Intelligent prediction method for die stamping manufacturing cost

Abstract

The present application relates to the technical field of die stamping manufacturing, and discloses a die stamping manufacturing cost intelligent prediction method. The method first acquires a multi-dimensional manufacturing data set composed of material characteristic parameters, process machining parameters and equipment operation parameters. Then, based on the material characteristic parameters, key process features are extracted, a process influence topology graph is generated according to the process machining parameters, and the equipment operation parameters are processed to obtain an equipment state evolution trend. Then, these data are input into a multi-factor evaluation model to generate a cost prediction requirement vector, and finally, a self-adaptive prediction strategy graph is constructed through a dynamic programming algorithm to output a manufacturing cost prediction scheme. The present application comprehensively considers multi-dimensional data, can effectively process data dynamic changes, improves the accuracy and reliability of die stamping manufacturing cost prediction, and helps enterprises to reasonably control costs, optimize production decisions and improve market competitiveness.

Description

Technical Field

[0001] This invention relates to the field of mold stamping manufacturing technology, specifically to a method for intelligent prediction of mold stamping manufacturing costs. Background Technology

[0002] In the die stamping manufacturing industry, cost control has always been one of the key factors for companies to maintain competitiveness. With increasingly fierce market competition and the continuous development of the manufacturing industry, accurately predicting the cost of die stamping manufacturing has become increasingly important.

[0003] Traditional methods for predicting manufacturing costs in die stamping largely rely on manual experience and simple calculation models. Manual judgment is limited by individual knowledge and practical experience, leading to significant differences in judgment among individuals and making it difficult to guarantee the accuracy and stability of predictions. For example, when assessing material costs, estimates are based solely on past prices of similar materials, ignoring factors such as market price fluctuations and differences in material properties. Similarly, when calculating processing costs, methods simply multiply conventional processing time and equipment rates without considering the impact of process details on processing efficiency.

[0004] Simple calculation models typically consider only one or a few cost-influencing factors, failing to fully reflect the complexity of the die stamping manufacturing process. For example, some models focus only on material costs and basic processing time, neglecting the significant impact of die wear and changes in equipment operating conditions on costs. During the stamping process, as the number of uses increases, the wear on the die intensifies, affecting not only product quality but also potentially reducing processing efficiency and increasing the cost of repairing or replacing the die. During equipment operation, different operating parameters lead to varying energy consumption, and the frequency of equipment failures is closely related to operating conditions. All of these factors directly or indirectly increase manufacturing costs, but traditional models struggle to incorporate them.

[0005] Furthermore, the die stamping manufacturing process involves various types of data, such as material property parameters, process parameters, and equipment operating parameters. These data are interconnected and constantly changing. Existing forecasting methods cannot effectively handle multi-dimensional and dynamically changing data, making it difficult to reflect cost change trends in real time and accurately. In actual production, material properties vary from batch to batch, process parameters may be adjusted according to product requirements, and equipment operating status can change at any time. Traditional methods are inadequate in the face of these complex situations, failing to provide reliable cost forecasts for enterprises in a timely manner, thus affecting their production decisions, cost control, and profit generation.

[0006] With the development of intelligent manufacturing technology and the continuous improvement of data acquisition and processing capabilities, new opportunities have been provided for solving the problem of cost prediction in die stamping manufacturing. However, the application of related technologies in the die stamping manufacturing field is not yet mature enough, and their advantages have not been fully utilized to achieve accurate cost prediction. Therefore, it is urgent to develop an intelligent cost prediction method and system for die stamping manufacturing that can comprehensively consider multi-dimensional manufacturing data, effectively handle dynamic data changes, and achieve high prediction accuracy. This is of great significance for improving the market competitiveness and economic benefits of die stamping manufacturing enterprises. Summary of the Invention

[0007] The purpose of this invention is to provide an intelligent prediction method for die stamping manufacturing costs, so as to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: an intelligent prediction method for die stamping manufacturing costs, the method comprising:

[0009] Acquire a multi-dimensional manufacturing data set; the multi-dimensional manufacturing data includes material property parameters, process parameters, and equipment operating parameters; the material property parameters include the material ductility coefficient and hardness distribution gradient, and the process parameters include stamping speed adjustment value and die wear index;

[0010] Based on the material property parameters, key process features are extracted through process parameter analysis. These key process features include material deformation threshold, stress concentration factor, and energy consumption fluctuation range.

[0011] Based on the process parameters, a process influence topology map is generated by a feature association network construction algorithm. The influence topology map includes parameter interaction paths and node influence weights.

[0012] The equipment operating parameters are processed into time series segments to generate a trend of equipment state evolution.

