Stamping die wear prediction method and device based on big data analysis
The method and device use big data analysis to accurately predict stamping die wear by integrating multiple factors, enhancing prediction accuracy and enabling real-time monitoring and management.
Patent Information
- Application Number
- CN202411577716.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-07
- Publication Date
- 2024-12-06
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Traditional methods for predicting stamping die wear rely on empirical judgment and lack comprehensive analysis of multiple factors, leading to inaccurate predictions that fail to meet actual production needs.
A method and device using big data analysis to predict stamping die wear by mining mold wear sequences, performing feature normalization, introducing a self-attention mechanism, constructing a stacked coding network with sparsity constraints, and visualizing a wear prediction curve for real-time monitoring and management.
Improves the accuracy of mold wear prediction by integrating multiple factors, enabling real-time monitoring and optimizing production processes through intelligent management decisions.
Smart Images

Figure CN119089178A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of wear prediction, and in particular to a method and device for predicting wear of a stamping die based on big data analysis. Background Art
[0002] As a key link in metal processing, the efficiency and stability of the stamping process directly affect the quality and production cost of the product. However, as the core component of this process, the stamping die is subjected to high-intensity and high-frequency stamping operations for a long time, and is very prone to wear. Traditional die wear prediction methods mainly rely on empirical judgment or speculation based on a single factor, lacking analysis of the comprehensive impact of multiple factors under complex production conditions, resulting in prediction results that are often not accurate enough and difficult to meet actual production needs. Therefore, how to accurately predict the wear state of stamping dies and realize real-time monitoring and early warning of die wear has become a technical problem that needs to be solved urgently in the manufacturing industry. Summary of the invention
[0003] The present application provides a stamping die wear prediction method and device based on big data analysis, which solves the technical problem of inaccurate stamping die wear prediction in the prior art.
[0004] In view of the above problems, the present application provides a stamping die wear prediction method and device based on big data analysis.
[0005] In a first aspect of the present application, a method for predicting stamping die wear based on big data analysis is provided, the method comprising: Industrial big data retrieval is performed for stamping scenarios to mine mold wear sequences, wherein each mold wear sequence corresponds to the entire wear cycle and is marked with stamping conditions; the mold wear sequence is traversed, feature normalization processing based on stamping conditions is performed, and a self-attention mechanism is introduced to determine a wear feature library, wherein the feature value is positively correlated with the self-attention level; data-driven training is performed based on the wear feature library and in combination with the mold wear sequence to construct a mold monitoring module; sparsity constraints are introduced to construct a stacked coding network, wherein the sparsity constraints are used to constrain the activation of network layers and network nodes within layers under preset input conditions; in combination with the mold monitoring module, the mold stamping data is monitored and received, transmitted to the stacked coding network, network selective activation and feature extraction prediction are performed, and a wear prediction curve is determined, wherein the wear prediction curve includes multiple wear feature sub-curves and assembly wear sub-curves; the wear prediction curve is terminal visualized to reversely guide mold stamping management.
[0006] The second aspect of the present application provides a stamping die wear prediction device based on big data analysis, the device comprising: A retrieval module, the retrieval module is used to perform industrial big data retrieval for stamping scenarios and mine mold wear sequences, wherein each mold wear sequence corresponds to a full wear cycle and is marked with stamping conditions; a processing module, the processing module is used to traverse the mold wear sequence, perform feature normalization processing based on stamping conditions, intervene in the self-attention mechanism, and determine a wear feature library, wherein the feature value is positively correlated with the self-attention level; a training module, the training module is used to perform data-driven training based on the wear feature library and the mold wear sequence to construct a mold monitoring module; a network construction module, the network construction module is used to introduce sparsity constraints and construct a stacked coding network, wherein the sparsity constraints are used to perform activation constraints on network layers and network nodes within layers under preset input conditions; a prediction module, the prediction module is used to monitor and receive mold stamping data in combination with the mold monitoring module, transmit it to the stacked coding network, perform network selective activation and feature extraction prediction, and determine a wear prediction curve, wherein the wear prediction curve includes multiple wear feature sub-curves and assembly wear sub-curves; a management module, the management module is used to perform terminal visualization of the wear prediction curve and reversely guide mold stamping management.
