Container vulnerable part identification and 3D display method and system

Abstract

The invention belongs to the technical field of container maintenance analysis and display, provides a container vulnerable part identification and 3D display method and system, and aims to select key factor fields with high importance from factor fields included in collected historical maintenance data through a random forest model, thereby facilitating accurate identification of vulnerable parts. According to the method, the association relationship between the component and the damage reason is analyzed, the association characteristics are generated according to the association relationship and the association rule, and based on the association characteristics, the key factor field and the current static characteristics of the target container, damage risk prediction is accurately performed on the component in the target container which is not maintained for a long time in advance through the damage prediction model. Therefore, the vulnerable part can be accurately identified. By visually displaying the 3D model and the vulnerable parts of the target container, the container structure and the vulnerable parts are displayed to a user more clearly, visually and vividly, the comprehensiveness of cognition of the user on the container structure and the vulnerable parts is improved, and the accuracy of maintenance decision is improved.

Description

Technical Field

[0001] This invention relates to the field of container repair analysis and display, specifically to a method and system for identifying and 3D displaying vulnerable components of containers. Background Technology

[0002] In the container shipping and logistics industry, container maintenance and repair are crucial for ensuring the safe transport of goods. However, traditional maintenance processes often lack systematic analysis of maintenance data, leading to inaccurate maintenance decisions and difficulty in controlling maintenance costs. Furthermore, customers and maintenance personnel often lack sufficient understanding of the specific structure of containers and the location of vulnerable components, which limits the improvement of maintenance efficiency and customer satisfaction.

[0003] Against this backdrop, while some studies have attempted to optimize maintenance processes through data analysis, these studies are typically limited to single data sources and simple statistical analyses, failing to fully utilize big data for comprehensive data analysis and making it difficult to accurately predict vulnerable components of containers. Furthermore, existing container display methods mostly employ two-dimensional images or physical models, lacking interactivity and realism, making it difficult for users to gain a deeper understanding of the container's structure and vulnerable components, thus affecting the accuracy of maintenance decisions. Therefore, how to develop 3D container displays based on the analysis of historical maintenance data is a crucial issue.

[0004] Based on this, this specification provides a method and system for identifying and 3D displaying vulnerable components of a container. Summary of the Invention

[0005] To address the existing problems of insufficient systematic analysis of maintenance data, difficulty in accurately identifying vulnerable components of containers, and inability to clearly display the container structure and the location of vulnerable components, this invention proposes a method and system for identifying and displaying vulnerable components of containers in 3D. By analyzing historical maintenance data, it identifies vulnerable components of containers and analyzes their correlation with the causes of damage, improving the user's comprehensive understanding of component failure. The 3D display clearly shows the container structure and vulnerable components, improving the accuracy of the user's understanding of vulnerable components and thus enhancing the accuracy of maintenance decisions.

[0006] This manual provides a method for identifying and 3D displaying vulnerable components of a container, including:

[0007] S1: Data Collection: Collect historical maintenance data for containers;

[0008] S2: Feature selection: Based on the historical maintenance data, a random forest model is used to determine the importance of each factor field included in the historical maintenance data, and the factor fields are filtered according to the importance of each factor field to determine the key factor fields;

[0009] S3: Predict component damage risk: Based on the historical maintenance data, determine the correlation between the historical damaged components of the container and the historical causes of damage. Based on the determined correlation and the pre-set correlation rules, generate correlation features, and use the correlation features, the key factor fields, and the current static features of the target container as input parameters for a pre-trained damage prediction model to predict the component damage risk of the target container.

[0010] S4: Identify vulnerable components: Based on the risk of damage to the components, identify the vulnerable components of the target container;

[0011] S5: Modeling and Visualization: Using 3D modeling software, construct a 3D model corresponding to the target container, and visualize the vulnerable parts and the 3D model.

[0012] Optionally, the method further includes:

[0013] Use the historical maintenance costs in the historical maintenance data as labels, and use other types of data in the historical maintenance data other than the historical maintenance costs as training samples;

[0014] The training samples are input into a linear regression model to determine the prediction results output by the linear regression model.

[0015] Based on the Huber loss function, ridge regression regularization, and elastic network feature selection, the target loss function of the linear regression model is determined, and the target loss is determined by using the target loss function based on the prediction results and labels.

[0016] Based on the target loss, a linear regression model is trained, and the trained linear regression model is used to predict the maintenance cost of the target container based on the maintenance information of the target container.

[0017] Optionally, S2 specifically includes:

[0018] Using big data analytics algorithms, the historical maintenance data is clustered according to the field values ​​corresponding to each factor field included in the historical maintenance data to obtain data groups; the field values ​​corresponding to each factor field in the data of each data group are the same;

[0019] Based on the data sets, a random forest model is constructed with the type of damaged component and the repair method as the objective variables;

[0020] Based on the random forest model, determine the Gini impurity and SHAP value corresponding to each factor field;

[0021] The importance of each factor field is determined based on the Gini impurity and SHAP value.

[0022] Based on the importance of each factor field, the factor fields are filtered to determine the key factor fields.

[0023] Optionally, the key factor fields include historical repair patterns, historical repair frequency, historical repair counts for each component, repair location information, and historical repair costs.

[0024] Optionally, the visualization of the vulnerable component and the 3D model in step S6 specifically includes:

[0025] Mark the vulnerable components on the 3D model;

[0026] The annotated 3D model is rendered using Physically Based Rendering (PBR) technology and displayed on the user interface.

[0027] Optionally, the generation of association features in step S3 based on the determined association relationships and pre-set association rules specifically includes:

[0028] Determine the historical frequency of each relationship in the historical maintenance data;

[0029] Construct a relationship matrix based on the historical frequency of each of the aforementioned relationships;

[0030] Based on the relationship matrix and the pre-set association rules, association features of each dimension corresponding to each association are generated; wherein, the association rules include confidence scores corresponding to various associations, and each dimension includes at least the damage frequency of historically damaged parts in the association, the causal association score of the association, and the historical occurrence frequency of the association, and the causal association score is the confidence score corresponding to the association in the association rules.

[0031] This specification provides a container vulnerable component identification and 3D display system. The system includes a data analysis module and a 3D display module. The data analysis module includes a data acquisition submodule, a feature selection submodule, a risk prediction submodule, and a vulnerable component identification submodule, wherein:

[0032] The data acquisition submodule is used to collect historical maintenance data of the container;

[0033] The feature selection submodule is used to determine the importance of each factor field included in the historical maintenance data using a random forest model based on the historical maintenance data, and to filter the factor fields according to the importance of each factor field to determine the key factor fields.

[0034] The risk prediction submodule is used to determine the correlation between historically damaged components and historical causes of damage to the container based on the historical maintenance data, generate correlation features based on the determined correlation and pre-set correlation rules, and use the correlation features, the key factor fields and the current static features of the target container as input parameters for a pre-trained damage prediction model to predict the component damage risk of the target container.

[0035] The vulnerable component determination submodule is used to determine the vulnerable components of the target container based on the risk of damage to the components.

[0036] The 3D visualization module is used to construct a 3D model of the target container using 3D modeling software, and to visualize the vulnerable parts and the 3D model.

[0037] Optionally, the system further includes a maintenance cost prediction module;

[0038] The maintenance cost prediction module is used to use historical maintenance costs from the historical maintenance data as labels and other types of data from the historical maintenance data besides the historical maintenance costs as training samples; input the training samples into a linear regression model to determine the prediction result output by the linear regression model; determine the target loss function of the linear regression model based on the Huber loss function, ridge regression regularization, and elastic network feature selection, and determine the target loss based on the prediction result and labels using the target loss function; train the linear regression model based on the target loss, and the trained linear regression model is used to predict the maintenance cost of the target container based on the maintenance information of the target container.

[0039] Optionally, the data analysis module further includes a data extraction submodule;

[0040] The data extraction submodule is used to cluster the historical maintenance data according to the field values ​​corresponding to each factor field included in the historical maintenance data using big data analysis algorithms to obtain data groups; the field values ​​corresponding to each factor field in the data of each data group are the same;

[0041] The feature selection submodule is specifically used to: construct a random forest model with damaged component type and repair method as target variables based on the data groups; determine the Gini impurity and SHAP value corresponding to each factor field based on the random forest model; determine the importance of each factor field based on the Gini impurity and SHAP value; and filter the factor fields based on their importance to determine key factor fields.

