Intelligent monitoring method for fatigue life of construction machinery load-bearing structure

Abstract

The present application belongs to the technical field of mechanical structure health monitoring and life prediction, in particular to an engineering machinery load-bearing structure fatigue life intelligent monitoring method, comprising the following steps: step one, establishing a high-fidelity geometric model; step two, establishing a finite element model; step three, constructing a finite element simulation dataset; step four, constructing a neural network proxy model; step five, obtaining working parameters and loading history; and step six, integrating a visual operation and maintenance platform. The engineering machinery load-bearing structure fatigue life intelligent monitoring method can intelligently deduce the evolution process of fatigue damage based on the working parameters and loading history of the structure, realize dynamic calculation of the remaining life and real-time early warning of the service risk, save computing resources, improve the efficiency of fatigue life evaluation, and solve the problems of dependence on offline simulation, response lag, and discontinuous evaluation in traditional structure life management, thereby effectively improving the intelligent operation and maintenance level of engineering machinery and providing protection for safe and efficient operation.

Description

Technical Field

[0001] This invention relates to the field of mechanical structure health monitoring and life prediction technology, and in particular to an intelligent monitoring method for fatigue life of load-bearing structures of engineering machinery. Background Technology

[0002] Key load-bearing components of construction machinery are subjected to frequent start-stop operations and significant cyclic loads over long periods, making them prone to fatigue damage and posing a potential risk of fracture failure. This type of fatigue damage is typically cumulative, insidious, and sudden; once structural failure occurs, it can lead to equipment malfunctions or even safety accidents, resulting in economic losses and operational risks. Therefore, conducting fatigue damage assessments and life predictions for key load-bearing components of in-service construction machinery is a critical engineering requirement for ensuring operational safety and equipment reliability.

[0003] Traditional structural life management methods primarily rely on periodic inspections and manual experience-based judgment. Existing monitoring methods generally suffer from limited monitoring coverage, unstable data quality, and high system costs. Furthermore, publicly available solutions rarely involve specific measurement point selection, and common methods still depend on empirical rules or statistical indicators. For example, CN103557890A and CN109524139B both rely on empirical point selection at key locations, without involving structural mechanics analysis or inversion model optimization, potentially leading to monitoring blind spots and difficulty in reflecting damage evolution. Simultaneously, existing methods lack effective damage characterization, visualization analysis, and safety status assessment processes for monitoring data, making it difficult to support accurate life prediction and maintenance decisions. While finite element simulation technology offers high computational accuracy, its long computation time and heavy reliance on hardware resources prevent rapid on-site assessment. For engineering components in service, operating conditions are often complex and loads are variable; a single simulation cannot cover all possibilities. Relying on resimulation for each assessment is extremely inefficient, and the assessment results lack timeliness. These problems limit the improvement of life management levels for critical load-bearing structures in engineering machinery, making it difficult to meet the operational requirements for equipment reliability and safety under various working environments.

[0004] While existing mainstream solutions possess the framework of digital twin visualization platforms, most remain at the level of dynamic display of real-time operational data and lack a core intelligent decision-making mechanism based on damage theory and failure modes. For example, CN119276728A, published on the Chinese patent website, infers equipment performance through machine learning, but due to the lack of a physical failure model, the accuracy and reliability of the results are difficult to guarantee. Another example is CN113190886A, which requires multiple calls to the CAE solver for DOE sampling, resulting in time-consuming calculations and dependence on specific software environments, thus lacking real-time performance and deployment flexibility. Therefore, the functionality of such platforms is limited to varying degrees, making it difficult to support the efficient, accurate, and intelligent lifespan management needs of critical load-bearing structures. Summary of the Invention

[0005] Based on existing technical problems, this invention proposes an intelligent monitoring method for the fatigue life of load-bearing structures of engineering machinery. It comprehensively utilizes finite element analysis, deep learning modeling, neural network deployment, and visualization human-computer interaction technologies, and integrates fatigue damage theory to construct an integrated solution with digital perception, intelligent inference, autonomous decision-making, and data closed-loop capabilities. It also improves the modeling process of digital twins and provides a new solution for the intelligent operation and maintenance of key load-bearing structures of engineering machinery.

