Steel structure mold service life prediction and maintenance decision system and method

Abstract

The embodiment of the invention provides a steel structure mold service life prediction and maintenance decision system and method. The method comprises the steps that multiple steel structure molds serve as standard parts, and multi-source data of the multiple steel structure molds are obtained and used for constructing knowledge maps of the multiple steel structure molds; the visual image information of the target mold, the production condition of the target mold and the mold material information serve as index conditions, and a machine learning model matched with the target mold is selected from the candidate model library in combination with the knowledge graph; training a machine learning model matched with the target mold by adopting the visual image information, the production condition of the target mold, the mold material information and the historical operation characteristics matched with the target mold in the knowledge graph; inputting real-time operation data of the target mold into the data driving model obtained by training to obtain a service life prediction value and a prediction damage state of the target mold; and executing a corresponding mold maintenance strategy based on the service life predicted value and the predicted damage state.

Description

Technical Field

[0001] The present application relates to the field of data prediction, and in particular to a steel structure mold life prediction and maintenance decision-making system and method. Background Art

[0002] At present, the service life of steel structure molds is mainly calculated based on empirical formulas.

[0003] In related technologies, when maintaining steel structure molds, empirical relationships between mold life and factors such as mold materials, process parameters, and production batches are summarized based on a large amount of actual production data and experimental results. For example, in stamping molds, empirical formulas may consider the relationship between parameters such as sheet thickness, stamping speed, and mold hardness and mold life. However, empirical formulas have poor versatility and are often only applicable to specific production conditions and mold types. Once production conditions change, such as the use of new mold materials or processing technologies, the original empirical formula may no longer be applicable and need to be re-established. In addition, the historical data on which the empirical formula is based is of varying quality and may contain noise, missing values, and other problems, which will also affect the accuracy of the empirical formula.

[0004] Therefore, it is urgent to design a technical solution to overcome at least one technical problem in the process of maintaining steel structure molds. Summary of the Invention

[0005] The main purpose of the embodiments of this application is to provide a steel structure mold life prediction and maintenance decision-making system and method, aiming to solve technical problems such as poor universality of empirical formulas and low quality of historical data in the process of maintaining steel structure molds.

[0006] In a first aspect, an embodiment of the present application provides a method for predicting the life of a steel structure mold and making maintenance decisions, comprising:

[0007] Using multiple steel structure molds as standard parts, multi-source data of multiple steel structure molds is obtained to construct a knowledge graph of multiple steel structure molds; material information, historical operation data, performance under specific process parameters, damage conditions, service life, mold vibration information, and shape characteristics of different steel structure molds in the multi-source data are associated in the form of a knowledge graph;

[0008] Scanning visual image information of the target mold; using the visual image information, the production conditions of the target mold, and mold material information as index conditions, and combining the knowledge graph to select a machine learning model that matches the target mold from a candidate model library;

[0009] The visual image information, the production conditions of the target mold, the mold material information, and the historical operation characteristics matching the target mold in the knowledge graph are used to train a machine learning model for matching the target mold to obtain a data-driven model for the target mold; the data-driven model is used to predict the service life and damage of the target mold;

[0010] Collecting real-time operating data of the target mold and inputting it into the data-driven model to obtain a predicted service life value and a predicted damage state of the target mold;

[0011] A corresponding mold maintenance strategy is executed based on the predicted service life value and the predicted damage state.

[0012] In a second aspect, an embodiment of the present application provides a steel structure mold life prediction and maintenance decision-making system, comprising:

[0013] A construction module is used to obtain multi-source data of various steel structure molds using various steel structure molds as standard parts, so as to construct a knowledge graph of the various steel structure molds; and to associate the material information, historical operation data, performance under specific process parameters, damage conditions, service life, mold vibration information, and shape characteristics of different steel structure molds in the multi-source data in the form of a knowledge graph;

[0014] A selection module is configured to scan visual image information of a target mold; use the visual image information, the production conditions of the target mold, and mold material information as index conditions, and select a machine learning model that matches the target mold from a candidate model library in combination with the knowledge graph;

[0015] a training module for training a machine learning model matching the target mold using the visual image information, the production conditions of the target mold, the mold material information, and historical operating characteristics matching the target mold in the knowledge graph to obtain a data-driven model for the target mold; the data-driven model is used to predict the service life and damage of the target mold;

[0016] A prediction module, configured to collect real-time operating data of a target mold and input the data into the data-driven model to obtain a predicted service life and damage status of the target mold;

[0017] A maintenance module is used to execute a corresponding mold maintenance strategy based on the predicted service life value and the predicted damage state.

[0018] In a third aspect, an embodiment of the present application further provides a terminal device, which includes a processor and a memory for storing computer programs; the processor is used to execute the computer program and implement the steel structure mold life prediction and maintenance decision-making method described in the first aspect or any embodiment of the present application when executing the computer program.

[0019] The embodiment of the present application provides a system and method for predicting the life of steel structure molds and making maintenance decisions. This method realizes comprehensive and accurate life prediction and damage monitoring of steel structure molds through a series of operations such as constructing a knowledge graph based on multi-source data, selecting models based on multiple information, customizing training data-driven models, real-time data input prediction, and scientifically executing maintenance strategies. It improves the scientificity and rationality of mold maintenance decisions, optimizes the allocation of maintenance resources, reduces the impact of mold failures on production, and improves production efficiency and product quality. At the same time, the application of knowledge graphs and data-driven models makes full use of historical data and prior knowledge, improves the generalization ability and adaptability of the model, and provides a more effective technical means for the management of steel structure molds. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A flow chart of a steel structure mold life prediction and maintenance decision-making method provided in an embodiment of the present application;

[0021] Figure 2 A schematic diagram of the module structure of a steel structure mold life prediction and maintenance decision-making system provided in an embodiment of the present application;

[0022] Figure 3 A schematic block diagram of the structure of a terminal device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0023] In response to the technical problems existing in the relevant technologies, the embodiments of the present application propose a system and method for predicting the life of steel structure molds and maintenance decisions. Specifically, first, the material information, historical operating data, performance under specific process parameters, damage conditions, service life, mold vibration information and shape characteristics of different steel structure molds are associated in the form of a knowledge graph to form a comprehensive understanding of the mold. For example, through the knowledge graph, you can intuitively see the damage pattern and life change of a certain mold material under a specific process, which helps to discover the potential relationship between different factors and provide richer information support for the optimization and maintenance of the mold. The construction of the knowledge graph enables knowledge between different molds to be shared and reused. For example, when designing a new mold, you can refer to the experience of similar molds in the knowledge graph to select appropriate materials and process parameters, reduce trial and error, and improve design efficiency and mold quality.

[0024] Secondly, a matching machine learning model is selected from a library of candidate models, using the visual image information of the scanned target mold, the production conditions, and the mold material information as index conditions. This approach selects the most appropriate model based on the specific characteristics of the target mold, improving the model's predictive accuracy. For example, for molds with complex shapes, a model capable of handling complex geometric features is selected to better capture changes in mold performance. The model selection is combined with a knowledge graph, leveraging the prior knowledge about molds in the knowledge graph to make the selected model more adaptable to different molds and production conditions. For example, when production conditions change, the knowledge graph can provide operating data for molds under similar conditions, helping the model better adapt to the new production environment.

[0025] Thirdly, a data-driven model is trained using the target mold's visual image information, production conditions, mold material information, and historical operating features matched in the knowledge graph. This allows for a customized prediction model specifically for the target mold. This model can more accurately reflect the actual operating conditions of the target mold, improving the accuracy of predictions of service life and damage. For example, by training the model on historical data from the target mold's specific materials and production conditions, it can more accurately predict the mold's damage level at different stages. Combining the model with historical operating features from the knowledge graph allows for full utilization of past mold operating data and the potential patterns within the data. Even for new molds, training can be performed using historical data from similar molds in the knowledge graph, reducing reliance on large amounts of new data and improving the model's generalization capabilities.

[0026] Next, real-time operating data of the target mold is collected and input into the data-driven model, enabling real-time predictions of the mold's service life and predicted damage status. For example, if the mold's vibration data is abnormal, the model can promptly predict potential damage and issue an early warning, preventing sudden mold failure and reducing production risks and costs. Based on the real-time service life predictions and predicted damage status, the mold maintenance strategy can be dynamically adjusted. For example, if the remaining life of the mold is predicted to be short, maintenance or replacement can be scheduled in advance to avoid production interruptions caused by mold failure. If the damage status is predicted to be minor, the maintenance cycle can be appropriately extended to optimize the allocation of maintenance resources.

[0027] Finally, mold maintenance strategies are implemented based on the predicted service life and damage status, making maintenance decisions more scientific and rational. The model's predictions can help avoid over- or under-maintenance, thereby improving mold reliability and service life. For example, for molds predicted to have minimal damage, unnecessary maintenance can be reduced, lowering maintenance costs. For molds predicted to have significant damage, timely measures can be taken to prevent further damage. A sound maintenance strategy can minimize the impact of mold failures on production and improve productivity. For example, proactive maintenance or mold replacement ensures production continuity, reduces downtime, and improves production efficiency and product quality.