[0042] Optionally, the 3D visualization display module includes a 3D modeling submodule, a user interaction submodule, and a physical rendering submodule;

[0043] The 3D modeling submodule is used to construct a 3D model corresponding to the target container using 3D modeling software; and to mark the vulnerable parts on the 3D model.

[0044] The physical rendering submodule is used to render the annotated 3D model using physically based rendering (PBR) technology.

[0045] The user interaction submodule is used to visually display the rendered 3D model on the user interaction interface.

[0046] Optionally, the system further includes a feedback module;

[0047] The feedback module is used to display a feedback page to users, receive feedback submitted by users through the feedback page, and optimize the visual display based on the feedback.

[0048] Optionally, the risk prediction submodule is specifically used to: determine the historical frequency of each correlation in the historical maintenance data; construct a correlation matrix based on the historical frequency of each correlation; and generate correlation features for each dimension corresponding to each correlation based on the correlation matrix and pre-set correlation rules; wherein the correlation rules include confidence levels corresponding to various correlations, and each dimension includes at least the damage frequency of historically damaged parts in the correlation, the causal correlation score of the correlation, and the historical frequency of the correlation, and the causal correlation score is the confidence level corresponding to the correlation in the correlation rules.

[0049] The above-mentioned technical solutions adopted in this specification can achieve the following beneficial effects:

[0050] This specification provides a method for identifying and 3D displaying vulnerable components of a shipping container. First, historical maintenance data of the container is collected. Based on this data, a random forest model is used to determine the importance of each factor field within the historical maintenance data. Then, based on the importance of each factor field, key factor fields are identified. The random forest model selects key factor fields that significantly influence the type of damaged component and the repair method, which helps to accurately identify vulnerable components, thereby reducing unexpected downtime and maintenance costs. Based on the historical maintenance data, the correlation between historically damaged components and their causes is determined. Based on the determined correlation and pre-set correlation rules, correlation features are generated. These correlation features, key factor fields, and the current static features of the target container are used as input parameters for a pre-trained damage prediction model to predict the component damage risk of the target container. Based on the component damage risk, vulnerable components of the target container are identified. By analyzing the correlation between components and their causes, and using the damage prediction model to accurately predict the damage risk of components in target containers that have not been repaired for a long time, vulnerable components in the target container are accurately identified based on the component damage risk. Next, 3D modeling software is used to construct a 3D model of the target container, and vulnerable components and the 3D model are visualized. Visualizing the 3D model and vulnerable components provides a clearer, more intuitive, and vivid presentation of the container structure and vulnerable parts to users, improving their comprehensive understanding of the container structure and vulnerable parts, thereby enhancing the accuracy of maintenance decisions.

[0051] This invention uses historical maintenance costs as labels and other types of data from historical maintenance data (excluding historical maintenance costs) as training samples. Based on the training samples and labels, a linear regression model is trained to predict container maintenance costs. Furthermore, the Huber loss function is introduced during the training of the linear regression model to suppress the interference of outliers, ridge regression regularization is introduced to address multicollinearity between input features and maintenance costs, and elastic network feature selection is introduced to automatically optimize the contribution of each input feature. Through the Huber loss function, ridge regression regularization, and elastic network feature selection, a loss function is constructed to better train the linear regression model for predicting maintenance costs.

[0052] Before constructing the random forest model, this invention can use big data analysis algorithms to cluster the historical maintenance data according to the field values ​​corresponding to each factor field included in the historical maintenance data, to obtain data groups. The field values ​​corresponding to each factor field in the data of each data group are the same, which provides data support for the subsequent construction of the random forest model and the identification of vulnerable parts in the container.

[0053] This invention can construct a random forest model with damaged component type and repair method as target variables based on each data set. By calculating the Gini impurity and SHAP value corresponding to each factor field, and determining the importance of each factor field based on the dual weights of Gini impurity and SHAP value, the key factor fields affecting damaged component type and repair method can be identified from each factor field, ensuring the interpretability and engineering applicability of the key factor fields.

[0054] This invention can annotate vulnerable components on a 3D model, and then render the annotated 3D model using physically based rendering (PBR) technology. By combining PBR technology and data visualization technology, it can intuitively present vulnerable components while ensuring high-fidelity visual effects in the 3D model. This allows users to understand the location of vulnerable components in the container, providing efficient visual support for container maintenance decisions.

[0055] This invention, when generating association features, determines the historical frequency of each association in historical maintenance data. Based on the historical frequency of each association, a relationship matrix is ​​constructed. According to the relationship matrix and pre-set association rules, association features corresponding to each association in various dimensions are generated. The association rules include confidence scores for various associations, and each dimension includes at least the damage frequency of historically damaged components in the association, the causal correlation score of the association, and the historical frequency of the association. The causal correlation score is the confidence score corresponding to the association in the association rule. By generating association features with several dimensions based on associations and association rules obtained from historical maintenance data, the damage prediction model can better predict the component damage risk of each component in the target container.

[0056] This invention provides a container vulnerable component identification and 3D display system. The system includes a feedback module that provides a feedback page for users to input their opinions on the visualization. The feedback model then optimizes the visualization based on the user's feedback to meet user needs and provide a better user experience. Attached Figure Description

[0057] Figure 1 This is a flowchart illustrating a method for identifying and 3D displaying vulnerable components of a container, as provided in this specification.

[0058] Figure 2 This is a schematic diagram of a 3D model of a highlighted vulnerable component provided in this specification;

[0059] Figure 3 This is a schematic diagram of a user interface provided in this specification;

[0060] Figure 4 This is a schematic diagram of a rotated 3D model provided in this specification;

[0061] Figure 5 This is a schematic diagram of a 3D model of a door opening and closing provided in this specification;

[0062] Figure 6 This is a schematic diagram of a 3D model of floor sliding out and floor sliding in, provided in this specification.

[0063] Figure 7 This is a schematic diagram illustrating the ascent and descent of a 3D model provided in this specification.

[0064] Figure 8 This is a schematic diagram of an information query page provided in this specification;

[0065] Figure 9 This is a schematic diagram of a container vulnerable component identification and 3D display system provided in this specification.

[0066] Figure 10 This is a schematic diagram of the structure of a data analysis module provided in this specification;

[0067] Figure 11 This is a schematic diagram of another container vulnerable component identification and 3D display system provided in this specification. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this specification will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of them. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this specification.

[0069] This specification provides a method and system for identifying and 3D displaying vulnerable components of a container. The technical solutions provided by the various embodiments of this specification are described in detail below with reference to the accompanying drawings.

[0070] Figure 1 This is a flowchart illustrating a method for identifying and 3D displaying vulnerable components of a container, as provided in this specification. The method specifically includes the following steps:

[0071] S1: Data Collection: Collect historical maintenance data for containers.

[0072] In this manual, the equipment used for demonstration can first collect data, specifically historical maintenance data of the container. This equipment can be a server, a system, one or more modules within a system, or electronic devices such as desktop computers or laptops. For ease of description, the following explanation will focus on a server as the primary execution entity, illustrating the container vulnerable component identification and 3D display method provided in this manual.

[0073] The aforementioned historical maintenance data can be all maintenance data for all containers in history, or it can be all maintenance data for a subset of containers in history. This subset of containers must include at least the target container, which refers to a container that has not been repaired within a target time period. This target time period can be preset, such as the past five years. Alternatively, the aforementioned historical maintenance data can be maintenance data for containers (i.e., all containers or a subset of containers) within a specified historical time period. This maintenance data can include data such as container size, container structure, cargo type, transportation route, maintenance time, maintenance cost, maintenance frequency, damaged parts, type of damaged parts, cause of damage, location of reported repairs, repair mode, number of repairs, customer information, and post-repair effectiveness evaluation. The aforementioned repair mode refers to whether it is partial repair or complete replacement. The aforementioned specified time period can be preset, such as the past five years, or it can be selected by the user. That is, the user can select a time period on the data selection page and use it as the specified time period. In other words, the user can flexibly select time periods on the data selection page and use the selected time period as the specified time period. All historical maintenance data for containers can be stored in a historical database. This database contains all past damage and repair data for all containers. Users can flexibly select time periods and extract all maintenance data for all containers within that specified time period from the historical database, using this data as the aforementioned historical maintenance data. Furthermore, when there are multiple containers, the aforementioned historical maintenance data can include several historical maintenance sub-data sets, each containing all maintenance data corresponding to each container.

[0074] S2: Feature selection: Based on the historical maintenance data, a random forest model is used to determine the importance of each factor field included in the historical maintenance data, and the factor fields are filtered according to the importance of each factor field to determine the key factor fields.

