A product forging data intelligent management system and method based on 5G cloud computing

Abstract

The application discloses a product forging data intelligent management system and method based on 5G cloud computing, and relates to the field of intelligent manufacturing.The system collects temperature, stress, load and image and other multi-source data through a process data module and marks them with weights; a multi-model reasoning module selects support vector regression, random forest and convolutional neural network in parallel prediction according to data characteristics; a prediction fusion module generates unified prediction results by using a nonlinear function based on confidence score; a cloud-edge collaborative scheduling module determines cloud or edge deployment in combination with task complexity, edge computing capability and network state; a parameter optimization module optimizes process parameters by using an EPQ-Opt algorithm based on energy consumption prediction and residual score; and a security guarantee module adjusts strategies and guarantees data security through consistency detection and robustness evaluation.The application significantly improves prediction accuracy and response speed, reduces energy consumption and enhances system security, and is suitable for high-end forging and intelligent manufacturing fields.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent manufacturing, and in particular to a product forging data intelligent management system and method based on 5G cloud computing. BACKGROUND

[0002] At present, in the forging process, the traditional control system mainly relies on a single data source to adjust the process parameters, and lacks comprehensive collection and effective fusion of multi-source data. The process data collection, processing and analysis capability of the existing system is limited, and only simple parameter estimation or feedback can be provided, resulting in insufficient accuracy of process control, especially when facing complex production conditions, the system is difficult to quickly respond and adjust, and cannot maintain production efficiency under high precision requirements.

[0003] In addition, the existing forging process system generally lacks intelligent analysis and optimization methods. Many systems rely on a single prediction model or empirical rules for reasoning and optimization, and cannot comprehensively consider the characteristics of different types of data, nor have an effective model fusion mechanism. Therefore, the accuracy of the prediction result and the self-adaptive ability of the system are weak, and the precision and efficiency of process optimization cannot be effectively improved. In addition, the traditional system also has problems in scheduling computing resources, which cannot be dynamically optimized according to the complexity of production tasks and resource status, resulting in waste of computing resources or response delay.

[0004] Therefore, an intelligent management system is needed to overcome the shortcomings of the prior art, provide comprehensive data collection, intelligent reasoning and multi-model fusion, and improve the accuracy of control and optimization efficiency of the forging process through cloud-edge collaborative scheduling and security mechanism. SUMMARY

[0005] The purpose of the present application is to provide a product forging data intelligent management system and method based on 5G cloud computing, which solves the limitations of inaccurate data collection, insufficient model reasoning and fusion capability, inflexible computing resource scheduling and insufficient security in the prior art, and significantly improves the accuracy of control and production efficiency of the forging process through intelligent data processing, multi-model fusion and optimization, cloud-edge collaborative scheduling and security mechanism, while ensuring the efficiency and data security of the system.

[0006] In order to achieve the above purpose, the present application realizes the following technical solutions:

[0007] On the one hand, the present application provides a product forging data intelligent management system based on 5G cloud computing, comprising:

[0008] A process data module is used to acquire multi-source process data in the forging process, and to evaluate the data credibility based on the running state of the collection equipment and the network transmission state, and to generate process feature data after weighted labeling.

[0009] a multi-model inference module configured to receive the process feature data, select a plurality of prediction sub-models with heterogeneous modeling methods according to the feature type identification and the integrity score, including a support vector regression model, a random forest model and a convolutional neural network model, and call the prediction sub-models in a parallel manner to output prediction results of the plurality of sub-models;

[0010] a prediction fusion module configured to receive the prediction results of the plurality of sub-models, calculate a confidence score based on the performance of each sub-model in a historical validation data set, and generate a unified prediction result by dynamically assigning fusion weights using a nonlinear normalization function according to the score result and a set fusion path control strategy;

[0011] a cloud-edge collaborative scheduling module configured to generate a deployment instruction according to the unified prediction result, an edge node computing resource state and a 5G network transmission state, determine a deployment path of the multi-model inference module based on a comparison analysis of the inference task complexity level, the edge node processing capability and the network stability, and output a deployment type determination result;

[0012] a parameter optimization module configured to construct a target function based on the unified prediction result, the target function integrating an energy consumption prediction value and a fusion model residual error score mechanism, and performing process parameter combination optimization using an optimization algorithm, and outputting the optimization result to an industrial control system;

[0013] a security assurance module connected with the multi-model inference module, the prediction fusion module and the cloud-edge collaborative scheduling module, respectively, configured to perform robustness evaluation on the sub-models, detect the credibility of the fusion result based on a historical data consistency score mechanism, automatically trigger model fusion strategy adjustment when the score is lower than a preset threshold, and implement access control and encryption protection on the data transmission process in the system.

