipment service life and reducing the risk of sudden failures.

[0070] The real-time feedback and decision support module depends on the data provided by the aforementioned crack detection, curvature analysis, crack growth path optimization, and crack growth prediction modules. Information such as crack location, growth path, and growth speed output by these modules will be used as inputs. Combined with the actual operating data of the equipment, real-time processing is performed through edge computing and intelligent decision algorithms to quickly generate decision support information.

[0071] In this embodiment, the real-time feedback and decision support module generates an equipment maintenance report by inputting real-time data into the decision support system and provides specific maintenance suggestions to the operator. Specifically, the real-time feedback and decision support module includes the following steps Real-time data collection and transmission: Real-time acquisition of the operating status data of the device, such as stress, temperature, vibration, etc., and through the edge computing platform, these data are combined with crack propagation information for processing. Through edge computing, this module can analyze the crack propagation speed, crack propagation trend, and possible affected areas in real time at the device site.

[0072] Decision support generation: According to the propagation trend information provided by the crack propagation prediction module and the real-time operating data, the decision support module can evaluate whether the device needs maintenance. Based on the prediction model and the real-time data of the device, the decision-making system will judge whether the crack has reached the critical value, whether shutdown and maintenance are required, or whether maintenance can be postponed.

[0073] Maintenance recommendation generation: Generate a maintenance report for the device, providing specific maintenance operation suggestions, such as whether immediate repair is required, or whether preventive maintenance should be carried out.

[0074] Interactive control interface: This module provides an intuitive graphical user interface (GUI), which visually displays the current state of the device, crack propagation situation, and prediction results to the operator. This interface not only shows the crack propagation speed, path, and affected area, but also can adjust the maintenance strategy and generate operation instructions according to user needs.

[0075] The decision support module combines machine learning methods to learn the historical operating data of the device and the historical crack propagation data, so as to optimize the prediction and maintenance strategies. Through the analysis of historical data, this module can identify the typical failure modes of the device, further improving the accuracy and response speed of decision-making.

[0076] The decision support module adopts a reinforcement learning model to generate the optimal operation strategy based on the current state of the device, crack propagation prediction, and historical maintenance records. Through continuous learning and feedback, this model can dynamically adjust the maintenance strategy to maximize the safety and stability of the device. The reward function of the reinforcement learning model may be designed based on the stability of device operation, crack propagation speed, and maintenance cost; ; where: is the reward value at time indicating the effect of the current strategy; represents the penalty term for crack propagation speed, reflecting the crack propagation speed; represents the stability evaluation of device operation, reflecting the safety of the device in the current state; is the adjustment coefficient, controlling the influence of various factors on decision-making.

[0077] The real-time feedback and decision support module can significantly improve the real-time performance and accuracy of crack propagation prediction, providing a scientific basis for maintenance decisions during equipment operation. Through close cooperation with the aforementioned various modules, this module can detect the risk of crack propagation in a timely manner during the actual operation of the equipment and take corresponding control measures.

[0078] Please refer to the appendix Figure 2 , an image intelligent analysis method for special equipment safety detection, includes the following steps; S1. Use a laser scanning device and a stereo vision camera device to obtain three-dimensional image data of the surface of special equipment and generate point cloud data; Specifically, a laser scanning device or a stereo vision camera device is used to obtain three-dimensional data of the surface of special equipment. These devices precisely capture the point cloud data of the equipment surface through laser beam reflection or multi-view shooting. The point cloud data contains the spatial coordinates of each point on the equipment surface. For equipment with complex shapes, traditional two-dimensional image methods cannot accurately detect cracks, while three-dimensional point cloud data can provide complete geometric information of the equipment surface.

[0079] S2. Based on the point cloud data, use the grid interpolation method to generate a three-dimensional surface model of the surface of special equipment; Specifically, the point cloud data is then transformed into a smooth and continuous three-dimensional surface model through Poisson surface reconstruction or grid interpolation methods. Poisson surface reconstruction technology uses the normal vectors and spatial positions of the point cloud to construct a seamless surface model, which is suitable for high-precision industrial inspection. The grid interpolation method triangulates the point cloud data and converts the discrete points into a three-dimensional grid. Through these methods, the generated three-dimensional surface model provides a reliable geometric basis for subsequent crack detection and crack propagation path optimization.

