Mold cost calculation method and device, equipment and storage medium

Abstract

The embodiment of the invention relates to the field of cost calculation, and discloses a mold cost calculation method and device, equipment and a storage medium, and the method comprises the steps: obtaining multi-modal data related to the manufacturing of a target mold; extracting feature data related to the manufacturing cost of the target mold from the multi-modal data; based on a preset time-space diagram network model, extracting time sequence process data from the feature data, analyzing a first relationship between the time sequence process data and the cost, and calculating a first cost according to the first relationship; and / or extracting spatial process data from the feature data based on the space-time diagram network model, analyzing a second relationship between the spatial process data and the cost, and calculating a second cost according to the second relationship. According to the method and the device, the manufacturing data of different links are gathered, and the first relationship between the process parameters and the cost along with the time and the second relationship between the correlation between different nodes and the cost are captured by using the preset time-space diagram network model, so that the cost calculation accuracy is improved.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of cost calculation, and particularly to a method, device, equipment and storage medium for calculating the cost of a mold. Background Art

[0002] The mold industry plays a crucial role in the manufacturing industry and is a key link in product forming. However, mold cost control faces multiple challenges, which seriously restrict the refined management and profitability of enterprises.

[0003] In the process of mold manufacturing, it involves multiple links such as design, processing, assembly, and inspection. The data of each link is often independent of each other, forming information islands. According to industry surveys, the utilization rate of manufacturing all-factor data is insufficient. The fragmentation of data makes it difficult for enterprises to comprehensively and systematically grasp the cost composition and change rules, and it is difficult to achieve refined cost management. Summary of the Invention

[0004] The purpose of the present invention is to provide at least a method, device, equipment and storage medium for calculating the cost of a mold, which can at least solve the technical problem that in the process of mold manufacturing, the utilization rate of manufacturing all-factor data is insufficient, it is difficult to comprehensively and systematically grasp the cost composition and change rules, and achieve refined cost management, and can at least achieve the effect of calculating the manufacturing cost of the mold more precisely.

[0005] To solve the above technical problem, at least one embodiment of the present application provides a method for calculating the cost of a mold, including: obtaining multimodal data related to the manufacture of a target mold; extracting feature data related to the manufacturing cost of the target mold from the multimodal data; based on a preset spatio-temporal graph network model, extracting temporal process data from the feature data, and analyzing the first relationship between the temporal process data and the cost as it changes over time, and calculating the first cost of the target mold according to the first relationship; and / or based on the spatio-temporal graph network model, extracting spatial process data from the feature data, and analyzing the second relationship between the spatial process data and the cost, and calculating the second cost of the target mold according to the second relationship, where the spatial process data includes interrelated process parameters and / or interrelated equipment.

[0006] At least one embodiment of the present application also provides a device for calculating the cost of a mold, including: a data acquisition module for obtaining multimodal data related to the manufacture of a target mold; a feature data extraction module for extracting feature data related to the manufacturing cost of the target mold from the multimodal data; a first cost calculation module, for extracting time series process data from the feature data based on a preset spatiotemporal graph network model, analyzing a first relationship between the time series process data and the cost over time, and calculating a first cost of the target mold according to the first relationship; and / or A second cost calculation module is used to extract spatial process data from the feature data based on the space-time graph network model, and analyze the second relationship between the spatial process data and the cost, and calculate the second cost of the target mold according to the second relationship, wherein the spatial process data includes interrelated process parameters and / or interrelated equipment.

[0007] At least one embodiment of the present application also provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned mold cost calculation method.

[0008] At least one embodiment of the present application further provides a computer-readable storage medium storing a computer program, wherein the computer program implements the above-mentioned mold cost calculation method when executed by a processor.

[0009] The mold cost calculation method, device, equipment and storage medium provided in the embodiments of the present application obtain multimodal data related to the target mold manufacturing: actively collect heterogeneous data (such as design drawings, processing procedures, processing parameters, etc.) covering multiple links such as design, processing, assembly, and testing, and bring together data originally scattered in different departments, systems and links, breaking the physical and logical "information islands" from the source, and providing a data basis for subsequent comprehensive analysis; identifying and extracting key features that are strongly correlated with costs, and improving the "value density" and pertinence of data; in response to the pain point of "difficult to fully grasp the cost structure and change rules", a preset spatiotemporal graph network model is used to capture the change rules of process parameters, equipment status, resource consumption, etc. over time, analyze the first relationship between the change rules and costs, and solve the problem that the dynamic change rules in the cost structure are difficult to capture; and / or analyze the second relationship between the mutual correlation between different nodes and costs, such as how design complexity affects the selection of processing parameters, how processing accuracy affects assembly difficulty and rework rate, etc., to quantify how a design decision affects the cost of all subsequent links through a chain reaction. The first cost, second cost or their combination obtained by the final cost calculation is no longer a simple addition of the costs of separate links, but a refined calculation result that integrates the dynamic interaction of the entire process and all factors. This enables enterprises to fully and systematically grasp the real cost structure and achieve refined cost management.

[0010] In some alternative embodiments, calculating the first cost of the target mold according to the first relationship and calculating the second cost of the target mold according to the second relationship include: Constructing a process-cost association matrix according to the first relationship and the second relationship, where the process-cost association matrix is used to quantitatively analyze the degree and direction of the influence of feature data on cost; Calculating the first cost and the second cost according to the process-cost association matrix.

[0011] In this embodiment, the matrix integrates multi-dimensional features such as design, process, equipment, and environment, captures the interaction effects between features through matrix operations, and dynamically updates the matrix parameters when the process is adjusted (such as replacing materials or new equipment), and immediately generates the cost prediction of the new solution.

[0012] In some alternative embodiments, constructing the process-cost association matrix according to the first relationship and the second relationship includes: abstracting each of the feature data and cost elements into each node in the spatio-temporal graph structure by using the spatio-temporal graph network model, where the cost elements include at least one of raw material procurement cost, processing cost, and processing loss cost; analyzing the historical sequence data of each node through the time series model in the spatio-temporal graph network model to obtain the first relationship between the feature data and the cost elements; aggregating the information of neighbor nodes through the graph neural network model in the spatio-temporal graph network model to obtain the second relationship between the feature data and the cost elements; constructing the process-cost association matrix according to the first relationship and the second relationship.

[0013] Abstracting each of the feature data and cost elements into each node in the spatio-temporal graph structure by using the spatio-temporal graph network model can present the originally complex and disordered feature data and cost element information in an intuitive and structured way. This structured representation helps subsequent analysis and processing of data, enabling the complex process-cost relationship to be presented in a clearer graph structure form, facilitating understanding and operation.

[0014] In some alternative embodiments, the method further includes: Extracting structured data from the feature data, where the type of the structured data includes at least one of geometric complexity, process path, and material properties; Calculating the third cost of the target mold according to the contribution degree of the structured data to the cost.

[0015] In this embodiment, the structured data is static index parameters related to cost, with clear formats and definitions. The data types include geometric complexity, process path, material properties, etc. These data usually exist in the forms of standard tables, database fields, etc. Each data item has a clear meaning and value range. The structured data provides a benchmark for cost calculation based on the essential characteristics of the mold, making the mold cost calculation result more comprehensive.

[0016] In some alternative embodiments, extracting the structured data from the feature data includes at least one of the following: extracting the geometric complexity through 3D point cloud curvature analysis; and extracting the relevant parameters under the process path based on the numerical control machining instruction library, where the relevant parameters include at least one of path length, feed rate, and cutting depth; and obtaining the material properties through dynamic hardness map analysis, where the dynamic hardness map is used to simulate the change of material hardness with position, temperature, time, or strain rate under actual machining conditions.

[0017] In this embodiment, the geometric complexity feature is extracted through 3D point cloud curvature analysis, accurately quantifying the mold geometric features, reflecting the complexity of the mold shape and its impact on machining difficulty and cost; the numerical control machining instruction library contains rich machining information. By extracting the relevant parameters under the process path, such as path length, feed rate, and cutting depth, etc., the process details in the mold manufacturing process can be comprehensively understood; the dynamic hardness map can simulate the change of material hardness with position, temperature, time, or strain rate under actual machining conditions. The material properties obtained through dynamic hardness map analysis are closer to the true performance of the material in the actual machining process, providing more accurate material property data for cost calculation and process design.

[0018] In some alternative embodiments, the method further includes: performing weighted summation on the first cost, the second cost, and the third cost to obtain the comprehensive cost of manufacturing the target mold.