[0014] Preferably, the process data module comprises:

[0015] a trusted acquisition unit configured to acquire multi-source process data of temperature, stress, load and image in the forging process, and quantitatively evaluate the data credibility based on the running state of the acquisition equipment and the network transmission state to output the weighted and labeled process data;

[0016] a semantic preprocessing unit configured to receive the weighted and labeled process data, perform data consistency analysis based on a stress-temperature prediction model constructed based on the thermal-mechanical coupling relationship, identify abnormal samples and perform interpolation completion on missing data, and perform time synchronization processing on image data and non-image data;

[0017] ​A feedback optimization unit is configured to receive prediction error information output by the stress-temperature prediction model, dynamically adjust feature screening rules and normalization parameter configurations, improve consistency and effectiveness of subsequent modeling data, and output process feature data.

[0018] Preferably, the stress-temperature prediction model comprises:

[0019] A regression model constructed based on forging process historical data, the regression model taking temperature, strain rate and loading rate at multiple time points in the same process as input variables and taking material stress as an output variable, for describing a nonlinear relationship of a thermal-mechanical process;

[0020] The regression model, when in use, dynamically selects a most matched sub-model or adopts a weighted fusion output result based on feature similarity between current process data and historical sample data, for identifying an abnormal data point and performing regression estimation completion on a missing value.

[0021] Preferably, the multi-model inference module comprises:

[0022] A plurality of structurally heterogeneous sub-models, including a vector regression model, a random forest model and a convolutional neural network model, respectively suitable for processing a subset of time-series sensor data, statistical feature data and image-type process feature data;

[0023] A model calling control unit configured to receive the process feature data, dynamically select a matched sub-model combination according to a type identifier and an integrity score result of each data feature, and perform a prediction task in a parallel manner to generate prediction output results of the plurality of sub-models.

[0024] Preferably, the prediction fusion module comprises:

[0025] A confidence evaluation unit configured to receive prediction output results of the plurality of sub-models and generate corresponding historical reliability scores according to performance of each sub-model in a historical verification data set, indicating prediction accuracy and stability of each sub-model in past samples;

[0026] A fusion weight calculation unit configured to calculate a fusion weight of each sub-model based on a function, formula being:

[0027] ;

[0028] Wherein: is a reliability score of an i-th sub-model obtained on a historical verification data set, indicating prediction accuracy or stability of the i-th sub-model in past samples; is a historical reliability score of the i-th sub-model, used for normalization weight calculation; is a historical reliability score of the i-th sub-model, used for normalization weight calculation; is a historical reliability score of the i-th sub-model, used for normalization weight calculation; is a historical reliability score of the i-th sub-model, used for normalization weight calculation. a fusion weight of each sub-model; a total number of sub-models currently participating in fusion; a response adjustment coefficient, being a positive real number, used to control the sensitivity of the impact of score difference on the weight; a score reference value, used to set the center value of the score function; a base number of natural logarithm;

[0029] a fusion output unit configured to perform weighted processing on the prediction output of each sub-model according to the fusion weight to generate a unified prediction result, expressed as:

[0030] ;

[0031] wherein: a unified prediction result obtained by weighted fusion of the outputs of the plurality of sub-models; a prediction output of the i-th sub-model for the current input data.

[0032] Preferably, the cloud-edge collaborative scheduling module comprises:

[0033] an inference awareness unit configured to receive the unified prediction result and analyze the complexity level of the inference task based on the number of sub-models involved in the prediction result, the calling frequency, and the data feature dimension;

[0034] a path decision unit configured to compare the complexity level of the inference task with preset edge node processing capability threshold and communication link stability threshold, and select an edge deployment path or a cloud deployment path according to the comparison result and output a deployment type determination result;

[0035] an inference deployment generation unit configured to construct a deployment instruction according to the deployment type determination result, the deployment instruction comprising a target node identifier, a model calling sequence, and a model loading priority, for controlling the deployment and execution of the multi-model inference module in the edge or the cloud.

[0036] Preferably, the parameter optimization module adopts an optimization algorithm to perform process parameter combination optimization, and the optimization algorithm is constructed based on a multi-objective fitness function as follows:

[0037] ;

[0038] wherein: indicates the combination of forging process parameters to be optimized; an energy consumption prediction value estimated based on the unified prediction result; ​​​​​a prediction quality score output by the fusion model residual score mechanism; a system reference energy consumption upper limit; 、 a minimum value and a maximum value respectively set for the system score mechanism; 、 a target weight coefficient for adjusting the weight relationship between the energy efficiency target and the prediction quality target;

[0039] the The optimization algorithm adopts a particle swarm optimization method to search for an optimal parameter combination under the constraint of a fitness function, and outputs the optimal parameter combination to an industrial control system to realize process control.

