An intelligent twin control method and system based on material process optimization

Abstract

The present invention discloses an intelligent twin control method and system based on material process optimization. A multi-node control mechanism is set along the time axis to collect equipment parameters, process parameters, and environmental parameters and serialize them into a multi-dimensional input sequence; the spectral characteristics and performance indicators of the material are synchronously acquired to construct a spectrum-performance output sequence; a multi-to-multi-parameter simulation model is established based on the Transformer architecture, and the multi-head attention mechanism is used to analyze the dependencies between parameters; a spectral-efficiency data set is extracted from the literature in combination with a semantic model, and real-time / simulation parameters are predicted and analyzed and parameter settings are optimized. Finally, the process adjustment is guided by the prediction results and parameters, and a closed-loop feedback iterative optimization simulation model is formed. The present invention realizes dynamic modeling, accurate prediction, and control analysis of material processes, significantly improves process optimization efficiency and material performance prediction accuracy, and reduces the cost of R&D trial and error.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control of material processes, and more specifically, to an intelligent twin control method and system based on material process optimization. Background Art

[0002] The optimization analysis of material technology is an important topic for efficient and high-quality production of materials. It involves multiple regulatory analysis dimensions, which directly affects the industry's competitiveness and sustainable development capabilities.

[0003] However, due to the variability of materials across different processes, traditional techniques for material production and process optimization often rely on extensive experimental data and manual adjustments for optimization and evaluation. This results in poor process repeatability and an inability to achieve real-time feedback and rapid adjustments. Although computational simulation and optimization methods have been developed in some high-end material manufacturing fields, these methods are unable to adapt to complex and changing process requirements due to limitations in computing resources and modeling accuracy. Furthermore, existing material optimization and analysis processes lack simulation and big data retrieval capabilities, making it difficult to leverage extensive production data and historical empirical data for comprehensive analysis and simulation evaluation. Furthermore, existing literature data struggles to translate into meaningful optimization decision-making information, resulting in inadequate data utilization. Furthermore, existing optimization methods mostly rely on manual experience and step-by-step adjustments, lacking automated and intelligent adjustment mechanisms. Especially when dealing with multivariable and multi-objective process optimization, manual adjustments are not only inefficient but also easily limited by the designer's experience and knowledge, making them unable to quickly adapt to changing process requirements. Furthermore, existing technologies lack spectral efficiency analysis processes and predictive models for material production, making it difficult to accurately predict material properties in the long and short term, and hindering the advancement of intelligent and automated material production, resulting in low material production efficiency. Summary of the Invention

[0004] The present invention overcomes the shortcomings of the prior art and proposes an intelligent twin control method and system based on material process optimization.

[0005] The first aspect of the present invention provides an intelligent twin control method based on material process optimization, comprising:

[0006] S1: According to the target material production process, multiple control nodes are set based on the time dimension, and the preparation equipment parameters, processing parameters and processing environment parameters are recorded;

[0007] S2: Sequence the preparation equipment parameters, processing parameters and processing environment parameters to obtain a multi-dimensional parameter sequence;

[0008] S3: Spectral characteristic data and performance data of the target material are measured experimentally, and the spectral characteristic data and performance data are serialized to obtain a spectral sequence and a performance sequence;

[0009] S4: Based on the Transformer model, a multi-to-multi-parameter simulation model is established. The parameter sequence is used as the input sequence, and the spectrum sequence and performance sequence are used as the output sequence to perform multi-parameter interaction modeling. The dependency analysis and prediction training of the input and output sequences are performed through the multi-head attention mechanism.

[0010] S5: Using the semantic model and target materials as search tags, content data is retrieved from big data literature and a spectrum-efficiency dataset is extracted. User simulation sequences or real-time parameter sequences are imported into the simulation model for spectrum-efficiency prediction. The rationality of the prediction results is evaluated in combination with the spectrum-efficiency dataset, and spectrum-efficiency prediction data and process optimization parameter sequences are obtained.