[0075] In this specification, the server can feature selection, that is, based on historical maintenance data, using a random forest model to determine the importance of each factor field included in the historical maintenance data, and filtering each factor field according to the importance of each factor field to determine the key factor fields. The aforementioned factor fields represent all types of data in the historical maintenance data, including types such as historical repair mode, historical repair frequency, historical repair count, historical repair time, cause of damage, type of damaged component, repair location information, and historical repair cost. The field value corresponding to each factor field is the specific data corresponding to each type in the historical maintenance data. Of course, the aforementioned factor fields can also be a subset of the data types in the historical maintenance data; that is, a specified number of types can be randomly selected from each type as factor fields, or a specified number of types can be manually selected from each type as factor fields. This specification does not impose specific limitations.

[0076] Specifically, the server can employ big data analytics algorithms to cluster historical maintenance data according to the field values ​​corresponding to each factor field, resulting in data groups. Based on each data group, a random forest model is used to determine the importance of each factor field, and the factor fields are then filtered based on their importance to identify key factor fields. The aforementioned big data analytics algorithm can be any existing or pre-defined algorithm for clustering analysis. This clustering analysis algorithm can be an unsupervised learning algorithm used to divide historical maintenance data into several data groups, where the data within each group exhibits similarity, meaning that the field values ​​corresponding to each factor field in the data within each group are identical.

[0077] In addition, after obtaining each data group, anomaly detection can be performed on each data group to handle any anomalies found. Specifically, the number of data points in each data group can be determined. For each data group, if the number of data points in the group is less than a first threshold, the group is considered anomaly. This group is then displayed to the processing personnel for manual verification and modification, and the processed data group is returned. The first preset threshold can be a pre-set value or determined based on the number of data points in each data group. Specifically, the server can determine the average or median number of data points in each data group and use this as the first threshold.

[0078] In addition, when there are multiple containers, the above data groups can be the data groups corresponding to each container. That is, the server can use big data analysis algorithms to cluster the historical maintenance data of each container according to the field values ​​of each factor field included in the historical maintenance data of that container, and obtain the data groups of that container.

[0079] The above describes a process where, based on each data set, a random forest model is used to determine the importance of each factor field. Then, based on the importance of each factor field, it is filtered to identify key factor fields. The server can construct a random forest model with damaged component type and repair method as objective variables for each data set. Based on the random forest model, the Gini impurity and SHAP value corresponding to each factor field are determined. The importance of each factor field is determined based on its Gini impurity and SHAP value. Finally, based on the importance of each factor field, it is filtered to identify key factor fields.

[0080] Random Forest is an ensemble learning method that uses multiple decision trees to make predictions and aggregates their results. It can also be used for feature selection. Random Forest models can be built based on datasets, and the model is constructed by calculating the Gini impurity of each factor field. Therefore, the Gini impurity for each factor field can be extracted from the random forest model. This Gini impurity represents the contribution of each factor field to reducing the impurity of the dataset (i.e., each dataset group) when splitting nodes. A higher Gini impurity indicates a more important factor field, and vice versa. To determine the Gini impurity of a factor field, the server can first determine the initial Gini impurity for the entire dataset. Specifically, this can be based on the proportion of each category and calculated using the formula for calculating Gini impurity. Based on the field values ​​of the factor fields, the datasets are divided, resulting in datasets where the field values ​​of the factor fields are the same for all data in each dataset. Based on the proportion of each category in each dataset, the formula for calculating Gini impurity is used to calculate the first Gini impurity for each dataset. The sum of the first Gini impurities for each data point is taken as the second Gini impurity, and the difference between the initial Gini impurity and the second Gini impurity is calculated and used as the Gini impurity for that factor field. Each data group has a corresponding category, which is a combination of damaged component types and repair methods in that data. Specifically, when determining the initial Gini impurity or the first Gini impurity, the server can determine various categories (combinations of various damaged component types and repair methods) based on each data group. For each category, the number of data points in each data group belonging to that category is determined as the first quantity, and the total number of data points in each data group is determined as the second quantity. The ratio of the first and second quantities is determined and used as the proportion of that category. Based on the proportion of each category, the formula for calculating Gini impurity is used to calculate the initial Gini impurity. The formula (1) for calculating the Gini impurity is as follows:

[0081]

[0082] Among them, I G (D) represents the Gini impurity, and J represents the number of classes. This represents the square of the proportion of category i.

[0083] The SHAP (SHapley Additive exPlanations) values ​​described above represent the contribution of each factor field to the classification result (i.e., the type of damaged component and the repair method). SHAP values ​​can be obtained based on a random forest model and a SHAP interpreter or SHAP framework.

[0084] Specifically, when determining the importance of each factor field, the server can, for each factor field, determine its Gini impurity and SHAP value based on a random forest model, and then determine the importance of that factor field based on the Gini impurity and SHAP value. Specifically, when determining the Gini impurity and SHAP value of a factor field based on the random forest model, the server can directly extract the Gini impurity of the factor field from the random forest model, and simultaneously use a SHAP interpreter or SHAP framework to calculate the SHAP value of the factor field from the random forest model.

[0085] When determining the importance of a factor field based on Gini impurity and SHAP value, the server can directly use the sum of the Gini impurity and SHAP value as the importance of the factor field. Alternatively, the server can weight the Gini impurity using a first weight and the SHAP value using a second weight, then use the sum of the weighted Gini impurity and the weighted SHAP value as the importance of the factor field. The first and second weights can be preset.

[0086] The above-described method filters factor fields based on their importance. To determine key factor fields, the server sorts the factor fields according to their respective importance, from highest to lowest, resulting in a factor sequence. Then, following the order of the factor sequence, a specified number of factor fields are selected as key factor fields. This specified number can be pre-set, such as five. At least one key factor field is required. These key factor fields may include historical repair patterns, historical repair frequency, historical repair counts, repair location information, and historical repair costs. It should be noted that the content included in the above key factor fields is only an example; this specification does not limit the key factor fields to only the content shown above, and other content may also be included.

[0087] In addition, when there are multiple containers, the aforementioned key factor fields can be the key factor fields corresponding to each container. That is, the server can use random forest feature selection technology to determine the key factor fields of each container based on the factor fields corresponding to each container.

[0088] S3: Predict component damage risk: Based on the historical maintenance data, determine the correlation between the historical damaged components and the historical causes of damage to the container. Based on the determined correlation and the pre-set correlation rules, generate correlation features, and use the correlation features, the key factor fields, and the current static features of the target container as input parameters for a pre-trained damage prediction model to predict the component damage risk of the target container.

[0089] S4: Identify vulnerable components: Based on the risk of damage to the components, identify the vulnerable components of the target container.

[0090] In this specification, the server can first predict the risk of component damage. That is, based on historical maintenance data, it determines the correlation between historically damaged components and their causes within the container. Based on the determined correlations and pre-set correlation rules, it generates correlation features. These correlation features, key factor fields, and the current static features of the target container are then used as input parameters to a pre-trained damage prediction model to predict the component damage risk of the target container. Next, the server can identify vulnerable components, that is, based on the component damage risk, it identifies the vulnerable components of the target container. Since historical maintenance data includes both historically damaged components and their causes, the server can directly determine the correlation between these historically damaged components and their causes from the historical maintenance data. This correlation can be represented in the form of "component-cause," such as door lock damage-collision. Furthermore, the aforementioned correlation refers to the mapping relationship between historically damaged components and their causes, that is, the damage to historically damaged components caused by historical causes. Components within a container may include the container body, door locks, hinges, etc. Damage can be caused by various factors, such as structural damage (e.g., container deformation, corrosion), functional component failure (e.g., door lock, hinge failure), sealing issues (e.g., aging of sealing strips), and wear and tear (e.g., floor wear). The aforementioned association rules can be pre-set, including confidence levels for various association relationships. These confidence levels characterize the likelihood that a cause within the association relationship will lead to component damage. The aforementioned association features can be generated based on each association relationship and association rule. Each association feature can include features across various dimensions, at least including the damage frequency of historically damaged components within the association relationship, the cause-relatedness score, and the historical frequency of the association relationship. It may also include the time the association relationship occurred, but this specification does not specify a particular dimension. The damage frequency of historically damaged components within the aforementioned association relationship can be the ratio of historically damaged components in any association relationship to all historically damaged components in historical maintenance data. The cause-relatedness score is the confidence level corresponding to the association relationship in the association rule. The historical frequency of the aforementioned association relationship is the ratio of the number of times any association relationship occurs among all association relationships to the total number of all association relationships.