[0040] Preferably, the security assurance module comprises:

[0041] a trustworthiness detection unit configured to compare the unified prediction result with historical data, specifically including the following steps:

[0042] obtaining a distribution consistency score of the current prediction result and the historical sample data;

[0043] if the consistency score is lower than a preset trustworthiness threshold, triggering an alarm signal and starting a fusion strategy adjustment mechanism;

[0044] starting a model self-adaptive adjustment mechanism to dynamically adjust the weight of the fusion model according to historical data and current prediction errors, and optimize subsequent prediction effects.

[0045] In another aspect, the present application provides a product forging data intelligent management method based on 5G cloud computing, applied to the product forging data intelligent management system based on 5G cloud computing as described above, comprising the following steps:

[0046] Step 1: obtaining multi-source process data in the forging process, the process data including temperature, stress, load and image data; and performing trustworthiness evaluation on the data based on the running state of the collection equipment and the network transmission state, to generate weighted labeled process feature data;

[0047] Step 2: receiving the process feature data, selecting a plurality of prediction sub-models with heterogeneous modeling methods according to the feature type identifier and the integrity score, the prediction sub-models including a support vector regression model, a random forest model and a convolutional neural network model, and performing inference in a parallel manner to generate prediction results of the plurality of sub-models;

[0048] Step 3: receiving the prediction results of the plurality of sub-models, calculating a confidence score based on the performance of each sub-model in the historical verification data set, and according to the score result and the set fusion path control strategy, assigning fusion weights using a nonlinear normalization function to generate a unified prediction result;

[0049] Step 4: generate deployment instructions based on the unified prediction results, edge node computing resource state and 5G network transmission state, and determine the deployment path of the multi-model inference module based on the comparison and analysis of the inference task complexity level, edge node processing capability and network stability, and output the deployment type determination result;

[0050] Step 5: based on the unified prediction results, construct a target function containing energy consumption prediction value and fusion model residual error scoring mechanism, and use Optimization algorithm to perform process parameter combination optimization, and output the optimization results to the industrial control system for process control;

[0051] Step 6: perform robustness evaluation on the sub-model, and perform credibility detection on the fusion result based on the historical data consistency scoring mechanism; when the score is lower than the preset threshold, automatically trigger the model fusion strategy adjustment, and implement access control and encryption protection on the data transmission process in the system.

[0052] The beneficial effects of the present application are: the present application comprehensively collects multi-source data such as temperature, stress, load and image of the forging process through the process data module, and combines the equipment running state and network transmission state to perform credibility quantitative evaluation, generates weighted labeled high reliability process feature data, and significantly improves the data quality. The multi-model inference module dynamically selects heterogeneous sub-models such as support vector regression, random forest and convolutional neural network according to the feature type identifier and data integrity score, and performs parallel prediction, fully utilizes the advantages of each model, and improves the prediction accuracy under complex working conditions. The prediction fusion module is based on the confidence score of the historical verification data, and uses a nonlinear normalization function to dynamically optimize the weight distribution, realizes high reliable weighted fusion of the prediction results, and is better than the traditional linear weighting method. The cloud edge collaborative scheduling module combines the characteristics of 5G network, intelligently generates deployment instructions and flexibly selects cloud or edge path through comprehensive analysis of inference task complexity, edge node computing capability and network stability, significantly reduces resource waste and response delay. The parameter optimization module uses Optimization algorithm to realize global optimization of process parameters, which effectively reduces energy consumption while ensuring prediction accuracy, and improves production efficiency. The security protection module automatically triggers the model fusion strategy adjustment through robustness evaluation and consistency detection, and performs access control and encryption protection on data transmission, which enhances the stability and security of the system. Therefore, through the collaborative application of multi-source data credible collection, intelligent inference of heterogeneous models, dynamic fusion optimization and cloud edge collaborative scheduling, the present application has made significant progress in process control accuracy, energy efficiency optimization and system security, and has obvious technical advantages and industrial application value. BRIEF DESCRIPTION OF DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative labor. Among them:

[0054] Figure 1 It is a schematic diagram of the overall structure of the system of the present application.

[0055] Figure 2 It is a flow chart of the method of the present application. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solutions and advantages of the embodiments of the present application more clear, the following will combine the drawings of the embodiments of the present application to clearly and completely describe the technical solutions of the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all. Based on the described embodiments of the present application, all other embodiments obtained by those skilled in the art belong to the scope of protection of the present application.