[0013] The key process characteristics, the influence topology diagram, and the equipment state evolution trend are input into the multi-factor evaluation model to generate a cost prediction demand vector.

[0014] Based on the predicted demand vector, an adaptive prediction strategy graph is constructed using a dynamic programming algorithm to output a manufacturing cost prediction scheme; the nodes of the adaptive prediction strategy graph represent prediction analysis modules, and the edges represent the analysis order and complexity weights.

[0015] Preferably, the extraction of key process features through process parameter analysis includes:

[0016] Outlier removal is performed on the material property parameters to generate standardized material data;

[0017] Based on a pre-defined baseline process parameter library, the basic deformation category is identified and the deformation stage is divided using a gradient matching algorithm.

[0018] The energy distribution characteristics of the deformation stage are extracted using a frequency domain decomposition algorithm, and the deformation deviation is generated by combining the reference threshold in the benchmark process parameter library.

[0019] The deformation deviation, stress concentration factor, and energy consumption fluctuation range are encoded as key process features.

[0020] Preferably, the step of generating the process influence topology map through the feature association network construction algorithm includes:

[0021] Redundant data filtering is performed on the process parameters to remove invalid adjustment noise;

[0022] The core processing parameters are divided into groups based on a parameter clustering algorithm, and the correlation between each group is calculated.

[0023] Node interaction strength is generated based on the correlation tightness and a preset influence threshold.

[0024] By combining the node interaction strength with the parameter influence weight, a multi-level influence topology graph is formed.

[0025] Preferably, the multi-factor evaluation model includes a data compression module and a weight mapping module, wherein the data compression module includes:

[0026] The material deformation threshold in the key process features is normalized to obtain the first compression vector;

[0027] The interaction strength of nodes in the topology graph is segmented and quantized to generate a second compressed vector.

[0028] The moving mean difference calculation is performed on the state evolution trend of the equipment to extract the state change rate feature and obtain the third compression vector;

[0029] The first, second, and third compressed vectors are merged into a low-dimensional analysis sequence through a data fusion layer.

[0030] Preferably, the weight mapping module includes:

[0031] Spatial dimension alignment is performed on the low-dimensional analysis sequence to generate a weighted correlation matrix;

[0032] The parameter weight features are extracted using a hierarchical attention mechanism to generate a weight distribution matrix;

[0033] The weight correlation matrix and the weight distribution matrix are superimposed using tensors to generate weighted fusion features.

[0034] By superimposing the weighted fusion features with the original low-dimensional analysis sequence through residual connections, a cost prediction demand vector is output.

[0035] Preferably, the step of constructing an adaptive prediction strategy graph using a dynamic programming algorithm includes:

[0036] Initialize node attributes based on the predictive analysis module and generate an edge weight matrix based on complexity weights;

[0037] The cost prediction demand vector is used as the node state, and the edge weight matrix is ​​composed of the time cost and complexity weight of the analysis order.

[0038] The path value of each node is updated iteratively through the cost function, and the edge weight matrix is ​​adjusted accordingly.

[0039] Generate the optimal prediction analysis sequence covering all nodes based on the adjusted edge weight matrix.

[0040] Preferably, the method for constructing the benchmark process parameter library includes:

[0041] Standard process samples under various typical material deformation states were collected, and baseline gradients and energy distribution thresholds were extracted.

[0042] Multi-scale transformation of the baseline gradient is performed to generate a multi-resolution gradient model;

[0043] The gradient model is classified according to the deformation category and associated with a reference threshold database;

[0044] The classified gradient model is stored as a baseline process parameter library, and the model is updated periodically based on new samples.

[0045] Preferably, the parameter optimization method of the parameter clustering algorithm includes:

[0046] Calculate the initial clustering granularity and minimum parameter density based on the distribution of historical processing data;

[0047] The parameter combinations are traversed using a grid search algorithm, and the parameter with the highest matching degree between the clustering result and the manually labeled parameter group is selected.

[0048] The clustering granularity and minimum parameter density are dynamically adjusted based on the matching degree to optimize the node interaction strength partitioning accuracy.

[0049] Preferably, the method for constructing the cost function includes:

[0050] The cost between nodes is defined as a balance coefficient between time cost and complexity weight;

[0051] The path value of each node is initialized to zero, and the starting cost value is a preset initial value;

[0052] The minimum cumulative generation value of each node based on its predecessor nodes is calculated using recursive equations, and the optimal path is recorded.

[0053] A complete predictive analysis sequence is generated by forward derivation based on the optimal path.