[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: First, industrial big data retrieval is performed for stamping scenarios to mine mold wear sequences, where each mold wear sequence corresponds to the full wear cycle and is marked with stamping conditions. Next, the mold wear sequence is traversed, feature normalization processing based on stamping conditions is performed, and the self-attention mechanism is introduced to determine the wear feature library, where the feature value is positively correlated with the self-attention level. Then, according to the wear feature library, combined with the mold wear sequence, data-driven training is performed to construct a mold monitoring module. At the same time, sparsity constraints are introduced to construct a stacked coding network, where the sparsity constraints are used to constrain the activation of network layers and network nodes within the layer under preset input conditions. Further, combined with the mold monitoring module, the mold stamping data is monitored and received, transmitted to the stacked coding network, and the network selective activation and feature extraction prediction are performed to determine the wear prediction curve, where the wear prediction curve contains multiple wear feature sub-curves and assembly wear sub-curves. Finally, the wear prediction curve is terminal visualized to reversely guide the mold stamping management. The technical problem of inaccurate wear prediction of stamping molds in the prior art is solved, and the technical effect of improving the accuracy of mold wear prediction is achieved. BRIEF DESCRIPTION OF THE DRAWINGS
[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0009] Figure 1 A schematic diagram of a stamping die wear prediction method based on big data analysis provided in an embodiment of the present application; Figure 2 A schematic diagram of the structure of a stamping die wear prediction device based on big data analysis provided in an embodiment of the present application.
[0010] Explanation of the accompanying drawings: retrieval module 11, processing module 12, training module 13, network construction module 14, prediction module 15, management module 16. DETAILED DESCRIPTION
[0011] The present application solves the technical problem of inaccurate stamping die wear prediction in the prior art by providing a stamping die wear prediction method and device based on big data analysis.
[0012] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.
[0013] It should be noted that the terms "including" and "having" are intended to cover non-exclusive inclusions. For example, a process, method, apparatus, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules that are not explicitly listed or are inherent to these processes, methods, products or devices.
[0014] Embodiment 1, as Figure 1 As shown, the present application provides a stamping die wear prediction method based on big data analysis, wherein the method includes: Industrial big data retrieval is performed for stamping scenarios to mine mold wear sequences, where each mold wear sequence corresponds to the entire wear cycle and is marked with stamping conditions.
[0015] In a modern industrial environment, each stamping operation generates a large amount of production data, including the use time of the mold, the number of stamping times, the stamping material, the process parameters (such as stamping force, temperature, etc.), and the wear of each mold under different stamping conditions. By using big data technology, these scattered data can be centralized and all data related to mold wear can be retrieved. The wear of the mold does not occur instantaneously, but is a gradual process. In order to describe the entire wear process of the mold, it is necessary to serialize the wear of each mold to form a mold wear sequence. The mold wear sequence covers the entire wear life cycle of the mold, that is, the whole process from the start of use of a new mold to its final wear limit; in each wear sequence, the corresponding stamping conditions are identified, such as key process parameters such as stamping force, material type, stamping speed, and mold temperature.
[0016] The die wear sequence is traversed, feature normalization processing based on stamping conditions is performed, and a self-attention mechanism is introduced to determine a wear feature library, wherein the feature value is positively correlated with the self-attention level.
[0017] The wear of the die is affected by a variety of different stamping conditions. By traversing the die wear sequence, the stamping conditions are normalized to eliminate the dependence of local features on the stamping conditions, so that the prediction results of subsequent analysis and prediction are not restricted by conditions. After normalization, the self-attention mechanism is used to screen and evaluate important features to establish a more predictive wear feature library. The self-attention mechanism will weight and screen all input features, giving priority to those features that contribute more to the wear prediction effect in historical data, while suppressing those features that have less impact or are irrelevant. For example, for some stamping conditions (such as material hardness, stamping force, etc.), these features may have a greater impact in the model than other conditions, so the self-attention mechanism will assign higher weights to these features. Under the action of the self-attention mechanism, the device can dynamically evaluate the contribution of each stamping condition and feature to wear prediction, and the features weighted by self-attention are extracted and stored in a wear feature library.