[0091] The aforementioned current static characteristics may include the basic characteristics of the target container and current maintenance data. Basic characteristics may include the service life of the target container, the ambient humidity, temperature, etc. Current maintenance data refers to the maintenance data of the target container at present. If no parts of the target container are being maintained, the current maintenance data will be empty. If a part of the target container is being maintained, the current maintenance data may include the name of the part being maintained, the cause of the damage, the maintenance time, etc. The aforementioned damage prediction model is a pre-trained model. This damage prediction model can be an XGBoost model, or other types of models; this specification does not specify any particular limitation. The aforementioned damage prediction model is used to predict the component damage risk of the container. This component damage risk includes the future damage risk corresponding to all components in the container. Furthermore, the aforementioned target container is a container that has not been maintained within the target time period. By predicting the damage risk of components in a container that has not been maintained within the target time period using the damage prediction model, vulnerable components in the container can be identified, thereby alerting the user to repair the container.

[0092] When generating association features based on the identified relationships and pre-set association rules, the server can determine the historical frequency of each association in historical maintenance data. A relationship matrix is ​​constructed based on the historical frequency of each association. Then, based on the relationship matrix and pre-set association rules, association features for each dimension corresponding to each association are generated. The elements in the relationship matrix represent the frequency of each association in the historical maintenance data.

[0093] In step S4 above, the server can classify components in the target container whose damage risk exceeds a preset threshold as vulnerable components. Specifically, for each component in the target container, the server can classify that component as a vulnerable component when its corresponding damage risk exceeds the preset threshold. Furthermore, the server can sort the components in the target container according to their damage risk, from highest to lowest, to obtain a component sequence. Following the order of the component sequence, a first number of components are determined from the sequence and designated as vulnerable components. This first number is a pre-set value.

[0094] In addition, when there are multiple containers, in S3, the server can first determine the correlation between the historical damaged parts and the historical causes of damage for each container based on the container's historical maintenance data. Based on the determined correlation and pre-set correlation rules, correlation features are generated. Then, based on the determined correlation of each container, the key factor fields of each container, and the current static features of the target container, a pre-trained damage prediction model is used to predict the component damage risk of the target container.

[0095] S5: Modeling and Visualization: Using 3D modeling software, construct a 3D model corresponding to the target container, and visualize the vulnerable parts and the 3D model.

[0096] In this specification, the server can model and visualize the target container, using 3D modeling software to construct a 3D model of the container and visualize the vulnerable components and the 3D model. The 3D modeling software can be any existing software for building 3D models, such as Autodesk Maya. When constructing the 3D model of the target container, the server can use the software to build a 3D model of the target container based on its actual size and structure. When visualizing the vulnerable components and the 3D model, the server can annotate the vulnerable components on the 3D model. The annotated 3D model is then rendered using Physically Based Rendering (PBR) technology and displayed on the user interface.

[0097] In the process of annotating vulnerable parts on a 3D model, the server can use different colors or textures to annotate the vulnerable parts based on their location on the 3D model. These different colors or textures represent different types of vulnerable parts, or the degree of damage to the vulnerable parts. Darker colors or more complex textures indicate a greater degree of damage (i.e., a higher risk of damage), while lighter colors or simpler textures indicate a less severe degree of damage. For example, annotating vulnerable parts on a 3D model using highlighting can be done as follows... Figure 2 As shown, Figure 2 This is a schematic diagram of a 3D model of a highlighted vulnerable component provided in this specification. Figure 2 The diagram includes sub-figures a and b. Sub-figure a shows a 3D model of the highlighted container floor, highlighting this vulnerable component. The container in sub-figure a is in an open state. To better display the container floor, sub-figure a separates the 3D model and the floor; the left side shows the 3D model, and the right side shows the floor (area a). Sub-figure b shows a 3D model of the highlighted container roof, highlighting this vulnerable component. The container in sub-figure b is in a closed state.

[0098] The aforementioned Physically Based Rendering (PBR) technology aims to more accurately simulate the interaction of light in a scene by following the laws of physics, thereby achieving more realistic and consistent material representation under different lighting conditions. It is an advanced rendering technique in modern computer graphics. The user interface can include the rendered 3D model and vulnerable components, with the vulnerable components labeled on the 3D model. Alternatively, the user interface can directly display vulnerable components and the 3D model, meaning that vulnerable components can be displayed in text form instead of labeled on the 3D model. Specifically, the name or abbreviation of the vulnerable component can be displayed in text form. Some common vulnerable components for door assemblies are shown in Table 1 below. Table 1 shows common vulnerable components for container door assemblies. In Table 1, "W / Hdw." stands for "With Hardware," indicating that hardware is included, and "W / O Hdw" stands for "Without Hardware," indicating that hardware is not included.

[0099] Table 1

[0100]

[0101]

[0102] In some embodiments of this specification, after identifying a vulnerable component, S4 can determine the association rule that the vulnerable component matches based on the aforementioned association rule, and display it on the user interface. That is, when a component associated with a relationship in an association rule is a vulnerable component, then the vulnerable component matches that association rule, and subsequently, all matched association rules can be displayed on the user interface.

[0103] In some embodiments of this specification, the server can pre-acquire historical maintenance data of all containers within a specified time period and divide the acquired historical maintenance data to obtain training data and test data. The ratio of the training data to the test data can be 7:3, but other ratios are also possible and are not specifically limited in this specification. When training the damage prediction model, the server can determine training samples based on the historical maintenance data within a first time period in the training data, and determine the labels corresponding to the training samples based on the historical maintenance data within a second time period in the training data. The first time period is earlier than the second time period, and both the first and second time periods are within the specified time period. Subsequently, the server can optimize the hyperparameters and train the damage prediction model based on the training samples and labels using 5-fold cross-validation.

[0104] In addition, the server can input training samples into the damage prediction model to be trained, and obtain the prediction results output by the damage prediction model. Based on the prediction results and labels, the damage prediction model to be trained is trained, and the trained damage prediction model is tested using test data. If the test is passed, the damage prediction model is used as the model for subsequent component damage risk prediction.

[0105] When determining training samples based on historical maintenance data within the first time period of the training data, key factor fields can be determined according to the process described in S2, based on the historical maintenance data within the first time period of the training data. Furthermore, based on the historical maintenance data within the first time period of the training data, the correlation between historically damaged components and historical causes of damage to the container is determined. Correlation features are generated based on the determined correlations and pre-set correlation rules. The correlation features, key factor fields, and the current static features of the sample container are used as training samples. These sample containers are those that have not been repaired within the target time period of the first time period, and the aforementioned current static features are the static features of the target container at the end of the first time period.

[0106] When determining the label corresponding to the training sample based on the historical maintenance data within the second time period in the training data, the server can determine the damaged parts of the sample container within the second time period based on the historical maintenance data within the second time period in the training data, and use them as the label corresponding to the training sample.

[0107] When training the damage prediction model based on the prediction results and labels, the server can determine the loss using a pre-set loss function based on the difference between the prediction results and the labels, and then train the damage prediction model based on this loss. The loss function can be the cross-entropy loss function, the association loss function, or a combination of multiple loss functions; this specification does not specify a particular limitation. The association loss function can be a loss function that captures the relationship, structure, or similarity between data, such as the center loss function.

[0108] When testing the trained damage prediction model using the test data mentioned above, evaluation metrics (i.e., AUC and F1-score) can be used to test the trained damage prediction model based on the test data to prevent model overfitting.

[0109] In some embodiments of this specification, the user interface may include, in addition to 3D models and vulnerable parts, information such as the risk of damage to vulnerable parts, the number of historical repairs of vulnerable parts, and the percentage of historical repairs of vulnerable parts. This specification does not make specific limitations on this.