[0054] Preferably, the present invention further includes an intelligent prediction system for die stamping manufacturing costs, the system comprising:

[0055] Multi-dimensional data acquisition module: used to acquire a multi-dimensional manufacturing data set, which includes material property parameters, process parameters and equipment operating parameters; wherein, the material property parameters include the material ductility coefficient and hardness distribution gradient, and the process parameters include the stamping speed adjustment value and the die wear index;

[0056] Process feature extraction module: configured to extract key process features based on the material property parameters by analyzing the process parameters. The key process features include material deformation threshold, stress concentration factor and energy consumption fluctuation range.

[0057] Topology construction module: used to generate a process influence topology map based on the process parameters using a feature association network construction algorithm. The influence topology map includes parameter interaction paths and node influence weights.

[0058] Equipment data processing module: Performs time series segmentation processing on the equipment operating parameters to generate equipment state evolution trends;

[0059] Multi-factor evaluation module: Input the key process characteristics, influence topology diagram and equipment status evolution trend into the multi-factor evaluation model to generate a cost prediction demand vector; the multi-factor evaluation model includes a data compression module and a weight mapping module;

[0060] Prediction strategy generation module: Based on the predicted demand vector, it constructs an adaptive prediction strategy graph using a dynamic programming algorithm and outputs a manufacturing cost prediction scheme; the nodes of the adaptive prediction strategy graph represent the prediction analysis module, and the edges represent the analysis order and complexity weights.

[0061] Compared with the prior art, the beneficial effects of the present invention are:

[0062] In terms of data processing, a multi-dimensional manufacturing dataset is acquired, covering material property parameters, process parameters, and equipment operating parameters, comprehensively collecting data on various factors affecting the manufacturing cost of die stamping. Among these, the material ductility coefficient and hardness distribution gradient, within the material property parameters, accurately reflect the impact of the material's inherent properties on the stamping process; the stamping speed adjustment value and die wear index, within the process parameters, reflect changes in key factors during processing; and the equipment operating parameters record the real-time status of the equipment during production. This rich data lays a solid foundation for subsequent accurate cost analysis and prediction. When faced with differences in material properties between different batches, this detailed data allows for a more accurate assessment of the material's impact on cost, avoiding cost estimation errors caused by material variations.

[0063] In the process feature extraction stage, outlier removal of material characteristic parameters generates standardized material data, ensuring data reliability. Based on a pre-set benchmark process parameter library, a gradient matching algorithm is used to identify basic deformation categories and classify deformation stages. Then, a frequency domain decomposition algorithm is used to extract energy distribution features and generate deformation deviation. Finally, multiple key features are encoded as key process features. This processing method can deeply explore the inherent characteristics of materials in the stamping process and accurately grasp the material's change patterns during the process. In actual production, by accurately obtaining key process features such as material deformation threshold, stress concentration factor, and energy consumption fluctuation range, potential problems in the process can be predicted in advance. For example, measures can be taken in advance in stress concentration areas to optimize mold design, reduce the risk of mold damage, thereby reducing the cost of maintenance and mold replacement, improving product quality, reducing scrap rate, and indirectly reducing production costs.

[0064] When constructing the process influence topology diagram, redundant data of process parameters is filtered to remove invalid adjustment noise. A parameter clustering algorithm is used to divide core processing parameter groups and calculate their correlation strength, thereby generating a multi-level influence topology diagram. This topology diagram clearly displays the parameter interaction paths and node influence weights, facilitating an intuitive understanding of the complex relationships between process parameters. When adjusting stamping speed, the topology diagram allows for a quick understanding of its impact on the die wear index and other related parameters, thereby optimizing the combination of process parameters, improving production efficiency, and reducing increased processing costs caused by unreasonable parameters.

[0065] In processing equipment operating parameters, time-series segmentation is used to generate equipment state evolution trends, enabling real-time tracking of changes in equipment operating status. By monitoring changes in equipment status, potential equipment failures can be detected early, allowing for timely maintenance and preventing production interruptions caused by sudden equipment failures. This reduces economic losses due to downtime, extends equipment lifespan, and lowers equipment replacement costs.

[0066] The data compression module in the multi-factor evaluation model processes key process characteristics, relevant data in the influence topology diagram, and equipment state evolution trends, merging them into a low-dimensional analysis sequence. This effectively reduces data dimensionality, improves data processing efficiency, and retains key information. The weight mapping module generates a cost forecast demand vector through a series of operations, accurately reflecting the degree of influence of each factor on costs. This multi-factor comprehensive evaluation method significantly improves the accuracy of cost forecasting compared to traditional single-factor or simple factor-based evaluation methods. In practical applications, enterprises can formulate more reasonable production plans based on accurate cost forecast results, avoiding resource waste or order losses caused by cost estimation errors.