[0080] S3. Use a three-dimensional convolutional neural network to analyze the three-dimensional surface model and extract the starting position of the crack; Specifically, the system uses a three-dimensional convolutional neural network (3D-CNN) to analyze the three-dimensional surface model and extract the starting position of the crack. Different from traditional two-dimensional image detection methods, 3D-CNN can process three-dimensional spatial data and automatically learn the spatial relationship from the equipment surface to the crack. Through depth convolution and pooling operations on the input data, the network can identify the crack regions on the surface, regardless of the size or shape of these cracks. The deep learning model is trained to effectively distinguish cracks from other details on the equipment surface and accurately identify the starting points of cracks.

[0081] S4. Curvature analysis, calculate the local curvature information of each point on the surface of special equipment, and determine the crack propagation trend according to Riemannian geometry theory; Specifically, the direction of crack propagation is often closely related to the curvature of the equipment surface. Specifically, cracks usually propagate along areas with smaller curvature. To judge the crack propagation trend, the system analyzes the crack propagation direction by calculating the local curvature of each point on the equipment surface. By locally analyzing the curved surface and calculating the principal curvature and Gaussian curvature, the system can understand the possible paths of crack propagation. Curvature analysis not only identifies the crack location but also provides data support for predicting the crack propagation path. Especially in complex surfaces and variable stress environments, curvature analysis can greatly improve the accuracy of crack propagation trend prediction. S5. Optimize the crack propagation path based on the variational method, and combine the local curvature information, stress distribution, and material properties of special equipment surfaces to calculate the optimal propagation trajectory of the crack along the geodesic direction; Specifically, based on the curvature information, stress distribution, and material properties of the equipment surface, optimizing the crack propagation path is the next key task. The system uses mathematical tools such as the variational method and geodesic calculation to calculate the optimal path for crack propagation along the equipment surface. The variational method finds the path with the minimum energy required for crack propagation by minimizing the objective function. As one of the key factors in optimization, cracks will propagate along the path of stress concentration. Combining the curvature information and material properties of the equipment surface, the system can predict the crack propagation route in complex geometries and variable environments.

[0082] S6. Establish a dynamic model of crack propagation using the optimal control method, predict the future crack propagation path, and calculate the crack propagation speed; Specifically, after optimizing the crack propagation path, the system predicts the future crack propagation trend through the optimal control theory. The speed and path of crack propagation are not only affected by the surface state of the equipment but also closely related to the external environment. By establishing a dynamic model of crack propagation, the system can make real-time predictions at all stages of crack propagation. The optimal control method provides the best control strategy for crack propagation and can adjust the prediction results of crack propagation according to real-time environmental and equipment operation data.

[0083] S7. Provide equipment maintenance suggestions based on the crack propagation prediction results and make real-time adjustments in combination with edge computing; Specifically, the system makes real-time adjustments to the crack propagation prediction results through the real-time feedback and decision support module and generates equipment maintenance decisions. The system will provide suggestions on whether to stop for maintenance, delay repair, or perform preventive maintenance based on real-time stress, temperature, and other data, combined with the crack propagation prediction results. Edge computing technology enables this process to be carried out on-site at the equipment, avoiding delays and errors in information transmission.

[0084] In practical applications, the system can collect the operation data of the device in real time and, combined with the aforementioned crack propagation prediction results, provide immediate and actionable maintenance suggestions for the operators. These suggestions help the device operators make decisions such as whether to perform repairs, adjust operating conditions, or replace components.

