Universal bridge crane safety bottom event identification method based on machine vision

Abstract

The invention discloses a universal bridge crane safety bottom event identification method based on machine vision, and relates to the technical field of hoisting machinery safety detection.The method comprises the steps that full-coverage image collection is conducted on a crane metal structure through a machine vision collection system, and the crane safety bottom event is identified through image preprocessing and a convolutional neural network classifier; visual features such as cracks, deformation and weld defects are automatically identified and quantified, the features are mapped into specific bottom events in an accident tree analysis model, the size, number, position and development trend of specific defects are calculated through image analysis, and quantitative indexes are converted into occurrence probability of the bottom events and input into the accident tree model. The accident tree model can reflect the current real health condition of the structure, the minimum cut set and the top event probability are calculated in real time, and the current failure path is dynamically revealed, so that accurate target guidance is provided for preventive maintenance, and post-event response of safety supervision is converted into beforehand early warning.

Description

Technical Field

[0001] This invention relates to the field of safety inspection technology for lifting machinery, specifically a method for identifying safety incidents on the bottom of a general-purpose bridge crane based on machine vision. Background Technology

[0002] As a key piece of equipment in industrial production and logistics transportation, the metal structure of general-purpose bridge cranes mainly includes main beams, end beams and legs. Under long-term exposure to alternating loads, impact loads and complex environments, these metal structures are prone to defects such as cracks, plastic deformation, weld cracking and corrosion wear. These defects are potential safety hazards and are the root cause of structural failure or even catastrophic accidents.

[0003] In existing technologies, computer technology is introduced into the crane safety assessment process, and the allowable stress method or limit state method are used to perform mechanical calculations to verify the structure. However, these methods still have significant shortcomings in the identification of safety-critical events. First, the probability of critical events in existing fault tree analysis models largely relies on historical statistical data or expert experience, lacking real-time and objective data sources. This results in static and lagging risk assessment results, making it difficult to reflect the true time-varying state of the structure. Second, computer vision-based defect detection only reaches the level of single image recognition and fails to form a deep linkage with the system's safety assessment model. The defect features identified by vision are not effectively quantified and mapped to the probability of critical events in the fault tree, causing a disconnect between the detection and assessment stages and making it impossible to directly link them to the determination of safety-critical events.

[0004] In summary, the existing technological system suffers from a disconnect between data flow and decision-making flow in the safety assessment of general-purpose bridge cranes. Therefore, there is an urgent need for an innovative method that deeply integrates machine vision with risk analysis models to achieve accurate, automatic, and real-time identification and risk assessment of safety incidents. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a general-purpose bridge crane safety event identification method based on machine vision. This method uses a machine vision acquisition system to perform full-coverage image acquisition of the crane's metal structure. Utilizing image preprocessing and a convolutional neural network classifier, it automatically identifies and quantifies visual features such as cracks, deformations, and weld defects. These features are mapped to specific events in a fault tree analysis model. Image analysis calculates the size, quantity, location, and development trend of specific defects. These quantified indicators are then converted into the probability of occurrence of the events and input into the fault tree model, enabling the model to reflect the current true health condition of the structure and thus providing precise target guidance for preventative maintenance.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a general-purpose bridge crane safety bottom event recognition method based on machine vision, wherein the specific steps of the method are as follows:

[0007] S1. The machine vision acquisition system is used to collect full-coverage image data of the detection area of ​​the metal structure of the general bridge crane to obtain the crane image sequence.

[0008] S2. Preprocess and extract features from the crane image sequence to identify and quantify visual features related to the safety of the metal structure, wherein the visual features correspond to the safety bottom events in the fault tree analysis.

[0009] S3. Quantize the extracted visual features into feature vectors and map them into a preset fault tree analysis model. By calculating the minimum cut set and minimum path set of the fault tree analysis model, evaluate the risk probability of each identified safety event.

[0010] S4. Perform structural mechanics verification based on the allowable stress method and the limit state method, calculate the safety factor, and generate a comprehensive safety assessment conclusion.