[0018] Furthermore, the determination of the wear feature library includes: Based on the preset influence, determine the local feature set that is strongly correlated with the stamping conditions; perform zero-mean normalization on the local feature set to determine the normalized feature set; integrate the local feature set that is not strongly correlated with the normalized feature set, determine the self-attention level based on the feature value and mark it, and determine the wear feature library, wherein the feature value is guided by the wear relevance.
[0019] Among all stamping conditions and features, not every feature has a significant impact on die wear. Therefore, firstly, based on the preset influence standard, the local feature set with strong correlation with stamping conditions is screened out; the preset influence is an indicator to measure the correlation between features and wear prediction; based on the preset influence, those features with strong correlation with die wear are screened out, and these features are combined into a local feature set. For example, features such as stamping pressure, die temperature, and material hardness may have a high correlation with die wear, so they will be included in the local feature set.
[0020] After determining the strongly correlated local feature set, these features are further subjected to zero-mean normalization, that is, each feature value is subtracted from the feature mean and then divided by the standard deviation. This normalization method ensures that the feature mean is 0 and the standard deviation is 1, thereby eliminating the impact of different feature value ranges. After zero-mean normalization, a normalized feature set is generated, and the feature values in the normalized feature set are all within the same scale range, which is conducive to the subsequent processing of the self-attention mechanism.
[0021] Features that are not strongly correlated still contribute to mold wear prediction. By integrating the local feature set with non-strong correlation with the normalized feature set, a more comprehensive feature set is formed. The self-attention mechanism is introduced to assign an attention weight to each feature in the feature set. The size of the weight reflects the feature value of the feature. Specifically, the expert team evaluates the feature value based on factors such as the correlation between each feature and the wear result and the performance in historical data. The higher the feature value, the greater the contribution of the feature to wear prediction. Based on the feature value, a self-attention level identifier is assigned to each feature. Features with higher levels will receive more attention and computing resources in subsequent model processing.
[0022] Through multi-level feature screening, normalization processing, and self-attention mechanism integration, a feature library that can accurately describe mold wear is constructed. The wear feature library contains a set of features that have been screened, normalized, and integrated, and each feature has been assigned a corresponding self-attention level; at the same time, the feature value in the wear feature library is guided by wear relevance, which means that the features in the feature library are sorted and screened according to their contribution to mold wear prediction.
[0023] According to the wear feature library and in combination with the mold wear sequence, data-driven training is performed to construct a mold monitoring module.
[0024] Combine the wear feature library with the mold wear sequence, perform data-driven training, and build a mold monitoring module that can monitor and predict mold wear in real time. Specifically, use the wear features in the wear feature library and the mold wear sequence to train the machine learning model; by using the features in the wear feature library as input variables and the wear values in the mold wear sequence as target variables, the model will gradually learn the wear pattern of the mold under specific conditions based on these input-output relationships; after training, the mold monitoring module is obtained.
[0025] Furthermore, the mold monitoring module takes monitoring trigger, monitoring process and monitoring feedback trigger as training targets.
[0026] Preferably, the mold monitoring module takes monitoring trigger, monitoring process and monitoring feedback trigger as training targets; wherein, monitoring trigger refers to the mechanism by which the mold monitoring module determines when to start wear monitoring during the production process, which is mainly used to define the conditions and timing for the start of monitoring to ensure that wear monitoring is performed in an appropriate time period to avoid resource waste or monitoring delays. For example, when the number of mold uses reaches a certain threshold, the stamping conditions change (such as changes in stamping pressure or material type), or the mold temperature is detected to exceed a specific range, the monitoring process will be automatically triggered; the monitoring process refers to continuous monitoring and data collection of the mold wear state after triggering monitoring, for example, a fixed time interval can be set, or periodic monitoring can be performed according to the number of stamping times to ensure the timeliness of the data; monitoring feedback trigger means that when the monitoring module finds that the mold wear has reached a certain critical state or the prediction results indicate that intervention measures need to be taken, the feedback mechanism will be triggered and the results will be fed back to the operator, that is, when the wear curve is detected to be close to or reaches the preset wear critical value, the feedback mechanism will be immediately triggered, such as sending an alarm to the operator. By setting the three key training targets of monitoring trigger, monitoring process and monitoring feedback trigger, the mold monitoring module can realize full-process intelligent monitoring from data collection to result feedback. Monitoring triggers ensure that monitoring is started at the right time to avoid unnecessary waste of resources; the monitoring process ensures the continuity of monitoring and the timeliness of data, improving the accuracy of predictions; the monitoring feedback trigger provides timely feedback and response mechanisms when mold wear reaches a critical point, ensuring the safety and effectiveness of production.