[0006] The present invention proposes an intelligent monitoring method for fatigue life of load-bearing structures of engineering machinery, including step one: establishing a high-fidelity geometric model;

[0007] Step 2: Establish a finite element model;

[0008] Step 3: Construct the finite element simulation dataset;

[0009] Step 4: Construct a neural network proxy model;

[0010] Step 5: Obtain working parameters and loading history;

[0011] Step 6: Integrate the visual operation and maintenance platform.

[0012] Preferably, in step one, the geometric model is constructed by 3D modeling software, and then exported to a corresponding file format according to requirements.

[0013] Through the above technical solution, relevant information of each component of the lifting device is obtained based on the design data. Then, 3D modeling software is used to construct a 3D model of the lifting device, such as that used in engineering machinery, to ensure sufficient detail and ensure that the model structure is consistent with the actual structure, so as to obtain an original model that can truly reflect the mechanical behavior of the lifting device.

[0014] Preferably, the geometric model is imported into the finite element simulation software in step two according to the file type requirements to simplify the geometric model.

[0015] The above technical solution involves dismantling the electrical equipment, hydraulic pipelines, and accessories arranged in the geometric model of the lifting device; modeling smaller welds and components in the main beam and telescopic beam as a whole, and modeling larger welds between the webs of the main beam and between the main beam and the plate as a separate component; ignoring smaller threaded holes and other structures; and modeling small chamfers, fillets, and other structures in the lifting device structure as right angles.

[0016] Preferably, after the geometric model is simplified, the material properties are defined, then the structure in the geometric model is meshed and mesh independence is verified, and finally relevant boundary conditions are set.

[0017] The above technical solution involves using high-strength steel for the main load-bearing components of the lifting device structure, nylon 66 for the buffer pad between the main beam and the telescopic beam, and ordinary carbon structural steel for the remaining components. Material parameters such as Poisson's ratio, elastic modulus, density, yield strength, and tensile strength are assigned to the corresponding materials.

[0018] Preferably, in step three, the integration step size is set according to the working state and working time, and an appropriate damping form is selected according to the structural characteristics and analysis requirements. Transient dynamic analysis is then performed on the entire cycle process of the geometric model to obtain the load spectrum.

[0019] Preferably, the load spectrum is obtained by jointly performing fatigue analysis using Ansys Workbench and nCode DesignLife, and the corresponding load is then obtained. Fatigue life ;

[0020] The maximum load that the geometric model may encounter under actual working conditions. and minimum load Add a certain amount of redundancy The upper limit load of the geometric model is then obtained. and lower limit load Simulation dataset input ;

[0021] Super Latin cube sampling was used at the set upper limit load. and lower limit load Extract a sufficient number of data points and obtain the fatigue life for each data point through simulation analysis. These input and output data are recorded in tabular form to form a load-life mapping dataset for finite element simulation. ,in, .

[0022] The above technical solution addresses the three different operating modes of the spreader during lifting operations due to variations in container size. For each operating mode, the maximum and minimum loads that may occur under actual working conditions are identified, and a certain amount of redundancy is added as the upper and lower limits of the simulation dataset input. Points are selected within the input range using hyper-Latin cube sampling, and the corresponding fatigue life is obtained through transient dynamics combined with Ansys Workbench and nCode DesignLife fatigue analysis. This constructs three load-life mapping datasets for finite element simulations. .

[0023] Preferably, the neural network architecture in step four is built using Python, based on the load-lifetime mapping dataset constructed in step three. The model is normalized and divided into training and testing sets according to a certain ratio. The model is then trained, validated, and its parameters are tuned to construct a load-lifetime surrogate model.

[0024] Through the above technical solutions, appropriate optimizers and loss functions are selected during the training process, and hyperparameters such as learning rate, training epochs, and batch size are set.

[0025] Preferably, during training, the loss function will calculate the predicted values ​​of the neural network model and the load-lifetime mapping dataset. The error between labels is backpropagated to update the parameters of the neural network model;

[0026] The neural network model is evaluated using mean absolute error (MAE), mean squared error (MSE), and coefficient of determination. As an evaluation indicator, if the indicator does not meet expectations, the size of the hyperparameters will be adjusted. When the accuracy of the neural network model meets the requirements, the neural network model will be saved as a target file format and the normalization information during training will be manually saved in the checkpoint of the target file to prevent prediction deviations in the inference stage caused by inconsistent normalization.