[0067] Based on the predicted demand vector, an adaptive prediction strategy graph is constructed using a dynamic programming algorithm to output a manufacturing cost prediction scheme. The nodes and edges of this adaptive prediction strategy graph explicitly represent the prediction analysis modules, analysis order, and complexity weights. Through continuous iterative optimization, an optimal prediction analysis sequence covering all nodes can be generated. This makes the prediction process more scientific and rational, enabling dynamic adjustments to the prediction strategy based on actual production conditions, adapting to different production scenarios and data changes, and further improving the reliability and practicality of cost prediction. Enterprises can use these accurate prediction schemes to rationally allocate production resources, optimize processes, and control production costs, gaining a more advantageous position in market competition and improving their economic efficiency and market competitiveness. Attached Figure Description

[0068] Figure 1 This is a schematic diagram illustrating the working principle of the intelligent prediction method for die stamping manufacturing costs described in this invention.

[0069] Figure 2 To generate a process impact topology diagram flowchart;

[0070] Figure 3 This is a flowchart of the data compression module;

[0071] Figure 4 This is a flowchart of the weight mapping module. Detailed Implementation

[0072] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] Please see Figures 1-4 This invention provides an intelligent prediction method for die stamping manufacturing costs, specifically including the following steps:

[0074] The acquired multi-dimensional manufacturing data encompasses material property parameters, process parameters, and equipment operating parameters. Material property parameters include the material's ductility coefficient and hardness distribution gradient; these two parameters reflect the material's inherent physical properties and significantly impact subsequent stamping processes. Process parameters include stamping speed adjustment values ​​and die wear indices, which directly relate to the actual operation of the stamping process. This multi-dimensional manufacturing data is collected through various sensors, data acquisition devices, and production recording systems, providing fundamental data support for subsequent analysis and prediction.

[0075] Based on the acquired material property parameters, key process characteristics are extracted through process parameter analysis. These key process characteristics include the material deformation threshold, stress concentration factor, and energy consumption fluctuation range. The material deformation threshold determines the critical value at which the material begins to undergo plastic deformation during the stamping process; the stress concentration factor reflects the degree of stress concentration within the material; and the energy consumption fluctuation range reflects the range of energy consumption variations during the stamping process. Through in-depth analysis of the material property parameters and the use of specific algorithms and models, these key process characteristics are extracted, providing important process basis for subsequent cost prediction.

[0076] Based on the process parameters, a process influence topology map is generated using a characteristic association network construction algorithm. This topology map includes parameter interaction paths and node influence weights. The parameter interaction paths illustrate the interaction relationships between different process parameters, while the node influence weights quantify the degree of influence of each parameter on other parameters and the entire stamping process. By utilizing the characteristic association network construction algorithm, the process parameters are processed and analyzed to generate a process influence topology map that intuitively reflects the relationships between parameters.

[0077] By segmenting equipment operating parameters into time series data, a trend of equipment state evolution can be generated. During operation, the equipment's state changes over time. By segmenting the equipment operating parameters into time series data and analyzing the characteristics and patterns of change in each segment, a trend of equipment state evolution can be generated. This helps to understand the changes in the equipment's operating state during the stamping process and provides equipment-related information for cost prediction.

[0078] The extracted key process features, the generated process influence topology diagram, and the equipment state evolution trend are input into a multi-factor evaluation model to generate a cost prediction demand vector. The multi-factor evaluation model comprehensively considers multiple factors such as process and equipment. Through processing and analysis of the input data, it generates a vector that reflects the cost prediction demand, providing crucial data support for subsequent cost prediction.

[0079] Based on the generated predicted demand vector, an adaptive prediction strategy graph is constructed using a dynamic programming algorithm, outputting a manufacturing cost prediction scheme. Nodes in the adaptive prediction strategy graph represent prediction analysis modules, and edges represent the analysis order and complexity weights. The dynamic programming algorithm constructs the optimal prediction strategy graph based on the predicted demand vector, combined with the analysis order and complexity weights, thereby outputting an accurate manufacturing cost prediction scheme.

[0080] The present invention will be further described below with reference to Examples 1 to 5:

[0081] Example 1:

[0082] This embodiment details the specific process of extracting key process features through process parameter analysis.

[0083] First, outlier removal is performed on the acquired material property parameters. In actual production, outliers may appear in material property parameters due to sensor errors, data transmission errors, etc. These outliers can affect the accuracy of subsequent analysis results, therefore, they need to be removed. An outlier detection algorithm based on statistical methods, such as the 3σ criterion, is employed. Assume the set of material property parameters is... ,in Indicates the first Calculate the mean of the set of material property parameter values. Standard deviation If a certain parameter value satisfy If any value is found to be outlier, it will be identified and removed. After outlier removal, standardized material data is generated.