[0085] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An image intelligent analysis system for special equipment safety detection, characterized in that: include; Data acquisition module, used to obtain three-dimensional image data of the surface of special equipment and generate point cloud data; Surface reconstruction module, used to generate a three-dimensional surface model of the surface of special equipment based on point cloud data; Crack detection module, used to analyze the three-dimensional surface model and identify the crack shape and starting position; The curvature analysis module is used to calculate the curvature information of each point on the surface of special equipment and determine the crack propagation direction based on the curvature information; The crack propagation path detection module is used to determine the crack propagation path based on the curvature information and crack morphology, and optimize the possible crack propagation trajectory in combination with the geodesic calculation method; Crack propagation prediction module, which is used to predict the future crack propagation trend based on the historical data of crack propagation and combined with the optimal control theory, and calculate the crack propagation speed; The real-time feedback and decision support module is used to provide equipment maintenance suggestions or control strategies for controlling the crack growth rate based on the crack growth prediction results.

2. The image intelligent analysis system for special equipment safety detection according to claim 1 is characterized in that: The data acquisition module includes a laser scanning device and a stereoscopic vision camera device. The laser scanning device is used to obtain three-dimensional coordinate information of the surface of special equipment through laser point cloud scanning. The stereoscopic vision camera device is used to obtain images of the surface of special equipment from multiple perspectives and generate a depth map using a depth estimation algorithm.

3. The image intelligent analysis system for special equipment safety detection according to claim 1 is characterized in that: The surface reconstruction module uses a grid interpolation method to convert point cloud data into a continuous and smooth three-dimensional surface model.

4. The image intelligent analysis system for special equipment safety detection according to claim 1 is characterized in that: The crack detection module analyzes the three-dimensional surface model based on a three-dimensional convolutional neural network, extracts feature information through a deep learning model, and performs crack classification.

5. The image intelligent analysis system for special equipment safety detection according to claim 1 is characterized in that: The curvature analysis module calculates the local curvature information of each point on the surface of the special equipment, and determines the expansion trend of the crack along the geodesic direction of the equipment surface according to the Riemann geometry theory.

6. The image intelligent analysis system for special equipment safety detection according to claim 1 is characterized in that: The crack propagation path detection module optimizes the crack propagation path based on the variational method, combines the curvature information of the surface of the special equipment and the crack morphology, calculates the optimal propagation trajectory of the crack along the geodesic direction, and determines the minimum energy path for crack propagation.

7. The image intelligent analysis system for special equipment safety detection according to claim 1 is characterized in that: The crack propagation prediction module adopts an optimal control method to predict the future propagation path of the crack and calculate the crack propagation speed by establishing a dynamic model of crack propagation.

8. The image intelligent analysis system for special equipment safety detection according to claim 1 is characterized in that: The real-time feedback and decision support module combines with the edge computing platform to perform real-time analysis of crack propagation and provide decision recommendations for equipment shutdown maintenance, preventive maintenance or adjustment of operating parameters based on the prediction results.

9. The image intelligent analysis system for special equipment safety detection according to claim 1, characterized in that: The real-time feedback and decision support module includes a graphical user interface for displaying the crack growth status, prediction results and maintenance suggestions of the equipment to the operator, and providing interactive control options.

10. An image intelligent analysis method for special equipment safety detection, according to the image intelligent analysis system for special equipment safety detection according to claims 1-9, characterized in that: The steps include: S1. Use laser scanning equipment and stereoscopic vision camera equipment to obtain three-dimensional image data of the surface of special equipment and generate point cloud data; S2. Generate a three-dimensional surface model of the special equipment surface using a grid interpolation method based on point cloud data; S3, using a three-dimensional convolutional neural network to analyze the three-dimensional surface model and extract the starting position of the crack; S4, curvature analysis, calculate the local curvature information of each point on the surface of special equipment, and determine the crack expansion trend based on Riemann geometry theory; S5. Optimize the crack propagation path based on the variational method, combine the local curvature information, stress distribution and material properties of the surface of special equipment, and calculate the optimal crack propagation trajectory along the geodesic direction; S6. Use the optimal control method to establish a dynamic model of crack propagation, predict the future crack propagation path, and calculate the crack propagation speed; S7. Provide equipment maintenance recommendations based on crack propagation prediction results, and make real-time adjustments in combination with edge computing.

Images