[0011] S5. Using COM component technology, the comprehensive security assessment conclusions, risk probabilities, and handling recommendations are presented as a structured report and output.

[0012] Furthermore, the machine vision acquisition system in S1 includes multiple industrial cameras and auxiliary light sources for acquiring the detection area of ​​the metal structure of a general-purpose bridge crane, wherein the detection area is the main beam, end beam, and weld area;

[0013] The acquisition frequency of the crane image sequence is synchronized with the crane's working cycle, covering the crane's no-load, rated load, and dynamic load conditions.

[0014] Furthermore, in S2:

[0015] Preprocessing the crane image sequence includes one or more of the following: image denoising, contrast enhancement, and geometric correction image preprocessing.

[0016] Feature extraction of the crane image sequence includes: using a U-Net-based semantic segmentation model to perform semantic segmentation on the preprocessed image sequence, and using edge detection to extract structural contours and identify the safety event types related to the detection area. The safety event types include one or more of weld cracks, loose connections, component deformation, and surface corrosion. Specifically, weld crack features are identified through linear texture analysis, loose connections are calculated through connection node displacement, component deformation is identified through contour curvature calculation, weld defects are identified through porosity and incomplete penetration area segmentation, and surface corrosion is identified through surface texture and color changes.

[0017] Furthermore, the quantization process of visual features in S3 is as follows:

[0018] Feature quantization: The identified detection area is quantified to obtain visual feature parameters, including one or more of the following: crack length, crack width, crack area, component deformation, and corrosion area percentage.

[0019] Feature mapping: Establish the correspondence between visual feature parameters and the safety bottom events in the fault tree, where weld cracks correspond to the bottom event. Loose connection corresponding to the bottom event Component deformation corresponds to the bottom event Surface corrosion corresponds to bottom events .

[0020] Furthermore, the fault tree analysis model in S3 is a hierarchical logical model constructed based on the failure mechanism of general bridge crane metal structures, including top events, intermediate events, bottom events, and logical relationships. Its construction process is as follows:

[0021] Determine the top event: Use the safety failure of the metal structure of a general-purpose bridge crane as the top event in the fault tree. ;

[0022] Identifying intermediate events: Through failure mechanism analysis, the top event is decomposed into intensity failures. Stiffness failure Fatigue failure Three intermediate events, which are transitional logical nodes between the top event and the bottom event, reflect the classification of failure types;

[0023] Define the bottom-level events: Combining machine vision-recognizable visual features, identify the safety bottom-level events corresponding to each intermediate event, including weld cracks. Loose connection Component deformation Surface corrosion Weld cracking Structural assembly deviation Material fatigue damage The bottom event is the most basic failure unit that causes the upper-level event to occur;

[0024] Establish logical relationships: Use AND gates and OR gates to describe the causal relationships between events, forming a fault tree logic expression, specifically:

[0025] The logical relationship between the top event and intermediate events is as follows: ,in This indicates that any intermediate event will lead to the occurrence of the top event;

[0026] Logical relationship between intermediate events and underlying events: , , ,in The representation of AND logic is that an intermediate event will only occur if all input events occur simultaneously.

[0027] Furthermore, in S3, the minimum cut set is the minimum combination of basic events that leads to the top event. Each minimum cut set represents a failure path. The more minimum cut sets there are, the greater the probability of the top event occurring. The minimum path set is the minimum combination of measures to prevent the top event from occurring. Each minimum path set represents a set of protection schemes, wherein:

[0028] Minimal cutset computation: A descending method is used, starting from the top event and decomposing the events layer by layer from top to bottom. The input events of each node are processed according to the logic gate type to obtain all minimal failure combinations that do not contain other cutsets. ,in, It is a minimal cut set. The number of AND gates in the fault tree. For the first The number of input events for an AND gate. For the first The first and the door One input event;

[0029] Minimum path set calculation: Using the up-step method, the fault tree is transformed into a dual tree, that is, AND gates in the original tree are transformed into OR gates, OR gates are transformed into AND gates, top events are transformed into bottom events, and bottom events are transformed into top events. After constructing the dual tree, the minimum cut set calculation is used to solve the problem and obtain the minimum set of bottom events that can prevent the top event from occurring.