[0084] Based on a pre-defined baseline process parameter library, a gradient matching algorithm is used to identify the basic deformation category and classify the deformation stages. The baseline process parameter library stores standard process parameters for various typical materials under different deformation states. The gradient matching algorithm identifies the basic deformation category by calculating the similarity between the gradient of the standardized material data and the baseline gradient in the baseline process parameter library. Assuming the standardized material data is... The baseline gradient is The similarity calculation formula is: ,in These are weighting coefficients, set according to the importance of different parameters. Based on similarity... The value of is used to determine the best-matching baseline gradient, thereby identifying the basic deformation category. Then, based on the deformation law of the material during the stamping process, the deformation stages are divided, such as the elastic deformation stage, the plastic deformation stage, etc.

[0085] A frequency domain decomposition algorithm is used to extract the energy distribution characteristics during the deformation stage, and the deformation deviation is generated by combining this with reference thresholds from a baseline process parameter library. The frequency domain decomposition algorithm transforms the time-domain signal into the frequency domain for analysis, providing a clearer view of the signal's energy distribution. Assuming the signal during the deformation stage is... Through Fourier transform Transform it to the frequency domain to obtain the energy distribution. The reference process parameter library stores the reference energy distribution thresholds for the corresponding deformation stages. The formula for calculating the deformation deviation is: .

[0086] Finally, deformation deviation, stress concentration factor, and energy consumption fluctuation range are encoded as key process features. Through a specific encoding method, these features are transformed into a computer-processable format, providing crucial process feature data for subsequent cost prediction.

[0087] Example 2:

[0088] In this embodiment, the focus is on generating a process influence topology map using a feature association network construction algorithm. This process plays a crucial role in the intelligent prediction of die stamping manufacturing costs, as it clearly presents the intrinsic relationships between process parameters, providing an important basis for subsequent cost prediction.

[0089] The first step is to filter redundant data from the process parameters. In actual production, the collected process parameters often contain redundant information and invalid adjustment noise. For example, the stamping speed adjustment value may experience minor fluctuations due to slight malfunctions in the equipment sensors, which constitute invalid adjustment noise. Furthermore, multiple parameters may reflect production conditions repeatedly in certain aspects, which is redundant data. To ensure accurate analysis, a filtering method based on the similarity of data trends is used. Process parameters such as stamping speed adjustment values ​​and die wear indexes over a period of time are plotted as line graphs to observe the trends of each parameter curve. If two curves show almost identical trends for most of the time, it can be preliminarily determined that one of the parameters is redundant. Then, the details of parameter changes under different operating conditions are further compared to identify and eliminate redundant parameters. This processing makes the subsequent analysis data more representative and accurate.

[0090] After filtering redundant data, a parametric clustering algorithm is used to divide the core processing parameters into groups. Hierarchical clustering is chosen here, as it can gradually merge parameters into different levels of clusters based on their similarity, much like building blocks. Taking parameters such as stamping speed adjustment, die temperature, and die wear index as examples, each parameter is initially treated as a separate class. The similarity between different parameters is then calculated, for example, using Euclidean distance, as shown in the formula:

[0091] ,

[0092] in Represents two parameter vectors and The Euclidean distance between the parameters is used. The smaller the distance, the more similar the two parameters are. Each time, the two closest clusters are merged to form a new cluster. This process is repeated until the preset number of clusters is reached or the inter-cluster distance no longer meets the merging condition. In this way, highly correlated parameters can be grouped together to form the core processing parameter group.

[0093] Next, the correlation between each group is calculated. For the defined core processing parameter groups, the correlation is determined by analyzing their variations across different production batches. For example, one group includes stamping speed adjustment values ​​and die wear index, while another group includes die temperature and stamping pressure. If, across multiple production batches, it is observed that when the stamping speed adjustment value in the first group increases, the die temperature in the second group also shows a significant upward trend, it indicates a high correlation between these two parameter groups. Mutual information is used to quantify this correlation; the greater the mutual information, the stronger the correlation between the two parameter groups.

[0094] The node interaction strength is generated based on the correlation strength and a preset influence threshold. The preset influence threshold is determined based on extensive historical production data and practical production experience. If the correlation strength is greater than this threshold, it indicates a significant mutual influence between the two sets of parameters, requiring the generation of a corresponding node interaction strength. For example, if the influence threshold is set to 0.5, and the mutual information calculation result of the two sets of parameters is 0.6, it means they meet the condition. In this case, the node interaction strength can be determined based on the difference between the correlation strength and the threshold; the larger the difference, the higher the interaction strength.