[0019] In this embodiment, the first cost is calculated based on the relationship between the time-series process data and the cost, reflecting the impact of the process change over time on the cost; the second cost is calculated based on the relationship between the spatial process data and the cost, reflecting the role of the spatial correlation between the process parameters and the equipment on the cost; the third cost is calculated according to the contribution degree of the structured data (such as geometric complexity, process path, material properties) to the cost, considering the impact of the inherent attributes of the mold itself on the cost. Through weighted summation, the cost information in these three different dimensions is integrated, avoiding the one-sidedness that may be caused by single-dimensional cost calculation, and enabling the comprehensive cost to more comprehensively reflect various cost factors in the mold manufacturing process.

[0020] In some alternative embodiments, the calculation formula for the comprehensive cost of the target mold includes the following:

[0021] Wherein, is the comprehensive cost of the target mold, , and respectively represent prediction functions based on XGBoost, LSTM, and graph convolutional network, is the structured data, is the time-series process data, is the spatial process data, , and are gating coefficients, which are dynamically adjusted through a preset attention mechanism.

[0022] In this embodiment, the attention mechanism can dynamically adjust the gating coefficients (i.e., weights) of XGBoost, LSTM, and GCN according to the characteristics of the input data. This dynamic adjustment enables the model to automatically assign different importances to different models in different situations. For example, when the structured data has a greater impact on the cost, the attention mechanism will increase the gating coefficient of XGBoost; when the change in the time-series process data has a significant impact on the cost, the gating coefficient of LSTM will be increased. This enables the method of the present application to adapt to complex and changing manufacturing environments, such as equipment status, material supply, process adjustment, etc. By dynamically adjusting the gating coefficients through the attention mechanism, the comprehensive cost calculation method can adapt to these complex and changing environments, adjust the model weights in real time, and ensure that the cost prediction always maintains a high level of accuracy.

[0023] In some alternative embodiments, after obtaining the comprehensive cost of manufacturing the target mold, it further includes: Based on the comprehensive cost and real-time production line data, a cost control strategy is calculated using a preset reinforcement learning algorithm. The cost control strategy includes at least one of setting process parameters, material selection, and optimizing processing paths. The reward function of the reinforcement learning algorithm includes the degree of cost reduction and process quality; The cost control strategy is verified using a preset digital twin model of the production line, and the cost control strategy is adjusted according to the verification results until the verification results meet the preset standards to obtain the final cost control strategy.

[0024] In this embodiment, the digital twin model of the production line is a virtual mapping of the actual production line, which can verify the cost control strategy without interfering with actual production. By simulating the implementation process of the strategy in the digital twin model, potential problems and risks, such as equipment conflicts and process mismatches, can be discovered in advance, avoiding adverse consequences such as increased costs, decreased quality, or production interruptions in actual production.

[0025] In some alternative embodiments, extracting the feature data related to the target mold manufacturing cost from the multimodal data includes: Based on the multimodal data, initial feature data related to cost calculation is extracted; The initial feature data is subjected to feature dimension compression to obtain the final feature data related to cost calculation.

[0026] In this embodiment, high-dimensional data is difficult to directly visualize. However, after reducing the data to two-dimensional or three-dimensional space through dimension compression, the relationship and distribution of the data can be more intuitively displayed. For example, visualization tools such as scatter plots and line charts can be used to show the relationship between the compressed feature data and the cost, helping to better understand the influencing factors of the cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] One or more embodiments are exemplarily illustrated by the pictures in the corresponding drawings, and these exemplary illustrations do not limit the embodiments.

[0028] Figure 1 is a flowchart of a mold cost calculation method provided by an embodiment of the present application; Figure 2 is a schematic structural diagram of a mold cost calculation system provided by an embodiment of the present application; Figure 3 is a schematic diagram of the optimization effect of a mold cost calculation system provided by another embodiment of the present application; Figure 4 is a schematic diagram of a mold cost calculation device provided by another embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0029] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the following will elaborate on each embodiment of this application in conjunction with the accompanying drawings. However, those of ordinary skill in the art can understand that in the embodiments of this application, many technical details are presented to help readers better understand this application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical solutions claimed in this application can still be implemented. The division of the following embodiments is for convenience of description and should not impose any limitation on the specific implementation of this application. The various embodiments can be combined and cross-referenced with each other on the premise of not being contradictory.

[0030] The mold industry plays a crucial role in manufacturing and is a key link in product shaping. However, mold cost control faces multiple challenges, severely restricting the refined management and profitability of enterprises. Through in-depth research on the industry status quo, it is found that there are three core dilemmas in current mold cost control: First of all, the industry generally faces a serious dilemma of relying on experience. According to data released by the China Die & Mould Industry Association in 2023, 78% of enterprises still use manual experience to estimate mold costs. This method that relies on manual experience is not only inefficient but also inaccurate, easily leading to a large deviation between cost estimation and reality. Many enterprises report that the method of amortizing mold costs is difficult because molds are part of the cost of producing product components, while general cost accounting takes finished products or semi-finished products as cost objects, which naturally causes difficulties in amortizing mold costs. This dilemma results in enterprises lacking accurate cost information support when making decisions and being unable to conduct scientific cost management and optimization.

[0031] Secondly, the lack of dynamic response ability has become a key factor restricting the efficiency of mold cost control. Traditional cost control systems often adopt batch processing and periodic adjustment methods, and the system adjustment lag time usually reaches 4 - 6 hours. In a rapidly changing market environment, this lag makes it difficult for enterprises to promptly respond to the impact of external factors such as raw material price fluctuations and market demand changes on costs. Especially against the backdrop of continuously rising raw material prices currently, cost control has become a core element in the operation and management of mold enterprises. Therefore, establishing a cost control system that can perceive in real time and respond quickly has become an urgent problem in the industry.

[0032] Third, the problem of data islands severely restricts the depth and breadth of mold cost management. In the mold manufacturing process, multiple links such as design, processing, assembly, and inspection are involved, and the data of each link is often independent of each other, forming information islands. According to industry surveys, the utilization rate of all-factor manufacturing data is insufficient. The fragmentation of data makes it difficult for enterprises to comprehensively and systematically grasp the cost composition and change rules, and it is difficult to achieve refined cost management. Calculating the cost of the original mold requires a large amount of manual data statistics, and the repetitive labor is serious. This not only increases the management cost but also reduces the decision-making efficiency.

[0033] To solve the above technical problem of insufficient accuracy in calculating the cost of molds, the present invention proposes a method for calculating the cost of molds. The following specifically describes the implementation details of the method for calculating the cost of molds in this embodiment. The following content is only the implementation details provided for convenient understanding and is not necessary for implementing this solution.

[0034] Embodiment 1: The method for calculating the cost of molds in this embodiment can be applied to an electronic device with communication, computing, and data storage capabilities, such as Figure 1 as shown, which includes: Step 110, obtaining multi-modal data related to the manufacturing of the target mold; In this embodiment, the multi-modal data refers to multi-source heterogeneous data in the manufacturing process of the target mold, which can be obtained from multiple departments related to manufacturing. For example: Design department: Obtain geometric information from materials such as the design drawings and 3D models of the target mold, such as the size, shape complexity, and structural features of the mold. These information directly affect the processing difficulty and material consumption of the mold, and thus affect the cost.

[0035] Production department: Collect process parameter records during the production process, including machining parameters such as cutting speed, feed rate, and cutting depth, as well as the running time, energy consumption, and maintenance records of the equipment. The process parameters determine the processing efficiency and product quality, and the operating conditions of the equipment affect the production efficiency and maintenance cost. In some alternative embodiments, sensors are installed on the production equipment to collect the running parameters of the equipment in real time, such as temperature, pressure, vibration, etc. These sensor data can provide more accurate and real-time equipment operation information, which helps to detect equipment failures and abnormal conditions in a timely manner and reduce the maintenance cost.

[0036] Purchasing department: Obtain the purchase price of raw materials, supplier information, quality inspection reports, etc. The cost of raw materials is an important part of the mold manufacturing cost. The prices and qualities of raw materials from different suppliers may vary, and the quality inspection report can reflect whether the raw materials meet the production requirements and avoid cost increases caused by raw material quality problems.

[0037] Quality inspection department: Collect quality inspection data of the mold, such as the inspection results of dimensional accuracy, surface roughness, hardness, etc. The quality inspection data can reflect whether the mold meets the design requirements. If it does not meet the requirements, rework or scrapping may be required, thus increasing the cost.

[0038] It is also possible to query relevant historical data and real-time data from databases such as the enterprise's production management system, quality management system, and procurement management system. The data in the database is usually sorted and classified to facilitate quick access to the required information.