[0027] Export the model weights and structure from the target file into a general model deployment format, and record the normalization information from the checkpoint in the target file.

[0028] The above technical solutions enable the correct reproduction of the input specifications during training when using the model for inference.

[0029] Preferably, the working parameters in step five include the load and cycle number of the geometric model;

[0030] The loading history includes the entire process of load changes experienced by the geometric model structure over time during operation, representing the loading of the geometric model in each work cycle. The cumulative total of the corresponding cycle periods, where, ;

[0031] Static analysis is performed based on the finite element model established in step three to calculate the critical points of the geometric model structure. A static inversion model of the neural network model is then established. By analyzing the stress at the critical points of the geometric model structure, the loads acting on the geometric model structure are inverted. This forms a stress-load mapping dataset for finite element simulation. Then, the neural network model architecture is established, and the stress-load mapping dataset from finite element simulation is used. Complete the training of the static inversion model;

[0032] When the static finite element inversion model no longer matches the actual dynamic load environment, a sliding window is used to determine the stability of the monitored stress values: the stress fluctuation variance is calculated for the stress values ​​within each window. When the variance of the fluctuation When the stress level is below a preset threshold, the geometric model structure is determined to be in a horizontal operation phase. At this time, the average stress within the window is taken as input, and the output is obtained through a static inversion model, which serves as the load on the geometric model for that operation cycle. The load during this period Under the action of stress, the cycle number is obtained by using the rain flow counting method based on the stress change sequence recorded by the stress sensor, thus completing the acquisition of working parameters for a single work cycle;

[0033] Considering the maximum dynamic load of the geometric model during the working cycle The stage is accompanied by a rapid increase in stress, and the minimum dynamic load at the end. The stress decreases rapidly during this phase; therefore, the transition point where the first derivative of stress changes from negative to positive is the boundary of the working period. Real-time forward difference calculation is then performed on the stress sequence. The first derivative of the stress is obtained and used to identify the working cycle boundary;

[0034] Alternatively, step five can be achieved by having the user input relevant information through the platform's human-computer interaction interface.

[0035] The above technical solution, by setting a minimum time span to avoid false triggering, and combining it with the aforementioned sliding window steady-state detection to avoid pseudo-cycle identification, stores the loading history for each cycle in a database for platform retrieval. This data is used for calculating the remaining lifespan of the spreader and visualizing the equipment's load status. Alternatively, users can input relevant information through the platform's human-machine interface.

[0036] Preferably, in step six, the integrated visual operation and maintenance platform calculates the remaining lifespan and compares it with the lifespan threshold to update the device status.

[0037] The geometric model described in step one is converted to a file format using 3ds Max, the target file format is exported, and it is imported into the main scene of Unity3D for visualization.

[0038] Meanwhile, Unity3D can use XChart charts to dynamically visualize the stress sequence collected in step five. The Unity3D backend uses Miner's linear cumulative damage theory to dynamically assess the remaining life of the structure. The formula for Miner's linear cumulative damage theory is:

[0039] ;

[0040] In the formula, Total fatigue damage, This represents the actual number of loops. This refers to the number of cycles allowed under the corresponding load.

[0041] damage Used to estimate remaining lifetime, where remaining lifetime When the value is 1, the geometric model is considered intact; remaining lifetime When the value is 0, failure occurs;

[0042] Based on the calculated remaining lifespan, the device status of the geometric model is updated in real time, and alarm thresholds are set. When the calculated remaining lifespan does not decrease to At that time, the equipment remained in normal service condition, but when it dropped to... It issues a fatigue failure risk warning in a timely manner. After the equipment has been repaired or replaced, the remaining life is updated to 1, and the equipment status is updated to normal service.

[0043] The above technical solution imports the geometric model into the Unity3D main scene. C# scripts are then written to leverage the Unity engine to rotate and reset objects in the 3D scene, facilitating user inspection of the geometric model. Simultaneously, a human-computer interaction interface is provided, allowing users to input current operating parameters, including lifting mode, lifting load, and cycle number. The platform supports repeated input of operating parameters, and all input data is managed as a load history record by a backend database, enabling the proxy model to call and dynamically update damage accumulation. The structural lifespan is quickly predicted using a deep learning proxy model, and the remaining lifespan is calculated using Miner's theory. A visualization module is then used to achieve real-time display and intelligent early warning of the equipment status.