[0095] Finally, the node interaction strength and parameter influence weights are combined to form a multi-level influence topology diagram. Parameter influence weights are determined through expert evaluation and feedback from actual production data. For example, experts, based on experience, determine that the stamping speed adjustment value has a significant impact on the die stamping manufacturing cost, thus assigning it a higher weight. The node interaction strength and parameter influence weights are combined and presented graphically. Nodes represent each core processing parameter group, edges represent the interaction relationships between them, the thickness of the edges indicates the magnitude of the node interaction strength, and the color intensity of the edges indicates the level of parameter influence weight, thereby forming a multi-level influence topology diagram that intuitively displays the complex relationships between process parameters.

[0096] Example 3:

[0097] This embodiment focuses on the data compression module in the multi-factor evaluation model. Its purpose is to transform complex data such as key process characteristics, influence topology diagrams, and equipment state evolution trends into a lower-dimensional analysis sequence that is easier to process, thereby improving the efficiency and accuracy of subsequent cost prediction.

[0098] First, the material deformation threshold in the key process characteristics is normalized to obtain the first compression vector. The material deformation threshold varies greatly under different materials and stamping processes. Taking two different mold materials as examples, the deformation threshold of one material may be in the range of 5-10, while that of the other may be in the range of 50-100. To unify the data scale, a normalization method is adopted. Assume the material deformation threshold is... Its range of values ​​is The normalization formula is ,here This is the result after normalization. The normalized material deformation thresholds are arranged in a certain order to form the first compression vector. For example, after normalization, values ​​such as 0.3, 0.5, and 0.7 are obtained and arranged sequentially to form a vector.

[0099] The interaction strength of nodes in the topology graph is segmented and quantized to generate a second compressed vector. Since the interaction strength is a continuously changing value, segmented quantization is necessary for computer processing. Based on experience and data distribution characteristics, the range of node interaction strength values ​​is divided into several intervals. For example, these intervals are [0, 0.2), [0.2, 0.4), [0.4, 0.6), [0.6, 0.8), and [0.8, 1]. For each node interaction strength value, it is determined which interval it belongs to, and then a corresponding code is assigned. If the node interaction strength value is 0.3, it is encoded as 1; if it is 0.7, it is encoded as 3. All the encoded values ​​of node interaction strength are arranged in order to obtain the second compressed vector.

[0100] The moving average difference (MAD) is used to calculate the evolution trend of the equipment state, extracting the rate of change features and obtaining a third compressed vector. The equipment state changes continuously over time, and the moving average difference calculation can effectively extract its rate of change features. A moving average window is set, with the window size selected based on the characteristics of the equipment operating data and experience; assuming five time points are used. The values ​​of the equipment operating parameters over the time series are... First, calculate the moving average. Taking these 5 values ​​as an example, the moving average As time progresses, the window slides forward one point in time and a new moving average is calculated. Then, the moving average difference is calculated, which is the difference between two adjacent moving averages. Arranging these moving average differences in chronological order forms the third compression vector, which reflects the rate of change of the device state.

[0101] Finally, the data fusion layer merges the first, second, and third compressed vectors into a low-dimensional analysis sequence. The data fusion layer uses a simple concatenation method, linking the three compressed vectors sequentially. For example, if the first compressed vector is [0.3, 0.5, 0.7], the second is [1, 2, 3], and the third is [0.1, -0.2, 0.3], the resulting low-dimensional analysis sequence is [0.3, 0.5, 0.7, 1, 2, 3, 0.1, -0.2, 0.3], providing processing data for the subsequent weight mapping module.

[0102] Example 4:

[0103] This embodiment details the weight mapping module in the multi-factor evaluation model, which aligns the spatial dimensions of the low-dimensional analysis sequence to generate a weight correlation matrix. The low-dimensional analysis sequence is formed by merging multiple compressed vectors, which may have different dimensions and features. Spatial dimension alignment is necessary to facilitate weight analysis. Assume the low-dimensional analysis sequence is... By padding with zeros or performing dimensional transformations, each element is made to have the same dimension. Then, a weighted correlation matrix is ​​generated based on the correlation between different elements. ,For example

[0104] ,

[0105] in Indicates the first The element of the first 1 eigenvalue, Indicates the first The mean of the elements. The number of features.

[0106] A hierarchical attention mechanism is used to extract parameter weight features and generate a weight distribution matrix. This mechanism assigns different weights to different elements based on their importance in the low-dimensional analysis sequence. Assume the low-dimensional analysis sequence is... The first-layer attention mechanism calculates the attention weight for each element. ,in The element is obtained through a neural network calculation. The relevant scores. The second-layer attention mechanism further processes the attention weights of the first layer to obtain the final weight distribution matrix. .

[0107] Tensor superposition of the weight correlation matrix and the weight distribution matrix generates a weighted fusion feature. Tensor superposition fuses information from the two matrices, yielding more comprehensive weight information. Assume the weight correlation matrix is... The weight distribution matrix is Weighted fusion features ,in This indicates tensor superposition operations.