[0039] Step 120: Extract feature data related to the manufacturing cost of the target mold from the multi-modal data; In this embodiment, for the text descriptions in the design drawings, text descriptions in the process documents, etc., natural language processing techniques can be used to extract keywords, phrases, and semantic information, and then obtain feature data. Feature extraction can be performed on the 3D model of the target mold, the image of the design drawing, etc., to extract shape features, contour features, texture features, etc. of the mold. These feature data can intuitively reflect the geometric complexity and structural characteristics of the mold. Statistical features such as mean, variance, maximum value, minimum value, etc. are extracted from numerical data such as process parameter records and equipment operation data during the production process as feature data. These statistical features can reflect the distribution and change trend of the data, and help analyze the relationship between process parameters and cost.

[0040] Step 130: Based on a preset spatio-temporal graph network model, extract temporal process data from the feature data, and analyze the first relationship between the temporal process data and the cost over time. Calculate the first cost of the target mold according to the first relationship; and / or Based on the spatio-temporal graph network model, extract spatial process data from the feature data, and analyze the second relationship between the spatial process data and the cost. Calculate the second cost of the target mold according to the second relationship, where the spatial process data includes interrelated process parameters and / or interrelated equipment.

[0041] In this embodiment, the spatio-temporal graph network model (STGCN) is used to model and analyze the temporal relationship and spatial relationship between feature data and cost. It is a pre-trained model that can capture the complex relationship between feature data and cost. The STGCN model includes a graph neural network and a temporal model, and combines the advantages of both to effectively capture the spatio-temporal dependence in the data. Specifically as follows: The time-series process data can be further subdivided into machining-stage time-series data and equipment-operation time-series data. The machining-stage time-series data can reflect the process parameter changes and cost consumption in different machining stages. For example, the cutting parameters and machining times in the rough machining stage and the finish machining stage are different, and their impacts on costs are also different. The equipment-operation time-series data can reflect the operation efficiency, energy consumption changes, etc. of the equipment, which helps to analyze the impact of equipment operation on costs. In addition to extracting the original values of the time-series process data, time-series features such as the trend, periodicity, and volatility of the time-series data can also be extracted. For example, by analyzing the time-series trend of the machining time, it can be judged whether the production efficiency is stable. By analyzing the periodic changes in the equipment energy consumption, the operation plan of the equipment can be optimized to reduce the energy consumption cost.

[0042] When analyzing the first relationship between the time-series process data changing over time and costs, through its internal memory unit and gating mechanism, the spatio-temporal graph network model remembers the key process parameter information in the past period of time, so as to extract the features reflecting the process change trend, analyze the impact of this feature on costs, and obtain the first relationship.

[0043] In some alternative embodiments, the spatio-temporal graph network model also extracts the spatial features of the spatial process data, such as the distribution features of process parameters, the layout features of equipment, etc. For example, by analyzing the spatial distribution of different process parameters during the mold machining process, it can be judged whether there is a problem of excessive local machining difficulty leading to increased costs; by analyzing whether the layout of the equipment is reasonable and whether there is a situation where the material transportation distance is too long resulting in increased costs. For the mutually related process parameters and equipment in the spatial process data, the spatio-temporal graph network model can establish a spatial association relationship model. Through the graph structure, the association relationship of the spatial process data can be more intuitively displayed, so as to analyze and obtain the second relationship between the spatial process data and costs.

[0044] This embodiment changes the traditional cost management mode mainly based on qualitative analysis and provides a dynamic cost control mode. Specifically, by analyzing the impacts of the time-series process data and the spatial process data on costs respectively, the modeling and analysis of the complex association relationship between the feature data and costs are realized, which can effectively capture the cost impacts of the process parameters changing over time and space, and provide a decision-making basis for the dynamic optimization of costs.

[0045] In actual use, it is possible to only analyze the first cost generated by the time-series process data, or only analyze the second cost corresponding to the spatial process data, or also comprehensively calculate by combining the first cost and the second cost. It depends on the actual production requirements, and specific details are not limited here.

[0046] In this embodiment, by obtaining multi-modal data related to target mold manufacturing: actively collecting heterogeneous data covering multiple links such as design, processing, assembly, and inspection (such as design drawings, processing programs, processing parameters, etc.), the data originally scattered in different departments, systems, and links are gathered together, breaking the physical and logical "information islands" from the source and providing a data basis for subsequent comprehensive analysis; identifying and extracting key features strongly related to cost, improving the "value density" and pertinence of the data; aiming at the pain point of "difficulty in comprehensively grasping the cost composition and change rules", using a preset spatio-temporal graph network model to capture the change rules of process parameters, equipment status, resource consumption, etc. over time, analyzing the first relationship between the change rules and cost, and solving the problem of difficult to capture the dynamic change rules in cost composition; and / or analyzing the second relationship between the mutual correlation between different nodes and cost, such as how design complexity affects the selection of processing parameters, how processing accuracy affects assembly difficulty and rework rate, etc., and quantifying the problem of how a design decision affects the cost of all subsequent links through a chain reaction. The finally calculated first cost, second cost, or their combination is no longer a simple superposition of fragmented link costs, but a refined calculation result integrating the dynamic interaction effects of the whole process and all elements. It enables enterprises to comprehensively and systematically grasp the true cost composition and achieve refined cost management.

[0047] In some alternative embodiments, calculating the first cost of the target mold according to the first relationship and calculating the second cost of the target mold according to the second relationship include: constructing a process-cost correlation matrix according to the first relationship and the second relationship, where the process-cost correlation matrix is used to quantitatively analyze the influence degree and influence direction of feature data on cost; calculating the first cost and the second cost according to the process-cost correlation matrix.

[0048] In this embodiment, the process-cost correlation matrix integrates multi-dimensional features such as design, process, equipment, and environment, and captures the interaction effects between features through matrix operations. For example, in this embodiment, the process-cost correlation matrix is a three-dimensional matrix, which includes: The m-axis corresponds to process features (such as injection pressure, cooling time, material melting point); The n-axis corresponds to cost elements (material cost, energy consumption cost, scrap loss); The t-axis corresponds to the time window (capturing dynamic evolution); Encoding the triple relationship of "process feature-cost element-time evolution" into a computable structure of a mathematical tool provides underlying support for dynamic cost control.

[0049] In some alternative embodiments, constructing the process-cost correlation matrix according to the first relationship and the second relationship includes: abstracting each of the feature data and cost elements into respective nodes in a spatio-temporal graph structure by using the spatio-temporal graph network model, where the cost elements include at least one of raw material procurement cost, processing cost, and processing loss cost; analyzing the historical sequence data of each node through the time series model in the spatio-temporal graph network model to obtain the first relationship between the feature data and the cost elements; aggregating the information of neighbor nodes through the graph neural network model in the spatio-temporal graph network model to obtain the second relationship between the feature data and the cost elements; and constructing the process-cost correlation matrix according to the first relationship and the second relationship.

[0050] Specifically, each feature data that affects the cost in the mold manufacturing process is abstracted as a node in the spatio-temporal graph. For example, feature data such as cutting speed, feed rate, cutting depth, equipment energy consumption, raw material procurement price, etc. are respectively used as different nodes. Cost elements such as raw material procurement cost, processing cost, processing loss cost, etc. are also abstracted as nodes in the spatio-temporal graph. Each cost element node corresponds to a specific cost category. The edges between the nodes represent the association relationships between the nodes, such as the collaborative relationship between process parameters, the causal relationship between feature data and cost elements, etc.

[0051] When analyzing the first relationship, models such as Long Short-Term Memory Network (LSTM), Gated Recurrent Unit (GRU), or Temporal Convolutional Network (TCN) can be selected. These models can effectively capture the long-term dependence relationships and short-term change patterns in time series data. For each feature data node and cost element node, collect its historical sequence data. For example, the historical data of cutting speed, the monthly data of raw material procurement price, the quarterly data of processing cost, etc. Use the time series model to model and analyze the historical sequence data of each node, and extract the features and patterns in the time dimension. By analyzing the correlation, causality, etc. between the feature data node and the cost element node in the time series, determine the first relationship between them. For example, how the change in cutting speed affects the processing cost over time.

[0052] In the spatio-temporal graph, neighbor nodes can include spatial neighbors: multiple nodes connected physically / associated with process flows, and logical neighbors: multiple nodes whose data correlation reaches a preset standard. For example, taking the molding process of a mold as an example, a stamping machine (a forming equipment node) and a subsequent polishing machine (a polishing equipment node) are continuous in the process flow, so the stamping machine and the polishing machine are physically connected nodes. There is a strong data correlation between the cutting speed (a process parameter node) and the surface roughness of the mold (a quality index node). Generally speaking, too high a cutting speed may cause an increase in the surface roughness of the mold, while too low a cutting speed may affect the processing efficiency. When the correlation coefficient between the cutting speed and the surface roughness reaches the preset standard (such as above 0.7) through data analysis, the surface roughness node is the logical neighbor of the cutting speed node.