[0057] As shown in Figure 1 , it is an embodiment of the present application, which provides a product forging data intelligent management system based on 5G cloud computing, comprising:

[0058] (1) Process data module

[0059] used for acquiring multi-source process data in the forging process, and evaluating the data credibility based on the running state of the acquisition equipment and the network transmission state, and generating process feature data after weighted labeling;

[0060] The process data module comprises:

[0061] A trusted acquisition unit is used for acquiring multi-source process data of temperature, stress, load and image in the forging process, and quantitatively evaluating the data credibility based on the running state of the acquisition equipment and the network transmission state, and outputting the weighted labeled process data;

[0062] A semantic preprocessing unit is used for receiving the weighted labeled process data, performing data consistency analysis based on the stress-temperature prediction model constructed based on the thermal-mechanical coupling relationship, identifying abnormal samples and performing interpolation completion on missing data, and simultaneously performing time synchronization processing on image data and non-image data;

[0063] A feedback optimization unit is used for receiving the prediction error information output by the stress-temperature prediction model, dynamically adjusting the feature selection rules and the normalization parameter configuration, improving the consistency and effectiveness of the subsequent modeling data, and outputting the process feature data.

[0064] As a preferred implementation of this embodiment, the stress-temperature prediction model includes:

[0065] A regression model constructed based on historical forging process data uses temperature, strain rate, and loading rate at multiple time points in the same process as input variables and material stress as output variable to describe the nonlinear relationship between thermal and mechanical processes;

[0066] When the regression model is used, based on the feature similarity between the current process data and the historical sample data, the most matching sub-model is dynamically selected or the weighted fusion output result is adopted to identify abnormal data points and perform regression estimation to complete the missing values.

[0067] During the forging process, the stress distribution of the material is significantly coupled with temperature changes. To improve the reliability of the collected data, this embodiment constructs a physical model based on the thermal-mechanical coupling relationship to detect and repair missing values ​​and anomalies in multi-source sensor data. Specifically, the system constructs the following thermodynamic relationship model using historical forging data sets:

[0068] ;

[0069] in: is the stress of the material during the forging process, is the temperature, For strain, is the strain rate. This model is used to determine whether the data collected by the sensor exceeds the reasonable data range; if the data collected by some sensors within a certain period of time cannot meet the model, the system will automatically trigger the data interpolation and repair process.

[0070] The implementation of this model uses empirical formulas based on finite element calculation models, or models trained through machine learning to perform data repair. The specific process is as follows: by comparing the similarity between current process data and historical data, abnormal data points are identified; for identified missing or abnormal data points, reasonable data are estimated using interpolation or regression models.

[0071] This process ensures the accuracy of data repair and improves data quality by combining physical models and machine learning technology, thereby providing reliable data support for subsequent process optimization and predictive analysis.

[0072] (2) Multi-model inference module

[0073] for receiving the process feature data, selecting a plurality of prediction sub-models with heterogeneous modeling methods according to the feature type identifier and the completeness score, including a support vector regression model, a random forest model, and a convolutional neural network model, calling the prediction sub-models in parallel, and outputting prediction results of the plurality of sub-models;

[0074] The multi-model inference module includes:

[0075] Multiple structurally heterogeneous sub-models, including at least a vector regression model, a random forest model, and a convolutional neural network model, which are respectively suitable for processing subsets of time series sensor data, statistical feature data, and image-based process feature data;

[0076] The model calling control unit is used to receive the process feature data, dynamically select matching sub-model combinations based on the type identification and integrity score results of each data feature, and execute prediction tasks in parallel to generate prediction output results of multiple sub-models.

[0077] In this embodiment, through the design of the above-mentioned multi-model reasoning module, the advantages of different models in processing multi-source process data such as time series, statistics and images can be fully utilized, the comprehensiveness and accuracy of the prediction results can be improved, and good stability and adaptability can be maintained under complex working conditions, thereby providing reliable data support for subsequent prediction fusion and process optimization.

[0078] (3) Prediction Fusion Module

[0079] Used to receive the prediction results of the multiple sub-models, calculate the confidence score based on the performance of each sub-model in the historical verification data set, and dynamically assign fusion weights using a nonlinear normalization function based on the score results and the set fusion path control strategy to generate a unified prediction result;

[0080] The prediction fusion module includes:

[0081] The confidence evaluation unit is used to receive the prediction output results of multiple sub-models and generate a corresponding historical reliability score based on the performance of each sub-model in the historical validation dataset, indicating the prediction accuracy and stability of each sub-model in the past samples;

[0082] The fusion weight calculation unit is used to calculate the fusion weight of each sub-model based on the following function:

[0083] ;

[0084] in: For the The reliability score obtained by each sub-model on the historical validation dataset indicates its prediction accuracy or stability in past samples; For the a historical reliability score of the i-th sub-model, used for normalization weight calculation; a fusion weight of the i-th sub-model; a fusion weight of the i-th sub-model; a total number of sub-models participating in fusion at present; a response adjustment coefficient, being a positive real number, used for controlling the sensitivity of score difference to weight; a score reference value, used for setting the center value of the score function; a base number of natural logarithm;

[0085] a fusion output unit, configured to perform weighted processing on the prediction output of each sub-model according to the fusion weight to generate a unified prediction result, with the expression being:

[0086] ;

[0087] wherein: the unified prediction result after weighted fusion of the prediction outputs of the plurality of sub-models; the prediction output of the i-th sub-model to the current input data.