[0110] To better display various types of information, the aforementioned user interface can display these types of information in separate areas. Specifically, the user interface may include a 3D container model display area, a vulnerable parts display area, and a vulnerable parts historical maintenance information display area. Furthermore, when the 3D model displayed in the 3D container model display area is a rendered 3D model with vulnerable parts labeled, the user interface may or may not include a vulnerable parts display area. The aforementioned 3D container model display area is used to display the 3D model corresponding to the container, i.e., a rendered 3D model or a rendered 3D model with vulnerable parts labeled. The vulnerable parts display area can display information about vulnerable parts, such as their name, type, location, and part number, and can be displayed in a list format. The order of the vulnerable parts in the list can be based on the risk of damage from highest to lowest, or it can be a random order; this manual does not specify a particular order. The aforementioned historical maintenance information display area for vulnerable components is used to display historical maintenance information for vulnerable components. This historical maintenance information may include the number of historical maintenance operations or the percentage of historical maintenance operations, and may also include the average maintenance cost. The percentage of historical maintenance operations refers to the ratio of the historical maintenance operations of this vulnerable component to the total historical maintenance operations of this vulnerable component across all containers. This historical maintenance information display area can present the historical maintenance information for vulnerable components in a list format, or in a graph format, such as a line graph showing the change in the number of historical maintenance operations or the percentage of historical maintenance operations over time. This area can also include a graph showing the change in the number of historical maintenance operations or the percentage of historical maintenance operations for each vulnerable component over time. This graph can be displayed as a line graph, with each graph corresponding to a vulnerable component. The horizontal axis of each graph represents time, and the vertical axis represents the number of historical maintenance operations or the percentage of historical maintenance operations.

[0111] In addition, the aforementioned user interface may also include a component damage risk display area. This area displays the component damage risk for each component, arranged in a list from highest to lowest risk. Furthermore, the component damage risk display area and the vulnerable component display area may be merged into a single vulnerable component display area, which would also display the damage risk for each vulnerable component. Of course, the vulnerable component display area, the vulnerable component historical maintenance information display area, and the component damage risk display area may also use other data visualization methods (such as pie charts) when displaying data; this specification does not impose specific limitations.

[0112] In some embodiments of this specification, the server may also use historical maintenance costs from historical maintenance data as labels, and other types of data besides historical maintenance costs from historical maintenance data as training samples. The training samples are input into a linear regression model to determine the prediction result output by the linear regression model. Based on the Huber loss function, ridge regression regularization, and elastic network feature selection, the target loss function of the linear regression model is determined, and the target loss is determined by applying the target loss function based on the prediction result and labels. The linear regression model is then trained based on the target loss, and the trained linear regression model is used to predict the maintenance cost of the target container based on the maintenance information of the target container.

[0113] The training samples mentioned above must include at least the number of historical repairs, and may also include data such as container size, container structure, cargo type, transportation route, historical repair time, historical repair frequency, historical damaged parts, types of historical damaged parts, causes of historical damage, historical repair location information, historical repair patterns, customer information, and post-repair effect evaluations. This specification does not impose specific limitations. The target loss function mentioned above is constructed using the Huber loss function, ridge regression regularization, and elastic network feature selection. That is, it introduces the Huber loss function, ridge regression regularization, and elastic network feature selection into the original least squares method. The Huber loss function enhances robustness, ridge regression regularization prevents overfitting, and elastic network feature selection ignores unimportant features. The introduction of ridge regression regularization can be achieved by adding L2 regularization to the Huber loss function, and the introduction of elastic network feature selection can be achieved by adding L1 regularization to the Huber loss function and L2 regularization. Therefore, the target loss function includes least squares, Huber loss function, L1 regularization, and L2 regularization. The aforementioned maintenance information includes at least the number of maintenance visits for the current target container, and may also include data such as container size, container structure, cargo type, transportation route, maintenance time, damaged parts, type of damaged parts, cause of damage, location of the reported maintenance, and maintenance mode. Furthermore, when there are multiple containers, the aforementioned training samples and labels can be generated based on the historical maintenance data of each container; that is, there can be multiple training samples and labels.

[0114] Based on this, the aforementioned user interface may also include a maintenance cost display area, which displays the predicted maintenance costs. In addition to displaying the predicted maintenance costs, the maintenance cost display area may also display the average maintenance cost, maintenance frequency, and average maintenance cost trend. This average maintenance cost trend may be a first trend graph showing changes over time, which can be displayed as a line graph, with the horizontal axis representing time and the vertical axis representing the average maintenance cost. Furthermore, this first trend graph may include the average maintenance cost of the container over a preset first time period. This preset first time period may be from several past years to the predicted year, or from several past months to the predicted month. The predicted year or month corresponds to the time period of the predicted maintenance costs. The specific number of past years or months, i.e., the number of years or months included in the first time period, can be set according to requirements and is not specifically limited in this specification. The average maintenance cost may be the average maintenance cost of the target container within a preset second time period, which may be pre-set, such as the past year. The aforementioned maintenance frequency can be the maintenance frequency of the target container within a preset third time period, which can also be pre-set, such as the past month. Of course, the aforementioned maintenance cost display area can also display the ratio of the predicted maintenance cost to the total maintenance cost, which can be the sum of historical maintenance costs and the predicted maintenance cost.

[0115] The aforementioned user interface can also display a second trend graph showing the average maintenance cost of all containers over time. This second trend graph can be displayed in the maintenance cost display area. The second trend graph can be displayed as a line graph. The horizontal axis of the second trend graph represents time, and the vertical axis represents the average maintenance cost, which is the average of the total maintenance cost of all containers. The second trend graph can also include the maintenance cost of containers within a preset fourth time period. This preset fourth time period can be the same as the preset first time period mentioned above. The specifics will not be elaborated here.

[0116] In addition, to enhance user interactivity and allow users to gain a deeper understanding of the container's structure and vulnerable components, the user interface can also include interactive animation control buttons and a container information display area. The container information display area shows the container's structural information, name, etc. The interactive animation control buttons include buttons for operations such as opening and closing doors, floor sliding out and in, and raising and lowering. Users can change the display state of the target container's 3D model by clicking these buttons. The server can use interactive driving technology to bind these interactive animation control buttons to the 3D model's actions, achieving comprehensive interactive functionality for the 3D model. This interactive driving technology results in an intuitive design that conforms to natural human vision. During the binding process, JavaScript's event binding mechanism is used to precisely associate operations such as opening and closing doors, floor sliding out and in, and raising and lowering with their corresponding buttons and 3D models. This is achieved by constructing an event-action response matrix, which supports flexible interaction logic. Users can trigger corresponding actions simply by clicking the buttons; the operation response is sensitive, the interaction logic is simple and easy to understand, and the convenience and smoothness of user operation are improved. Furthermore, users can rotate the 3D model and view detailed information about different parts of the model by clicking and dragging it. The interactive animation control buttons may also include a back button, which allows the 3D model to revert to the previous state or the initial state.

[0117] In some embodiments of this specification, the above-described user interface may be as follows: Figure 3 As shown, Figure 3 This is a schematic diagram of a user interface provided in this specification. Figure 3 The user interface shown includes a 3D container model display area, a vulnerable parts display area, a vulnerable parts historical maintenance information display area, and a maintenance cost display area. It should be noted that... Figure 3 This is just one example of a user interface; a user interface may include more than just this. Figure 3 The display area shown is only an example of the user interface, which includes the aforementioned 3D container model display area, vulnerable parts display area, vulnerable parts historical maintenance information display area, and maintenance cost display area. Figure 3Area 1 is a 3D container model display area. This area includes a 3D model in an open position. Area 1 also includes interactive animation control buttons labeled "WHOLE," "DOORS," "FLOOR," and "CROSSMEMBER." The "WHOLE" button is a back button; the "DOORS" button controls the opening and closing of the 3D model's doors; the "FLOOR" button controls the sliding of the floorboards; and the "CROSSMEMBER" button controls the raising and lowering of the 3D model. Additionally, Area 1 includes a "MoreReports" button, which leads to an information query page.

[0118] Figure 3 Area 2 is the display area for vulnerable components. This area only showcases vulnerable components from the contained components of the doors, displayed using part codes: “LBH”, “HGP”, “DFA”, “GTO”, “DSA”, “GIO”, “HGA”, “DRT”, “DSH”, “DST”, “DHC”, “DRL”, “DSS”, “LCK”, “DSC”, “LBL”, “DHR”, “LBR”, “GRS”, “DSO”, and “LBC”. It should be noted that due to the limitations of Area 2… Figure 3 Area 2 in the table only displays the part codes of some vulnerable components. Other vulnerable components' part codes can be displayed by sliding. The specific descriptions of the part codes shown in Area 2 are consistent with those in Table 1 above, and will not be repeated here.