[0027] A sparsity constraint is introduced to construct a stacked coding network, wherein the sparsity constraint is used to constrain the activation of network layers and network nodes within the layers under preset input conditions.
[0028] The sparsity constraint aims to make the network activation pattern more sparse by reducing redundant connections and information in the network. Specifically, the sparsity constraint forces the network to activate only a portion of important nodes or neurons when processing input data, rather than allowing all nodes to participate in the operation at the same time. This not only improves the network's computational efficiency, but also improves the model's generalization ability and prevents overfitting. The stacked coding network is used for efficient extraction and prediction of wear features. Through the sparsity constraint, the stacked coding network can automatically filter out the most critical features for the prediction results during the training process and suppress the influence of irrelevant or redundant features.
[0029] Furthermore, the stacked coding network includes an input layer, a multi-level coding layer and a top-level classifier, and the stacked coding network is constructed, including: Determine multiple analysis classes based on data type, prediction type and wear feature type; traverse the multiple analysis classes, combine with the wear feature library, configure a multi-level coding layer with the goal of wear feature identification and wear prediction, and the prediction type includes long-term prediction, short-term prediction and biased prediction; determine the mapping relationship between the network nodes of the multi-level coding layer and the multiple analysis classes, and configure the top-level classifier.
[0030] The stacked coding network is composed of an input layer, a multi-level coding layer and a top-level classifier. Specifically, the stacked coding network needs to classify the data according to different target tasks, and divide the wear prediction task into multiple specific analysis categories according to the data type (such as images, electrical signals, etc.), prediction type (such as long-term prediction, short-term prediction and bias prediction) and wear feature type (such as stamping force features, temperature features, etc.).
[0031] After determining the analysis class, the stacked coding network needs to configure a multi-level coding layer according to each analysis class. The main function of the multi-level coding layer is to extract features from the wear data layer by layer, so as to achieve accurate wear prediction. Specifically, traverse multiple analysis classes, combine the wear feature library, take wear feature identification and wear prediction as the goal, and dynamically configure the multi-level coding layer of each analysis class to ensure that each layer can effectively extract feature information useful to the analysis class.
[0032] According to the feature type of each analysis class, it is mapped to different network nodes in the multi-level encoding layer, which ensures that each feature can be processed and analyzed correctly. For example, features related to punching force may be mapped to a specific node in one layer, while features related to temperature are mapped to nodes in another layer.
[0033] After all features are extracted from the multi-level coding layer, the final wear prediction classification is performed through the top-level classifier. The main function of the top-level classifier is to perform the final classification and prediction of the features extracted from the multi-level coding layer to output the wear prediction results. The top-level classifier can be of various types, such as soft maximum classifier, support vector machine or decision tree; the classification targets of the top-level classifier include the results of long-term wear prediction, short-term wear prediction and bias prediction. Each prediction task will output the corresponding wear state, such as predicting the future life of the mold, short-term wear conditions, or wear trends under special conditions; the top-level classifier will output the wear prediction curve according to the results of each prediction task.
[0034] Furthermore, the method further comprises: Introducing the wear relationship: , where V is the wear volume, K is the wear coefficient, P is the die load, S is the stamping sliding distance, and T is the time; assisting the wear relationship, the prediction training of the multi-level coding layer is performed.
[0035] The wear relationship formula provides a physical framework to assist the multi-level coding layer in predictive training by associating the relationship between the wear volume and key variables (wear coefficient, die load, stamping sliding distance, time). Before predictive training, the variables in the wear relationship formula are extracted from the historical data, and the wear volume is calculated in combination with the material properties of the die; the multi-level coding layer extracts features from the wear data layer by layer, and introduces the wear relationship formula to constrain the features in the feature extraction and prediction of each layer. For example, the predicted results of the wear volume are compared with the wear volume obtained by the physical formula, and the model training is guided by calculating the error. In the training of the multi-level coding layer, the wear relationship formula is not only applicable to short-term wear prediction, but also to long-term wear trend prediction. For example, the short-term prediction can directly calculate the wear amount in a short period of time through the formula, and the long-term prediction can estimate the cumulative wear amount through the formula, thereby optimizing the weight adjustment of the multi-level coding layer; under specific stamping conditions (such as extremely high die load or special materials), the change of wear volume may deviate from the normal change trend. At this time, the physical formula can provide a reliable physical basis for biased prediction, so as to adjust the focus of feature extraction.