[0028] It is particularly important to note that in order to address the problem of poor versatility of empirical formulas, the embodiments of the present application first associate information such as mold materials, process parameters, production batches, mold shapes, and damage patterns by constructing a knowledge graph related to the mold. In this way, when production conditions change, such as when a new mold material is adopted, the life of the new mold material can be inferred based on the performance data of similar materials in the knowledge graph and the impact of the mold shape on the life, thereby supplementing and correcting the empirical formula so that it can adapt to the new production conditions. Data is collected in real time during the operation of the mold. When production conditions change, such as when the stamping speed changes, the impact on the mold life is analyzed based on the real-time collected mold vibration and stress data, and the empirical formula is dynamically corrected in a timely manner to enhance the adaptability of the empirical formula to different production conditions. Thus, the embodiments of the present application combine the empirical formula with a prediction model based on machine learning, integrating the physical meaning of the empirical formula with the ability of the machine learning model to process complex data. At the same time, optimization algorithms such as genetic algorithms and particle swarm optimization algorithms are used to optimize model parameters to improve the model's adaptability to different production conditions and mold types, thereby enhancing the versatility of the empirical formula. In particular, we integrate multi-source data, including mold design data, manufacturing data, and maintenance records, to fully understand the factors affecting mold performance and lifespan. We use data analysis techniques such as principal component analysis and factor analysis to extract key features and simplify dimensions from multi-source data. We then incorporate factors that significantly impact mold lifespan into empirical formulas, improving their accuracy and versatility.

[0029] To address the quality issues of historical data, the embodiments of this application further evaluate the integrity, accuracy, and consistency of historical data, check for missing values, outliers, and noise, and use statistical methods or machine learning algorithms to clean and repair the data, thereby controlling the data quality as a whole. In addition, cluster analysis can be used to identify outliers in the data, data interpolation methods can be used to fill missing values, and data smoothing can be used to reduce the impact of noise, improve data quality, provide a more reliable data foundation for empirical formulas, and thus improve the accuracy of empirical formulas.

[0030] Embodiments of the present application provide a system and method for predicting and maintaining steel structure mold lifespans. This method can be applied to terminal devices, such as mobile phones, virtual reality devices, tablet computers, laptop computers, desktop computers, wearable devices, and other electronic devices. The terminal device can be a server connected to a tower crane or a server cluster. This connection can be implemented via hardware circuitry or a communication module.

[0031] The following is a detailed description of some embodiments of the present application in conjunction with the accompanying drawings. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Figure 1 , Figure 1 A flow chart of a method for predicting the life of a steel structure mold and making maintenance decisions provided in an embodiment of the present application.

[0032] like Figure 1 As shown, the steel structure mold life prediction and maintenance decision-making method includes the following steps S101 to S106.

[0033] Step S101: Using a variety of steel structure molds as standard parts, obtain multi-source data of the various steel structure molds to construct a knowledge graph of the various steel structure molds.

[0034] In the embodiments of this application, standard parts refer to representative steel structure molds with clear characteristics. As the object of data collection and analysis, their data is typical and universal, and can be used to build the basic framework of the knowledge graph. Among them, multi-source data can be obtained from different channels and using different methods, including the composition and mechanical properties of the mold material, historical operating conditions, performance indicators under specific processes, damage location and type, actual service life, vibration signal characteristics, and shape and geometry parameters.

[0035] Furthermore, in step S101, the material information, historical operation data, performance under specific process parameters, damage conditions, service life, mold vibration information and shape characteristics of different steel structure molds in the multi-source data are associated in the form of a knowledge graph.

[0036] A knowledge graph is a structured network used to represent and store knowledge. It consists of nodes and edges, with nodes representing entities and edges representing relationships between entities. A knowledge graph graphically displays entities and their relationships. In this scenario, mold-related data is abstracted into entities (such as mold materials and process parameters) and relationships (such as "using a certain material" and "producing a certain damage under a certain process"). This facilitates the discovery of potential connections between the data, provides knowledge support for subsequent model selection and training, and addresses the difficulty of empirical formulas in associating multiple factors. In this step, material information includes the material composition, physical properties (such as hardness, strength, and toughness), and chemical properties of the steel structure mold. This information, as nodes in the knowledge graph, helps understand the basic characteristics of the mold. Different materials may perform differently under the same process. Historical operating data, such as mold operating time, operating frequency, and number of downtimes, reflects mold usage. Connected as nodes with other information, this data can be used to analyze the impact of usage patterns on mold performance and lifespan. Performance under specific process parameters, such as temperature, pressure, and processing speed, includes molding accuracy and surface quality. By associating them in the form of a knowledge graph, we can clearly see the effect of different process conditions on mold performance, which is convenient for optimizing the process. Damage conditions are used to record the type and degree of damage such as wear, cracks, and deformation of the mold. After being associated with other data, the factors that cause damage can be found, such as whether specific process parameters or operating time are prone to cause certain damage. As a key node, service life can be associated with other information to analyze which factors have a greater impact on service life, providing a basis for predicting service life. Mold vibration information includes that vibration may affect the stability and life of the mold. Association with other data can explore the relationship between vibration and materials, process parameters, etc., and take measures to reduce vibration hazards. Shape features include the size and contour of the mold. Different shape features may be subjected to different stresses during operation. Combining with other information helps analyze the impact of shape on mold performance.

[0037] Exemplarily, when implementing step S101, the architecture of "data gene encoding-intelligent association-dynamic graph growth" can be adopted to break through the static modeling limitations of traditional knowledge graph construction. First, the corresponding digital genes are labeled for each type of data. For example, the material specification document is parsed through natural language processing to generate material entities (such as "Cr12MoV steel" nodes), and the shape features are automatically extracted from the mold CAD model using a three-dimensional point cloud deep learning algorithm (such as "thin wall curvature radius 15mm" as a shape node), and the time series vibration signals in the historical operation data are converted into frequency domain feature nodes (such as "10kHz energy value 0.8V") through Fourier transform. Then, a causal association engine is constructed to mine the implicit relationship between data through a causal inference algorithm. For example, when it is found that the stamping speed of a batch of molds is greater than 200 times per minute, the association edge weight of its carbon steel material node and the "edge crack" damage node increases significantly, and a causal chain of "high-speed stamping → material fatigue → crack" is automatically generated. Therefore, knowledge embedding technology (such as the TransE model) is used to map process parameters (such as "injection temperature 230°C") and performance (such as "cavity wear rate 0.05mm / thousand times") to a unified vector space, so that cross-domain data can be semantically associated. Finally, edge computing nodes are deployed to collect mold vibration data in real time, and the pre-trained graph convolutional neural network (GCN) is used to identify emerging vibration abnormality patterns in real time, and automatically create temporary association edges of "abnormal vibration frequency-potential damage location". When new materials (such as nanocrystalline steel) are introduced, the "material-life" reasoning rules in the existing graph are reused through transfer learning to generate a predictive relationship chain that adapts to the new entity. This dynamic construction method makes the knowledge graph not only a static data warehouse, but also a digital twin knowledge base with self-evolution capabilities. It can automatically absorb new data and iterate the association logic, providing timely knowledge support for subsequent model matching and life prediction.

[0038] Step S102: Scan the visual image information of the target mold.

[0039] Step S103: Using the visual image information, the production conditions of the target mold, and the mold material information as index conditions, and combining the knowledge graph, a machine learning model that matches the target mold is selected from the candidate model library.

[0040] Step S104: Using the visual image information, the production conditions of the target mold, the mold material information, and the historical operation characteristics matching the target mold in the knowledge graph, a machine learning model for matching the target mold is trained to obtain a data-driven model of the target mold.

[0041] In an embodiment of the present application, the data-driven model is used to predict the service life and damage of the target mold.

[0042] During the target mold production process, in order to accurately predict and optimize mold performance, it is necessary to select and train appropriate machine learning models based on multi-dimensional information.

[0043] In step S102, the visual image information of the target mold is scanned. For example, for a stamping mold, its structure is usually more complex, and there are multiple key parts, such as punches and dies. When scanning, a high-precision industrial camera can be used to set the appropriate shooting angle and resolution. For example, for a stamping mold of an automobile cover, an industrial camera can be installed above the mold, and a multi-angle shooting method is used to first shoot the overall image from directly above to obtain the general shape and outline of the mold. Then shoot from the side to capture the details of the mold edge, such as the wear of the cutting edge, scratches on the surface, etc. In this way, the visual image information of the stamping mold is fully scanned to provide clear and detailed image data for subsequent analysis.

[0044] In another example, a large steel structure mold production workshop performed visual image scanning on a steel structure mold used in heavy machinery manufacturing. First, a high-precision industrial camera was used to take a preliminary photograph of the mold. The camera was mounted on an adjustable bracket, and the camera's position and angle were adjusted to capture the mold from all angles. For example, the mold's overall appearance was captured from directly above to obtain the mold's general outline and the relative positions of its components. Lateral images were taken at different angles to capture detailed surface features such as weld shapes and surface wear. Critical areas of the mold, such as the steel beams and connecting components that bear the primary load, were scanned at a microscopic level using a high-resolution microscope. The microscope's probe was aimed at these critical areas to perform a detailed scan of the surface microstructure, recording information such as tiny cracks and wear spots. For example, microscopic scanning can reveal microscopic crack propagation in stress-concentrated areas of the steel beam, providing critical information for assessing the mold's fatigue life. Simultaneously, a 3D scanner was used to perform a comprehensive 3D image scan of the mold. A 3D scanner emits a laser beam to obtain the three-dimensional coordinate information of the mold surface and construct a 3D model of the mold. During the scanning process, the scanner generates point cloud data, which is processed to produce a precise 3D image of the mold. For example, 3D scanning can accurately measure the size, shape, and relative position of mold components, and detect any dimensional or shape errors during the manufacturing process.