[0030] Furthermore, the risk probability assessment of the safety bottom event in S3 includes:

[0031] Risk probability calculation: Boolean algebra is used to integrate the influence of multiple visual feature parameters to calculate the risk probability of a safety-low event. ,in, For the first The probability of a safety event is given, with values ​​ranging from [0, 1]. A higher value indicates a higher risk. This represents the number of visual feature parameters corresponding to the event. For the first The bottom event The probability of a feature parameter exceeding the standard is calculated based on the deviation between the actual value of that parameter and the safety threshold. This is a risk amplification factor, with a value ranging from 1.2 to 2.5. For the first The combined value of the actual characteristic parameters of each base event For the first Safety threshold of characteristic parameters for each bottom event.

[0032] Furthermore, the specific steps of S4 are as follows:

[0033] Based on the visual features extracted from S200, the actual working stress at that location was calculated using finite element analysis. ;

[0034] Will Compared with the allowable stress obtained by the allowable stress method Compare with, and simultaneously with, the ultimate stress obtained through the limit state method. Compare;

[0035] Define the fusion security factor The ,in, and These are the weighting coefficients, and , representing the confidence weights of the allowable stress method and the limit state method, respectively. When the comprehensive safety assessment conclusion is marked as high risk, When the comprehensive safety assessment conclusion is marked as medium risk, At that time, the comprehensive safety assessment conclusion was marked as low risk.

[0036] Compared with existing technologies, this machine vision-based method for identifying safety events on the bottom of a general-purpose bridge crane has the following advantages:

[0037] This invention utilizes a machine vision acquisition system to capture full-coverage images of the crane's metal structure. Image preprocessing and a convolutional neural network classifier are then used to automatically identify and quantify visual features such as cracks, deformations, and weld defects. These features are mapped to specific base events in a fault tree analysis model. Image analysis is used to calculate the size, quantity, location, and development trend of specific defects. These quantified indicators are converted into the probability of occurrence of base events and input into the fault tree model. This allows the fault tree model to reflect the current true health condition of the structure and to calculate the minimum cut set and apex event probability in real time, dynamically revealing the current failure path. This provides precise target guidance for preventative maintenance, transforming safety supervision from reactive response to proactive early warning.

[0038] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0040] Figure 1 This is a flowchart illustrating the operation of a machine vision-based general-purpose bridge crane safety event recognition method.

[0041] Figure 2 This is a flowchart of fault tree analysis and risk probability assessment in a machine vision-based general bridge crane safety incident identification method;

[0042] Figure 3 This is a flowchart illustrating the construction of a fault tree analysis model in a machine vision-based general bridge crane safety event identification method. Detailed Implementation

[0043] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0044] Example

[0045] This embodiment discloses a general-purpose bridge crane safety event recognition method based on machine vision, aiming to solve the problem of the disconnect between the detection and assessment stages in existing crane safety assessments, such as... Figure 2 As shown, the method comprises the following steps: S1. A machine vision acquisition system is used to collect full-coverage image data of the detection area of ​​the metal structure of a general-purpose bridge crane, resulting in a crane image sequence. S2. The image sequence is preprocessed and features are extracted to identify and quantify visual features related to the safety of the metal structure. S3. The extracted visual features are quantized into feature vectors and mapped to a preset fault tree analysis model. The risk probability of each identified safety event is evaluated by calculating the minimum cut set and minimum path set. S4. Structural mechanics verification is performed based on the allowable stress method and the limit state method to calculate the safety factor and generate a comprehensive safety assessment conclusion. S5. The comprehensive safety assessment conclusion, risk probability, and handling recommendations are output as a structured report using COM component technology. The entire process realizes automatic, real-time identification and dynamic risk assessment of safety events, providing a basis for preventive maintenance of cranes and effectively improving the operational safety of crane machinery.