[0044] The beneficial effects of this invention are as follows:

[0045] This invention can intelligently deduce the evolution of fatigue damage based on the structure's service parameters and loading history, achieving dynamic calculation of remaining service life and real-time early warning of service risks. The neural network proxy model constructed in this invention saves computational resources and improves the efficiency of fatigue life assessment compared to traditional finite element simulation calculations, possessing good real-time performance and practicality. The model is deployed using a standardized and universal format, facilitating integration and cross-platform calls, thus improving the system's flexibility and scalability. The evaluation system effectively integrates damage theory and failure mechanisms by introducing finite element simulation calculation datasets and the Miner criterion, providing theoretical support for the intelligent deduction process and ensuring the accuracy and reliability of remaining service life prediction results. Furthermore, combined with a visualization platform developed based on the Unity3D engine, it can intuitively display the equipment's geometry and health status, with a user-friendly interface, enhancing the platform's operability. This method addresses the problems of reliance on offline simulation, response lag, and discontinuous assessment in traditional structural life management, effectively improving the intelligent operation and maintenance level of engineering machinery and ensuring its safe and efficient operation. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of an intelligent monitoring method for fatigue life of load-bearing structures of engineering machinery proposed in this invention;

[0047] Figure 2 This is a neural network architecture diagram of an intelligent monitoring method for fatigue life of load-bearing structures of engineering machinery proposed in this invention;

[0048] Figure 3 This is a neural network training loss curve for an intelligent monitoring method for fatigue life of load-bearing structures of engineering machinery proposed in this invention.

[0049] Figure 4 This is a comparison of the neural network prediction effects of the intelligent monitoring method for fatigue life of load-bearing structures of engineering machinery proposed in this invention;

[0050] Figure 5 This is a client-side main page diagram of a lifting device fatigue life prediction system based on an intelligent monitoring method for fatigue life of load-bearing structures of engineering machinery proposed in this invention. Detailed Implementation

[0051] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0052] Reference Figure 1-5 A method for intelligent monitoring of fatigue life of load-bearing structures in engineering machinery, such as Figure 1 As shown, the process includes step one: establishing a high-fidelity geometric model.

[0053] In step one, the geometric model is constructed using 3D modeling software, and then exported to the corresponding file format according to requirements.

[0054] Step 2: Establish a finite element model.

[0055] The geometric model is imported into the finite element simulation software in step two according to the file type requirements for simplification. After simplification, material properties are defined, the structure in the geometric model is meshed, mesh independence is verified, and finally, relevant boundary conditions are set. Specifically, this is implemented as follows:

[0056] Based on the established geometric model, finite element simulation software was used to simplify the model, retaining key components that affect the mechanical performance of the lifting device while reasonably simplifying minor factors. Common simplification measures include removing or closing structural details that do not participate in the main stress, such as chamfers, bolt holes, nameplate grooves, and small fillets; using equivalent rod elements or simplified blocks to replace regularly shaped components such as pins and tie rods that are much smaller than the overall structure; simplifying multiple welded components into a single solid to reduce the treatment of contact surfaces; and removing accessories such as wires and pipes. Subsequently, material properties were defined, the structure was meshed, and mesh independence verification was performed. Local refinement strategies were used in key stress areas to improve calculation accuracy. Finally, relevant boundary conditions were set to prepare for subsequent simulation calculations.

[0057] In one embodiment, the STEP file exported in step one is imported into Ansys Workbench for model simplification. Electrical equipment, hydraulic pipelines, and their accessories in the lifting device's geometric model are removed. Smaller welds and components in the main beam and telescopic beam are modeled as a whole, while larger welds between the main beam webs and between the main beam and the mounting plate are modeled as separate components. Smaller threaded holes and other structures are ignored. Small chamfers and fillets in the lifting device structure are modeled as right angles. The main load-bearing components in the lifting device structure are made of high-strength steel, the buffer pad between the main beam and the telescopic beam is made of nylon 66, and the remaining components are made of ordinary carbon structural steel. Material parameters such as Poisson's ratio, elastic modulus, density, yield strength, and tensile strength are assigned to the corresponding materials. Subsequently, contact conditions are generated based on the actual situation of the lifting device, fixed support constraints are set, and mesh generation is performed. The mesh is mainly hexahedral, with tetrahedral meshes used for locally complex structures. Mesh independence verification is conducted to confirm the reasonableness of the mesh generation.