[0045] Furthermore, a thermal imager can be used to scan the mold's temperature distribution during operation, generating thermal images of the mold. Thermal imagers can detect temperature differences on the mold surface, revealing the distribution of thermal stress during operation. For example, in welded areas of a mold, thermal imagers can identify areas of abnormal temperature, allowing assessment of weld quality and the extent of the heat-affected zone.

[0046] By using these various devices to perform visual image scanning of steel structure molds, comprehensive and accurate mold image information can be obtained. For example, this image information can be combined with machine learning algorithms to predict mold wear and potential failures, allowing for the development of appropriate maintenance plans to ensure proper mold operation and product quality.

[0047] Specifically, in the above step S103, a matching machine learning model can be selected based on the index condition.

[0048] First, the visual image information, the production conditions of the target mold, and the mold material information are used as core index conditions. Visual image information includes intuitive features such as the mold's appearance shape, surface texture, and geometric dimensions, which can be collected and acquired through industrial cameras, 3D scanners and other equipment. These image information can reflect the structural form and processing quality of the mold. The production conditions of the target mold cover actual production parameters such as stamping frequency, injection temperature, and working pressure. These parameters directly affect the operating status and performance of the mold. Mold material information includes the type of material (such as alloy steel, carbon steel, etc.), mechanical properties (elastic modulus, yield strength, etc.), and thermophysical properties (thermal conductivity, linear expansion coefficient, etc.). The material properties play a decisive role in the service life and bearing capacity of the mold.

[0049] The knowledge graph stores a wealth of knowledge about molds, including the structural characteristics, performance parameters, and applicable scenarios of different types of molds, as well as application cases and effect evaluations of various machine learning models in the mold field. Taking visual image information as an example, the image recognition technology in the knowledge graph can be used to perform feature matching between the visual image of the target mold and the mold images already in the knowledge graph, screening out molds with similar structural morphology and their corresponding machine learning models as candidates. For the production conditions and mold material information of the target mold, the semantic retrieval function of the knowledge graph is used to find machine learning models that perform well under similar production conditions and material properties. By comprehensively considering these three types of index conditions, machine learning models with a high degree of match with the target mold are preliminarily screened out from the candidate model library, narrowing the range of model selection and improving the efficiency and pertinence of subsequent model training.

[0050] Specifically, in the mold manufacturing sector, leveraging machine learning to improve the accuracy of mold performance predictions relies on precisely selecting the appropriate model. This process involves clarifying indexing criteria and using knowledge graphs to screen candidate models, which are discussed in detail below.

[0051] Visual image information, as a visual representation of the target mold, carries a wealth of detail. The target mold's external shape encompasses not only its overall outline but also complex curved structures, concave and convex features, and other details. For example, automotive panel molds are characterized by large dimensions and complex curved surfaces, and their external shape directly impacts the mold quality of the automotive parts. High-definition imaging of the mold surface using industrial cameras can capture micron-level surface texture defects such as scratches and pits. These subtle flaws can cause stress concentration during mold use, impacting mold life. 3D scanners, on the other hand, construct a 3D geometric model of the mold and precisely measure dimensional parameters of each component. For example, the cavity dimensional accuracy of an injection mold directly determines the dimensional accuracy of the plastic product. The acquired visual image information undergoes preprocessing, including image enhancement, noise reduction, and feature extraction. Edge detection algorithms are used to identify the mold's contour boundaries, and texture analysis algorithms are used to extract surface texture features, providing accurate image feature vectors for subsequent data matching with knowledge graphs.

[0052] Production conditions are environmental variables that affect the target mold's operation and have a direct and dynamic impact on mold performance. The stamping frequency determines the mold's workload. High-frequency stamping subjects the mold to frequent impact loads, accelerating mold wear. For example, during continuous high-speed stamping, repeated impacts can cause fatigue cracks at the contact point between the punch and the die. Injection temperature not only affects the plastic's fluidity but also alters the mold's thermal stress distribution. High-temperature injection molding causes thermal expansion of the mold material. Improper temperature control can cause mold deformation and compromise product quality. Operating pressure is a key indicator of a mold's load-bearing capacity. Die-casting molds operate under high pressure, subjecting them to immense internal pressure. Uneven pressure distribution can cause localized overload in the mold. Real-time collection of these production parameters, combined with dynamic data monitoring through a sensor network, allows for time-series alignment of production condition data with mold operating status data, providing an accurate data foundation for analyzing the impact of production conditions on mold performance.

[0053] The target mold material is the material foundation that determines mold performance. Different types of materials have unique performance characteristics. Alloy steel offers high strength, high wear resistance, and good toughness, making it suitable for stamping molds that withstand heavy loads. Carbon steel, while less expensive, has relatively weaker overall performance and is often used for simpler molds with lower performance requirements. Mechanical properties of a material, such as the elastic modulus, determine the degree of elastic deformation of the mold under load, while yield strength is a key indicator for determining whether the mold has undergone plastic deformation. Thermal conductivity, a thermophysical property, influences the mold's heat dissipation efficiency. In injection molds, good thermal conductivity facilitates rapid cooling of plastic products, improving production efficiency. The coefficient of linear expansion affects the mold's dimensional stability during temperature fluctuations. Controlling the material's coefficient of linear expansion is particularly important in precision mold manufacturing. Through material composition analysis, mechanical property testing, and thermophysical property testing, detailed mold material information is obtained and a material performance database is established to facilitate comparison with material data in the knowledge graph.

[0054] The knowledge graph graphically integrates the vast amount of knowledge in the mold field. It contains multiple types of nodes, such as mold type nodes (stamping molds, injection molds, die-casting molds, etc.), structural feature nodes (single-process molds, compound molds, progressive molds, etc.), performance parameter nodes (mold life, molding accuracy, production efficiency, etc.), and machine learning model nodes (neural network models, decision tree models, support vector machine models, etc.). Nodes are connected by directed edges, forming a complex knowledge network. For example, the "stamping mold" node is connected to the "single-process mold" node through the "structural type" edge, and the "neural network model" node is connected to the "predicted mold life" node through the "application scenario" edge. The knowledge graph draws data from a wide range of sources, including industry standards and specifications, academic research results, and corporate production experience. Through data cleaning, knowledge extraction, knowledge fusion and other technologies, the scattered knowledge is structured and stored in the graph database, providing rich knowledge resources for model screening.

[0055] When using image recognition technology from knowledge graphs to extract features from visual images of target molds, novel models can more effectively capture mold image features. For example, by using image recognition technology from knowledge graphs to extract features from visual images of target molds, a Transformer-based visual model can be pre-trained on a large-scale mold image dataset. The Transformer model is capable of capturing long-range dependencies in mold images. Compared to traditional convolutional neural networks, its self-attention mechanism allows it to more effectively learn universal feature representations of mold images. When the image of the target mold is input into a trained Transformer-based model, the model extracts high-level semantic features from the image, such as the mold's shape. The Transformer model can understand the mold's shape structure from a global perspective. For texture features, it can capture subtle texture information on the mold surface. Regarding structural features, the Transformer model can analyze the structural relationships between different parts of the mold.

[0056] Furthermore, a capsule network can be combined to further process image features. Capsule networks can model individual objects and their spatial relationships in mold images, accurately representing the mold's structure and shape through interactions between capsules. Capsule networks organize image features into capsules with a specific hierarchy, with each capsule representing a mold part or feature. A dynamic routing algorithm between capsules enables more accurate identification of mold features. The knowledge graph is then searched for mold image nodes with similar features. The features extracted by this model can more accurately match mold images with similar features in the knowledge graph. For example, if the target mold is an injection mold with complex curved surfaces, the features extracted by the Transformer and capsule network can be used to identify examples of injection molds with similar surfaces in the knowledge graph. The Transformer and capsule network-based models used in these examples can more accurately predict mold wear, providing stronger support for candidate model selection. This novel model combination provides a more comprehensive understanding of mold image features, improving the accuracy and applicability of model selection, and providing a more reliable basis for subsequent mold analysis and prediction.

[0057] The semantic search function of the knowledge graph is used to screen the production conditions and mold material information of the target mold. The production condition parameters and material performance parameters are converted into semantic vectors, and these parameters are semantically understood and encoded using natural language processing technology. The knowledge graph is searched for case nodes with similar production conditions and material properties to the target mold, and machine learning models successfully applied in these cases are obtained. For example, if the production conditions of the target mold are high stamping frequency, high-temperature injection molding environment, and the material is high-strength alloy steel, the knowledge graph is searched for mold cases with similar production conditions and material properties. If the random forest model is used in these cases to predict mold failures effectively, the random forest model can be used as a candidate model. By comprehensively considering the three types of index conditions: visual image information, production conditions, and material information, the candidate models are cross-validated and prioritized, and the machine learning model that best suits the target mold is ultimately determined, laying a solid foundation for subsequent data-driven model training.