[0053] When analyzing the second relationship, models such as graph convolutional network (GCN), graph attention network (GAT), or graph isomorphism network (GIN) can be selected. These models can effectively aggregate the information of neighbor nodes and capture the spatial relationships between nodes. By aggregating the information of neighbor nodes, analyze the spatial relationship between the feature data node and the cost element node. For example, the cutting speed and the feed rate, as neighbor nodes, jointly affect the processing cost node, thereby determining their second relationship.

[0054] According to the first relationship and the second relationship, determine the influence degree and influence direction of each feature data node on the cost element node. Quantify the influence degree into specific values, such as correlation coefficients, weights, etc., and represent the influence direction as positive or negative. For example, initialize a matrix of size n×m, where n is the number of feature data nodes and m is the number of cost element nodes. Fill the elements in the matrix according to the first relationship and the second relationship analyzed. If the influence degree of the cutting speed node on the processing cost node is 0.8 and the influence direction is positive, then fill in +0.8 at the corresponding position in the matrix.

[0055] In this embodiment, the structure of the spatio-temporal graph network model (STGCN) is as shown in the code: ```python class STGCN(nn.Module): def __init__(self): self.gcn = GraphConv(in_dim=128, out_dim=64) self.tcn = TemporalConv(64, 32) def forward(self, graph_data): spatial_feat = self.gcn(graph_data) temporal_feat = self.tcn(spatial_feat) return temporal_feat ``` In this model, the GraphConv layer is used to capture the correlation relationships in the spatial dimension, such as the mutual influence between different process parameters and the cooperation relationships between different devices. The TemporalConv layer is used to capture the variation laws in the temporal dimension, such as the trend of process parameters changing over time and the dynamic evolution of cost elements. Through the combination of the two, STGCN can comprehensively and deeply analyze the complex relationship between processes and costs. By constructing a process-cost correlation matrix using the STGCN model, the impact degree and direction of process parameters on costs can be quantitatively analyzed. This matrix provides a clear direction and basis for cost optimization, guiding enterprises to carry out process improvement and cost control in a targeted manner.

[0056] In this embodiment, by using the spatio-temporal graph network model to abstract each feature data and cost element into each node in the spatio-temporal graph structure, the originally complex and disordered feature data and cost element information can be presented in an intuitive and structured way. This structured representation helps to analyze and process the data subsequently, enabling the complex process-cost relationship to be presented in a clearer graph structure form, which is convenient for understanding and operation.

[0057] In some alternative embodiments, the method further includes: extracting structured data from the feature data, where the type of the structured data includes at least one of geometric complexity, process path, and material properties; and calculating the third cost of the target mold according to the contribution degree of the structured data to the cost.

[0058] In this embodiment, the structured data is static index parameters related to the cost, with clear formats and definitions. The data types include geometric complexity, process path, material properties, etc. These data usually exist in the forms of standard tables, database fields, etc., and each data item has a clear meaning and value range. The structured data provides a benchmark for cost calculation based on the essential characteristics of the mold, making the mold cost calculation result more comprehensive.

[0059] In some alternative embodiments, extracting the structured data from the feature data includes at least one of the following: extracting the geometric complexity through 3D point cloud curvature analysis; and extracting the relevant parameters under the process path based on a numerical control machining instruction library, where the relevant parameters include at least one of path length, feed rate, and cutting depth; and obtaining the material properties through dynamic hardness map analysis, wherein the dynamic hardness map is used to simulate the change of material hardness with position, temperature, time, or strain rate under actual machining conditions.

[0060] In this embodiment, the geometric complexity feature is extracted through 3D point cloud curvature analysis, accurately quantifying the geometric features of the mold, reflecting the complexity of the mold shape and its impact on machining difficulty and cost; the numerical control machining instruction library contains rich machining information. By extracting relevant parameters under the process path, such as path length, feed rate, and cutting depth, etc., the process details in the mold manufacturing process can be comprehensively understood; the dynamic hardness map can simulate the change of material hardness with position, temperature, time, or strain rate under actual machining conditions. The material properties obtained through dynamic hardness map analysis are closer to the true performance of the material during actual machining, providing more accurate material property data for cost calculation and process design.

[0061] These three types of features together constitute the feature space for mold cost calculation, covering the main influencing factors in the mold manufacturing process. Through dynamic feature engineering, the system can dynamically adjust feature weights and calculation methods according to real-time data and historical experience to adapt to changes in different scenarios and requirements. This dynamic characteristic enables the system to have good adaptability and robustness, capable of coping with various changes and challenges in the mold manufacturing process. As shown in Table 1 below: Table 1

[0062] Table 1 shows the structured data obtained by simultaneously using 3D point cloud curvature analysis, G-code semantic parsing, and dynamic hardness map analysis. The dimensions in this table refer to the dimensions through which the geometric complexity is calculated when analyzing the complexity of the set, such as degree of bending, spatial flatness, etc.; specifically, what dimensions are included under the process path, such as path length, feed rate, cutting depth, etc. The dimensions of material features can include hardness, heat resistance, etc. Its contribution degree refers to the degree of influence on cost. This structured data provides a benchmark for cost calculation based on the essential features of the mold.

[0063] In some alternative embodiments, the method further includes: performing a weighted sum of the first cost, the second cost, and the third cost to obtain the comprehensive cost of manufacturing the target mold.

[0064] In this embodiment, the first cost is calculated based on the relationship between the chronological process data and the cost, reflecting the impact of the process change over time on the cost; the second cost is calculated based on the relationship between the spatial process data and the cost, reflecting the effect of the spatial correlation between the process parameters and the equipment on the cost; the third cost is calculated according to the contribution degree of the structured data (such as geometric complexity, process path, material properties) to the cost, considering the impact of the inherent attributes of the mold itself on the cost. By weighted summation, the cost information in these three different dimensions is integrated, avoiding the one-sidedness that may be caused by the single-dimensional cost calculation, and enabling the comprehensive cost to more comprehensively reflect various cost factors in the mold manufacturing process.

[0065] In some alternative embodiments, the calculation formula for the comprehensive cost of the target mold is as follows:

[0066] Wherein, is the comprehensive cost of the target mold, , and respectively represent the prediction functions based on XGBoost, LSTM, and graph convolutional network, is the structured data, is the chronological process data, is the spatial process data, , and are gating coefficients, which are dynamically adjusted through a preset attention mechanism.

[0067] In this embodiment, XGBoost is used to model the relationship between the structured data and the cost. By constructing multiple decision trees, XGBoost can effectively capture the non-linear relationships and interactions in the data, improving the accuracy and stability of the prediction.

[0068] LSTM (Long Short-Term Memory network) is a chronological model in the spatio-temporal graph neural network STGCN, which is good at processing chronological data and is used to model the first relationship between the chronological process data and the cost. LSTM can effectively capture the time-dependent patterns, identify the rules and trends in the chronological data, and provide a chronological basis for cost prediction.

[0069] GCN (Graph Convolutional Network) is a graph neural network model in the spatio-temporal graph neural network STGCN, which is used to process graph-structured data and is used to model the second relationship between the correlation relationship between process parameters and the cost. By constructing a correlation graph between process parameters, GCN can capture the mutual influence and synergy between parameters, and provide a correlation basis for cost prediction.

[0070] The gating coefficients , and Dynamically adjusted through the attention mechanism, which reflects the importance and contribution degree of different data sources and models in the current scenario. This dynamic adjustment mechanism enables the model to automatically adjust the weights of each sub-model according to the quality, relevance, and reliability of different data, improving the overall prediction accuracy and robustness.

[0071] The attention mechanism can dynamically adjust the gating coefficients (i.e., weights) of XGBoost, LSTM, and GCN according to the characteristics of the input data. This dynamic adjustment enables the model to automatically assign different importances to different models in different situations. For example, when structured data has a greater impact on cost, the attention mechanism will increase the gating coefficient of XGBoost; when the changes in time-series process data have a significant impact on cost, the gating coefficient of LSTM will be increased. This makes the method of the present application adaptable to complex and variable manufacturing environments, such as equipment status, material supply, process adjustment, etc. By dynamically adjusting the gating coefficients through the attention mechanism, the comprehensive cost calculation method can adapt to these complex and variable environments, adjust the model weights in real time, and ensure that the cost prediction always maintains a high level of accuracy.