[0036] Combined with the mold monitoring module, the mold stamping data is monitored and received, transmitted to the stacked coding network, network selective activation and feature extraction prediction are performed, and the wear prediction curve is determined, wherein the wear prediction curve includes multiple wear feature sub-curves and assembly wear sub-curves.
[0037] The mold monitoring module monitors the operating status of the mold in real time and receives mold stamping data, including but not limited to stamping pressure, mold temperature, stamping speed, etc.; the mold stamping data is transmitted to the stacked coding network. The stacked coding network extracts features at different levels through multiple coding layers, and combines the wear relationship formula to generate a wear prediction curve. The wear prediction curve includes multiple wear feature sub-curves and assembly wear sub-curves. The multiple wear feature sub-curves represent the wear conditions of the mold under different wear characteristics, and the assembly wear sub-curve reflects the overall wear condition of the mold.
[0038] Furthermore, the network selective activation and feature extraction prediction include: The stacked coding network receives the mold monitoring data after triggering and returning the monitoring data; the mold monitoring data is traversed, and the top-level classifier performs data recognition and prediction task classification to determine a multi-level sparse coding layer; based on the multi-level sparse coding layer, the mold monitoring data is processed and integrated and output accordingly.
[0039] The monitoring trigger is used to collect and transmit relevant monitoring data. After the stacked coding network receives the mold monitoring data, the network first needs to traverse and analyze the data, and the top-level classifier classifies the tasks. The top-level classifier is responsible for identifying various types of data transmitted to the network and classifying the tasks according to different prediction targets. After the top-level classifier classifies the tasks, the corresponding multi-level sparse coding layer is configured according to different prediction tasks, and the key information in the mold monitoring data is extracted layer by layer through the sparse activation mechanism. Sparsity constraints are introduced in the multi-level coding layer so that only a small number of nodes in each layer are activated. Different task types will activate different coding layers. For example, long-term wear prediction may activate a deeper coding layer to extract high-level features of wear trends; short-term prediction may only activate a shallow coding layer to extract fast-changing features under the current working conditions. Features are extracted step by step according to the coding layers at each level. At each level, the network will extract and process only relevant data features through the sparse activation mechanism. For example, the temperature and punching pressure of the mold may only activate certain nodes, while features such as sliding distance or time may activate other nodes. After being processed by the multi-level sparse coding layer, the network integrates the extracted mold monitoring data features and outputs the final prediction results. This output process not only includes the prediction of the wear state, but also combines the task classification results of the previous classifier to generate wear prediction curves under different tasks.
[0040] Furthermore, each level of the multi-level coding layer includes an encoder and a decoder, the encoder maps the mold monitoring data to a low-dimensional sparse representation, and the decoder reconstructs the original data from the low-dimensional sparse representation.
[0041] Each level of the multi-level coding layer contains an encoder and a decoder. The main function of the encoder is to map high-dimensional input data to a lower-dimensional sparse representation, thereby extracting the core features of the mold monitoring data. The encoder compresses the input data layer by layer through the neurons in the network, gradually reducing the dimension of the data. During the compression process, the encoder will use sparsity constraints to ensure that the output low-dimensional representation is sparse. For example, under certain mold working conditions, punching pressure and temperature may be the decisive factors, while other parameters (such as material hardness) may have little effect under the current working conditions. The encoder retains the features closely related to wear prediction (such as punching pressure and temperature) through sparse processing, and compresses the secondary features; the role of the decoder is to reconstruct the original data from the low-dimensional sparse representation through reverse mapping to ensure that no key information is lost during the compression process. The decoder remaps the low-dimensional sparse representation compressed by the encoder to the original data space through layer-by-layer expansion.