[0119] Figure 3Area 3 is the historical maintenance information display area for vulnerable parts. This area only displays the historical maintenance information of the top 10 vulnerable parts. That is, the vulnerable parts are arranged in descending order of risk of damage, and the historical maintenance information of the top 10 vulnerable parts is displayed in a list format. Due to the limited display area of ​​the list, the list is scrollable, and the display area of ​​this list can display 6 vulnerable parts (i.e., "GTO", "GTA", "LBR", "DHC", "LBC", "DHR"). Each row of the list displays the information of one vulnerable part, including the part code, historical maintenance information, and a bar chart corresponding to the percentage of historical maintenance frequency. The historical maintenance information includes the average maintenance cost and the percentage of historical maintenance frequency. Taking "GTO" as an example, the average maintenance cost of this vulnerable part is "$6.13" and the percentage of historical maintenance frequency is "20.54%". Specific information for vulnerable parts will not be elaborated here. In addition, Area 3 also displays line graphs showing the historical number of repairs for some vulnerable components over time. The horizontal axis of this line graph represents time, which is preset, and the vertical axis represents the historical number of repairs. Area 3 only shows line graphs of the historical number of repairs for four vulnerable components: “GTO”, “GTA”, “LBR”, and “DHC”, over time.

[0120] Figure 3 Area 4 is the maintenance cost display area. This area displays the target container's maintenance cost, average maintenance cost, maintenance frequency, the ratio of predicted maintenance cost to total maintenance cost, and the average maintenance cost trend. The maintenance cost is "$5,506,243.71" in Area 4, the average maintenance cost is "$9.86" in Area 4, the maintenance frequency is "558,536" in Area 4, and the ratio of predicted maintenance cost to total maintenance cost is "8.45%" in Area 4. The average maintenance cost trend is displayed in the form of a line graph, and the line graph shows the average maintenance cost for some months in 2024 and 2025. The horizontal axis of the line graph represents time, and the vertical axis represents the average maintenance cost.

[0121] In some embodiments of this specification, users can rotate the 3D model in the 3D container model display area of ​​the user interface by dragging and dropping with the mouse. Specifically, it can be done as follows: Figure 4 As shown, Figure 4 This is a schematic diagram of a rotated 3D model provided in this specification. Figure 4 Two rotated 3D models were mainly showcased. Figure 4 This is merely one example of a rotated 3D model; the rotated 3D model in this specification is not limited to any other type. Figure 4 The two scenarios shown.

[0122] Users can control the opening and closing of doors in the 3D model by clicking the button marked "DOORS". Specifically, it can be done as follows: Figure 5 As shown, Figure 5 This is a schematic diagram of a 3D model of a door opening and closing provided in this specification. Figure 5 The 3D model displayed on the left shows the door in its open state, while the 3D model displayed on the right is a schematic diagram of a certain moment in the process of the door closing. Users can control the 3D model to change from the open state to the closed state, or vice versa, by clicking the button marked "DOORS".

[0123] Users can control the sliding of the floor in and out of the 3D model by clicking the button marked "FLOOR". Specifically, it can be done as follows: Figure 6 As shown, Figure 6 This is a schematic diagram of a 3D model showing the floor sliding out and sliding in, as provided in this specification. Figure 6 The left side of the image shows the 3D model's floor sliding out, while the right side shows a diagram of a specific moment during the floor's sliding-in process. Users can control the 3D model's floor to slide in and out by clicking the button marked "FLOOR".

[0124] Users can control the 3D model's ascent and descent by clicking the button marked "CROSSMEMBER," specifically as follows: Figure 7 As shown, Figure 7 This is a schematic diagram illustrating the ascent and descent of a 3D model provided in this specification. Figure 7 The 3D model displayed on the left is in an ascending state, while the 3D model displayed on the right is a schematic diagram of a certain moment in the process of transitioning from the ascending state to the descending state, that is, a schematic diagram of a moment during the descent. Users can control the ascent and descent of the 3D model by clicking the button marked "CROSSMEMBER".

[0125] In some embodiments of this specification, the information query page described above allows users to query corresponding historical maintenance data by selecting the value range corresponding to each factor field. Specifically, as shown below... Figure 8 As shown, Figure 8 This is a schematic diagram of an information query page provided in this specification. Figure 8This includes subplots c and d. As shown in subplot c, the user-selectable factor fields include "Begin Date," "End Date," "Region," "Area," "Segment," "Port," "Depot," "Customer," "Equipment Group," "Equipment Type," "Assembly," "ComponentGroup," and "Component." Clicking the "Export" button exports the underlying data—an Excel file—that matches the range of values ​​for the selected factor fields. Region 1 in subplot c includes "Begin Date" and "End Date," with "Begin Date" set to 2021-01-01 and "End Date" set to 2025-06-11. The lower left area of ​​sub-chart c displays an overview of the query results, including TotalAmount (USD) (total repair cost, in US dollars), TotalFrequency (total number of repairs), and Average Repair Cost (USD). The Average Repair Cost (USD) is calculated by dividing the total repair cost (i.e., total repair expenses) by the total number of repairs. The total repair cost can be further illustrated using a pie chart, showing the breakdown of repair costs and percentages for each responsible party, including the amount and percentage borne by the User, Owner, and Insurance. The lower right area of ​​sub-chart c (area 2 in sub-chart c) displays the trend of the total repair cost over time using a line chart. The horizontal axis represents time, and the vertical axis represents the repair cost (in US dollars), showing trends for four time dimensions: Yearly, Monthly, Weekly, and Daily. Users can switch between viewing the yearly, monthly, weekly, and daily trends by clicking the tabs (Yearly, Monthly, Weekly, Daily). The line chart can also display the historical average repair cost across different time dimensions using dashed lines. Additionally, the information query page can include sub-chart d, where the "Repair Cost of Selected Customers" area displays the total or average repair cost for the selected customers (i.e., "TSLINE", "EGREN", and "WH" in sub-chart d) over the years in a bar chart format. The horizontal axis represents the year, and the vertical axis represents the repair cost (USD). Users can switch tabs to view "Total" (total repair cost) and "Average" (average repair cost) separately.Users can click the "Export Customer Total / Avg Price" button to export the customer's total repair cost and average repair cost data over the years, which will be displayed in an Excel file.

[0126] In some embodiments of this specification, to address the complexity of the target container model structure, model structure optimization techniques can be employed during 3D model construction to rationally layer and simplify the data of the 3D model. While ensuring the functional integrity and detail of the 3D model, unnecessary polygon counts are reduced, data redundancy is eliminated, and the loading speed and operating efficiency of the 3D model are significantly improved. Furthermore, clear hierarchical management ensures that operations on different parts of the 3D model do not interfere with each other, facilitating precise operation of specific components by the user. Specifically, the server can use 3D modeling software, combined with non-uniform rational B-splines (NURBS) curves, to reconstruct the geometry of key components of the target container based on its actual size and structure, constructing a 3D model of the target container and ensuring the accuracy of the constructed 3D model. Simultaneously, a topology optimization algorithm based on the SIMP (Solid Isotropic Material with Penalization) method is used to automatically reduce redundant polygons, optimizing the constructed 3D model.

[0127] In some embodiments of this specification, when rendering the annotated 3D model in S5 using physically based rendering (PBR) technology, WebGL (Web Graphics Library) technology and PBR technology can be used to render the 3D model on the user interaction page. By deeply applying WebGL, browser rendering limitations are overcome, enabling real-time rendering of the 3D model in various mainstream browsers, decoupling rendering speed and accuracy from the local environment. It should be noted that the aforementioned user interaction page is displayed in a browser. WebGL is a JavaScript API for rendering interactive 2D and 3D graphics in web browsers, requiring no additional plugin support. Based on this, the aforementioned WebGL can be optimized code. The server can fully exploit the hardware acceleration potential by optimizing the way the box details are expressed in the WebGL code, ultimately achieving a clear presentation of the box's complex decoupled details and repair process at an ultra-high frame rate. Through the aforementioned WebGL technology and PBR technology, the 3D model is rendered with vibrant colors and clear outlines, providing repair workers with an intuitive and smooth visual presentation, significantly improving the operating experience. The aforementioned physically based rendering (PBR) technology can be implemented through a physically based rendering engine. This engine, based on a physically based PBR material shading system, achieves lighting effects close to those of a real industrial environment through HDR (High Dynamic Range) environment mapping, giving 3D models higher visual fidelity. PBR technology, which simulates real-world lighting behavior based on physical laws, calculates lighting effects using 3D models and physical principles to achieve more realistic and natural image rendering. Specifically, PBR technology can intuitively reflect the Fresnel effect on the surface of a container under different light intensities, simulating the realistic texture of various materials during container repair, such as the gloss of metallic containers.