[0108] Residual connections are used to superimpose the weighted fusion features onto the original low-dimensional analysis sequence, outputting a cost prediction demand vector. Residual connections avoid information loss during weight mapping, ensuring model accuracy. Assume the weighted fusion features are... The original low-dimensional analysis sequence is Cost forecast demand vector This method outputs a cost forecast demand vector, providing crucial data support for subsequent cost forecasting.

[0109] Example 5:

[0110] This embodiment mainly describes the specific process of constructing an adaptive prediction strategy graph using a dynamic programming algorithm. It initializes node attributes based on the prediction analysis module and generates an edge weight matrix based on complexity weights. The prediction analysis module is a node in the adaptive prediction strategy graph, and each node has different attributes, such as its function and computational complexity. Node attributes are initialized according to the specific functions and characteristics of the prediction analysis module. The complexity weight reflects the computational complexity of each prediction analysis module and is set according to the actual situation. The edge weight matrix represents the connection weights between nodes, and the edge weights consist of the time cost of the analysis order and the complexity weight. Assume the nodes... and nodes The edge weight between them is The time cost is The complexity weight is ,but .

[0111] The cost forecasting demand vector is used as the node state, and the edge weight matrix consists of the time cost and complexity weights of the analysis order. The cost forecasting demand vector contains information on various aspects such as process and equipment, and using it as the node state can provide input data for each predictive analysis module. The edge weight matrix determines the connection relationships and weights between nodes, affecting the order and complexity of the predictive analysis.

[0112] The path value of each node is iteratively updated using a cost function, and the edge weight matrix is ​​adjusted accordingly. The cost function measures the path cost from the starting point to each node, and the cost value between nodes is defined as a balance coefficient between time cost and complexity weights. Assume the balance coefficient is... ,node To the node The value of Initialize the path value of each node to zero, and set the starting cost to a preset initial value (e.g., ...). Through recursive equations Calculate the minimum cumulative generation value for each node based on its predecessor nodes, and record the optimal path, where Represents a node The minimum cumulative cost value, Represents a node The minimum cumulative cost value is calculated. Based on the calculated minimum cumulative cost value, the edge weight matrix is ​​adjusted to make the edge weights more consistent with the actual prediction requirements.

[0113] The optimal predictive analysis sequence covering all nodes is generated based on the adjusted edge weight matrix. This optimal sequence is obtained by finding the optimal path from the starting point to the ending point. Graph search algorithms such as Dijkstra's algorithm can be used to find the optimal path. The execution order of the predictive analysis modules is determined based on the optimal path, thus outputting an accurate manufacturing cost prediction scheme.

[0114] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0115] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent prediction of die stamping manufacturing costs, characterized in that, include: Acquire a multi-dimensional manufacturing data set; the multi-dimensional manufacturing data includes material property parameters, process parameters, and equipment operating parameters; the material property parameters include the material ductility coefficient and hardness distribution gradient, and the process parameters include stamping speed adjustment value and die wear index; Based on the material property parameters, key process features are extracted through process parameter analysis. These key process features include material deformation threshold, stress concentration factor, and energy consumption fluctuation range. Based on the process parameters, a process influence topology map is generated by a feature association network construction algorithm. The influence topology map includes parameter interaction paths and node influence weights. The equipment operating parameters are processed into time series segments to generate a trend of equipment state evolution. The key process characteristics, the influence topology diagram, and the equipment state evolution trend are input into the multi-factor evaluation model to generate a cost prediction demand vector. Based on the predicted demand vector, an adaptive prediction strategy graph is constructed using a dynamic programming algorithm to output a manufacturing cost prediction scheme; the nodes of the adaptive prediction strategy graph represent prediction analysis modules, and the edges represent the analysis order and complexity weights.

2. The intelligent prediction method for die stamping manufacturing costs according to claim 1, characterized in that, The extraction of key process features through process parameter analysis includes: Outlier removal is performed on the material property parameters to generate standardized material data; Based on a pre-defined baseline process parameter library, the basic deformation category is identified and the deformation stage is divided using a gradient matching algorithm. The energy distribution characteristics of the deformation stage are extracted using a frequency domain decomposition algorithm, and the deformation deviation is generated by combining the reference threshold in the benchmark process parameter library. The deformation deviation, stress concentration factor, and energy consumption fluctuation range are encoded as key process features.

3. The intelligent prediction method for die stamping manufacturing costs according to claim 1, characterized in that, The generation of the process influence topology map through the feature association network construction algorithm includes: Redundant data filtering is performed on the process parameters to remove invalid adjustment noise; The core processing parameters are divided into groups based on a parameter clustering algorithm, and the correlation between each group is calculated. Node interaction strength is generated based on the correlation tightness and a preset influence threshold. By combining the node interaction strength with the parameter influence weight, a multi-level influence topology graph is formed.