[0011] S6: Optimize the production process of the target material through spectral effect prediction data and process optimization parameter sequence, and dynamically optimize the simulation model based on the actual optimized parameter results as training data.

[0012] In this solution, S1 is specifically:

[0013] Based on the material production process, multiple control time points are set in a production process, and each control time point is used as a control node;

[0014] In a production process, the processing parameters are monitored in real time through equipment sensors and environmental sensors. For each control node, the preparation equipment parameters, processing parameters and processing environment parameters are recorded.

[0015] In this solution, S2 is specifically:

[0016] Based on the time dimension, each parameter in the preparation equipment parameters, processing technology parameters and processing environment parameters is serialized, the sequence point corresponds to each control node, and a multi-dimensional parameter sequence is obtained.

[0017] In this solution, S3 is specifically:

[0018] Through material experiments, test nodes are set and the spectrum of the target material is measured. Multi-segment layered information is extracted from the spectrum, and the spectrum information is stored as characteristic values ​​to obtain spectral characteristic data.

[0019] Spectral information includes pixel spacing, wavelength response, reflectivity, transmittance, and absorptivity of electromagnetic waves of different wavelengths;

[0020] Determine the performance status of the target material based on the test node, and serialize the obtained performance information to obtain a performance sequence;

[0021] Performance information includes material strength, electrical conductivity, and heat resistance.

[0022] In this solution, S4 is specifically:

[0023] Based on the Transformer model, a multi-to-multi-parameter simulation model is established, with parameter sequences as input sequences and spectral sequences and performance sequences as output sequences. The nonlinear coupling relationship between the multiple variables of the input and output sequences is analyzed, and a multi-head attention mechanism is introduced to perform linear layer transformations on different parameters in the input sequence, and obtain the corresponding query, key, and value values ​​respectively. The attention score is then obtained through the multi-head attention calculation formula.

[0024] The long-term dependency between input and output sequences is analyzed through attention scores, and long-term prediction training of input and output sequences is performed. The prediction optimization of the simulation model is performed using the mean square error as the loss function.

[0025] An LSTM model is constructed in the simulation model. The weight values ​​of the input parameters are set according to the long-term dependencies and the attention scores of different parameters. The input and output sequences are respectively imported into the LSTM model for short-term prediction training. The mean square error is used as the loss function, and the weight value is used as the prediction weight of the input sequence for the output sequence. Through the backpropagation algorithm, the LSTM model parameters are automatically adjusted to reduce the value of the loss function until the loss function converges to the preset value or the number of training iterations is reached.

[0026] In this solution, S5 is specifically:

[0027] Generate text-based material content data based on the target material, perform word segmentation and keyword annotation on the material content data using the TF-IDF algorithm, and generate search tags;

[0028] Deploy the SciBERT semantic analysis model to match and filter relevant content from big data documents using search tags to obtain search content data;

[0029] Based on the search content data, word segmentation and word context association are performed to match preparation parameters, process parameters, spectral characteristics and performance characteristics phrases, and spectral characteristic data and performance characteristic data are screened out. The screened data are integrated to obtain the spectrum-efficiency data set.

[0030] In this solution, S5 is specifically:

[0031] Set the user simulation process sequence based on the target material production process;

[0032] Constructing a real-time parameter sequence based on the parameters of the current target material real-time production record;

[0033] Import the user simulation process sequence or real-time parameter sequence into the simulation to perform sequence prediction and obtain the output prediction sequence;

[0034] The output prediction sequence is subjected to spectrum-efficacy characteristic data analysis to generate predicted spectrum-efficacy data;

[0035] Multi-dimensional parameter deviation calculations were performed on the predicted spectrum-efficiency data and the spectrum-efficiency data set. The prediction weights of different parameters in the simulation model were introduced to evaluate the deviation. The rationality of the simulation model prediction and the degree of optimization of the process parameters were evaluated through the parameter deviation.

[0036] If the deviation range is within the preset range, the predicted spectrum-efficiency data is used as the target optimization data, multiple optimization parameter sequences are set for prediction analysis, and finally the optimal result is screened out, and the spectrum-efficiency prediction data and process optimization parameter sequence in the optimal result are recorded.