[0128] In some embodiments of this specification, the server may also employ damage feature visualization technology, using a heatmap mapping algorithm to mark the component damage risk of vulnerable parts on the 3D model. Additionally, the server may perform a time-axis-based dynamic simulation of the corrosion process on the 3D model and display it on the user interface to intuitively demonstrate the aging trend of the target container.

[0129] In some embodiments of this specification, to make the 3D model's movements more natural and fluid, specifically to make the 3D model's movements more natural and fluid when responding to the aforementioned interactive animation control buttons, advanced animation transition techniques can be employed. Specifically, state machines and quaternion interpolation algorithms are used to achieve non-linear mapping of the 3D model across multiple dimensions and perspective changes. Furthermore, the Tween.js library is used to implement Bézier curve motion trajectory control, making the animation movements more natural. Simultaneously, an animation progress feedback system is used to display the 3D model's state changes in real time, accurately reproducing various dynamic characteristics during the 3D model's movement. The server can use animation transition techniques to control the animation of each component, using the Tween.js engine (i.e., the Tween.js library) to obtain animation progress in real time, update component positions, and mark the animation completion status.

[0130] The Tween.js library provides powerful tweening capabilities, enabling precise control over the animation of individual parts of a 3D model. Regarding easing functions, the easing module defines various easing effects, such as linear easing, which keeps the 3D model at a constant speed during movement, and quadratic easing (easeInQuad, easeOutQuad, and easeInOutQuad). easeInQuad starts the 3D model's movement slowly and gradually accelerates; easeOutQuad starts the movement quickly and decelerates near the end; and easeInOutQuad slows the 3D model at the beginning and end of the movement, while increasing the speed in the middle. These different easing functions simulate motion states that better conform to the physical laws of the real world, making the 3D model's movements more realistic.

[0131] Specifically, in terms of timeline control, Tween.js's timeline management mechanism allows for precise setting of the duration of each animation action using its `duration` property. For example, in the animation of the floor sliding out, setting `duration` to 500 milliseconds ensures the floor completes the sliding motion within 0.5 seconds. Simultaneously, combined with the `onUpdate` callback function, the current animation progress of the 3D model is retrieved in real-time with each frame update, and its position, rotation angle, and other properties are updated accordingly, achieving smooth movement of the 3D model. For instance, in the door opening animation, the `onUpdate` function adjusts the door's rotation angle in real-time based on the animation progress, ensuring a smooth and natural opening process.

[0132] In addition, the onComplete callback function is triggered when the animation is complete, and is used to perform some operations after the animation ends. For example, after the floor slides in, the onComplete function marks the floor's state as "in place", and can also trigger relevant prompts (such as "completed") and display them to the user to inform the user that the operation is complete, ensuring the accuracy and consistency of the 3D model's state under complex operations, and providing the user with a stable and reliable operating experience.

[0133] In some embodiments of this specification, considering the dependence of 3D rendering itself on the performance of the hardware graphics card, the server can optimize the device performance during the actual delivery of the 3D model. For example, strategies such as resource preloading and dynamic memory management can be adopted to further improve the running performance of the 3D model on the device, so that users can perform repair operations smoothly regardless of what device they use.

[0134] In some embodiments of this specification, in step S1 above, after collecting historical maintenance data, the server can perform data cleaning on the collected historical maintenance data and execute the processes shown in S2 to S5 above on the cleaned data to avoid outliers or missing values, ensuring the accuracy and completeness of the data, and providing data support for subsequent big data analysis, key factor field determination, and component damage risk prediction. The data cleaning includes outlier detection and missing value handling, and may also include other data processing methods, which are not specifically limited in this specification. Outlier detection includes detecting null values ​​and data exceeding a preset range, and missing value handling includes filling with the average value or filling with a preset value.

[0135] Based on the above Figure 1 The present invention illustrates a method for identifying and 3D displaying vulnerable components of a container. This specification may also provide a system for identifying and 3D displaying vulnerable components of a container, such as... Figure 9 As shown, Figure 9 This diagram illustrates a container vulnerable component identification and 3D display system provided in this specification. The system includes a data analysis module 100 and a 3D display module 101. The data analysis module 100 includes a data acquisition submodule 1001, a feature selection submodule 1002, a risk prediction submodule 1003, and a vulnerable component determination submodule 1004. The data acquisition submodule 1001 is used to collect historical maintenance data of the container. The execution process of the data acquisition submodule 1001 is similar to the execution process of S1 described above, and will not be repeated here.

[0136] The aforementioned feature selection submodule 1002 is used to determine the importance of each factor field included in the historical maintenance data using a random forest model, and to filter each factor field according to its importance to determine the key factor fields. The execution process of the aforementioned feature selection submodule 1002 is similar to the execution process of S2, and will not be described again here.

[0137] The aforementioned risk prediction submodule 1003 is used to determine the correlation between historically damaged components and historical causes of damage to the container based on historical maintenance data. Based on the determined correlation and pre-set correlation rules, it generates correlation features and uses these features, key factor fields, and the current static features of the target container as input parameters to a pre-trained damage prediction model to predict the component damage risk of the target container. The execution process of the aforementioned risk prediction submodule 1003 is similar to that of S3 described above, and will not be repeated here.

[0138] The aforementioned vulnerable component determination submodule 1004 is used to determine the vulnerable components of the target container based on the component damage risk. The execution process of the aforementioned vulnerable component determination submodule 1004 is similar to the execution process of S4, and will not be described again here.

[0139] The aforementioned 3D visualization module 101 is used to construct a 3D model corresponding to the target container using 3D modeling software, and to visualize the vulnerable parts and the 3D model. The execution process of the aforementioned 3D visualization module 101 is similar to the execution process of S5, and will not be described again here.

[0140] In some embodiments of this specification, the system further includes a maintenance cost prediction module 102, as detailed below. Figure 11 As shown, the maintenance cost prediction module 102 uses historical maintenance costs from historical maintenance data as labels and other types of data from historical maintenance data besides historical maintenance costs as training samples. The training samples are input into a linear regression model to determine the prediction result output by the linear regression model. Based on the Huber loss function, ridge regression regularization, and elastic network feature selection, the target loss function of the linear regression model is determined, and the target loss is determined using the target loss function based on the prediction result and labels. The linear regression model is trained based on the target loss, and the trained linear regression model is used to predict the maintenance cost of the target container based on its maintenance information.

[0141] In some embodiments of this specification, the data analysis module further includes a data extraction submodule 1005, which can be described as follows: Figure 10 and Figure 11As shown, the data extraction submodule 1005 uses a big data analysis algorithm to cluster historical maintenance data according to the field values ​​corresponding to each factor field included in the historical maintenance data, obtaining data groups. The field values ​​corresponding to each factor field in the data within each data group are the same. Then, the feature selection submodule 1002 specifically constructs a random forest model with damaged component type and maintenance method as target variables based on each data group. Based on the random forest model, the Gini impurity and SHAP value corresponding to each factor field are determined. Based on the Gini impurity and SHAP value, the importance of each factor field is determined. Based on the importance of each factor field, the factor fields are filtered to determine the key factor fields.

[0142] In some embodiments of this specification, the risk prediction submodule 1003 can determine the historical frequency of each correlation in historical maintenance data. A correlation matrix is ​​constructed based on the historical frequency of each correlation. Correlation features for each dimension are generated based on the correlation matrix and pre-set correlation rules. The correlation rules include confidence levels corresponding to various correlations, and each dimension includes at least the damage frequency of historically damaged components in the correlation, the causal correlation score of the correlation, and the historical frequency of the correlation. The causal correlation score is the confidence level corresponding to the correlation in the correlation rules.

[0143] In some embodiments of this specification, the 3D visualization display module 101 includes a 3D modeling submodule 1011, a user interaction submodule 1012, and a physically based rendering submodule 1013, as detailed below. Figure 9 and Figure 11 As shown. The 3D modeling submodule 1011 can be used to construct a 3D model corresponding to the target container using 3D modeling software. Vulnerable parts are annotated on the 3D model. The physically based rendering submodule 1013 is used to render the annotated 3D model using physically based rendering (PBR) technology. The user interaction submodule 1012 is used to visually display the rendered 3D model on the user interface.

[0144] In some embodiments of this specification, the system further includes a feedback module 103, as detailed below. Figure 11 As shown, the above-mentioned feedback module 103 is used to display the feedback page to the user, receive the feedback submitted by the user through the feedback page, and optimize the visual display based on the feedback, such as optimizing the layout of each display area in the user interaction interface, or optimizing the display format of the display area, etc.