[0058] Step 3: Construct the finite element simulation dataset.

[0059] First, the integration step size is set according to the working state and working time, and an appropriate damping form is selected according to the structural characteristics and analysis requirements. Transient dynamic analysis is then performed on the entire cycle process of the geometric model to obtain the load spectrum.

[0060] Then, the load spectrum was analyzed using Ansys Workbench and nCode DesignLife in conjunction with fatigue analysis to obtain the corresponding loads. Fatigue life .

[0061] The maximum load that the geometric model may encounter under actual working conditions. and minimum load Add a certain amount of redundancy The upper limit load of the geometric model is then obtained. and lower limit load Simulation dataset input The goal is to make the simulation data as close as possible to actual working conditions.

[0062] Finally, hyper-Latin cube sampling was used at the set upper limit load. and lower limit load Extract a sufficient number of data points and obtain the fatigue life for each data point through simulation analysis. These input and output data are recorded in tabular form to form a load-life mapping dataset for finite element simulation. ,in, .

[0063] In some real-world applications, lifting operations involve three different modes depending on the container size. For each mode, the maximum and minimum loads that may occur in the actual working conditions are identified, and a certain amount of redundancy is added as the upper and lower limits of the simulation dataset input. Points are selected within the input range using hyper-Latin cube sampling. The corresponding fatigue life is obtained through transient dynamics combined with Ansys Workbench and nCode DesignLife fatigue analysis, thus constructing three load-life mapping datasets for finite element simulations. .

[0064] Step 4: Construct a neural network proxy model; build the neural network architecture from Step 4 using Python, based on the load-lifetime mapping dataset constructed in Step 3. The model is normalized and divided into training and testing sets according to a certain ratio. It is then used for training, validation, and hyperparameter tuning. Specifically, during training, an appropriate optimizer and loss function are selected, and hyperparameters such as learning rate, epochs, and batch size are set. A load-lifetime surrogate model is then constructed.

[0065] During training, the loss function will calculate the predicted values ​​of the neural network model and the load-lifetime mapping dataset. The error between labels is backpropagated to update the parameters of the neural network model;

[0066] The neural network model is evaluated using mean absolute error (MAE), mean squared error (MSE), and coefficient of determination. As an evaluation metric, if the metric does not meet expectations, the hyperparameter size is adjusted. When the accuracy of the neural network model meets the requirements, the neural network model is saved as a target file format, and the normalization information during training is manually saved in the checkpoint of the target file to prevent prediction deviations in the inference stage caused by inconsistent normalization.

[0067] Export the model weights and structure from the target file into a general model deployment format, and record the normalization information in the checkpoint of the target file so that the input specifications during training can be correctly restored when using the model for inference.

[0068] In this embodiment, a four-layer ANN model is established, and the model architecture is as follows: Figure 2 As shown, the input layer is a one-dimensional load on a lifting device, the hidden layer has two layers, each with 64 neurons, and the output is the logarithm of the one-dimensional fatigue life. The activation function used is ReLU. Using the dataset constructed in step three, data preprocessing is performed on the finite element calculations. Logarithmic values ​​were taken and Max-Min normalization was performed. Load-life surrogate models were then trained under three different operating modes. The comparison between the training loss curve and prediction performance under the 20-foot container lifting operating mode is shown in the figure below. Figure 3 and Figure 4 As shown, the weights and structures of the three trained models are exported into a general model deployment format, and the normalization information is saved.

[0069] Step 5: Obtain working parameters and loading history; the working parameters in Step 5 include the load and cycle number of the geometric model;

[0070] When the geometric model is a lifting device, the working parameters refer to the lifting device load and cycle number, and the loading history refers to the entire process of load changes experienced by the lifting device structure over time during operation. This refers to the load on the geometric model for each working cycle. The cumulative total of the corresponding cycle periods, where, ;

[0071] Based on the finite element model established in step three, a static analysis is performed to calculate the critical points of the geometric model structure, such as the lifting device. A static inversion model of the neural network model is then established. By analyzing the stress at the critical points of the geometric model structure, the loads acting on the geometric model structure are inverted. Specifically, the maximum and minimum loads that may occur in actual working conditions are examined, and a certain amount of redundancy is added as the upper and lower limits of the dataset, so that the finite element simulation data can cover the actual working conditions as much as possible.