[0058] In step S104, visual image information, production conditions, and mold material information of the target mold are collected. Historical operational features matching the target mold, such as historical vibration data, stress curves, and fault records, are extracted from the knowledge graph. This data is integrated to construct a training dataset encompassing multiple feature dimensions. For example, data such as mold geometry, stamping frequency, and material yield strength are associated with corresponding labeled data such as mold life and wear level to form structured training samples.

[0059] The selected machine learning model is trained using the integrated training dataset. During training, the model's algorithm learns from the data, exploring the potential relationships between various features and mold performance indicators. For example, for a model predicting mold life, the various features in the training data are learned to establish a mathematical model linking these features with mold life. Simultaneously, the model is optimized using methods such as cross-validation and regularization to prevent overfitting or underfitting, thereby improving its generalization and predictive accuracy. During training, model parameters, such as the number of neural network layers and nodes and the depth of the decision tree, are continuously adjusted to achieve optimal training results. Through multiple rounds of iterative training, the model accurately predicts key performance indicators such as the mold's operating status and lifespan based on the input mold feature information, ultimately resulting in a data-driven model suitable for the target mold.

[0060] It's important to note that, taking graph neural networks as an example, steel structure molds are complex structures with physical connections and mechanical transmission relationships between their components. These relationships can naturally be represented using a graph structure. Each component can be considered a node in the graph, and the connections between components are edges. Graph neural networks can perform information propagation and feature learning on this graph structure, effectively capturing the interactions and dependencies between the various components of steel structure molds. For example, the stress state of a steel beam node is not only related to its own properties but also influenced by other connected beams and connectors. GNNs can learn about these influences through message passing mechanisms.

[0061] Take the Deep Reinforcement Learning (DRL) model as an example. DRL combines deep learning and reinforcement learning methods and is suitable for scenarios requiring dynamic decision-making. For steel structure molds, the mold's operating status can be considered the environment, and maintenance decisions (such as whether to replace a component and when to perform maintenance) can be considered actions. By interacting with the environment, the model continuously learns the optimal decision-making strategy to maximize long-term rewards (such as extending mold life and reducing maintenance costs).

[0062] During training, a state space can be defined, including the mold's visual image features, production conditions, material information, and historical operating characteristics; an action space can also be defined, such as different maintenance operations. A reward function can be defined, such as providing rewards based on extending the mold's lifespan or reducing failures. Deep reinforcement learning algorithms such as Deep Q-Networks (DQN) and Proximal Policy Optimization (PPO) can be employed. During training, the agent (model) continuously attempts different actions in the environment and updates the parameters of the policy network based on reward feedback. Through extensive interaction and training, the model is able to make optimal maintenance decisions based on the current mold state, thereby indirectly predicting and optimizing the mold's operating status and lifespan.

[0063] Step S105 : collecting real-time operating data of the target mold and inputting it into the data-driven model to obtain a predicted service life value and a predicted damage state of the target mold.

[0064] As an optional example, in step S105, stress cycle features, vibration features, heat conduction characteristics, and real-time operating condition features are extracted from the real-time operation data through the feature extraction layer; the extracted feature data are pre-calculated with the steel structure mold failure mode model in the knowledge graph to obtain the first remaining service life probability distribution of each component in the target mold; the extracted feature data are adapted to each prediction module in the data-driven model, and input into the prediction module that matches the change characteristics of each feature data for change trend prediction to obtain the change trend information of each feature data; the change trend information is mapped to each component of the target mold and the connection between each component to correct the first remaining service life probability distribution and obtain the second remaining service life probability distribution of each component; based on the second remaining service life probability distribution, the dynamic service life prediction value of each component in the target mold that changes with time and the corresponding predicted damage state are marked.

[0065] For example, in the mold life prediction and maintenance decision-making system, step S105 achieves accurate prediction of the service life and damage status of the target mold through a series of data processing and model calculations.

[0066] During the operation of the target mold, its key components are subjected to periodic stress, and the stress cycle characteristics are closely related to mold fatigue damage. Based on Miner's linear fatigue cumulative damage theory, by analyzing real-time stress data, characteristics such as stress amplitude, stress ratio, and number of cycles are extracted. The stress amplitude reflects the range of stress changes in each stress cycle, the stress ratio reflects the proportional relationship between tensile stress and compressive stress, and the number of cycles records the cumulative number of stress cycles. These characteristics can quantify the degree of damage to the mold under fatigue loads and provide an important basis for predicting the life of the mold. For example, in a stamping mold, the punch is repeatedly subjected to impact loads. By extracting the stress cycle characteristics, its fatigue damage process can be determined.

[0067] It's understandable that the vibration signal of a mold during operation contains a wealth of information about its operating status. Using time-frequency analysis methods such as fast Fourier transform (FFT) or wavelet transform, the time-domain vibration signal is converted to the frequency domain, extracting characteristics such as the dominant frequency and frequency band energy. When the mold is operating normally, its vibration signal exhibits specific frequency components and energy distribution patterns. However, when the mold exhibits faults such as wear or looseness, the vibration frequency and energy distribution change. By analyzing these changes in vibration characteristics, the health of the mold's internal structure can be determined. For example, when the ejector pin of an injection mold is worn, the high-frequency components of the vibration signal increase, and the energy distribution also changes accordingly.

[0068] Thermal conductivity characteristics are primarily used to reflect changes in the mold's thermal state during thermal processing. By analyzing mold temperature data, characteristics such as temperature gradient and heat flux density are calculated. In processes such as injection molding and die-casting, mold temperature uniformity has a significant impact on product quality and mold life. The temperature gradient reflects the temperature difference between different parts of the mold, while the heat flux density represents the amount of heat transferred through a unit area per unit time. When the mold's local temperature is too high or the heat flux is unevenly distributed, the mold material's performance degrades, accelerating thermal fatigue damage to the mold. Extracting thermal conductivity characteristics helps to promptly detect thermal anomalies in the mold and prevent mold failure due to thermal factors.

[0069] Real-time working condition features comprehensively consider the mold's operating environment and operating conditions, such as operating hours, workpiece weight, and stamping frequency. These features are closely related to the mold's actual usage and directly affect the mold's wear and fatigue progression. For example, prolonged continuous operation or processing heavy workpieces increases the load on the mold, accelerating mold wear. Excessive stamping frequency causes the mold to experience more frequent impacts, increasing the risk of fatigue damage. Extracting real-time working condition features integrates mold operating data with actual operating scenarios, making predictions more accurate.

[0070] The steel structure mold failure mode model in the knowledge graph is constructed based on a large amount of historical data and industry experience, and includes the correlation between various failure modes and feature data. The extracted feature data is pre-calculated with the failure mode model. The principle is to evaluate the possibility of different failure modes in each component of the target mold through pattern matching and data analysis. For example, if the stress cycle characteristics of a certain mold component match the characteristics of fatigue crack generation in the failure mode model, it can be preliminarily determined that the component has a high risk of fatigue cracking. Through this pre-calculation, the first remaining service life probability distribution of each component in the target mold can be obtained, and a preliminary prediction of the life of each mold component can be made from a macro perspective, providing a basis for subsequent in-depth analysis.

[0071] The data-driven model comprises multiple prediction modules tailored to different features and prediction objectives. Each module employs a distinct algorithm and structure, suited to processing different types of feature data and their varying characteristics. The extracted feature data is adapted to the prediction module, selecting the most appropriate prediction module based on the data's characteristics (such as time series characteristics and numerical variation patterns). For example, for stress cycle and vibration data with distinct time series characteristics, time series prediction models such as long short-term memory (LSTM) networks or gated recurrent units (GRU) are employed. For structured real-time operating data, models such as random forests or support vector machines are used. The adapted prediction module predicts the changing trends of the feature data, leveraging the patterns learned from historical data to predict future trends. For example, the growth trend of the stress amplitude at a specific part of the mold over the next several operating cycles or the change in vibration frequency can be predicted. This trend information reflects the dynamic evolution of the mold's operating state and provides key information for more accurate prediction of mold life and damage status.

[0072] Furthermore, the changing trend information of the feature data is mapped to the various components of the target mold and their connections. This principle is based on the mechanical transmission relationships and failure mechanisms of the mold structure. Mold components are interconnected, and changes in the state of one component can affect connected components. For example, increased vibration of a mold component can cause its connections to bear greater dynamic loads, accelerating wear. By mapping this changing trend information to the mold structure, the mutual influence between components can be analyzed. The impact of multiple factors on mold life can be comprehensively considered, thereby revising the first remaining useful life probability distribution to obtain a more accurate second remaining useful life probability distribution. This revision considers the integrity of the mold structure and the inter-component connections, making the prediction results more realistic and enabling more accurate predictions of the remaining useful life of each mold component. Based on the second remaining useful life probability distribution, the dynamic life prediction values ​​and corresponding predicted damage states of each component in the target mold are annotated over time, transforming the probability distribution into a concrete and visual prediction result. By setting appropriate thresholds, the probability distribution is converted into a specific predicted useful life time and damage severity level. For example, when the probability of a component's remaining service life is above 90%, its remaining service life is marked as long and its damage status is slight; when the probability is less than 10%, its remaining service life is marked as short and its damage status is severe.