[0046] S1, Image Data Acquisition

[0047] This embodiment addresses the safety monitoring needs of the metal structure of a general-purpose bridge crane. A machine vision acquisition system is used to collect image data of the detection area. The specific operation is as follows:

[0048] The machine vision acquisition system consists of multiple industrial cameras and auxiliary light sources. Based on the structural characteristics and inspection requirements of the crane's metal structure, the industrial cameras are rationally arranged in key inspection areas such as the main beam, end beams, and weld seams of the crane to ensure full coverage of the inspection area. The auxiliary light source adopts an adaptive lighting design, which automatically adjusts the brightness and illumination angle according to the on-site lighting conditions to avoid strong light reflection or insufficient light affecting the image quality and ensure the clear presentation of structural details and potential defects in the image.

[0049] The image acquisition process is synchronized with the crane's work cycle, and the acquisition frequency is dynamically adjusted according to the crane's operating speed and the complexity of the working conditions to ensure complete coverage of the three core working conditions: no-load, rated load, and dynamic load. Under no-load conditions, baseline images of the metal structure are acquired when the crane is running without load; under rated load conditions, images of the structure are acquired when the crane is under rated weight load; under dynamic load conditions, image sequences of the structure under dynamic loads during the crane's lifting, moving, and braking processes are acquired. Through multi-condition image acquisition, the state characteristics of the metal structure under different load conditions are comprehensively captured, providing complete data support for subsequent safety event identification, ultimately resulting in a crane image sequence encompassing each working condition and each detection area.

[0050] S2, Image Preprocessing and Feature Extraction

[0051] The acquired crane image sequences were systematically preprocessed and feature extracted to identify and quantify visual features related to the safety of metal structures. The specific steps are as follows:

[0052] To address issues such as noise, insufficient contrast, and geometric distortion in image sequences, a combination of preprocessing techniques is employed for image optimization. First, image denoising is performed, using adaptive filtering algorithms to remove interfering signals such as Gaussian noise and salt-and-pepper noise while preserving structural details and potential defect information. Next, contrast enhancement is applied using histogram equalization to improve the grayscale difference between structural and defective regions, making subtle defects easier to identify. Finally, geometric correction is performed, adjusting for geometric deviations caused by shooting angle and lens distortion based on the calibration parameters of the industrial camera, ensuring consistency between the structural dimensions in the image and their actual dimensions, providing an accurate image foundation for subsequent feature quantization. These preprocessing operations can be used individually or in combination, depending on the image quality, to achieve the best image optimization results.

[0053] Feature extraction and security event type identification

[0054] A U-Net-based semantic segmentation model is used to perform semantic segmentation on the preprocessed image sequence, achieving accurate separation between the detection region and the background region, focusing on feature analysis of key structural regions such as main beams, end beams, and welds. Based on this, an edge detection algorithm is combined to extract structural contour information. Through comprehensive analysis of contour features and region features, the type of safety incident is identified.

[0055] Specifically, weld cracks are identified through linear texture analysis, utilizing the linear grayscale variation characteristics of the crack area to segment the crack region from the image; loose connections are identified by calculating the displacement of connection nodes, comparing the positional changes of connection nodes under different working conditions to determine if loosening exists; component deformation is identified by calculating the curvature of the contour, comparing the curvature of the extracted structural contour with that of a standard contour, and determining deformation if the difference exceeds a set range; porosity and incomplete penetration areas in weld defects are identified through semantic segmentation results, completing region segmentation based on the difference in their grayscale values ​​compared to normal weld areas; surface corrosion is identified through surface texture and color changes, utilizing the rough surface and dark color characteristics of corroded areas, combined with texture analysis and color space conversion technology to achieve accurate identification of corroded areas. Through these methods, one or more types of safety-related events, such as weld cracks, loose connections, component deformation, and surface corrosion, can be identified.

[0056] Visual feature quantization

[0057] The detection areas corresponding to identified safety incidents are quantitatively analyzed to obtain visual feature parameters. For weld cracks, parameters such as crack length, crack width, and crack area are quantified; for component deformation, the amount of deformation is quantified; and for surface corrosion, the proportion of corrosion area is quantified. Through precise parameter quantification, visual features are transformed into quantitative indicators that can be used for subsequent risk assessment, providing data support for calculating the risk probability of safety incidents.