[0042] Furthermore, determining the wear prediction curve includes: Based on the strong correlation between features, the feature synergy relationship is determined; based on the feature synergy relationship, the wear prediction data is simultaneously coupled to determine the coupled prediction data; curve coordinates are constructed with multiple feature axes and time axes, and the wear prediction data and the coupled prediction data are converted from digital to graphical to determine the wear prediction curve.
[0043] In wear prediction, different features do not exist independently, but may be strongly correlated with each other. For example, there may be some synergy between punching force, temperature and mold material properties. By performing correlation analysis on historical data, we can identify the combination of features that are closely related to wear. For example, punching force and temperature may show strong correlation under specific working conditions. By calculating the correlation coefficient (such as the Pearson correlation coefficient), we can quantify the strength of the correlation between different features and thus determine the synergistic relationship of the features.
[0044] According to the synergistic relationship of the features, the wear prediction data is coupled simultaneously. Simultaneous coupling refers to coupling multiple features on the same time axis so that the synergistic effect of these features can be integrated and analyzed at the same time point. Through simultaneous coupling, related features can be combined to generate coupled prediction data. These data not only include the changes of each single feature, but also reflect the wear changes under the synergistic effect of multiple features. For example, the increase in punching force and the change in temperature may affect the wear trend at the same time. Through coupling, a wear data point that reflects the influence of multiple features can be generated at the same time.
[0045] Since the effects of multiple features on wear are synergistic, a feature axis is created for each important feature. For example, features such as punching force, temperature, and material hardness all have their own feature axes, which can reflect the role and changes of different features in the wear process. The time axis is used to reflect the changes in wear prediction over time. The coupled prediction data and the original wear prediction data are mapped to the feature axis and the time axis respectively, forming a curve chart with time dimension and multi-feature dimension. Each point not only represents the wear volume or trend, but also contains the feature combination corresponding to that moment.
[0046] The wear prediction curve is terminally visualized to provide reverse guidance for die stamping management.
[0047] By visually displaying the wear prediction curve on the terminal, operators and managers can be provided with intuitive real-time information, enabling them to understand the wear status of the mold in a timely manner and make corresponding management decisions. By reversely guiding mold stamping management, operators can optimize the production process, adjust stamping process parameters, and formulate accurate maintenance and replacement plans based on the prediction results. This intelligent wear management method not only improves the service life of the mold, but also optimizes production efficiency and reduces maintenance costs.
[0048] In summary, the embodiments of the present application have at least the following technical effects: First, industrial big data retrieval is performed for stamping scenarios to mine mold wear sequences, where each mold wear sequence corresponds to the full wear cycle and is marked with stamping conditions. Next, the mold wear sequence is traversed, feature normalization processing based on stamping conditions is performed, and the self-attention mechanism is introduced to determine the wear feature library, where the feature value is positively correlated with the self-attention level. Then, according to the wear feature library, combined with the mold wear sequence, data-driven training is performed to construct a mold monitoring module. At the same time, sparsity constraints are introduced to construct a stacked coding network, where the sparsity constraints are used to constrain the activation of network layers and network nodes within the layer under preset input conditions. Further, combined with the mold monitoring module, the mold stamping data is monitored and received, transmitted to the stacked coding network, and the network selective activation and feature extraction prediction are performed to determine the wear prediction curve, where the wear prediction curve contains multiple wear feature sub-curves and assembly wear sub-curves. Finally, the wear prediction curve is terminal visualized to reversely guide the mold stamping management. The technical problem of inaccurate wear prediction of stamping molds in the prior art is solved, and the technical effect of improving the accuracy of mold wear prediction is achieved.