[0088] In the embodiment, through the design of the prediction fusion module, the advantages of each sub-model on different data features can be fully utilized, the prediction results of each sub-model are reasonably weighted based on the reliability score of the historical verification data, the overall precision and stability of the fusion result are improved, the prediction deviation possibly caused by a single model is effectively reduced under complex working conditions, and thus a more reliable basis is provided for subsequent process optimization and parameter adjustment.

[0089] (Four) Cloud-edge collaborative scheduling module

[0090] configured to generate a deployment instruction according to the unified prediction result, the state of edge node computing resources and the state of 5G network transmission, and determine a deployment path of the multi-model inference module based on comparison and analysis of the complexity level of the inference task, the processing capability of the edge node and the stability of the network, and output a deployment type determination result;

[0091] The cloud-edge collaborative scheduling module comprises:

[0092] an inference perception unit, configured to receive the unified prediction result, and analyze the complexity level of the inference task based on the number of sub-models involved in the prediction result, the calling frequency and the data feature dimension;

[0093] ​​A path decision unit is configured to compare the inference task complexity level with preset edge node processing capability threshold and communication link stability threshold, select an edge deployment path or a cloud deployment path according to a comparison result, and output a deployment type determination result.

[0094] An inference deployment generation unit is configured to construct a deployment instruction according to the deployment type determination result, the deployment instruction including a target node identifier, a model calling sequence, and a model loading priority, for controlling deployment and execution of the multi-model inference module at the edge or the cloud.

[0095] In the embodiment, through the design of the cloud-edge collaborative scheduling module, intelligent switching and optimal deployment of the inference task at the cloud and the edge can be realized on the basis of comprehensive analysis of the inference task complexity, the edge node computing resource state, and the 5G network transmission stability. The design not only effectively improves the resource utilization rate and the response speed of the system, but also maintains high running stability in the case of network fluctuation or resource limitation, thereby guaranteeing the real-time performance and reliability of the prediction and optimization task.

[0096] (Five) Parameter optimization module

[0097] The parameter optimization module is configured to construct a target function based on the unified prediction result, the target function integrating an energy consumption prediction value and a fusion model residual error scoring mechanism, and adopt An optimization algorithm to perform process parameter combination optimization and output the optimization result to an industrial control system;

[0098] Further, the parameter optimization module adopts An optimization algorithm to perform process parameter combination optimization, and the optimization algorithm is based on a multi-objective fitness function constructed as follows:

[0099] ;

[0100] Wherein: represents a combination of forging process parameters to be optimized; represents an energy consumption prediction value estimated based on the unified prediction result; represents a prediction quality score output by the fusion model residual error scoring mechanism; represents a system reference energy consumption upper limit; , respectively represent a minimum value and a maximum value set by the system scoring mechanism; , represents a target weight coefficient for adjusting the weight relationship between the energy efficiency target and the prediction quality target;

[0101] The parameter optimization module is configured to construct a target function based on the unified prediction result, the target function integrating an energy consumption prediction value and a fusion model residual error scoring mechanism, and adopt ​The optimization algorithm adopts the particle swarm optimization method to search for the optimal parameter combination under the constraint of the fitness function, and outputs the optimal parameter combination to the industrial control system to realize process control.

[0102] The following is The pseudo code of the optimization algorithm shows the basic implementation logic of the particle swarm optimization:

[0103] # Optimized pseudo code

[0104] Initialize swarm of particles with random @i

[0105] For each @i:

[0106] Compute E_pred(@i)

[0107] Compute S_qual(@i)

[0108] Evaluate fitness F(@i)

[0109] Repeat until convergence:

[0110] For each particle:

[0111] Update velocity and position

[0112] Evaluate new F(@i)

[0113] Update personal and global best

[0114] Return θ* with max F(θ)

[0115] In this embodiment, by The application of the optimization algorithm can significantly improve the efficiency and product quality of the forging process, and the specific effects include: energy efficiency improvement: by optimizing the parameter combination, the energy consumption is reduced. Prediction accuracy improvement: through multi-model fusion optimization, the prediction accuracy of process parameters is improved. Process quality stability improvement: the optimized process parameters significantly improve the consistency and reliability of the forging process, ensuring the stability of product quality.