[0058] S3. Fault Tree Analysis and Risk Probability Assessment

[0059] The quantified visual feature parameters are mapped to a pre-defined fault tree analysis model. The risk assessment of safety-related events is then calculated and analyzed using this model. Figure 2 As shown, the specific process is as follows:

[0060] S3.1 Fault Tree Analysis Model Construction

[0061] The pre-defined fault tree analysis model is a hierarchical logical model built based on the failure mechanism of general bridge crane metal structures. It includes top events, intermediate events, bottom events, and logical relationships, such as... Figure 3 As shown, the specific construction process is as follows:

[0062] Top event determination: The top event in the fault tree is the safety failure of the metal structure of a general-purpose bridge crane. This refers to the final failure state that needs to be prevented.

[0063] Intermediate event analysis: Through failure mechanism analysis, the top event is decomposed into intensity failures. Stiffness failure Fatigue failure Three intermediate events serve as transitional logical nodes between the top and bottom events, reflecting the classification of failure types and clarifying the propagation paths of different failure modes.

[0064] Definition of bottom-level events: Combining machine vision-recognizable visual features and metal structure failure mechanisms, the safety bottom-level events corresponding to each intermediate event are identified, including weld cracks. Loose connection Component deformation Surface corrosion Weld cracking Structural assembly deviation Material fatigue damage The bottom event is the basic failure unit that leads to the occurrence of the upper event, and is used to characterize the factors that cause failure of metal structures.

[0065] Logical relationship establishment: AND gates and OR gates are used to describe the causal relationships between events, forming a fault tree logic expression. The logical relationship between the top event and intermediate events is as follows: Where U represents OR logic, meaning that the occurrence of any intermediate event will lead to the occurrence of the top event; the logical relationship between intermediate events and bottom events is as follows: , , ,in The representation of AND logic is that an intermediate event will only occur if all input events occur simultaneously.

[0066] S3.2, Eigenvector Mapping

[0067] The visual feature parameters obtained from S2 quantization are converted into feature vectors. Based on a predefined correspondence, these visual feature parameters are associated with the safety-level events in the fault tree. Specifically, the visual feature parameters corresponding to weld cracks are mapped to the underlying events. The visual feature parameters corresponding to the loose connection are mapped to the underlying event. The visual feature parameters corresponding to component deformation are mapped to the underlying events. ; Visual feature parameters corresponding to surface corrosion are mapped to underlying events This completes the mapping of visual features to events at the bottom of the accident tree.

[0068] S3.3 Calculation of Minimal Cut Set and Minimal Path Set

[0069] Minimal Cut Set Calculation: The minimum cut set of the fault tree is calculated using a descending method, starting from the top event and decomposing the events layer by layer from top to bottom. The input events of each node are processed according to the logic gate type. For OR gates, their input events are treated as independent branches for further decomposition; for AND gates, all their input events are grouped into one branch for decomposition, until the bottom event is reached. Finally, the minimum failure combination that does not contain any other cut sets is obtained, i.e., the minimum cut set, whose expression is: ,in It is a minimal cut set. The number of AND gates in the fault tree. For the first The number of input events for an AND gate. For the first The first and the door There are 10 input events, and each minimal cut set represents a failed path. The more minimal cut sets there are, the greater the probability that the top event will occur.

[0070] Minimum path set calculation: The minimum path set is calculated using an upward method. First, the fault tree is transformed into a dual tree, i.e., AND gates in the original tree are converted into OR gates, OR gates are converted into AND gates, and top events are converted into bottom events, and bottom events are converted into top events, thus constructing a dual tree. Using the same method as the minimum cut set calculation, the minimum set of bottom events that can prevent the top event from occurring is obtained, i.e., the minimum path set. Each minimum path set represents a set of protection schemes. Through the calculation of minimum cut sets and minimum path sets, the paths leading to structural failure and the measures to prevent failure are clarified.