[0073] This dynamic annotation intuitively presents the health status of each mold component, providing strong support for mold maintenance decisions. Based on the annotation results, staff can plan mold maintenance plans in advance, focusing on monitoring or preemptively replacing components with short lifespans and high damage risks to avoid production interruptions due to mold failures. Furthermore, the prediction results are fed back into the mold design process, helping to optimize mold structure design, improve mold reliability and service life, reduce production costs, and increase production efficiency.

[0074] In summary, step S105 achieves efficient prediction of the service life and damage status of the target mold through the data-driven model, which has important application value in intelligent management and maintenance of molds.

[0075] Step S106: Execute the corresponding mold maintenance strategy based on the service life prediction value and the predicted damage status. In step S106, the service life prediction value and the predicted damage status obtained by the model prediction are visualized in an intuitive way. For example, through tools such as charts (such as line charts, bar charts, heat maps), dashboards, etc., the trend of changes in the remaining service life of the mold and the distribution of damage status of each part are presented. For example, a heat map is used to display the damage probability distribution of the mold surface, and the darker the color, the higher the probability of damage; a line chart is used to track the changes in the mold service life prediction value over time, so that the staff can intuitively understand the health status of the mold.

[0076] Optionally, the prediction results can be further analyzed in depth, and a corresponding maintenance strategy can be developed based on the actual usage and production plan of the target mold. If the predicted mold lifespan falls below a set threshold, mold repair or replacement can be scheduled in advance to avoid production interruptions caused by mold failure. For areas with high predicted damage, monitoring frequency can be increased, and local repairs or reinforcement treatments can be performed as necessary. Furthermore, the prediction results can be fed back into the mold design process, providing data support for subsequent mold optimization and improving mold reliability and service life.

[0077] In the embodiments of the present application, a series of operations such as constructing a knowledge graph based on multi-source data, selecting a model based on multiple information, customizing the training data-driven model, real-time data input prediction, and scientifically executing maintenance strategies are implemented to achieve comprehensive and accurate life prediction and damage monitoring of steel structure molds. This improves the scientificity and rationality of mold maintenance decisions, optimizes the allocation of maintenance resources, reduces the impact of mold failures on production, and improves production efficiency and product quality. At the same time, the application of knowledge graphs and data-driven models makes full use of historical data and prior knowledge, improves the generalization ability and adaptability of the model, and provides a more effective technical means for the management of steel structure molds.

[0078] The present application may further optionally train a machine learning model for target mold matching using the visual image information, the production conditions of the target mold, the mold material information, and the historical operating characteristics matching the target mold in the knowledge graph to obtain a data-driven model for the target mold. Furthermore, the physical quantities of the target mold may be pre-calculated based on the low-frequency modes calculated by finite element analysis, and the physical quantities may be introduced into the data-driven model as physical constraints to correct prediction errors in the data-driven model caused by differences between the target mold and the standard part. The physical quantities may include at least the natural vibration frequency, stress distribution, and structural strength of components of different shapes of different molds.

[0079] Specifically, after the mold data-driven model is constructed, physical constraints based on finite element analysis are introduced to further improve model prediction accuracy and compensate for prediction errors caused by differences between the target mold and the standard part. First, the target mold is modeled. A precise mesh is created based on the target mold's actual geometric dimensions, assembly relationships, and material properties to ensure that the model accurately reflects the mold's structural characteristics. For example, for complex mold shapes, a highly adaptable tetrahedral mesh is used for fine meshing, while for regular shapes, a hexahedral mesh is used to ensure computational accuracy while improving efficiency. After modeling is complete, boundary conditions and load cases are set. Boundary conditions simulate the actual mold installation and fixation methods in production, such as fixed constraints and hinge constraints. Load cases are set based on the production conditions of the target mold, including stamping pressure, injection pressure, and temperature loads. Based on this, low-frequency modal analysis is performed. Low-frequency modal analysis can capture the mold's natural vibration characteristics in the low-frequency range. This modal information is crucial for understanding the mold's dynamic response and structural stability. Through calculation, the mold's low-frequency modal parameters, such as each order's natural frequency and mode shape, are obtained. These parameters will serve as the basis for subsequent precalculation of physical quantities. Based on the low-frequency modes obtained by finite element analysis, the key physical quantities of the target mold are further precalculated. Based on the results of low-frequency modal analysis, the mold's natural vibration frequency under different operating conditions is determined. In actual operation, if the external excitation frequency is close to the mold's natural vibration frequency, it may cause resonance and cause damage to the mold. Therefore, accurately calculating the natural vibration frequency can provide an important reference for subsequent monitoring of the mold's operating status. By calculating the mold's natural vibration frequency under different assembly states and different material properties, a natural vibration frequency database is constructed to facilitate rapid query and analysis.

[0080] The finite element method is used to calculate the stress distribution of the mold during operation, combining the load conditions and material properties of the mold. The magnitude and direction of stress in different parts of the mold and at different times are analyzed to identify areas of stress concentration. For example, stress concentration is often prone to occur at corners and joints of the mold. By accurately calculating the stress distribution, measures can be taken in advance to optimize or strengthen the structure to prevent mold failure due to excessive stress. For components of different shapes in the mold, such as thin plates, beams, and columns, the structural strength is calculated using the corresponding strength theory based on their geometric shape and material mechanical properties. The load-bearing capacity and safety factor of the components are evaluated by considering the tensile, compression, bending, and torsion stress conditions of the components under complex loads. For example, for the mold's support beam, by calculating its bending stress and deformation under maximum load, it is determined whether it meets the design requirements, providing data support for mold structure design and improvement.

[0081] Finally, the precalculated physical quantities are introduced as physical constraints into the data-driven model. Physical constraints serve as both limiting and guiding constraints during the training and prediction process of the data-driven model. For example, when predicting mold vibration, the natural vibration frequency is used as a constraint. If the model's predicted vibration frequency deviates from the natural vibration frequency range, the model output is adjusted to better align with physical reality.

[0082] For stress distribution and structural strength constraints, when the model predicts the stress and deformation of mold components, if the predicted results exceed the reasonable range calculated through finite element analysis, the model parameters are corrected according to the physical constraints. Through continuous iterative training, the data-driven model gradually corrects the prediction errors caused by the differences between the target mold and the standard part while considering the physical constraints. For example, due to the difference in material properties between the target mold and the standard part, the model's predicted stress and deformation may be inaccurate. After introducing physical constraints, the model can self-adjust according to the actual stress distribution and structural strength, improving the accuracy of the prediction.

[0083] By introducing the physical quantities calculated by finite element analysis into the data-driven model as physical constraints, the model can be effectively modified, enabling the model to better adapt to the characteristics of the target mold, improving the accuracy of mold status prediction and life assessment, and providing a more reliable basis for mold optimization design, fault warning and maintenance decision-making.

[0084] Further optionally, in the above steps, based on the low-frequency modes calculated by finite element analysis, the physical quantities of the target mold are pre-calculated, and the physical quantities are introduced into the data-driven model as physical constraints, including: determining the shape type of each component in the target mold according to the geometric size, assembly relationship and material properties of the target mold; wherein the shape type includes a regular shape matching a single component in the standard part, and an irregular shape composed of a combination of multiple components in the standard part; for components with regular shapes, hexahedral meshes are used for division; for components with irregular shapes, tetrahedral meshes are used for division to obtain a grid structure model of the target mold; boundary conditions and load conditions of the grid structure model are set; and low-frequency modal calculation is performed on the set grid structure model. The inherent characteristic parameters of the target mold in the low-frequency range are obtained as the physical quantities; when predicting the vibration of the target mold, the inherent vibration frequency is used as a constraint condition. When the vibration frequency predicted by the data-driven model deviates from the inherent vibration frequency range, the corresponding model parameters and / or model structure in the data-driven model are adjusted to make the prediction results of the data-driven model more consistent with the actual physical situation; when predicting the stress distribution, deformation and / or structural strength of the target mold components, the stress distribution of the target mold components and the structural strength of components with different shapes are used as constraints. If the prediction results of the data-driven model exceed the reasonable range obtained by finite element analysis, through continuous iterative training, the data-driven model is gradually corrected while considering the physical constraints.

[0085] This approach, based on the integration of finite element analysis (FEM) and data-driven models, aims to improve the accuracy of target mold performance predictions. First, component shapes are classified based on mold geometry, assembly relationships, and material properties. A mesh structure model is constructed using a hexahedral mesh for regular shapes and a tetrahedral mesh for irregular shapes to ensure the model closely matches the actual mold structure. By setting boundary conditions and load conditions, low-frequency modal calculations are performed on the model to obtain intrinsic characteristic parameters. These parameters reflect the mold's inherent physical properties, such as natural vibration frequency and stress distribution patterns. These physical quantities are introduced as constraints into the data-driven model, leveraging the principle of constraining the model output with physical laws. When predicting mold vibration, the natural vibration frequency serves as a reference. If the model's predicted value deviates, the model parameters or structure are adjusted to ensure that the results conform to physical laws and avoid predictions that violate actual conditions. When predicting component stress distribution, deformation, and structural strength, the FEM calculation's reasonable range is used as a constraint. If the model prediction exceeds this range, iterative training corrects the error, allowing the model to learn the correct prediction pattern within the physical constraints. In this way, the method effectively combines the physical accuracy of finite element analysis and the flexibility of data-driven models, avoiding the prediction deviation caused by the lack of physical constraints of data-driven models, improving the reliability and practicality of the prediction results, and providing a more scientific basis for mold design optimization, fault warning and maintenance decisions, reducing production risks, and improving production efficiency and product quality.