[0072] In some optional embodiments, after obtaining the comprehensive cost of manufacturing the target mold, it further includes: based on the comprehensive cost and real-time production line data, using a preset reinforcement learning algorithm to calculate a cost control strategy, the cost control strategy includes at least one of the setting of process parameters, material selection, and processing path optimization, and the reward function of the reinforcement learning algorithm includes the degree of cost reduction and process quality; using a preset digital twin model of the production line to verify the cost control strategy, and adjusting the cost control strategy according to the verification result until the verification result meets the preset standard to obtain the final cost control strategy.

[0073] In this embodiment, the digital twin model of the production line is a virtual mapping of the actual production line, which can verify the cost control strategy without interfering with actual production. By simulating the implementation process of the strategy in the digital twin model, potential problems and risks, such as equipment conflicts, process mismatches, etc., can be discovered in advance, avoiding adverse consequences such as increased costs, decreased quality, or production interruptions in actual production.

[0074] The preset reinforcement learning algorithm is used to generate an optimal cost control strategy according to the cost prediction result. In this embodiment, a deep Q-network (DQN) is used as the reinforcement learning framework to achieve dynamic optimization and adjustment of costs. The implementation of the DQN network is as shown in the code: ```python class DQNAgent: def __init__(self): self.memory = ReplayBuffer(10000) self.q_net = QNetwork(256) def choose_action(self, state): state = torch.FloatTensor(state) return self.q_net(state).argmax() ``` In this implementation, the DQNAgent class contains an experience replay buffer and a Q-network. The experience replay buffer is used to store historical experiences, and the Q-network is used to learn the mapping relationship from states to actions. The choose_action method selects the optimal action based on the current state through the Q-network.

[0075] In the context of mold cost control, the environmental state of reinforcement learning consists of the current cost situation, process parameters, equipment status, etc., and the actions include various cost control measures such as adjusting process parameters, replacing materials, optimizing processing paths, etc. The reward function is defined as a comprehensive index of cost reduction and process quality, considering both the cost control effect and the impact of process quality.

[0076] Reinforcement learning continuously optimizes its own algorithm type through a trial-and-error approach. In mold manufacturing, reinforcement learning algorithms can find the optimal manufacturing solution by continuously adjusting processing parameters. For example, by adjusting the cutting speed and feed rate, the optimal balance between processing efficiency and cost can be achieved. Reinforcement learning (RL) is a machine learning technique that enables robots to make intelligent decisions by learning from experience. By obtaining programmed rewards or punishments, the AI model of the robot is driven to continuously improve during the trial-and-error process.

[0077] Through the reinforcement learning optimizer, the system can dynamically adjust the cost control strategy according to real-time data and historical experience, achieving cost minimization and process quality optimization. This data-driven decision-making method is more scientific and accurate than traditional manual experience judgment, and can effectively improve the effect and efficiency of cost control.

[0078] In some alternative embodiments, extracting the feature data related to the target mold manufacturing cost from the multi-modal data includes: extracting initial feature data related to cost calculation based on the multi-modal data; performing feature dimension compression on the initial feature data to obtain the final feature data related to cost calculation.

[0079] In this embodiment summary, high-dimensional data is difficult to directly visualize. However, by compressing the dimensions to reduce the data to two or three dimensions, the relationships and distributions between the data can be more intuitively displayed. For example, visualization tools such as scatter plots and line charts can be used to show the relationship between the compressed feature data and the cost, helping to better understand the influencing factors of the cost.

[0080] Specifically, in the data processing process, a method combining PCA (Principal Component Analysis) and Autoencoder is adopted to achieve feature dimension compression. This not only reduces the computational complexity, improves the system operation efficiency, but also reduces data redundancy and improves the quality and effectiveness of feature representation.

[0081] Embodiment Two: Based on the technology of the above embodiment, this embodiment provides an application example. This application provides a die cost calculation system. This system applies the cost calculation method described in the above embodiment.

[0082] Specifically, the overall architecture diagram of the system is as Figure 2 shown. In this architecture, The physical layer collects various sensor data, equipment status data, and environmental parameter data through the OPC UA protocol to achieve real-time perception of the entire manufacturing process; The edge computing node is responsible for preliminary data processing and analysis to relieve the computing pressure on the backend; The Kafka streaming platform serves as the core data bus of the system, responsible for the efficient transmission and processing of data, and supports the real-time response ability of the system; The feature engineering pipeline cleans, transforms, and extracts features from the original data to provide standardized and high-quality input data for subsequent model calculations; The hybrid prediction model combines the advantages of multiple machine learning algorithms to achieve accurate prediction of die costs; The dynamic adjustment engine generates and executes cost control strategies based on the prediction results and real-time data; Digital twin verification simulates and verifies the control strategy to ensure its effectiveness and security; Finally, the process optimization suggestions present the optimization results to the user in an understandable form to support decision-making.

[0083] Based on the above architecture, the system processes 12 types of heterogeneous data sources, including design data, process data, equipment data, environmental data, quality data, etc. These data come from diverse sources and have different formats, and are processed and integrated through a unified data governance framework. First, for structured data, technologies such as data cleaning, standardization, and feature engineering are adopted to extract meaningful features and metrics. For unstructured data, such as design drawings, machining programs, inspection reports, etc., deep learning models are used for semantic understanding and information extraction. For time series data, time series analysis and prediction technologies are adopted to mine time-dependent patterns. For spatial data, such as 3D models of molds and machining paths, spatial analysis technologies are adopted to extract geometric features and topological relationships.

[0084] Through multi-modal data fusion, the system realizes a comprehensive perception and understanding of the entire mold manufacturing process, providing rich data support for subsequent analysis and decision-making. In the data processing process, a method combining PCA (Principal Component Analysis) and Autoencoder is adopted to achieve feature dimension compression, with a compression rate of 85%. This not only reduces the computational complexity and improves the system operation efficiency, but also reduces data redundancy and improves the quality and effectiveness of feature representation.

[0085] Among them, feature engineering technology is the core link of machine learning and has a decisive impact on the model performance. In this embodiment, a dynamic feature engineering method is designed for multi-source heterogeneous data in the mold manufacturing process to maximize the mining of data value. The system defines three major categories of key features: geometric complexity, process path, and material properties, as shown in the following table:

[0086] Geometric complexity: The feature is extracted through 3D point cloud curvature analysis, reflecting the complexity of the mold shape and its impact on machining difficulty and cost. Research shows that the geometric complexity of the mold directly affects the complexity of the machining process and thus affects the machining cost. Curvature analysis can quantify the bending degree of the mold surface, identify high-complexity regions, and provide a geometric basis for cost estimation.

[0087] Process path: The feature is extracted through G-code semantic parsing, reflecting the planning and execution of the machining path. G-code is the core instruction in numerical control machining and contains rich process information. Through semantic parsing, the system can extract key parameters such as path length, feed rate, and cutting depth from the code, providing a process basis for cost calculation. The optimization of the process path is of great significance for reducing machining costs, especially in the production mode of multi-variety and small-batch.

[0088] Material properties: Features are extracted through dynamic hardness map analysis, reflecting the changes in the physical properties of materials during the processing. Material properties directly affect the processing difficulty and cost, and different materials require different processing technologies and parameter settings. Through dynamic hardness map analysis, the system can identify the hardness distribution and change rules of materials, providing a material basis for process parameter optimization. In die manufacturing, material selection and performance analysis are key links, directly affecting the quality and cost of dies.

[0089] These three types of features together constitute the feature space for die cost calculation, covering the main influencing factors in the die manufacturing process. Through dynamic feature engineering, the system can dynamically adjust the feature weights and calculation methods according to real-time data and historical experience, adapting to changes in different scenarios and requirements. This dynamic characteristic enables the system to have good adaptability and robustness, capable of coping with various changes and challenges in the die manufacturing process.

[0090] The Spatio-Temporal Graph Convolutional Network (STGCN) is the core algorithm module of the system, used to model and analyze the spatio-temporal correlation relationship between process parameters and cost elements. The STGCN model combines the advantages of graph neural networks and temporal models, and can effectively capture the spatio-temporal dependencies in the data.