[0116] In this embodiment, The optimization algorithm has the following advantages:

[0117] The algorithm comprehensively considers energy efficiency and prediction accuracy, and can achieve all-round optimization of the forging process; through global optimization, it avoids the problem of local optimal solutions and ensures the optimal selection of process parameters; it can flexibly adapt to different production environments and process conditions, and has high adaptability and scalability.

[0118] Through the parameter optimization module of this embodiment, use The optimization algorithm significantly improved the energy efficiency and prediction accuracy of the forging process. This method not only reduces energy consumption but also improves product quality, and has broad industrial application prospects.

[0119] (6) Security Assurance Module

[0120] It is respectively connected to the multi-model reasoning module, prediction fusion module and cloud-edge collaborative scheduling module to perform robustness evaluation on the sub-model, conduct credibility detection on the fusion results based on the historical data consistency scoring mechanism, and automatically trigger the model fusion strategy adjustment when the score is lower than the preset threshold, while implementing access control and encryption protection for the data transmission process in the system.

[0121] The security modules include:

[0122] The credibility detection unit is used to compare the unified prediction results with historical data, specifically including the following steps:

[0123] Obtain the distribution consistency score between the current prediction result and the historical sample data;

[0124] If the consistency score is lower than a preset credibility threshold, an alarm signal is triggered and a fusion strategy adjustment mechanism is initiated;

[0125] Start the model adaptive adjustment mechanism to dynamically adjust the weight of the fusion model based on historical data and current prediction errors to optimize subsequent prediction results.

[0126] In this embodiment, the design of the aforementioned security assurance module enables real-time testing of prediction result credibility during the prediction process. It also triggers timely adaptive model adjustments when anomalies occur or confidence levels drop, effectively ensuring the stability and reliability of the prediction results. Furthermore, combined with access control and encryption protection during data transmission, the system's anti-interference capabilities and data security are significantly enhanced in complex industrial environments, providing a more reliable guarantee for overall process optimization.

[0127] like Figure 2 FIG. 1 is another embodiment of the present invention, which provides a method for intelligent management of product forging data based on 5G cloud computing, and is applied to the above-mentioned intelligent management system for product forging data based on 5G cloud computing, and includes the following steps:

[0128] Step 1: Obtain multi-source process data during the forging process, including temperature, stress, load and image data; and based on the running state of the acquisition device and the network transmission state, the data is evaluated for credibility to generate weighted labeled process feature data;

[0129] Step 2: Receive the process feature data, select a plurality of prediction sub-models with heterogeneous modeling methods according to the feature type identification and the integrity score, the prediction sub-models include support vector regression model, random forest model and convolutional neural network model, and execute inference in parallel to generate prediction results of the plurality of sub-models;

[0130] Step 3: Receive the prediction results of the plurality of sub-models, calculate the confidence score based on the performance of each sub-model in the historical verification data set, and according to the score result and the set fusion path control strategy, use a nonlinear normalization function to assign fusion weights to generate a unified prediction result;

[0131] Step 4: Generate deployment instructions according to the unified prediction result, edge node computing resource state and 5G network transmission state, and based on the comparison analysis of the inference task complexity level, edge node processing capability and network stability, determine the deployment path of the multi-model inference module, and output the deployment type judgment result;

[0132] Step 5: Based on the unified prediction result, construct a target function containing energy consumption prediction value and fusion model residual error score mechanism, and use Optimization algorithm to perform process parameter combination optimization, and output the optimization result to the industrial control system for process control;

[0133] Step 6: Perform robustness evaluation on the sub-model, detect the credibility of the fusion result based on the historical data consistency score mechanism; when the score is lower than the preset threshold, automatically trigger the model fusion strategy adjustment, and implement access control and encryption protection on the data transmission process in the system.

[0134] In summary, the present application realizes intelligent, refined and efficient management of the whole forging process by multi-source credible collection of process data, heterogeneous multi-model parallel inference based on feature identification, dynamic prediction fusion of nonlinear normalized weights, cloud-edge collaborative scheduling combining task complexity and network state, process parameter optimization integrating energy consumption prediction and residual score mechanism, and secure and reliable robustness guarantee. Compared with the prior art, the present application significantly improves the accuracy of process parameter prediction and the real-time response capability of the system, effectively reduces the production energy consumption, and maintains high data security and system stability in complex industrial environments. The technical solution has good universality and expandability, and is particularly suitable for intelligent manufacturing, high-end forging and other industrial scenarios with high requirements for precision and real-time performance, and has obvious technical progress, significant economic benefits and broad industrial application prospects.