[0086] In an optional example, in the above steps, performing low-frequency modal calculation on the set grid structure model to obtain inherent characteristic parameters of the target mold in the low-frequency range as the physical quantity includes:

[0087] In combination with the load conditions and material properties of the target mold, and in combination with the theories of elasticity and plasticity, a finite element solver is used to calculate a first stress cloud map during the operation of the target mold; the first stress cloud map includes at least: the stress magnitude and direction of different components of the target mold at various times; the stress concentration coefficient of each component in the first stress cloud map is calculated using a formula of elasticity theory to identify stress concentration areas in the target mold; the stress concentration areas include at least: the corners of the target mold and the connections between different components; the stress concentration areas are locally meshed, and the historical damage positions of the standard parts matched by the target mold are used as load excitations. Force simulation calculations are performed through the multi-physical field coupling effect to obtain a second stress cloud map with a smaller distribution mesh granularity; the second stress cloud map is merged into the first stress cloud map to obtain the stress distribution of the target mold.

[0088] Specifically, in mold design and operational evaluation, accurately determining the target mold's stress distribution is crucial for ensuring its performance and service life. Combining the target mold's load conditions and material properties with multi-stage finite element method calculations allows for in-depth analysis of the mold's stress state. The following details the techniques and methods employed in each calculation step.

[0089] First, comprehensively collect information on the load conditions to which the target mold is subjected, including static loads (such as gravity and preload), dynamic loads (such as stamping impact and vibration loads), and thermal loads (such as thermal stress generated by temperature changes during the injection molding process). Determine the magnitude, direction, and duration of the load through real-time monitoring by sensors or according to production process parameter settings. For material properties, retrieve parameters such as the elastic modulus, Poisson's ratio, yield strength, and thermal expansion coefficient of the target mold material from the material performance database. If new materials are involved, accurate data can be obtained through material mechanics experiments such as tensile tests and compression tests. Use finite element analysis software to import the three-dimensional geometric model of the target mold into the software. Select an appropriate meshing strategy based on the structural characteristics of the mold and the analysis accuracy requirements. For areas with complex structures and large stress gradients (such as fillets and thin-walled areas of the mold), use a refined mesh; for areas with relatively simple structures, use a coarser mesh to improve calculation efficiency while ensuring calculation accuracy.

[0090] After meshing is complete, boundary conditions and load conditions are applied. For example, for fixed mold components, fixed constraint boundary conditions are set; for mold surfaces subject to pressure, corresponding pressure loads are applied. The finite element software's solver, based on theories of elasticity and plasticity, calculates the stress magnitude and direction of different components at various times during the target mold's operation, thereby generating a first stress contour map. During the calculation process, the software's post-processing capabilities can be used to visually display the stress distribution in visual forms such as contour maps and vector diagrams, facilitating analysis and understanding.

[0091] The stress cloud map generated by finite element analysis can visually demonstrate the stress distribution trend across the mold. Observing areas with darker colors and larger values ​​in the stress cloud map provides a preliminary indication of potential stress concentration areas. Due to sudden geometric changes at mold corners and joints between different components, stress lines often converge, often creating high-risk areas for stress concentration. Using the measurement tools in finite element software, we can obtain stress values ​​in these areas and compare them with stress values ​​in other parts of the mold to further identify areas of stress concentration.

[0092] In addition to visual observation, numerical analysis methods can also be used, such as calculating the stress concentration factor. For common geometric shapes (such as circular holes and fillets), the stress concentration factor can be calculated based on the theoretical formula of elastic mechanics to assess the degree of stress concentration. Simultaneously, incorporating the Saint-Venant principle in material mechanics, the impact range of the stress concentration area can be analyzed to determine its impact on the overall performance of the mold. Furthermore, reference can be made to previous design experience and experimental data from similar molds to verify the rationality of the identified stress concentration area.

[0093] Collect historical damage location data for the target mold's matching standard parts. This data can come from the company's maintenance records, failure analysis reports, and other sources. This historical damage location data is integrated with the primary stress contour map. Preprocessing operations such as coordinate transformation and data alignment align the two to the same coordinate system and time scale. Furthermore, based on the mold's actual operating conditions and material properties, the data is screened and cleaned to remove outliers and noise, ensuring data accuracy and reliability.

[0094] The stress concentration areas identified in the first stress cloud map are the focus of research, and local mesh refinement is performed in these areas in the finite element software. More sophisticated unit types (such as high-order units) are used to improve calculation accuracy. The historical damage positions of the standard parts are used as references for boundary conditions or load excitations to perform secondary finite element calculations. During the calculation process, the nonlinear characteristics of the material (such as plastic deformation, fatigue damage) and the multi-physics field coupling effects (such as thermal-structural coupling and fluid-solid coupling) are taken into account to more realistically simulate the complex stress conditions of the mold in actual operation, thereby obtaining a second stress cloud map with a smaller distribution granularity.

[0095] Finally, a data fusion algorithm (such as weighted averaging, Kalman filtering, or DS evidence theory) is used to fuse the second stress contour map with the first. The weighted averaging method assigns different weights to the two stress contour maps based on their accuracy and reliability, and then performs a weighted summation. The Kalman filtering method uses a state-space model to optimally estimate the two stress contour maps. DS evidence theory utilizes evidence theory to address uncertainty and integrate the information from the two stress contour maps. When selecting a fusion algorithm, it is important to consider the actual mold conditions and data characteristics to ensure that the fusion result accurately reflects the mold's true stress distribution.

[0096] The fused stress distribution data is visualized through the finite element software's post-processing module, generating a final stress distribution cloud map and data report. The fusion results can be verified through experimental testing, such as attaching strain gauges to the mold and performing photoelastic experiments to measure the mold's actual stress distribution. This is then compared and analyzed with the fused calculation results. Based on these comparisons, the calculation model and fusion algorithm are adjusted and optimized to ensure the accuracy and reliability of the target mold's stress distribution calculations. This enables a systematic and comprehensive calculation of the target mold's stress distribution, providing a crucial basis for mold design improvements, fault prediction, and maintenance decisions, effectively improving mold performance and service life.

[0097] In another example, in the above steps, low-frequency modal calculation is performed on the set grid structure model to obtain the inherent characteristic parameters of the target mold in the low-frequency range as the physical quantity, which can also be implemented as follows:

[0098] For components of different shapes in the target mold, the corresponding structural strength is calculated based on the geometric shape and material mechanical properties of each component in combination with the corresponding strength theory; the force type of each component is divided, and the structural strength of each component is preliminarily checked based on the force type; among them, for components subjected to unidirectional tensile loads, the first strength theory is used for strength verification; for components under complex stress states, the fourth strength theory is used for strength verification; based on the tensile, compression, bending, and torsional stress conditions of each component under complex loads, the maximum load-bearing upper limit and safety factor of each component are calculated; the load-bearing components in the target mold are identified, and with the load-bearing components as the center, the maximum load-bearing upper limit and safety factor are used to make correlation corrections to the structural strength of other components; the topology optimization method is used to calculate the stress distribution of each component under different load conditions through finite element analysis, and the components with concentrated stress and bearing the main load are identified as load-bearing components; the joint load-bearing range after the combination of each component is evaluated, and the structural strength of each component in the target mold is corrected based on the said joint load-bearing range to avoid structural strength mismatch between connected components.

[0099] Specifically, during mold design and performance evaluation, low-frequency modal calculations and structural strength analysis of the target mold are critical steps in ensuring mold reliability and service life. Accurately model the various components within the target mold. Through 3D modeling, the complex geometry of the component is digitally represented, including detailed features such as curved surfaces, holes, and chamfers. This generates a highly accurate geometric model, providing an accurate geometric foundation for subsequent mechanical analysis.

[0100] Establish a material properties database to store the mechanical properties of various mold materials, such as elastic modulus, Poisson's ratio, yield strength, and tensile strength. For new or special materials, obtain their mechanical properties data through experimental testing. Experimental methods include tensile testing, compression testing, bending testing, and torsion testing. These tests can accurately determine the mechanical response of materials under different stress conditions. When performing structural strength calculations, retrieve the corresponding material performance parameters from the database to ensure calculation accuracy.

[0101] Based on the actual stress conditions of the component, select an appropriate strength theory for calculation. Common strength theories include the first strength theory (maximum tensile stress theory), the second strength theory (maximum elongation line strain theory), the third strength theory (maximum shear stress theory), and the fourth strength theory (shape change specific energy theory). For example, for components subjected to unidirectional tensile loads, the first strength theory can be used for strength verification; for components under complex stress states, the fourth strength theory can more accurately reflect the material's failure. By comparing the material's allowable stress with the calculated component stress, it is determined whether the component's structural strength meets the requirements.

[0102] Use finite element analysis software to perform mechanical analysis on the component. After importing the component's geometric model into the software, meshing is performed, discretizing the component into multiple finite element units. Based on the component's actual operating conditions, appropriate boundary conditions and loads are applied, such as fixed constraints, displacement constraints, pressure loads, and temperature loads. Finite element calculations are used to determine the stress and strain distribution of the component under different loads. Based on the calculation results, the location and magnitude of the component's maximum stress are determined, which serves as a reference for the maximum load-bearing upper limit.