[0071] S3.4 Calculation of the probability of safety-lowering events

[0072] The Boolean algebra method is used to integrate the influence of multiple visual feature parameters to calculate the risk probability of each safety-low event. The calculation formula is as follows: ,in, For the first The probability of a safety event is given, with values ​​ranging from [0,1]. A larger value indicates a higher risk. This represents the number of visual feature parameters corresponding to the event. For the first The bottom event The probability of a feature parameter exceeding the standard is calculated based on the deviation between the actual value of that parameter and the safety threshold. This is a risk amplification factor, with a value ranging from 1.2 to 2.5. For the first The actual feature parameter composite value of a base event is obtained by weighted summation of the visual feature parameters corresponding to that base event. For the first The safety threshold of the feature parameters of each basic event, by directly linking the quantification results of visual feature parameters with the risk probability, realizes the dynamic calculation of risk probability, which can truly reflect the current risk status of the structure.

[0073] S4. Structural Mechanics Verification and Comprehensive Safety Assessment

[0074] Based on feature extraction results and risk assessment data, structural mechanics verification is performed using the allowable stress method and the limit state method. The safety factor is calculated, and a comprehensive safety assessment conclusion is generated. The specific steps are as follows:

[0075] Actual working stress calculation

[0076] Based on the visual features extracted from S2, including parameters such as component deformation and crack size, the actual working stress at key locations in the metal structure inspection area was calculated using finite element analysis. In the finite element analysis process, an accurate finite element model is established based on the material properties, geometric model, and structural state reflected by visual features of the crane's metal structure. The structural defects corresponding to the visual features are integrated into the model to simulate the stress distribution under actual loads, and finally the actual working stress at key locations is obtained through inversion.

[0077] Allowable stress and ultimate stress calculation

[0078] Allowable stress calculation: Allowable stress is calculated using the allowable stress method. The specific process is as follows: Analyze and calculate the loads on the crane under various working conditions, increase the load with an appropriate dynamic coefficient, and combine them according to the load combination table to obtain the combined load; based on structural mechanics methods, use this combined load to calculate the internal forces of the components and determine the combined load effect; based on the load effect (internal force) acting on the components or parts, and combined with the principles of mechanics of materials, calculate the stress at the verification point or verification section, and combine it with any stress caused by local effects (internal forces) to obtain the combined design stress; based on the yield strength and safety factor of the material, determine the allowable stress. This serves as the upper limit for the allowable structural strength.

[0079] Ultimate stress calculation: The ultimate stress is calculated using the limit state method. The specific process is as follows: Calculate each specified load, increase the load with an appropriate dynamic coefficient, and multiply it by the corresponding sub-load coefficient of the calculated load in the load combination. Then, combine the loads according to the load combination table to obtain the combined load. In highly dangerous situations, the combined load also needs to be multiplied by a high-risk factor to finally obtain the design load. Calculate the internal forces of the components based on the design load to determine the load effect. Calculate the stress based on the load effect (internal force) acting on the component or part, and combine it with other stresses caused by local effects (internal forces) calculated using an appropriate load coefficient to obtain the composite design stress. Divide the ultimate strength of the material by the resistance coefficient to obtain the specified stress limit value, i.e., the ultimate stress. .

[0080] Safety factor calculation and comprehensive safety assessment conclusion generation

[0081] Define the fusion security factor Combining the calculation results of the allowable stress method and the limit state method, the calculation formula is as follows: ,in, and These are the weighting coefficients, and , representing the confidence weights of the allowable stress method and the limit state method, respectively. Based on the calculation results of the fused safety factor, a comprehensive safety assessment conclusion is generated: when When the comprehensive safety assessment conclusion is marked as high-risk, it indicates that the structure currently has serious safety hazards and must be immediately taken out of service and repaired; when When the comprehensive safety assessment conclusion is marked as medium risk, it indicates that the structure has certain safety risks and requires inspection and maintenance within a specified period; when When the overall safety assessment conclusion is marked as low risk, it indicates that the structure is currently in good safety condition and can continue to operate normally, but regular monitoring is required.