[0135] In the description of the present application, the description of the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, the skilled person in the art can combine and combine the different embodiments or examples described in the present application and the features of the different embodiments or examples without contradiction.

[0136] Any process or method descriptions in flow charts or described elsewhere herein can be understood as representing code modules, segments, or portions of code that include one or more executable instructions for implementing specific logic functions or other processes. And the scope of preferred embodiments of the present application includes additional implementation in which the functions are performed in different orders, in substantially simultaneous fashion, or in reverse order, depending on the functionality involved.

[0137] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any skilled person in the art can easily think of various changes or replacements within the technical scope disclosed in the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A product forging data intelligent management system based on 5G cloud computing, characterized in that: include: The process data module is used to obtain multi-source process data during the forging process, evaluate the data credibility based on the operating status of the acquisition equipment and the network transmission status, and generate process feature data after weighted annotation; a multi-model inference module, configured to receive the process feature data, select a plurality of prediction sub-models with heterogeneous modeling methods according to the feature type identifier and the completeness score, including a support vector regression model, a random forest model, and a convolutional neural network model, call the prediction sub-models in parallel, and output prediction results of the plurality of sub-models; A prediction fusion module is configured to receive the prediction results of the multiple sub-models, calculate a confidence score based on the performance of each sub-model in a historical validation dataset, and dynamically assign fusion weights using a nonlinear normalization function based on the score results and a set fusion path control strategy to generate a unified prediction result; A cloud-edge collaborative scheduling module is used to generate deployment instructions based on the unified prediction results, edge node computing resource status, and 5G network transmission status, and determine the deployment path of the multi-model reasoning module based on the comparative analysis of the complexity level of the reasoning task, the processing capability of the edge node, and the network stability, and output the deployment type determination result; Parameter optimization module, used to construct an objective function based on the unified prediction result, the objective function integrates the energy consumption prediction value and the fusion model residual scoring mechanism, using The optimization algorithm performs process parameter combination optimization and outputs the optimization results to the industrial control system; The security assurance module is connected to the multi-model reasoning module, the prediction fusion module and the cloud-edge collaborative scheduling module respectively, and is used to perform robustness evaluation on the sub-model, conduct credibility detection on the fusion results based on the historical data consistency scoring mechanism, and automatically trigger the model fusion strategy adjustment when the score is lower than the preset threshold, while implementing access control and encryption protection for the data transmission process in the system.

2. The product forging data intelligent management system based on 5G cloud computing according to claim 1 is characterized in that: The process data module includes: The trusted acquisition unit is used to collect multi-source process data such as temperature, stress, load, and image during the forging process, and quantitatively evaluates the data credibility based on the operating status of the acquisition equipment and the network transmission status, and outputs weighted and annotated process data; The semantic preprocessing unit receives weighted annotated process data, performs data consistency analysis based on a stress-temperature prediction model constructed based on the thermal-mechanical coupling relationship, identifies abnormal samples, and interpolates missing data. It also performs time synchronization processing on image data and non-image data. The feedback optimization unit is used to receive the prediction error information output by the stress-temperature prediction model, dynamically adjust the feature screening rules and normalization parameter configuration, improve the consistency and effectiveness of subsequent modeling data, and output process feature data.

3. The product forging data intelligent management system based on 5G cloud computing according to claim 2 is characterized in that: The stress-temperature prediction model includes: A regression model constructed based on historical forging process data uses temperature, strain rate, and loading rate at multiple time points in the same process as input variables and material stress as output variable to describe the nonlinear relationship between thermal and mechanical processes; When the regression model is used, based on the feature similarity between the current process data and the historical sample data, the most matching sub-model is dynamically selected or the weighted fusion output result is adopted to identify abnormal data points and perform regression estimation to complete the missing values.

4. The product forging data intelligent management system based on 5G cloud computing according to claim 1 is characterized in that: The multi-model reasoning module includes: Multiple structurally heterogeneous sub-models, including support vector regression models, random forest models, and convolutional neural network models, are suitable for processing subsets of time series sensor data, statistical feature data, and image-based process feature data, respectively; The model calling control unit is used to receive the process feature data, dynamically select matching sub-model combinations based on the type identification and integrity score results of each data feature, and execute prediction tasks in parallel to generate prediction output results of multiple sub-models.