[0103] The safety factor is typically calculated based on the material's ultimate stress and the component's operating stress. Ultimate stress can be determined based on mechanical property test results, such as yield strength or tensile strength. Operating stress is obtained through finite element analysis or theoretical calculations. The safety factor is calculated as follows: Safety Factor = Ultimate Stress / Operating Stress. Based on mold requirements and industry standards, a reasonable safety factor range should be established to ensure component safety during operation.

[0104] Using topology optimization methods and finite element analysis to calculate the stress distribution of components under different load conditions, components with concentrated stress and bearing the primary load are identified as load-bearing components. Alternatively, load-bearing components can be identified from a structural and functional perspective, combining mold design principles with actual workflows. For example, in a stamping mold, the punch and die are typically the primary load-bearing components; in an injection mold, the cavity and core bear the pressure of the plastic melt and can be considered load-bearing components.

[0105] With load-bearing components at the core, a mechanical inter-component model is established. The multibody dynamics module of finite element analysis software is used to simulate the interactions and force transmission relationships between components. Based on the maximum load-bearing limit and safety factor of the load-bearing components, the structural strength of other components is adjusted. For example, by increasing the thickness of non-load-bearing components and optimizing the connection structure, their load-bearing capacity is improved, ensuring strength matching between components. Furthermore, parametric design techniques are used to easily modify component structural parameters, enabling rapid design optimization.

[0106] Using system-level finite element analysis, we treat the target mold as a whole and build an assembly model encompassing all components. We accurately simulate the connections between components within the model, such as bolts, welding, and interference fits. By applying different load conditions, we calculate the stress, strain distribution, and deformation of the entire mold, assessing the combined load-bearing capacity of each component. Based on the mold's actual operating requirements and performance indicators, we determine the appropriate combined load-bearing capacity boundaries.

[0107] Based on the assessment results of the combined load-bearing range, the structural strength of each component in the target mold is reconciled. Optimization algorithms, such as genetic algorithms and particle swarm optimization, are used to optimize the structural parameters of the components, using the combined load-bearing range as a constraint. During the optimization process, the interactions and synergies between components are considered to ensure that the revised structural strength meets the combined load-bearing requirements while achieving overall mold lightweighting and cost optimization. Furthermore, through multiple iterative calculations and simulation verification, the structural strength of the components is continuously adjusted to avoid structural strength mismatches between connected components.

[0108] By comprehensively applying the above methods, we can systematically perform low-frequency modal calculations and structural strength analysis on the target mold, providing strong support for mold design optimization, performance evaluation, and fault prevention, ensuring the stable and reliable operation of the mold in actual production.

[0109] As an optional embodiment, various mold data, such as vibration, stress, and temperature, are collected in real time during mold operation. This real-time data is combined with historical data to analyze changes in the mold's state during actual operation and promptly update the empirical formula. When production conditions change, the empirical formula can be quickly adjusted based on real-time data feedback. For example, when the stamping speed changes, the impact on the mold life can be analyzed based on real-time mold vibration and stress data, and the empirical formula can be dynamically corrected to improve its adaptability to different production conditions.

[0110] Specifically, after step S104, similar material performance data and similar component failure modes that match the target mold can also be obtained through the knowledge graph; based on the obtained similar material performance data and similar component failure modes, a damage analysis model is used to predict the periodic changes and abnormal fluctuations during the operation of the target mold to obtain a wear rate prediction value of the target mold during the operation process; the wear rate prediction value is a dynamic value that fluctuates based on changes in operating time; the wear rate prediction value is used to update the formula parameters or formula structure in the empirical formula of the target mold.

[0111] Further optionally, after using the wear rate prediction value to update the formula parameters or formula structure in the empirical formula of the target mold, the updated empirical formula can be introduced into the data-driven model as a constraint condition to correct the prediction error of the data-driven model.

[0112] Specifically, during the operation of the target mold, data such as vibration, stress, and temperature are key indicators reflecting its working status. In order to achieve real-time data collection, a variety of sensors need to be deployed in key parts of the mold. High-precision stress sensors are installed at the core stress-bearing components of the mold, such as the punch and die of the stamping mold. These sensors can capture the stress changes of the mold in the stress-bearing process in real time, accurate to every tiny stress fluctuation. For vibration monitoring of the mold, a three-axis vibration sensor is used to collect vibration data in all directions of the mold during operation, including parameters such as vibration frequency and amplitude, so as to promptly detect abnormal vibrations caused by imbalance, wear and other problems of the mold. In terms of temperature monitoring, temperature sensors are arranged in the cooling channels, friction parts and other heat-prone areas of the mold to obtain the temperature distribution on the surface and inside of the mold in real time to avoid degradation of mold material performance or product quality defects due to excessive temperature.

[0113] These sensors rapidly transmit collected real-time data to the data processing center via a high-speed data transmission network. Advanced encryption technology and anti-interference measures are used during the data transmission process to ensure data accuracy and integrity. The data processing center is equipped with a high-performance data acquisition system capable of receiving, storing, and initially processing data in real time with millisecond response speeds, providing a reliable data foundation for subsequent data analysis.

[0114] By combining real-time data with historical data and applying big data analysis and machine learning algorithms, we can deeply explore the patterns of changes in mold operating status. Historical data includes information such as mold operating data at different production stages and under different production conditions, as well as the corresponding mold lifespan, and is an important reference for analyzing mold status changes.

[0115] Using time series analysis, we analyze the temporal trends of mold vibration, stress, temperature, and other data, identifying periodic changes and abnormal fluctuations during mold operation. Using clustering algorithms from machine learning, we classify data with similar operating characteristics and identify correlations between different data categories and mold lifespan. For example, we found that when the mold vibration frequency is within a certain range and the stress exceeds a certain threshold, the mold wear rate increases significantly, significantly shortening its lifespan.

[0116] Based on the above data analysis results, the existing empirical formula was updated. Empirical formulas are typically derived from extensive experiments and production practices and are used to describe the relationship between mold operating parameters and mold life. During the update process, new influencing factors and parameters were introduced, and the coefficients and weights in the formula were adjusted to enable the empirical formula to more accurately reflect the changes in the mold's state during actual operation. For example, the original empirical formula for mold life only considered factors such as the number of stampings and mold material. Data analysis revealed that mold vibration and stress have a more significant impact on mold life. Therefore, parameters related to vibration and stress were added to the empirical formula, and the coefficients of each parameter were redefined to improve the accuracy of the empirical formula.

[0117] Real-time data feedback plays a key role when production conditions change, such as changes in stamping speed, raw material changes, or ambient temperature fluctuations. For example, when stamping speed increases, mold vibration and stress significantly change. Real-time mold vibration and stress data collected are immediately transmitted to the data analysis system. The system compares and analyzes this data with historical data and updated empirical formulas to assess the impact of stamping speed changes on mold life.

[0118] In summary, in this embodiment, when there is a deviation between the prediction results of the data-driven model and the empirical formula, the model is adjusted based on the empirical formula as a constraint to make the model prediction more consistent with the actual physical conditions and production experience, thereby improving the accuracy and reliability of the model prediction, providing more scientific and powerful support for mold production and maintenance decisions, and realizing efficient management of the entire life cycle of the mold.

[0119] See also Figure 2 , Figure 2 The embodiment of the present application provides a steel structure mold life prediction and maintenance decision system, which includes the following modules:

[0120] A construction module is used to obtain multi-source data of multiple steel structure molds using multiple steel structure molds as standard parts to construct a knowledge graph of the multiple steel structure molds; the material information, historical operation data, performance under specific process parameters, damage status, service life, mold vibration information and shape characteristics of different steel structure molds in the multi-source data are associated in the form of a knowledge graph; a selection module is used to scan the visual image information of the target mold; the visual image information, the production conditions of the target mold, and the mold material information are used as index conditions, and a machine learning model matching the target mold is selected from a candidate model library in combination with the knowledge graph; a training module is used to use the visual image information, the production conditions of the target mold, the mold material information and the historical operation characteristics matching the target mold in the knowledge graph to train the machine learning model matching the target mold to obtain a data-driven model for the target mold; the data-driven model is used to predict the service life and damage status of the target mold; a prediction module is used to collect real-time operation data of the target mold and input it into the data-driven model to obtain a predicted service life value and predicted damage status of the target mold; a maintenance module is used to execute a corresponding mold maintenance strategy based on the predicted service life value and predicted damage status. In some embodiments, the steel structure mold life prediction and maintenance decision system can be applied to a terminal device. It should be noted that for ease of description and brevity, the specific operating process of the steel structure mold life prediction and maintenance decision system described above can refer to the corresponding process in the aforementioned steel structure mold life prediction and maintenance decision method embodiment, and will not be repeated here.

[0121] See also Figure 3 , Figure 3 This is a schematic block diagram of the structure of a terminal device provided in an embodiment of the present application. Figure 3 As shown, terminal device 300 includes a processor 301 and a memory 302, which are connected via a bus 303, such as an I2C bus. Specifically, processor 301 is used to provide computing and control capabilities to support the operation of the entire terminal device. Processor 301 can be a central processing unit, or it can be other general-purpose processors, digital signal processors, application-specific integrated circuits, field programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. Specifically, memory 302 can be a Flash chip, a read-only memory disk, an optical disk, a USB flash drive, or a mobile hard disk, etc.