[0049] Embodiment 2 is based on the same inventive concept as the stamping die wear prediction method based on big data analysis in the above embodiment. Figure 2 As shown, the present application provides a stamping die wear prediction device based on big data analysis, wherein the device includes: A retrieval module 11 is used to perform industrial big data retrieval for stamping scenarios and mine mold wear sequences, wherein each mold wear sequence corresponds to a full wear cycle and is marked with stamping conditions; a processing module 12 is used to traverse the mold wear sequence, perform feature normalization processing based on stamping conditions, intervene in the self-attention mechanism, and determine a wear feature library, wherein the feature value is positively correlated with the self-attention level; a training module 13 is used to perform data-driven training based on the wear feature library and the mold wear sequence to construct a mold monitoring module; a network construction module 14 is used to The network construction module 14 is used to introduce sparsity constraints and construct a stacked coding network, wherein the sparsity constraints are used to perform activation constraints on network layers and network nodes within layers under preset input conditions; the prediction module 15 is used to combine with the mold monitoring module, monitor and receive mold stamping data, transmit it to the stacked coding network, perform network selective activation and feature extraction prediction, and determine the wear prediction curve, wherein the wear prediction curve includes multiple wear feature sub-curves and assembly wear sub-curves; the management module 16 is used to perform terminal visualization of the wear prediction curve and reversely guide mold stamping management.
[0050] Furthermore, the processing module 12 is used to execute the following method: Based on the preset influence, determine the local feature set that is strongly correlated with the stamping conditions; perform zero-mean normalization on the local feature set to determine the normalized feature set; integrate the local feature set that is not strongly correlated with the normalized feature set, determine the self-attention level based on the feature value and mark it, and determine the wear feature library, wherein the feature value is guided by the wear relevance.
[0051] Furthermore, the training module 13 is used to perform the following method: The mold monitoring module takes monitoring trigger, monitoring process and monitoring feedback trigger as training targets.
[0052] Furthermore, the network construction module 14 is used to execute the following method: Determine multiple analysis classes based on data type, prediction type and wear feature type; traverse the multiple analysis classes, combine with the wear feature library, configure a multi-level coding layer with the goal of wear feature identification and wear prediction, and the prediction type includes long-term prediction, short-term prediction and biased prediction; determine the mapping relationship between the network nodes of the multi-level coding layer and the multiple analysis classes, and configure the top-level classifier.
[0053] Furthermore, the network construction module 14 is used to execute the following method: Introducing the wear relationship: , where V is the wear volume, K is the wear coefficient, P is the die load, S is the stamping sliding distance, and T is the time; assisting the wear relationship, the prediction training of the multi-level coding layer is performed.
[0054] Furthermore, the prediction module 15 is used to perform the following method: The stacked coding network receives the mold monitoring data after triggering and returning the monitoring data; the mold monitoring data is traversed, and the top-level classifier performs data recognition and prediction task classification to determine a multi-level sparse coding layer; based on the multi-level sparse coding layer, the mold monitoring data is processed and integrated and output accordingly.
[0055] Furthermore, the prediction module 15 is used to perform the following method: Each level of the multi-level coding layer includes an encoder and a decoder. The encoder maps the mold monitoring data to a low-dimensional sparse representation, and the decoder reconstructs the original data from the low-dimensional sparse representation.
[0056] Furthermore, the prediction module 15 is used to perform the following method: Based on the strong correlation between features, the feature synergy relationship is determined; based on the feature synergy relationship, the wear prediction data is simultaneously coupled to determine the coupled prediction data; curve coordinates are constructed with multiple feature axes and time axes, and the wear prediction data and the coupled prediction data are converted from digital to graphical to determine the wear prediction curve.
[0057] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0058] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.
[0059] This specification and drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.
Claims
1. Stamping die wear prediction method based on big data analysis, It is characterized in that The method comprises: Industrial big data retrieval is performed for stamping scenarios to mine die wear sequences, where each die wear sequence corresponds to the entire wear cycle and is marked with stamping conditions; Traversing the die wear sequence, performing feature normalization processing based on stamping conditions, introducing a self-attention mechanism, and determining a wear feature library, wherein the feature value is positively correlated with the self-attention level; According to the wear feature library and the mold wear sequence, data-driven training is performed to construct a mold monitoring module; Introducing sparsity constraints to construct a stacked coding network, wherein the sparsity constraints are used to constrain the activation of network layers and network nodes within the layers under preset input conditions; Combined with the mold monitoring module, the mold stamping data is monitored and received, and transmitted to the stacked coding network, and network selective activation and feature extraction prediction are performed to determine the wear prediction curve, wherein the wear prediction curve includes multiple wear feature sub-curves and an assembly wear sub-curve; The wear prediction curve is terminally visualized to provide reverse guidance for die stamping management.