[0082] S5, Structured Report Output

[0083] Utilizing COM component technology, a structured report is compiled, integrating comprehensive safety assessment conclusions, risk probabilities of various safety incidents, minimum cut set and minimum path set analysis results, and targeted remedial recommendations. The report clearly lists the inspection area, identified safety incident types, risk levels of each incident, structural mechanics verification data, safety factors, and corresponding remedial measures, such as emergency repair plans for high-risk conditions, maintenance priorities for medium-risk conditions, and periodic monitoring requirements for low-risk conditions. The structured report can be displayed through a visual interface or exported as a document, facilitating staff viewing, archiving, and subsequent maintenance decisions, thus effectively linking inspection results with maintenance measures.

[0084] This embodiment utilizes a machine vision acquisition system to achieve full coverage and multi-condition image acquisition of the metal structure inspection area of ​​a general-purpose bridge crane. After preprocessing and feature extraction, visual features related to safety-critical events such as weld cracks and loose connections are accurately identified and quantified. These quantified features are mapped to a fault tree analysis model. Through minimum cut set and minimum path set calculations and risk probability assessment, the critical path to structural failure and the risk status of each critical event are identified. Structural mechanics verification is performed using the optimized allowable stress method and limit state method. A fusion safety factor is calculated, generating a scientific comprehensive safety assessment conclusion. A structured report is output using COM component technology. This method achieves deep integration of machine vision and risk analysis models, solving the problems of disconnect between detection and assessment and static risk assessment in existing technologies. It enables real-time and accurate identification of safety-critical events and risk assessment, providing precise guidance for the preventative maintenance of cranes and effectively improving the operational safety and reliability of lifting machinery.

[0085] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A general-purpose bridge crane safety event recognition method based on machine vision, characterized in that, The specific steps of this method are as follows: S1. The machine vision acquisition system is used to collect full-coverage image data of the detection area of ​​the metal structure of the general bridge crane to obtain the crane image sequence. S2. Preprocess and extract features from the crane image sequence to identify and quantify visual features related to the safety of the metal structure, wherein the visual features correspond to the safety bottom events in the fault tree analysis. S3. Quantize the extracted visual features into feature vectors and map them into a preset fault tree analysis model. By calculating the minimum cut set and minimum path set of the fault tree analysis model, evaluate the risk probability of each identified safety event. S4. Perform structural mechanics verification based on the allowable stress method and the limit state method, calculate the safety factor, and generate a comprehensive safety assessment conclusion. S5. Using COM component technology, the comprehensive security assessment conclusions, risk probabilities, and handling recommendations are presented as a structured report and output.

2. The method for identifying safety bottom events of a general-purpose bridge crane based on machine vision according to claim 1, characterized in that, The machine vision acquisition system in S1 includes multiple industrial cameras and auxiliary light sources, used to acquire the detection area of ​​the metal structure of a general bridge crane, the detection area being the main beam, end beam, and weld area. The acquisition frequency of the crane image sequence is synchronized with the crane's working cycle, covering the crane's no-load, rated load, and dynamic load conditions.

3. The method for identifying safety bottom events of a general-purpose bridge crane based on machine vision according to claim 1, characterized in that, In S2: Preprocessing the crane image sequence includes one or more of the following: image denoising, contrast enhancement, and geometric correction image preprocessing. Feature extraction of the crane image sequence includes: using a U-Net-based semantic segmentation model to perform semantic segmentation on the preprocessed image sequence, and using edge detection to extract structural contours and identify the safety event types related to the detection area. The safety event types include one or more of weld cracks, loose connections, component deformation, and surface corrosion. Specifically, weld crack features are identified through linear texture analysis, loose connections are calculated through connection node displacement, component deformation is identified through contour curvature calculation, weld defects are identified through porosity and incomplete penetration area segmentation, and surface corrosion is identified through surface texture and color changes.