[0091] The structure of the STGCN model is shown in the code: ```python class STGCN(nn.Module): def __init__(self): self.gcn = GraphConv(in_dim=128, out_dim=64) self.tcn = TemporalConv(64, 32) def forward(self, graph_data): spatial_feat = self.gcn(graph_data) temporal_feat = self.tcn(spatial_feat) return temporal_feat ``` In this model, the GraphConv layer is used to capture the correlation relationships in the spatial dimension, such as the mutual influence between different process parameters and the cooperation relationships between different devices. The TemporalConv layer is used to capture the changing patterns in the time dimension, such as the trend of process parameters changing over time and the dynamic evolution of cost elements. Through the combination of the two, STGCN can comprehensively and deeply analyze the complex relationship between processes and costs.

[0092] Spatio-temporal graph neural networks have received extensive attention in recent years. By integrating graph neural networks and various time learning methods, they can extract complex spatio-temporal dependencies. In urban computing prediction, the STGNN framework has shown strong performance, being able to effectively process spatio-temporal data and make accurate predictions. In multivariate time series prediction, models based on dynamic adaptive spatio-temporal graphs have shown superior performance. These studies indicate that spatio-temporal graph neural networks have significant advantages in processing data with spatio-temporal characteristics, which provides a theoretical basis for the application of STGCN in mold cost control.

[0093] Through the STGCN model, the system can construct a process-cost correlation matrix to quantitatively analyze the degree and direction of the impact of process parameters on costs. This matrix provides a clear direction and basis for cost optimization, guiding enterprises to carry out process improvement and cost control in a targeted manner.

[0094] The hybrid prediction model is the core algorithm module of the system, used to accurately predict the mold cost. The model combines the advantages of multiple machine learning algorithms to effectively model complex non-linear relationships. The mathematical expression of the hybrid prediction model is:

[0095] where is the comprehensive cost of the target mold, , and represent the prediction functions based on XGBoost, LSTM, and graph convolutional network respectively, is the structured data, is the time-series process data, is the spatial process data, , and are gating coefficients, dynamically adjusted through a preset attention mechanism.

[0096] XGBoost is an ensemble learning algorithm, well-known for its high performance and interpretability. In this embodiment, XGBoost is used to model the relationship between structured data (such as material properties, equipment parameters, etc.) and cost. By constructing multiple decision trees, XGBoost can effectively capture the non-linear relationships and interactions in the data, improving the accuracy and stability of predictions.

[0097] LSTM (Long Short-Term Memory network) is a recurrent neural network, proficient in processing time series data. In this embodiment, LSTM is used to model the relationship between time series data (such as parameter changes during the processing, environmental condition changes, etc.) and cost. LSTM can effectively capture time-dependent patterns, identify the regularities and trends in time series data, providing a time series basis for cost prediction.

[0098] GCN (Graph Convolutional Network) is a graph neural network, used to process graph-structured data. In this embodiment, GCN is used to model the relationship between the association relationships among process parameters and cost. By constructing an association graph among process parameters, GCN can capture the mutual influences and synergistic effects among parameters, providing an association basis for cost prediction.

[0099] The gating coefficients α, β, and γ are dynamically adjusted through the attention mechanism, reflecting the importance and contribution degrees of different data sources and models in the current scenario. This dynamic adjustment mechanism enables the model to automatically adjust the weights of each sub-model according to the quality, relevance, and reliability of different data, improving the overall prediction accuracy and robustness.

[0100] The advantage of the hybrid prediction model lies in its ability to combine the characteristics of different models, complement their respective deficiencies, and improve the comprehensiveness and accuracy of predictions. XGBoost provides interpretability and stability, LSTM captures time series features, and GCN models association relationships. The three are combined to form a comprehensive and accurate prediction framework. This hybrid method has significant advantages in complex system modeling and can effectively address various challenges in die cost prediction.

[0101] The reinforcement learning optimizer is the core control module of the system, used to generate the optimal cost control strategy according to the cost prediction results. In this embodiment, the Deep Q-Network (DQN) is adopted as the reinforcement learning framework to achieve dynamic optimization and adjustment of costs.

[0102] The implementation of the DQN network is as shown in the code: ```python class DQNAgent: def __init__(self): self.memory = ReplayBuffer(10000) self.q_net = QNetwork(256) def choose_action(self, state): state = torch.FloatTensor(state) return self.q_net(state).argmax() ``` In this implementation, the DQNAgent class contains an experience replay buffer and a Q-network. The experience replay buffer is used to store historical experiences, and the Q-network is used to learn the mapping relationship from states to actions. The choose_action method selects the optimal action based on the current state through the Q-network.

[0103] In the context of mold cost control, the environmental state of reinforcement learning consists of the current cost situation, process parameters, equipment status, etc., and the actions include various cost control measures such as adjusting process parameters, replacing materials, and optimizing the processing path. The reward function is defined as a comprehensive index of cost reduction and process quality, considering both the cost control effect and the impact of process quality.

[0104] Reinforcement learning continuously optimizes its own algorithm type through a trial-and-error approach. In mold manufacturing, the reinforcement learning algorithm can find the optimal manufacturing solution by continuously adjusting processing parameters. For example, by adjusting the cutting speed and feed rate, the optimal balance between processing efficiency and cost can be achieved. Reinforcement learning (RL) is a machine learning technique that enables robots to make intelligent decisions by learning from experience. By obtaining programmed rewards or punishments, the AI model of the robot is driven to continuously improve during the trial-and-error process.

[0105] Through the reinforcement learning optimizer, the system can dynamically adjust the cost control strategy according to real-time data and historical experience, achieving cost minimization and process quality optimization. This data-driven decision-making method is more scientific and accurate than traditional manual experience judgment and can effectively improve the effectiveness and efficiency of cost control.

[0106] To verify the effectiveness of the system in this embodiment, an actual application test was conducted in a large mold enterprise. The experimental environment configuration is as shown in Table 2 below:

[0107] Table 2 (Configuration Table) The hardware platform used in the experiment is the NVIDIA DGX A100 cluster, which provides powerful computing capabilities and supports the efficient operation of deep learning and complex algorithms. The data scale reaches 15TB of CAD data and 240 million sensor data, covering the whole process of mold design, processing, assembly, inspection, etc., providing rich data support for the training and testing of the system.

[0108] The comparison benchmarks selected are SAP MES (Manufacturing Execution System) and SolidWorks PLM (Product Lifecycle Management) systems, which are widely used management software in the current mold industry and have high market recognition and application value. By comparing with these mature systems, the advantages and innovativeness of the method proposed in this embodiment can be objectively evaluated.

[0109] To verify the performance advantages of the system, this embodiment compares the traditional system and this system in terms of cost prediction accuracy, abnormal response time, and dynamic adjustment success rate. The results are shown in Table 3:

[0110] Table 3 (Performance Comparison) Cost prediction accuracy: The mean absolute error (MAE) of the traditional system is 87,000 yuan, while the MAE of this system is only 31,000 yuan, and the error is reduced by 64.4%. This shows that this system has significant advantages in cost prediction, can estimate the mold cost more accurately, and provides a reliable basis for cost control. The accuracy of cost prediction is crucial for the decision-making of enterprises. Accurate prediction can guide enterprises to make more reasonable resource allocation and cost management.

[0111] Abnormal response time: The abnormal detection delay of the traditional system is 4 hours, while the abnormal detection delay of this system is only 8.7 seconds, and the response time is shortened by 99.4%. This shows that this system has significant advantages in real-time monitoring and rapid response, can detect and handle cost anomalies in a timely manner, and prevent the expansion of problems and the increase of losses. In a rapidly changing market environment, the ability to respond in a timely manner is crucial for the survival and development of enterprises.

[0112] Dynamic adjustment success rate: The dynamic adjustment success rate of the traditional system is 68%, while the dynamic adjustment success rate of this system reaches 92.3%, and the success rate is increased by 35.7%. This shows that this system has significant advantages in the generation and execution of cost control strategies and can more effectively achieve the cost control goal. The dynamic adjustment ability is the core competitiveness of the intelligent cost control system and determines the system's ability to respond to changes and challenges.

[0113] These comparison results fully demonstrate the effectiveness and superiority of the method proposed in this embodiment, and verify the practical application value of the system. Through the application of artificial intelligence and machine learning technologies, the system has achieved intelligent, dynamic, and refined cost control, significantly improving the enterprise's cost management level and competitiveness.

[0114] To more intuitively display the optimization effect of the system, this embodiment conducts a comparative analysis of each component of the mold cost, and the results are as Figure 3 shown. It can be seen from the figure that the system has achieved significant optimization in each cost component: 1. Material cost: The material cost under the traditional method is 42 units, and after optimization, it is reduced to 35 units, a decrease of 16.7%. This shows that the system can effectively reduce the material cost by optimizing material selection, reducing material waste, etc. Material cost is the main cost component in mold manufacturing, accounting for about 40% of the total cost. Reducing the material cost is of great significance for overall cost control.