[0072] Hyper-Latin cubic sampling was used to extract data points, and the stress values ​​at critical points were calculated through static simulation. The calculated stress was used as input to the neural network dataset, and the corresponding load was used as output. The data was recorded in tabular form to form a stress-load mapping dataset for finite element simulation. Then, the neural network model architecture is established, and the stress-load mapping dataset from finite element simulation is used. Complete the training of the static inversion model. Install strain gauges or stress sensors at the critical points of the actual lifting equipment to monitor the stress magnitude at these points online. Because the stress changes of the lifting equipment exhibit distinct phased characteristics within a single work cycle, and the lifting and unloading phases are affected by dynamic loads.

[0073] When the static finite element inversion model no longer matches the actual dynamic load environment, it is necessary to identify the operational phase and use a sliding window to determine the stability of the monitored stress values: a preset fixed time step is taken as a window, and the number of data points in each window is the same. As the work progresses, the window is slid forward, continuously updating the stress value data points corresponding to the time step, and the stress fluctuation variance is calculated for the stress values ​​within each window. When the variance of the fluctuation When the stress level is below a preset threshold, the geometric model structure is determined to be in a horizontal operation phase. At this time, the average stress within the window is taken as input, and the output is obtained through a static inversion model, which serves as the load on the geometric model for that operation cycle. The load during this period Under the influence of stress, the cycle number is obtained by using the rain flow counting method based on the stress change sequence recorded by the stress sensor, thus completing the acquisition of working parameters for a single operation cycle.

[0074] Furthermore, considering the maximum dynamic load during the operation cycle of the geometric model... The stage is accompanied by a rapid increase in stress, and the minimum dynamic load at the end. The stress decreases rapidly during this phase; therefore, the transition point where the first derivative of stress changes from negative to positive is the boundary of the working period. Real-time forward difference calculation is then performed on the stress sequence. The first derivative of the stress is obtained and used to identify the working cycle boundary. Simultaneously, a minimum time span is set to avoid false identification, and combined with the aforementioned sliding window steady-state detection, pseudo-cycle identification is avoided. The loading history for each cycle is stored in a database for the platform to retrieve and use in calculating the remaining life of the spreader.

[0075] Alternatively, step five can be achieved by having the user input relevant information through the platform's human-computer interaction interface.

[0076] Step Six: Integrate the visual operation and maintenance platform. The integrated visual operation and maintenance platform in Step Six calculates the remaining lifespan and compares it with the lifespan threshold, updating the device status accordingly.

[0077] Use 3ds Max to convert the geometric model described in step one to a file format such as STEP, export it as FBX format and import it into the main scene of Unity3D for visualization.

[0078] Meanwhile, Unity3D can use XChart charts to dynamically visualize the stress sequence collected in step five, thus providing an intuitive presentation of the stress change process and facilitating a comprehensive understanding of the structural stress state by the user. Furthermore, combining the load-life surrogate model constructed in step four with the working parameters and loading history obtained in step five, the Unity3D backend uses Miner's linear cumulative damage theory to dynamically assess the remaining life of the structure. The formula for Miner's linear cumulative damage theory is:

[0079] ;

[0080] In the formula, Total fatigue damage, This represents the actual number of loops. This refers to the number of cycles allowed under the corresponding load.

[0081] damage Used to estimate remaining lifetime, where remaining lifetime When the value is 1, the geometric model is considered intact; remaining lifetime When the value is 0, failure occurs;

[0082] Based on the calculated remaining lifespan, the device status of the geometric model is updated in real time, and alarm thresholds are set. When the calculated remaining lifespan does not decrease to At that time, the equipment remained in normal service condition, but when it dropped to... It issues a fatigue failure risk warning in a timely manner. After the equipment has been repaired or replaced, the remaining life is updated to 1, and the equipment status is updated to normal service.

[0083] In this embodiment, the system supports both online and offline local operation modes. In online mode, the Flask framework serves as the backend service platform, while the frontend is built using the Unity engine and a Vue.js-based web frontend to achieve a multi-terminal collaborative human-computer interaction interface. The system frontend receives user-inputted device operating parameters and transmits them to the backend server for processing. The backend performs predictions using a deployed neural network model and estimates the remaining lifetime using the Miner criterion, then feeds the calculation results back to the frontend for visualization. Any frontend can independently input operating parameters, and the backend processes and pushes the results uniformly upon receiving the data. Simultaneously, a MySQL relational database configured in the system ensures data consistency between the frontend and backend, as well as data synchronization between the various frontends.