4. The method for identifying safety bottom events of a general-purpose bridge crane based on machine vision according to claim 1, characterized in that, The quantization process of visual features in S3 is as follows: Feature quantization: The identified detection area is quantified to obtain visual feature parameters, including one or more of the following: crack length, crack width, crack area, component deformation, and corrosion area percentage. Feature mapping: Establish the correspondence between visual feature parameters and the safety bottom events in the fault tree, where weld cracks correspond to the bottom event. Loose connection corresponding to the bottom event Component deformation corresponds to the bottom event Surface corrosion corresponds to bottom events .

5. The method for identifying safety bottom events of a general-purpose bridge crane based on machine vision according to claim 4, characterized in that, The fault tree analysis model in S3 is a hierarchical logical model built based on the failure mechanism of general bridge crane metal structures. It includes top events, intermediate events, bottom events, and logical relationships. Its construction process is as follows: Determine the top event: Use the safety failure of the metal structure of a general-purpose bridge crane as the top event in the fault tree. ; Identifying intermediate events: Through failure mechanism analysis, the top event is decomposed into intensity failures. Stiffness failure Fatigue failure Three intermediate events, which are transitional logical nodes between the top event and the bottom event, reflect the classification of failure types; Define the bottom-level events: Combining machine vision-recognizable visual features, identify the safety bottom-level events corresponding to each intermediate event, including weld cracks. Loose connection Component deformation Surface corrosion Weld cracking Structural assembly deviation Material fatigue damage ; Establish logical relationships: Use AND gates and OR gates to describe the causal relationships between events, forming a fault tree logic expression, specifically: The logical relationship between the top event and intermediate events is as follows: ,in Representation or logic; The logical relationship between intermediate events and bottom events: , , ,in Representation and logic.

6. The method for identifying safety bottom events of a general-purpose bridge crane based on machine vision according to claim 5, characterized in that, In S3, the minimum cut set is the minimum combination of bottom events that leads to the top event, and each minimum cut set represents a failure path. The minimum path set is the minimum combination of measures to prevent the top event, and each minimum path set represents a protection scheme. Minimal cutset computation: A descending method is used, starting from the top event and decomposing the events layer by layer from top to bottom. The input events of each node are processed according to the logic gate type to obtain all minimal failure combinations that do not contain other cutsets. ,in, It is a minimal cut set. The number of AND gates in the fault tree. For the first The number of input events for an AND gate. For the first The first and the door One input event; Minimum path set calculation: Using the up-step method, the fault tree is transformed into a dual tree, that is, AND gates in the original tree are transformed into OR gates, OR gates are transformed into AND gates, top events are transformed into bottom events, and bottom events are transformed into top events. After constructing the dual tree, the minimum cut set calculation is used to solve the problem and obtain the minimum set of bottom events that can prevent the top event from occurring.

7. The method for identifying safety bottom events of a general-purpose bridge crane based on machine vision according to claim 1, characterized in that, The risk probability assessment of the safety bottom event in S3 includes: Risk probability calculation: Boolean algebra is used to integrate the influence of multiple visual feature parameters to calculate the risk probability of a safety-low event. ,in, For the first The probability of a single safe event ranges from [0, 1]. This represents the number of visual feature parameters corresponding to the event. For the first The bottom event The probability of a feature parameter exceeding the limit. This is a risk amplification factor, with a value ranging from 1.2 to 2.

5. For the first The combined value of the actual characteristic parameters of each base event For the first Safety threshold of characteristic parameters for each bottom event.

8. The method for identifying safety bottom events of a general-purpose bridge crane based on machine vision according to claim 1, characterized in that, The specific steps of S4 are as follows: Based on the visual features extracted from S200, the actual working stress at that location was calculated using finite element analysis. ; Will Compared with the allowable stress obtained by the allowable stress method Compare with, and simultaneously with, the ultimate stress obtained through the limit state method. Compare; Define the fusion security factor The ,in, and These are the weighting coefficients, and , representing the confidence weights of the allowable stress method and the limit state method, respectively. When the comprehensive safety assessment conclusion is marked as high risk, When the comprehensive safety assessment conclusion is marked as medium risk, At that time, the comprehensive safety assessment conclusion was marked as low risk.

Images