[0037] In this solution, S6 is specifically:

[0038] Conduct feasibility analysis on target materials according to the process optimization parameter sequence, and set feasible optimization parameters to optimize the production process;

[0039] After process optimization, the result parameter data of material production are recorded, serialized and imported into the simulation model as real data for predictive training and dynamic optimization.

[0040] A second aspect of the present invention further provides an intelligent twin control system based on material process optimization, the system comprising: a memory and a processor, wherein the memory includes an intelligent twin control program based on material process optimization, and when the intelligent twin control program based on material process optimization is executed by the processor, the following steps are implemented:

[0041] S1: According to the target material production process, multiple control nodes are set based on the time dimension, and the preparation equipment parameters, processing parameters and processing environment parameters are recorded;

[0042] S2: Sequence the preparation equipment parameters, processing parameters and processing environment parameters to obtain a multi-dimensional parameter sequence;

[0043] S3: Spectral characteristic data and performance data of the target material are measured experimentally, and the spectral characteristic data and performance data are serialized to obtain a spectral sequence and a performance sequence;

[0044] S4: Based on the Transformer model, a multi-to-multi-parameter simulation model is established. The parameter sequence is used as the input sequence, and the spectrum sequence and performance sequence are used as the output sequence to perform multi-parameter interaction modeling. The dependency analysis and prediction training of the input and output sequences are performed through the multi-head attention mechanism.

[0045] S5: Using the semantic model and target materials as search tags, content data is retrieved from big data literature and a spectrum-efficiency dataset is extracted. User simulation sequences or real-time parameter sequences are imported into the simulation model for spectrum-efficiency prediction. The rationality of the prediction results is evaluated in combination with the spectrum-efficiency dataset, and spectrum-efficiency prediction data and process optimization parameter sequences are obtained.

[0046] S6: Optimize the production process of the target material through spectral effect prediction data and process optimization parameter sequence, and dynamically optimize the simulation model based on the actual optimized parameter results as training data.

[0047] The third aspect of the present invention also provides a computer-readable storage medium, which includes an intelligent twin control program based on material process optimization. When the intelligent twin control program based on material process optimization is executed by a processor, the steps of the intelligent twin control method based on material process optimization as described in any one of the above items are implemented.

[0048] The present invention can achieve the following technical effects:

[0049] This invention builds a simulation model based on intelligent digital twins to virtually recreate real-world experiments and production processes, reducing repetitive and time-consuming experimental processes. Through simulation and optimization algorithms, the system can quickly predict results under different process conditions, shortening experimental cycles and reducing costs. This improves the efficiency and reliability of process optimization.

[0050] In order to address the problem that traditional material process optimization methods usually rely on a small amount of experimental data and fail to fully explore and utilize the potential of big data analysis, resulting in the value of data not being fully utilized, the present invention is based on a semantic analysis model to conduct real-time mining and analysis of a large amount of historical experimental data, real-time production data and literature data, set a spectral effect feature set based on the target material, and further reasonably evaluate the prediction results, effectively improving the simulation model's prediction and optimization parameter capabilities.

[0051] Through a simulation model based on digital twins, the present invention can provide precise, efficient and intelligent support in the optimization of material processing, significantly overcome various limitations in existing technologies, and promote technological innovation in the field of materials manufacturing. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A flow chart of an intelligent twin control method based on material process optimization of the present invention is shown;

[0053] Figure 2 Shown is a schematic diagram of the response spectrum of the present invention;

[0054] Figure 3A block diagram of an intelligent twin control system based on material process optimization of the present invention is shown. DETAILED DESCRIPTION

[0055] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features therein can be combined with each other.

[0056] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.

[0057] Figure 1 A flow chart of an intelligent twin control method based on material process optimization of the present invention is shown.