2. The stamping die wear prediction method based on big data analysis according to claim 1, It is characterized in that The determination of the wear feature library comprises: Based on the preset influence, determine the local feature set that is strongly related to the stamping conditions; Performing zero-mean normalization processing on the local feature set to determine a normalized feature set; The local feature set with non-strong correlation is integrated with the normalized feature set, and the self-attention level is determined and marked based on the feature value to determine the wear feature library, wherein the feature value is guided by the wear correlation.
3. The stamping die wear prediction method based on big data analysis as claimed in claim 1, It is characterized in that The mold monitoring module takes monitoring trigger, monitoring process and monitoring feedback trigger as training targets.
4. The stamping die wear prediction method based on big data analysis according to claim 1, It is characterized in that The stacked coding network comprises an input layer, a multi-level coding layer and a top-level classifier. The stacked coding network is constructed, including: Determine multiple analysis categories based on data type, prediction type, and wear feature type; Traversing the multiple analysis classes, combining the wear feature library, taking wear feature identification and wear prediction as the goal, configuring a multi-level coding layer, wherein the prediction types include long-term prediction, short-term prediction and biased prediction; Determine the mapping relationship between the network nodes of the multi-level coding layer and the multiple analysis classes, and configure the top-level classifier.
5. The stamping die wear prediction method based on big data analysis as claimed in claim 4, It is characterized in that The method further comprises: Introducing the wear relationship: , where V is the wear volume, K is the wear coefficient, P is the die load, S is the punch sliding distance, and T is the time; The wear relationship is assisted to perform prediction training of the multi-level coding layer.
6. The stamping die wear prediction method based on big data analysis as claimed in claim 4, It is characterized in that The network selective activation and feature extraction prediction include: Triggering and transmitting monitoring data back, the stacked coding network receives the mold monitoring data; Traversing the mold monitoring data, the top-level classifier performs data recognition and prediction task classification, and determines a multi-level sparse coding layer; Based on the multi-level sparse coding layer, the mold monitoring data is processed and integrated and outputted accordingly.
7. The stamping die wear prediction method based on big data analysis according to claim 6, It is characterized in that Each level of the multi-level coding layer includes an encoder and a decoder. The encoder maps the mold monitoring data to a low-dimensional sparse representation, and the decoder reconstructs the original data from the low-dimensional sparse representation.
8. The stamping die wear prediction method based on big data analysis according to claim 1, It is characterized in that The step of determining the wear prediction curve comprises: Determine the feature synergy relationship based on the strong correlation between features; Based on the characteristic synergistic relationship, the wear prediction data is coupled simultaneously to determine the coupled prediction data; The curve coordinates are constructed with multiple characteristic axes and a time axis, and the wear prediction data and the coupling prediction data are converted into digital images to determine the wear prediction curve.
9. Stamping die wear prediction device based on big data analysis, It is characterized in that The device for implementing the stamping die wear prediction method based on big data analysis according to any one of claims 1 to 8 comprises: A retrieval module, which is used to perform industrial big data retrieval for stamping scenarios and mine die wear sequences, wherein each die wear sequence corresponds to a full wear cycle and is marked with stamping conditions; A processing module, the processing module is used to traverse the die wear sequence, perform feature normalization processing based on stamping conditions, intervene in the self-attention mechanism, and determine the wear feature library, wherein the feature value is positively correlated with the self-attention level; A training module, the training module is used to perform data-driven training based on the wear feature library and the mold wear sequence to construct a mold monitoring module; A network construction module, wherein the network construction module is used to introduce sparsity constraints to construct a stacked coding network, wherein the sparsity constraints are used to constrain the activation of network layers and network nodes within the layers under preset input conditions; A prediction module, the prediction module is used to monitor and receive the die stamping data in combination with the die monitoring module, transmit the data to the stacked coding network, perform network selective activation and feature extraction prediction, and determine a wear prediction curve, wherein the wear prediction curve includes a multi-wear feature sub-curve and an assembly wear sub-curve; A management module is used to perform terminal visualization on the wear prediction curve and reversely guide mold stamping management.
Citation Information
Patent Citations
Turbofan engine remaining service life prediction method based on improved stacked sparse auto-encoder and attention echo state network
CN113743016A
Cutter wear state prediction method and device and storage medium
CN114742322A
Measurement device of ring size
KR102069278B1