[0145] In some embodiments of this specification, the data analysis module 100 may further include a data cleaning submodule 1006, such as... Figure 10 As shown, Figure 10 This is a schematic diagram of the structure of a data analysis module provided in this specification. The data cleaning submodule 1006 is used to clean the historical maintenance data collected by the data collection submodule 1001 and send the cleaned historical maintenance data to the data extraction submodule 1005.

[0146] In some embodiments of this specification, such as Figure 11 As shown, Figure 11 This is a schematic diagram of another container vulnerable component identification and 3D display system provided in this specification. Figure 11 The system shown is in Figure 9 Based on the system shown, it also includes the aforementioned data extraction submodule 1005, the aforementioned maintenance cost prediction module 102, and the feedback module 103. Of course, the system may also include a data security module 104, which can be used for data encryption and network security protection to ensure data security. Specifically, advanced data encryption and network security measures can be adopted to ensure the security and stability of system data and prevent data leakage and unauthorized access.

[0147] It should be noted that the specific embodiments described above enable those skilled in the art to more fully understand the present invention, but do not limit the present invention in any way. Therefore, although the present invention has been described in detail with reference to the accompanying drawings and embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention. In short, all technical solutions and improvements that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the present invention patent.

Claims

1. A method for identifying and 3D displaying vulnerable components of a container, characterized in that, include: S1: Data Collection: Collect historical maintenance data for containers; S2: Feature selection: Based on the historical maintenance data, a random forest model is used to determine the importance of each factor field included in the historical maintenance data, and the factor fields are filtered according to the importance of each factor field to determine the key factor fields; S3: Predict component damage risk: Based on the historical maintenance data, determine the correlation between the historical damaged components and the historical causes of damage to the container. Based on the determined correlation and the pre-set correlation rules, generate correlation features. Use the correlation features, the key factor fields, and the current static features of the target container as input parameters for a pre-trained damage prediction model to predict the component damage risk of the target container. S4: Identify vulnerable components: Based on the risk of damage to the components, identify the vulnerable components of the target container; S5: Modeling and Visualization: Using 3D modeling software, construct a 3D model corresponding to the target container, and visualize the vulnerable parts and the 3D model.

2. The method for identifying and 3D displaying vulnerable components of a container according to claim 1, characterized in that, The method further includes: Use the historical maintenance costs in the historical maintenance data as labels, and use other types of data in the historical maintenance data other than the historical maintenance costs as training samples; The training samples are input into a linear regression model to determine the prediction results output by the linear regression model. Based on the Huber loss function, ridge regression regularization, and elastic network feature selection, the target loss function of the linear regression model is determined, and the target loss is determined by using the target loss function based on the prediction results and labels. Based on the target loss, a linear regression model is trained, and the trained linear regression model is used to predict the maintenance cost of the target container based on the maintenance information of the target container.

3. The method for identifying and 3D displaying vulnerable components of a container according to claim 1, characterized in that, S2 specifically includes: Using big data analytics algorithms, the historical maintenance data is clustered according to the field values ​​corresponding to each factor field included in the historical maintenance data to obtain data groups; the field values ​​corresponding to each factor field in the data of each data group are the same; Based on the data sets, a random forest model is constructed with the type of damaged component and the repair method as the objective variables; Based on the random forest model, determine the Gini impurity and SHAP value corresponding to each factor field; The importance of each factor field is determined based on the Gini impurity and SHAP value. Based on the importance of each factor field, the factor fields are filtered to determine the key factor fields.

4. The method for identifying and 3D displaying vulnerable components of a container according to claim 1, characterized in that, The key factor fields include historical repair patterns, historical repair frequency, historical repair counts for each component, repair location information, and historical repair costs.

5. The method for identifying and 3D displaying vulnerable components of a container according to claim 1, characterized in that, The visualization of the vulnerable components and the 3D model in step S6 specifically includes: The vulnerable components are marked on the 3D model; The annotated 3D model is rendered using Physically Based Rendering (PBR) technology and displayed on the user interface.

6. The method for identifying and 3D displaying vulnerable components of a container according to claim 1, characterized in that, The generation of association features in step S3, based on the determined association relationships and pre-set association rules, specifically includes: Determine the historical frequency of each relationship in the historical maintenance data; Construct a relationship matrix based on the historical frequency of each of the aforementioned relationships; Based on the relationship matrix and the pre-set association rules, association features of each dimension corresponding to each association are generated; wherein, the association rules include confidence levels corresponding to various associations, and each dimension includes at least the damage frequency of historically damaged parts in the association, the causal association score of the association, and the historical occurrence frequency of the association, and the causal association score is the confidence level corresponding to the association in the association rules.

7. A container vulnerable component identification and 3D display system according to any one of claims 1 to 6, characterized in that, The system includes a data analysis module and a 3D display module. The data analysis module includes a data acquisition submodule, a feature selection submodule, a risk prediction submodule, and a vulnerable component identification submodule, wherein: The data acquisition submodule is used to collect historical maintenance data of the container; The feature selection submodule is used to determine the importance of each factor field included in the historical maintenance data based on the historical maintenance data using a random forest model, and to filter the factor fields according to the importance of each factor field to determine the key factor fields. The risk prediction submodule is used to determine the correlation between historically damaged components and historical causes of damage to the container based on the historical maintenance data, generate correlation features based on the determined correlation and pre-set correlation rules, and use the correlation features, the key factor fields and the current static features of the target container as input parameters for a pre-trained damage prediction model to predict the component damage risk of the target container. The vulnerable component determination submodule is used to determine the vulnerable components of the target container based on the risk of damage to the components. The 3D visualization module is used to construct a 3D model of the target container using 3D modeling software, and to visualize the vulnerable parts and the 3D model.

8. A container vulnerable component identification and 3D display system according to claim 7, characterized in that, The system also includes a maintenance cost prediction module; The maintenance cost prediction module is used to use historical maintenance costs from the historical maintenance data as labels and other types of data from the historical maintenance data besides the historical maintenance costs as training samples; input the training samples into a linear regression model to determine the prediction result output by the linear regression model; determine the target loss function of the linear regression model based on the Huber loss function, ridge regression regularization, and elastic network feature selection, and determine the target loss based on the prediction result and labels using the target loss function; train the linear regression model based on the target loss, and the trained linear regression model is used to predict the maintenance cost of the target container based on the maintenance information of the target container.

9. A container vulnerable component identification and 3D display system according to claim 7, characterized in that, The data analysis module also includes a data extraction submodule; The data extraction submodule is used to cluster the historical maintenance data according to the field values ​​corresponding to each factor field included in the historical maintenance data using big data analysis algorithms to obtain data groups; the field values ​​corresponding to each factor field in the data of each data group are the same; The feature selection submodule is specifically used to construct a random forest model with the type of damaged parts and the repair method as target variables based on the data groups; and to determine the Gini impurity and SHAP value corresponding to each factor field based on the random forest model. The importance of each factor field is determined based on the Gini impurity and SHAP value. Based on the importance of each factor field, the factor fields are filtered to determine the key factor fields.

10. A container vulnerable component identification and 3D display system according to claim 7, characterized in that, The 3D visualization display module includes a 3D modeling submodule, a user interaction submodule, and a physical rendering submodule; The 3D modeling submodule is used to construct a 3D model corresponding to the target container using 3D modeling software; and to mark the vulnerable parts on the 3D model. The physical rendering submodule is used to render the annotated 3D model using physically based rendering (PBR) technology. The user interaction submodule is used to visually display the rendered 3D model on the user interaction interface.

11. A container vulnerable component identification and 3D display system according to claim 7, characterized in that, The system also includes a feedback module; The feedback module is used to display a feedback page to users, receive feedback submitted by users through the feedback page, and optimize the visual display based on the feedback.

12. The container vulnerable component identification and 3D display system according to claim 7, characterized in that, The risk prediction submodule is specifically used to determine the historical frequency of each correlation in the historical maintenance data; and to construct a correlation matrix based on the historical frequency of each correlation. Based on the relationship matrix and the pre-set association rules, association features of each dimension corresponding to each association are generated; wherein, the association rules include confidence levels corresponding to various associations, and each dimension includes at least the damage frequency of historically damaged parts in the association, the causal association score of the association, and the historical occurrence frequency of the association, and the causal association score is the confidence level corresponding to the association in the association rules.

Images