[0115] 2. Processing cost: The processing cost under the traditional method is 33 units, and after optimization, it is reduced to 28 units, a decrease of 15.2%. This shows that the system can effectively reduce the processing cost by optimizing the processing technology, improving the processing efficiency, etc. Processing cost is the second largest cost component in mold manufacturing, accounting for about 30% of the total cost. Reducing the processing cost is of great significance for improving the competitiveness of the enterprise.

[0116] 3. Energy consumption cost: The energy consumption cost under the traditional method is 15 units, and after optimization, it is reduced to 9 units, a decrease of 40%. This shows that the system can effectively reduce the energy consumption cost by optimizing equipment operation, reducing energy waste, etc. Although the energy consumption cost accounts for a relatively small proportion in the total cost, reducing energy consumption can not only save costs, but also reduce environmental pollution and improve the social responsibility image of the enterprise.

[0117] These optimization results show that the system in this embodiment can not only achieve the reduction of the overall cost, but also accurately optimize different cost components, improving the refinement level of cost control. This comprehensive and accurate cost control ability is difficult to achieve by traditional methods.

[0118] In summary, this embodiment in the mold cost calculation system is specifically manifested in three aspects: methodological innovation, technological innovation, and application innovation: 1. Methodological innovation: The theory of spatio-temporal graph modeling of process features This embodiment proposes a new methodological framework - the theory of spatio-temporal graph modeling of process features, providing new ideas and tools for mold cost control. The core of this methodological framework is the spatio-temporal graph convolutional network (STGCN), which can extract complex spatio-temporal dependencies by integrating graph neural networks and various temporal learning methods.

[0119] The STGCN model can simultaneously consider the spatial correlation and temporal evolution of process parameters, comprehensively capturing the complex relationship between processes and costs. In the spatial dimension, STGCN captures the mutual influence and correlation between different process parameters through graph convolution operations; in the temporal dimension, STGCN captures the dynamic change laws of process parameters and cost elements through temporal convolution operations. Through the combination of space and time, STGCN can analyze the relationship between processes and costs more comprehensively and deeply, providing more accurate predictions and more effective strategies for cost control.

[0120] The innovation of the spatio-temporal graph modeling theory for process characteristics lies in that it not only focuses on the impact of a single process parameter on cost, but also on the interaction and synergy between process parameters, and how these impacts and effects change over time and space. This comprehensive and dynamic modeling method can more accurately reflect the complexity and variability of the mold manufacturing process, providing a more scientific and accurate basis for cost control.

[0121] 2. Technological Innovation: Dynamic Adjustment Engine Driven by Streaming Computing This embodiment provides a new technical architecture - a dynamic adjustment engine driven by streaming computing, providing powerful technical support for mold cost control. The core of this technical architecture is streaming computing technology, which can perform real-time analysis on large-scale flowing data in a continuously changing process, capture valuable information, and send the results to the next computing node.

[0122] The dynamic adjustment engine driven by streaming computing monitors various data in the manufacturing process in real time, such as equipment status, processing parameters, environmental conditions, etc., promptly discovers cost anomalies, and automatically generates and executes adjustment measures according to preset rules or algorithm models. This real-time monitoring and rapid response capability enables the system to promptly discover and handle cost problems, preventing the problems from expanding and losses from increasing.

[0123] The innovation of the dynamic adjustment engine driven by streaming computing lies in that it breaks the periodicity and lag of traditional cost control, achieving real-time and forward-looking cost control. The system response time has been shortened from the traditional 4 - 6 hours to 8.7 seconds. With this real-time response capability, the system can promptly discover and handle cost anomalies, preventing the problems from expanding and losses from increasing. In a rapidly changing market environment, the real-time response capability is crucial for the survival and development of enterprises.

[0124] 3. Application Innovation: Closed-loop Control System Verified by Digital Twin This embodiment constructs a new application mode - a closed-loop control system for digital twin verification, providing a brand-new application scenario for mold cost control. The core of this application mode is digital twin technology, which creates a virtual replica of a physical object or system to enable simulation, testing, and optimization in a virtual environment.

[0125] The closed-loop control system for digital twin verification creates a digital twin model by mapping the actual manufacturing process into a virtual environment. This model can not only reflect the state and changes of the actual process in real time but also be used to simulate and test various cost control strategies and predict their effects and impacts. Through digital twin verification, the system can evaluate the effectiveness and safety of various cost control strategies before implementation, reducing implementation risks and costs.

[0126] The innovation of the closed-loop control system for digital twin verification lies in achieving the closed-loop management of cost control, forming a complete process of "prediction - decision - execution - evaluation". Through digital twin verification, the system can continuously learn and optimize, improving the effectiveness and efficiency of cost control. By associating the digital twin model with the production system to achieve synchronous operation and real-time monitoring, virtual simulation of products during the design and manufacturing processes can be realized to improve product quality, production efficiency, and product quality stability.

[0127] Through the mold cost calculation system of this embodiment, three major breakthroughs have been achieved in the field of mold cost control: First, the cost prediction accuracy has broken through the industry benchmark (<5% error). The cost prediction error rate of traditional methods is as high as 15.6%, while this system reduces the error rate to 3.1%, with an 80% reduction in error. This shows that the system has significant advantages in cost prediction, can estimate mold costs more accurately, and provides a reliable basis for cost control. The accuracy of cost prediction is crucial for enterprise decision-making. Accurate prediction can guide enterprises to conduct more reasonable resource allocation and cost management.

[0128] Second, the dynamic response speed reaches the sub-second level. The abnormal detection delay of the traditional system is 4 hours, while the abnormal detection delay of this system is only 8.7 seconds, with a 99.4% reduction in response time. This shows that the system has significant advantages in real-time monitoring and rapid response, can detect and handle cost anomalies in a timely manner, and prevent problems from expanding and losses from increasing. In a rapidly changing market environment, the ability to respond in a timely manner is crucial for the survival and development of enterprises.

[0129] Third, an interpretable process optimization knowledge base has been constructed. The system can not only achieve cost prediction and control but also provide interpretable optimization suggestions and knowledge to help users understand and apply these optimization measures. This interpretability makes the system not only a black-box tool but also a knowledge platform, which can promote users' learning and growth.

[0130] These breakthroughs fully demonstrate the application value of artificial intelligence and machine learning technologies in the field of mold cost control, and showcase the advantages and potential of data-driven decision-making. By applying these advanced technologies to mold cost control, a transformation has been achieved from experience-driven to data-driven, from passive response to proactive prevention, and from extensive management to refined management.

[0131] Embodiment Three: Another embodiment of the present application relates to a mold cost calculation device. The implementation details of the mold cost calculation device in this embodiment will be specifically described below. The following content is only the implementation details provided for convenience of understanding and is not necessary for implementing this solution. The schematic diagram of the mold cost calculation device in this embodiment can be as Figure 4 shown, including a data acquisition module 410, a feature data extraction module 420, a first cost calculation module 430, and a second cost calculation module 440.

[0132] The data acquisition module 410 is used to acquire multimodal data related to the manufacturing of the target mold; The feature data extraction module 420 is used to extract feature data related to the manufacturing cost of the target mold from the multimodal data; The first cost calculation module 430 is used to extract temporal process data from the feature data based on a preset spatio-temporal graph network model, and analyze the first relationship between the temporal process data and cost over time, and calculate the first cost of the target mold according to the first relationship; and / or The second cost calculation module 440 is used to extract spatial process data from the feature data based on the spatio-temporal graph network model, and analyze the second relationship between the spatial process data and cost, and calculate the second cost of the target mold according to the second relationship, where the spatial process data includes interrelated process parameters and / or interrelated equipment.

[0133] It is worth mentioning that each module involved in this embodiment is a logical module. In practical applications, a logical unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. In addition, to highlight the innovative part of the present application, units that are not closely related to solving the technical problems proposed in the present application are not introduced in this embodiment, but this does not mean that there are no other units in this embodiment.

[0134] In some optional embodiments, the mold cost calculation device further includes: An association matrix construction module, configured to construct a process-cost association matrix according to the first relationship and the second relationship, where the process-cost association matrix is used to quantitatively analyze the influence degree and influence direction of feature data on cost; An associated cost calculation module, configured to calculate the first cost and the second cost according to the process-cost association matrix.