[0084] Furthermore, the Unity frontend supports interactive viewing of 3D models, while the Vue.js frontend presents the device's lifespan status through a web interface, intuitively displaying the predicted remaining lifespan as a graphical health bar, enhancing the system's flexibility and visualization capabilities. This dual-frontend architecture improves the system's operability and deployment adaptability, making it suitable for various intelligent operation and maintenance scenarios for industrial equipment.

[0085] In offline mode, only local prediction is supported on the Unity client. The client's main interface is as follows: Figure 5 As shown. Similar to the online client interface, the geometric model is imported into the Unity3D main scene. C# scripts are written to utilize the Unity engine to rotate and reset the objects in the 3D scene, facilitating user inspection of the geometric model. The client features a human-computer interaction interface, allowing users to input current working parameters, including lifting mode, lifting load, and cycle number. The platform supports repeated input of working parameters, and all input data is managed uniformly by the backend database as a load history record, facilitating the proxy model's invocation and dynamic updating of damage accumulation. The structural lifespan is quickly predicted through a deep learning proxy model, and the remaining lifespan is calculated using Miner's theory. Combined with a visualization module, the equipment status is displayed in real time and intelligently warned.

[0086] The system forms a closed-loop process from data input, life prediction, status update to interactive feedback, which effectively improves the health management efficiency and safety reliability of the spreader structure.

[0087] This invention can intelligently deduce the evolution of fatigue damage based on the structure's service parameters and loading history, achieving dynamic calculation of remaining service life and real-time early warning of service risks. The neural network proxy model constructed in this invention saves computational resources and improves the efficiency of fatigue life assessment compared to traditional finite element simulation calculations, possessing good real-time performance and practicality. The model is deployed using a standardized and universal format, facilitating integration and cross-platform calls, thus improving the system's flexibility and scalability. The evaluation system effectively integrates damage theory and failure mechanisms by introducing finite element simulation calculation datasets and the Miner criterion, providing theoretical support for the intelligent deduction process and ensuring the accuracy and reliability of remaining service life prediction results. Furthermore, combined with a visualization platform developed based on the Unity3D engine, it can intuitively display the equipment's geometry and health status, with a user-friendly interface, enhancing the platform's operability. This method addresses the problems of reliance on offline simulation, response lag, and discontinuous assessment in traditional structural life management, effectively improving the intelligent operation and maintenance level of engineering machinery and ensuring its safe and efficient operation.

[0088] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An intelligent monitoring method for fatigue life of a construction machinery load-bearing structure, characterized by comprising the following steps: Step 1, establishing a high-fidelity geometric model; Step 2, establishing a finite element model; Step 3, constructing a finite element simulation dataset; Step 4, constructing a neural network proxy model; Step 5, obtaining working parameters and loading history; the working parameters in Step 5 include load and cycle number of the geometric model; Step 6, integrating a visual operation and maintenance platform. In Step 1, the geometric model is constructed by a three-dimensional modeling software, and then the geometric model is exported into a corresponding file format according to requirements. The geometric model is imported into a finite element simulation software in Step 2 according to file type requirements, and the geometric model is simplified. After the geometric model is simplified, material properties are defined, the structure in the geometric model is meshed, mesh independence verification is carried out, and relevant boundary conditions are set. In Step 3, integral step is set according to working state and working time, appropriate damping form is selected according to structure characteristics and analysis requirements, transient dynamics analysis is carried out on the whole cycle process of the geometric model, and load spectrum is obtained. The model weight and structure in the target file are exported into a general model deployment format, and the normalization information in the checkpoint of the target file is recorded. In Step 6, the integrated visual operation and maintenance platform calculates the remaining life and compares it with the life threshold, and updates the equipment state. The load history includes the full history of the load experienced by each of the geometric model structures over time during the course of the work, for each work cycle of the geometric model with the accumulation of the corresponding cycle number, wherein ; Based on the finite element model established in step three, static analysis is carried out to calculate the dangerous points of the geometric model structure, and a static inversion model of the neural network model is established, and the load borne by the geometric model structure is inverted through the stress at the dangerous point of the geometric model structure , forming a stress-load mapping data set of finite element simulation , the neural network model architecture is re-established, and the stress-load mapping data set of finite element simulation is used to complete the training of the static inversion model; When the static finite element inversion model and the actual dynamic load environment will no longer match, the sliding window is used to monitor the stress value to determine stability: the stress fluctuation variance is calculated for the stress value in each window When the fluctuation variance is lower than the preset threshold value, it is determined that the geometric model structure is in the horizontal operation stage, at which time the stress mean value in the window is taken as the input, the output is obtained by the statics inversion model, and is taken as the load of the geometric model of the operation cycle Under the load of the cycle, according to the stress change sequence recorded by the stress sensor, the cycle number is obtained by using the rain flow counting method, and the acquisition of the single operation cycle work parameter is completed; Considering the maximum dynamic load of the geometry model working cycle The minimum dynamic load at the end of the stage with rapid stress rise The stage is accompanied by rapid stress drop, so the transition point of the first derivative of stress from negative to positive value is the working cycle boundary. Real-time forward difference calculation is performed on the stress sequence: , to obtain the first derivative of stress for identifying the working cycle boundary; The file format of the geometric model in Step 1 is converted by 3DMax, and the target file format is exported and imported into the main scene of Unity3D for visual display.