4. The intelligent prediction method for die stamping manufacturing costs according to claim 1, characterized in that, The multi-factor evaluation model includes a data compression module and a weight mapping module. The data compression module includes: The material deformation threshold in the key process features is normalized to obtain the first compression vector; The interaction strength of nodes in the topology graph is segmented and quantized to generate a second compressed vector. The moving mean difference calculation is performed on the state evolution trend of the equipment to extract the state change rate feature and obtain the third compression vector; The first, second, and third compressed vectors are merged into a low-dimensional analysis sequence through a data fusion layer.

5. The intelligent prediction method for die stamping manufacturing costs according to claim 4, characterized in that, The weight mapping module includes: Spatial dimension alignment is performed on the low-dimensional analysis sequence to generate a weighted correlation matrix; The parameter weight features are extracted using a hierarchical attention mechanism to generate a weight distribution matrix; The weight correlation matrix and the weight distribution matrix are superimposed using tensors to generate weighted fusion features. By superimposing the weighted fusion features with the original low-dimensional analysis sequence through residual connections, a cost prediction demand vector is output.

6. The intelligent prediction method for die stamping manufacturing costs according to claim 1, characterized in that, The construction of the adaptive prediction strategy graph using a dynamic programming algorithm includes: Initialize node attributes based on the predictive analysis module and generate an edge weight matrix based on complexity weights; The cost prediction demand vector is used as the node state, and the edge weight matrix is ​​composed of the time cost and complexity weight of the analysis order. The path value of each node is updated iteratively through the cost function, and the edge weight matrix is ​​adjusted accordingly. Generate the optimal prediction analysis sequence covering all nodes based on the adjusted edge weight matrix.

7. The intelligent prediction method for die stamping manufacturing costs according to claim 2, characterized in that, The method for constructing the benchmark process parameter library includes: Standard process samples under various typical material deformation states were collected, and baseline gradients and energy distribution thresholds were extracted. Multi-scale transformation of the baseline gradient is performed to generate a multi-resolution gradient model; The gradient model is classified according to the deformation category and associated with a reference threshold database; The classified gradient model is stored as a baseline process parameter library, and the model is updated periodically based on new samples.

8. The intelligent prediction method for die stamping manufacturing costs according to claim 3, characterized in that, The parameter optimization method of the parameter clustering algorithm includes: Calculate the initial clustering granularity and minimum parameter density based on the distribution of historical processing data; The parameter combinations are traversed using a grid search algorithm, and the parameter with the highest matching degree between the clustering result and the manually labeled parameter group is selected. The clustering granularity and minimum parameter density are dynamically adjusted based on the matching degree to optimize the node interaction strength partitioning accuracy.

9. The intelligent prediction method for die stamping manufacturing costs according to claim 6, characterized in that, The method for constructing the cost function includes: The cost between nodes is defined as a balance coefficient between time cost and complexity weight; The path value of each node is initialized to zero, and the starting cost value is a preset initial value; The minimum cumulative generation value of each node based on its predecessor nodes is calculated using recursive equations, and the optimal path is recorded. A complete predictive analysis sequence is generated by forward derivation based on the optimal path.

10. A smart prediction system for die stamping manufacturing costs, characterized in that, include: Multi-dimensional data acquisition module: used to acquire a multi-dimensional manufacturing data set, which includes material property parameters, process parameters and equipment operating parameters; wherein, the material property parameters include the material ductility coefficient and hardness distribution gradient, and the process parameters include the stamping speed adjustment value and the die wear index; Process feature extraction module: configured to extract key process features based on the material property parameters by analyzing the process parameters. The key process features include material deformation threshold, stress concentration factor and energy consumption fluctuation range. Topology construction module: used to generate a process influence topology map based on the process parameters using a feature association network construction algorithm. The influence topology map includes parameter interaction paths and node influence weights. Equipment data processing module: Performs time series segmentation processing on the equipment operating parameters to generate equipment state evolution trends; Multi-factor evaluation module: Input the key process characteristics, influence topology diagram and equipment status evolution trend into the multi-factor evaluation model to generate a cost prediction demand vector; the multi-factor evaluation model includes a data compression module and a weight mapping module; Prediction strategy generation module: Based on the predicted demand vector, it constructs an adaptive prediction strategy graph using a dynamic programming algorithm and outputs a manufacturing cost prediction scheme; the nodes of the adaptive prediction strategy graph represent the prediction analysis module, and the edges represent the analysis order and complexity weights.

Images