[0135] In some alternative embodiments, the association matrix construction module includes: A spatio-temporal graph construction unit, configured to abstract each of the feature data and cost elements into nodes in a spatio-temporal graph structure by using the spatio-temporal graph network model, where the cost elements include at least one of raw material procurement cost, processing cost, and processing loss cost; A first relationship construction unit, configured to analyze the historical sequence data of each node through a time series model in the spatio-temporal graph network model to obtain the first relationship between the feature data and the cost elements; A second relationship construction unit, configured to aggregate information of neighbor nodes through a graph neural network model in the spatio-temporal graph network model to obtain the second relationship between the feature data and the cost elements; An association matrix construction unit, configured to construct the process-cost association matrix according to the first relationship and the second relationship.

[0136] In some alternative embodiments, the apparatus further includes: A structured data extraction module, configured to extract structured data from the feature data, where the type of the structured data includes at least one of geometric complexity, process path, and material properties; A third cost calculation module, configured to calculate a third cost of the target mold according to the contribution degree of the structured data to the cost.

[0137] In some alternative embodiments, the structured data extraction module includes at least one of the following: A geometric complexity extraction unit, configured to extract the geometric complexity through 3D point cloud curvature analysis; and A process path extraction unit, configured to extract relevant parameters under the process path based on a numerical control machining instruction library, where the relevant parameters include at least one of path length, feed rate, and cutting depth; and A material property extraction unit, configured to obtain the material property through dynamic hardness map analysis, where the dynamic hardness map is used to simulate the change of material hardness with position, temperature, time, or strain rate under actual machining conditions.

[0138] In some alternative embodiments, the apparatus further includes: A comprehensive cost calculation module is used to perform weighted summation on the first cost, the second cost, and the third cost to obtain the comprehensive cost of manufacturing the target mold.

[0139] In some alternative embodiments, it further includes: A cost control strategy generation module is used to calculate a cost control strategy based on the comprehensive cost and real-time production line data by using a preset reinforcement learning algorithm. The cost control strategy includes at least one of setting process parameters, material selection, and machining path optimization. The reward function of the reinforcement learning algorithm includes the degree of cost reduction and process quality. A control strategy verification module is used to verify the cost control strategy by using a digital twin model of the preset production line, and adjust the cost control strategy according to the verification result until the verification result meets the preset standard to obtain the final cost control strategy.

[0140] In some alternative embodiments, the feature data extraction module includes: An initial feature data extraction unit is used to extract initial feature data related to cost calculation based on the multimodal data. A feature dimension compression unit is used to compress the feature dimension of the initial feature data to obtain the final feature data related to cost calculation.

[0141] Embodiment 4: Another embodiment of the present application relates to an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the mold cost calculation method in the above embodiments.

[0142] Wherein, the memory and the processor are connected in a bus manner. The bus may include any number of interconnected buses and bridges, and the bus connects various circuits of one or more processors and the memory together. The bus may also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art, and thus will not be further described herein. The bus interface provides an interface between the bus and the transceiver. The transceiver may be an element or multiple elements, such as multiple receivers and transmitters, and provides a unit for communicating with various other devices on the transmission medium. The data processed by the processor is transmitted on the wireless medium through the antenna. Further, the antenna also receives data and transmits the data to the processor.

[0143] The processor is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. The memory can be used to store the data used by the processor when executing operations.

[0144] Embodiment Five: Another embodiment of the present application relates to a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the method embodiments described above are implemented.

[0145] That is, those skilled in the art can understand that all or part of the steps of implementing the methods in the above embodiments can be completed by instructing relevant hardware through a program. The program is stored in a storage medium and includes several instructions for causing a device (which can be a single-chip microcomputer, a chip, etc.) or a processor to execute all or part of the steps of the methods described in the various embodiments of the present application. The foregoing storage medium includes: USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks, or optical discs, etc., which can store program codes.

[0146] Those of ordinary skill in the art can understand that the above embodiments are specific embodiments for implementing the present application, and in actual applications, various changes can be made in form and details without departing from the spirit and scope of the present application.

Claims

1. A method for calculating the cost of a mold, characterized in that Including: Obtaining multi-modal data related to the manufacturing of a target mold; Extracting feature data related to the manufacturing cost of the target mold from the multi-modal data; Based on a preset spatio-temporal graph network model, extracting temporal process data from the feature data, and analyzing a first relationship between the temporal process data and the cost over time, and calculating a first cost of the target mold according to the first relationship; And / or Based on the spatio-temporal graph network model, extracting spatial process data from the feature data, and analyzing a second relationship between the spatial process data and the cost, and calculating a second cost of the target mold according to the second relationship, where the spatial process data includes interrelated process parameters and / or interrelated equipment.

2. The mold cost calculation method according to claim 1, characterized in that Calculating the first cost of the target mold according to the first relationship and calculating the second cost of the target mold according to the second relationship includes: Constructing a process-cost association matrix according to the first relationship and the second relationship, where the process-cost association matrix is used to quantitatively analyze the influence degree and influence direction of feature data on the cost; Calculating the first cost and the second cost according to the process-cost association matrix.

3. The mold cost calculation method according to claim 2, characterized in that, The constructing the process-cost association matrix according to the first relationship and the second relationship includes: Using the spatio-temporal graph network model to abstract each of the feature data and cost elements into nodes in a spatio-temporal graph structure, where the cost elements include at least one of raw material procurement cost, processing cost, and processing loss cost; Analyzing the historical sequence data of each node through the temporal model in the spatio-temporal graph network model to obtain the first relationship between the feature data and the cost elements; Aggregating the information of neighbor nodes through the graph neural network model in the spatio-temporal graph network model to obtain the second relationship between the feature data and the cost elements; Constructing the process-cost association matrix according to the first relationship and the second relationship.

4. The mold cost calculation method according to claim 1, wherein The method further includes: Extracting structured data from the feature data, where the type of the structured data includes at least one of geometric complexity, process path, and material properties; Calculating a third cost of the target mold according to the contribution degree of the structured data to the cost.

5. The mold cost calculation method according to claim 4, wherein The extracting the structured data from the feature data includes at least one of the following: Extracting the geometric complexity through 3D point cloud curvature analysis; and Extracting relevant parameters under the process path based on a numerical control machining instruction library, where the relevant parameters include at least one of path length, feed rate, and cutting depth; And Obtaining the material properties through dynamic hardness map analysis, where the dynamic hardness map is used to simulate the change of material hardness with position, temperature, time, or strain rate under actual processing conditions.

6. The mold cost calculation method according to claim 4, wherein The method further includes: Performing weighted summation on the first cost, the second cost, and the third cost to obtain a comprehensive cost of manufacturing the target mold.

7. The mold cost calculation method according to claim 6, wherein The calculation formula of the comprehensive cost of the target mold includes the following: Among them, is the comprehensive cost of the target mold, , and respectively represent prediction functions based on XGBoost, LSTM, and graph convolutional network, is the structured data, is the time-series process data, is the spatial process data, , and are gating coefficients, which are dynamically adjusted through a preset attention mechanism.

8. The mold cost calculation method according to claim 6, characterized in that, After obtaining the comprehensive cost of manufacturing the target mold, it further includes: Based on the comprehensive cost and real-time production line data, a cost control strategy is calculated using a preset reinforcement learning algorithm. The cost control strategy includes at least one of setting process parameters, material selection, and machining path optimization. The reward function of the reinforcement learning algorithm includes the degree of cost reduction and process quality; The preset digital twin model of the production line is used to verify the cost control strategy, and the cost control strategy is adjusted according to the verification result until the verification result meets the preset standard to obtain the final cost control strategy.

9. The mold cost calculation method according to any one of claims 1-8, characterized in that The extraction of feature data related to the target mold manufacturing cost from the multi-modal data includes: Based on the multi-modal data, initial feature data related to cost calculation is extracted; The initial feature data is compressed in terms of feature dimensions to obtain the final feature data related to cost calculation.

10. A mold cost calculation device, characterized in that, Including: A data acquisition module for acquiring multi-modal data related to target mold manufacturing; A feature data extraction module for extracting feature data related to the target mold manufacturing cost from the multi-modal data; A first cost calculation module for extracting temporal process data from the feature data based on a preset spatio-temporal graph network model, and analyzing the first relationship between the temporal process data and cost over time, and calculating the first cost of the target mold according to the first relationship; and / or A second cost calculation module for extracting spatial process data from the feature data based on the spatio-temporal graph network model, and analyzing the second relationship between the spatial process data and cost, and calculating the second cost of the target mold according to the second relationship, wherein the spatial process data includes interrelated process parameters and / or interrelated equipment.

11. An electronic device, characterized in that, Including: At least one processor; And, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the mold cost calculation method according to any one of claims 1 to 9.

12. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the mold cost calculation method according to any one of claims 1 to 9.

Images