5. The product forging data intelligent management system based on 5G cloud computing according to claim 1 is characterized in that: The prediction fusion module includes: The confidence evaluation unit is used to receive the prediction output results of multiple sub-models and generate a corresponding historical reliability score based on the performance of each sub-model in the historical validation dataset, indicating the prediction accuracy and stability of each sub-model in the past samples; The fusion weight calculation unit is used to calculate the fusion weight of each sub-model based on the following function: ； in: For the The reliability score of each sub-model obtained on the historical validation dataset; For the Historical reliability scores of each sub-model; For the The fusion weight of each sub-model; is the total number of sub-models currently involved in the fusion; is the response adjustment coefficient; is the benchmark value for the score; is the base of natural logarithms; A fusion output unit is used to output the fusion weights according to the fusion weights. Prediction output for each sub-model Perform weighted processing to generate a unified prediction result, the expression is: ； in: Output a unified prediction result after weighted fusion for multiple sub-models; For the The predicted output of each sub-model for the current input data.

6. According to the 5G cloud computing-based product forging data intelligent management system of claim 1, it is characterized in that: The cloud-edge collaborative scheduling module includes: The reasoning perception unit is used to receive the unified prediction results and analyze the complexity level of the reasoning task based on the number of sub-models involved in the prediction results, the call frequency, and the data feature dimensions; The path decision unit is used to compare the complexity level of the inference task with the preset edge node processing capability threshold and communication link stability threshold, select the edge deployment path or the cloud deployment path based on the comparison result, and output the deployment type determination result; The inference deployment generation unit is used to build deployment instructions based on the deployment type determination results. The deployment instructions include the target node identifier, model call sequence, and model loading priority, and are used to control the deployment and execution of multi-model inference modules on the edge or in the cloud.

7. The product forging data intelligent management system based on 5G cloud computing according to claim 1 is characterized in that: The parameter optimization module adopts The optimization algorithm performs process parameter combination optimization, the The optimization algorithm is constructed based on the following multi-objective fitness function, which is expressed as: ； in: Indicates the combination of forging process parameters to be optimized; The energy consumption forecast value estimated based on the unified forecast results; The prediction quality score output by the residual scoring mechanism of the fusion model; It is the reference energy consumption upper limit of the system; 、 These are the minimum and maximum values ​​set for the system scoring mechanism; 、 is the target weight coefficient, which is used to adjust the weight relationship between the energy efficiency target and the prediction quality target; described The optimization algorithm adopts the particle swarm optimization method to search for the optimal parameter combination under the constraints of the fitness function, and outputs the optimal parameter combination to the industrial control system to achieve process control.

8. The product forging data intelligent management system based on 5G cloud computing according to claim 1 is characterized in that: The security assurance module includes: The credibility detection unit is used to compare the unified prediction results with historical data, specifically including the following steps: Obtain the consistency score between the current prediction result and the distribution of historical sample data; If the consistency score is lower than a preset credibility threshold, an alarm signal is triggered and a fusion strategy adjustment mechanism is initiated; Start the model adaptive adjustment mechanism to dynamically adjust the weight of the fusion model based on historical data and current prediction errors to optimize subsequent prediction results.

9. A method for intelligent management of product forging data based on 5G cloud computing, applied to a product forging data intelligent management system based on 5G cloud computing as claimed in any one of claims 1 to 8, characterized in that: The following steps are involved: Step 1: Acquire multi-source process data during the forging process, wherein the process data includes temperature, stress, load, and image data; and performing a credibility assessment on the data based on the operating status of the acquisition equipment and the network transmission status, and generating weighted annotated process feature data; Step 2: Receive the process feature data, select multiple prediction sub-models with heterogeneous modeling methods based on the feature type identifier and the completeness score, the prediction sub-models including a support vector regression model, a random forest model, and a convolutional neural network model, and perform inference in parallel to generate prediction results for the multiple sub-models; Step 3: Receive the prediction results of the multiple sub-models, calculate the confidence score of each sub-model based on its performance in the historical validation dataset, and assign fusion weights using a nonlinear normalization function based on the score results and the set fusion path control strategy to generate a unified prediction result; Step 4: Generate deployment instructions based on the unified prediction results, edge node computing resource status, and 5G network transmission status. Determine the deployment path for the multi-model inference module based on a comparative analysis of the complexity level of the inference task, edge node processing capabilities, and network stability, and output a deployment type determination result. Step 5: Based on the unified prediction results, an objective function including energy consumption prediction value and fusion model residual scoring mechanism is constructed. The optimization algorithm performs process parameter combination optimization and outputs the optimization results to the industrial control system for process control; Step 6: Perform robustness evaluation on the sub-models and conduct credibility check on the fusion results based on the historical data consistency scoring mechanism; when the score is lower than the preset threshold, the model fusion strategy adjustment is automatically triggered, and access control and encryption protection are implemented for the data transmission process in the system.

Images