2. The method of claim 1, wherein: Meanwhile, the stress sequence collected in Step 5 is dynamically visualized by XChart charts in Unity3D, and the remaining life of the structure is dynamically evaluated in the background of Unity3D by using Miner linear cumulative damage theory, wherein the formula of Miner linear cumulative damage theory is:

3. A method of intelligent monitoring of fatigue life of a construction machine load bearing structure according to claim 2, characterized in that: ​ 4. The method of claim 3, wherein: ​ 5. A method of intelligent monitoring of fatigue life of a construction machine load bearing structure according to claim 4, characterized in that: ​ 6. A method of intelligent monitoring of fatigue life of a construction machine load bearing structure according to claim 5, characterized in that: The load spectrum is analyzed by Ansys workbench and nCode DesignLife, and the fatigue life under the corresponding load is obtained ;​ the maximum load and minimum load that can occur in the actual operating conditions of the geometry model additional redundancy upper and lower load limits of the geometry model are obtained simulation data set input Super Latin cube sampling was used at the set upper limit load. and lower limit load Extract a sufficient number of data points and obtain the fatigue life for each data point through simulation analysis. These input and output data are recorded in tabular form to form a load-life mapping dataset for finite element simulation. ,in, .

7. A method of intelligent monitoring of fatigue life of a construction machine load bearing structure according to claim 6, characterized in that: establishing the neural network architecture in the fourth step by Python, based on the load-life mapping dataset constructed in the third step Normalization processing is performed, and the training set and the test set are divided in a certain proportion. The model is trained, verified, and parameter-adjusted to construct a load-life proxy model.

8. A method of intelligent monitoring of fatigue life of a construction machine load bearing structure according to claim 7, characterized in that: During training, the loss function will calculate the error between the neural network model's predictions and the load-life mapping dataset propagated back to update the parameters of the neural network model; The neural network model adopts mean absolute error (MAE), mean square error (MSE) and coefficient of determination (R2) during evaluation As an evaluation index, if the index does not reach the expectation, the hyperparameter size is adjusted, when the accuracy of the neural network model reaches the requirement, the neural network model is saved as a target file format, and the normalization information during training is manually saved in the checkpoint of the target file. ​ 9. A method of intelligent monitoring of fatigue life of a construction machine load bearing structure according to claim 8, characterized in that: ​ ​ ​ ； wherein is the total fatigue damage, is the actual number of cycles, is the number of cycles allowed under the corresponding load; damage for estimating the remaining life, wherein the remaining life is 1, the geometric model is considered intact; the remaining life is 0, i.e. failure occurs; According to the calculated residual life, the device state of the geometric model is updated in real time, and an alarm threshold is set When the calculated residual life does not decrease to , the device state remains normal service, when it decreases to , a fatigue failure risk warning is issued, and after the device is repaired or replaced, the residual life is updated to 1 and the device state is updated to normal service.

Images