[0122] Those skilled in the art will understand that Figure 3The structure shown in is only a block diagram of a part of the structure related to the embodiment of the present application, and does not constitute a limitation on the terminal device to which the embodiment of the present application is applied. The specific server may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components. Among them, the processor is used to run the computer program stored in the memory, and implement any one of the steel structure mold life prediction and maintenance decision-making methods provided in the embodiment of the present application when executing the computer program. It should be noted that technical personnel in the relevant field can clearly understand that for the convenience and simplicity of description, the specific working process of the terminal device described above can refer to the aforementioned steel structure mold life prediction and maintenance decision-making method embodiment, and will not be repeated here.

Claims

1. A steel structure mold life prediction and maintenance decision-making method, characterized in that: include: Using a variety of steel structure molds as standard parts, we can obtain multi-source data of various steel structure molds to build a knowledge graph of various steel structure molds; Associating material information, historical operation data, performance under specific process parameters, damage conditions, service life, mold vibration information, and shape characteristics of different steel structure molds in the multi-source data in the form of a knowledge graph; Scanning visual image information of the target mold; using the visual image information, the production conditions of the target mold, and mold material information as index conditions, and combining the knowledge graph to select a machine learning model that matches the target mold from a candidate model library; The visual image information, the production conditions of the target mold, the mold material information, and the historical operation characteristics matching the target mold in the knowledge graph are used to train a machine learning model for matching the target mold to obtain a data-driven model for the target mold; the data-driven model is used to predict the service life and damage of the target mold; Collecting real-time operating data of the target mold and inputting it into the data-driven model to obtain a predicted service life value and a predicted damage state of the target mold; A corresponding mold maintenance strategy is executed based on the predicted service life value and the predicted damage state.

2. The method according to claim 1, characterized in that After training a machine learning model matching the target mold using the visual image information, the production conditions of the target mold, the mold material information, and the historical operation features matching the target mold in the knowledge graph to obtain a data-driven model of the target mold, the method further includes: Precalculating physical quantities of the target mold based on low-frequency modes calculated by finite element analysis, and introducing the physical quantities as physical constraints into the data-driven model to correct prediction errors in the data-driven model caused by differences between the target mold and the standard part; The physical quantities include at least: natural vibration frequencies of different molds, stress distribution, and structural strengths of components of different shapes.

3. The method according to claim 2, characterized in that The method of precalculating the physical quantities of the target mold based on the low-frequency modes calculated by finite element analysis and introducing the physical quantities as physical constraints into the data-driven model includes: Determine the shape type of each component in the target mold based on the geometric dimensions, assembly relationships, and material properties of the target mold. Shape types include regular shapes that match a single component in a standard part, and irregular shapes composed of multiple components in a standard part. For parts with regular shapes, hexahedral meshes are used for division; for parts with irregular shapes, tetrahedral meshes are used for division to obtain the mesh structure model of the target mold; Setting boundary conditions and load conditions for the grid structure model; performing low-frequency modal calculation on the set grid structure model to obtain inherent characteristic parameters of the target mold in the low-frequency range as the physical quantity; When predicting the vibration of the target mold, the natural vibration frequency is used as a constraint condition. When the vibration frequency predicted by the data-driven model deviates from the natural vibration frequency range, the corresponding model parameters and / or model structure in the data-driven model are adjusted to make the prediction result of the data-driven model more consistent with the actual physical situation; When predicting the stress distribution, deformation and / or structural strength of the target mold component, the stress distribution of the target mold component and the structural strength of components of different shapes are used as constraints. If the prediction result of the data-driven model exceeds the reasonable range obtained by finite element analysis, through continuous iterative training, the data-driven model can gradually correct the prediction error while taking into account the physical constraints.

4. The method according to claim 3, characterized in that The step of performing low-frequency modal calculation on the set grid structure model to obtain the inherent characteristic parameters of the target mold in the low-frequency range as the physical quantity includes: A finite element solver is used to calculate a first stress nephogram during the operation of the target mold, based on the load conditions and material properties of the target mold, and in conjunction with elasticity and plasticity theories. The first stress nephogram includes at least the magnitude and direction of stresses of different components of the target mold at various times. Calculating the stress concentration coefficient of each component in the first stress cloud map using a formula of elastic mechanics theory to identify stress concentration areas in the target mold; the stress concentration areas include at least: corners of the target mold and joints between different components; Local mesh refinement is performed on the stress concentration area, and the historical damage position of the standard part matched by the target mold is used as the load excitation. The force simulation calculation is performed through the multi-physics field coupling effect to obtain a second stress cloud map with a smaller distribution mesh granularity; The second stress cloud map is fused with the first stress cloud map to obtain the stress distribution of the target mold.

5. The method according to claim 3, characterized in that The step of performing low-frequency modal calculation on the set grid structure model to obtain the inherent characteristic parameters of the target mold in the low-frequency range as the physical quantity includes: For components of different shapes in the target mold, the corresponding structural strength is calculated based on the geometric shape and material mechanical properties of each component and combined with the corresponding strength theory; Classify the stress types of each component and conduct preliminary verification of the structural strength of each component based on the stress type. For components subjected to unidirectional tensile loads, the first strength theory is used for strength verification; for components under complex stress states, the fourth strength theory is used for strength verification. Calculate the maximum load limit and safety factor of each component based on the tensile, compression, bending, and torsion stress conditions of each component under complex loads; Identify the load-bearing components in the target mold and, with these components as the center, use the maximum load limit and safety factor to make correlation corrections to the structural strength of other components. Use topology optimization methods and finite element analysis to calculate the stress distribution of each component under different load conditions, identifying components with concentrated stress and bearing the main load as load-bearing components. The combined load-bearing range of each component after assembly is evaluated, and based on the combined load-bearing range, the structural strength of each component in the target mold is corrected for correlation to avoid structural strength mismatch between connected components.

6. The method according to claim 1, characterized in that After training a machine learning model matching the target mold using the visual image information, the production conditions of the target mold, the mold material information, and the historical operation features matching the target mold in the knowledge graph to obtain a data-driven model of the target mold, the method further includes: Obtain similar material performance data and similar component failure modes matching the target mold through the knowledge graph; Based on the acquired data on similar material properties and similar component failure modes, a damage analysis model is used to predict the periodic changes and abnormal fluctuations during the operation of the target mold to obtain a predicted wear rate value for the target mold during operation; the wear rate prediction value is a dynamic value that fluctuates based on changes in operation time; The wear rate prediction value is used to update the formula parameters or formula structure in the empirical formula of the target mold.

7. The method according to claim 6, characterized in that After the wear rate prediction value is used to update the formula parameters or formula structure in the empirical formula of the target mold, the method further includes: The updated empirical formula is introduced into the data-driven model as a constraint condition to correct the prediction error of the data-driven model.

8. The method according to claim 1, characterized in that The real-time operating data of the target mold is collected and input into the data-driven model to obtain a predicted service life value and a predicted damage state of the target mold, including: Extracting stress cycle features, vibration features, heat conduction characteristics, and real-time operating condition features from the real-time operating data through a feature extraction layer; Pre-calculate the extracted feature data with the steel structure mold failure mode model in the knowledge graph to obtain a first remaining service life probability distribution of each component in the target mold; Adapting each extracted feature data to each prediction module in the data-driven model, and inputting it into a prediction module that matches the change characteristics of each feature data to perform change trend prediction, so as to obtain change trend information of each feature data; Mapping the change trend information to the components of the target mold and the connections between the components to modify the first remaining useful life probability distribution and obtain a second remaining useful life probability distribution for each component; Based on the second remaining service life probability distribution, the dynamic service life prediction value of each component in the target mold that changes with time and the corresponding predicted damage state are marked.

9. A steel structure mold life prediction and maintenance decision system, characterized in that: The system comprises: A construction module is used to obtain multi-source data of various steel structure molds using various steel structure molds as standard parts, so as to construct a knowledge graph of the various steel structure molds; and to associate the material information, historical operation data, performance under specific process parameters, damage conditions, service life, mold vibration information, and shape characteristics of different steel structure molds in the multi-source data in the form of a knowledge graph; A selection module is configured to scan visual image information of a target mold; use the visual image information, the production conditions of the target mold, and mold material information as index conditions, and select a machine learning model that matches the target mold from a candidate model library in combination with the knowledge graph; a training module for training a machine learning model matching the target mold using the visual image information, the production conditions of the target mold, the mold material information, and historical operating characteristics matching the target mold in the knowledge graph to obtain a data-driven model for the target mold; the data-driven model is used to predict the service life and damage of the target mold; A prediction module, configured to collect real-time operating data of a target mold and input the data into the data-driven model to obtain a predicted service life and damage status of the target mold; A maintenance module is used to execute a corresponding mold maintenance strategy based on the predicted service life value and the predicted damage state.

10. A terminal device, characterized in that: The terminal device includes a processor and a memory; The memory is used to store computer programs; The processor is used to execute the computer program and implement the steel structure mold life prediction and maintenance decision-making method according to any one of claims 1 to 8 when executing the computer program.

Images