[0058] like Figure 1 As shown, the first aspect of the present invention provides an intelligent twin control method based on material process optimization, comprising:

[0059] S1: According to the target material production process, multiple control nodes are set based on the time dimension, and the preparation equipment parameters, processing parameters and processing environment parameters are recorded;

[0060] S2: Sequence the preparation equipment parameters, processing parameters and processing environment parameters to obtain a multi-dimensional parameter sequence;

[0061] S3: Spectral characteristic data and performance data of the target material are measured experimentally, and the spectral characteristic data and performance data are serialized to obtain a spectral sequence and a performance sequence;

[0062] S4: Based on the Transformer model, a multi-to-multi-parameter simulation model is established. The parameter sequence is used as the input sequence, and the spectrum sequence and performance sequence are used as the output sequence to perform multi-parameter interaction modeling. The dependency analysis and prediction training of the input and output sequences are performed through the multi-head attention mechanism.

[0063] S5: Using the semantic model and target materials as search tags, content data is retrieved from big data literature and a spectrum-efficiency dataset is extracted. User simulation sequences or real-time parameter sequences are imported into the simulation model for spectrum-efficiency prediction. The rationality of the prediction results is evaluated in combination with the spectrum-efficiency dataset, and spectrum-efficiency prediction data and process optimization parameter sequences are obtained.

[0064] S6: Optimize the production process of the target material through spectral effect prediction data and process optimization parameter sequence, and dynamically optimize the simulation model based on the actual optimized parameter results as training data.

[0065] It should be noted that the simulation model is specifically a many-to-many parameter prediction neural network based on Transformer.

[0066] According to an embodiment of the present invention, the S1 is specifically:

[0067] Based on the material production process, multiple control time points are set in a production process, and each control time point is used as a control node;

[0068] In a production process, the processing parameters are monitored in real time through equipment sensors and environmental sensors. For each control node, the preparation equipment parameters, processing parameters and processing environment parameters are recorded.

[0069] It should be noted that the preparation equipment parameters include the equipment model and specifications, equipment operating time, temperature control accuracy, pressure control accuracy, and time control accuracy; the processing technology parameters include heating time, cooling time, reaction time, atmosphere, equipment speed and rotation speed, temperature, pressure, flow rate and other parameters; the processing process environmental parameters include ambient temperature and humidity, cleanliness, vibration and noise parameters, etc.

[0070] According to an embodiment of the present invention, S2 specifically includes:

[0071] Based on the time dimension, each parameter in the preparation equipment parameters, processing technology parameters and processing environment parameters is serialized, the sequence point corresponds to each control node, and a multi-dimensional parameter sequence is obtained.

[0072] It should be noted that the parameter sequence includes multiple parameters, which is specifically related to the number of parameter types.

[0073] According to an embodiment of the present invention, S3 specifically includes:

[0074] Through material experiments, test nodes are set and the spectrum of the target material is measured. Multi-segment layered information is extracted from the spectrum, and the spectrum information is stored as characteristic values ​​to obtain spectral characteristic data.

[0075] Spectral information includes pixel spacing, wavelength response, reflectivity, transmittance, and absorptivity of electromagnetic waves of different wavelengths;

[0076] Determine the performance status of the target material based on the test node, and serialize the obtained performance information to obtain a performance sequence;

[0077] Performance information includes material strength, electrical conductivity, and heat resistance.

[0078] It should be noted that the test node is specifically a test cycle or test time point. The corresponding characteristic data is serialized based on the test node, and the obtained spectral sequence and performance sequence include multi-dimensional spectral effect parameters. The spectral effect parameters can be correlated with the multi-dimensional parameter sequence, sequence feature learning and corresponding multi-parameter prediction. The target material can be measured based on different production process steps or based on the completion status, depending on the specific production process optimization requirements. Material experiments can be spectrally measured based on a step-by-step Fourier interferometer.

[0079] Figure 2 Shown is a schematic diagram of the response spectrum of the present invention; Figure 2 The following is a schematic diagram of the response spectrum of a certain material.

[0080] According to an embodiment of the present invention, the S4 is specifically:

[0081] Based on the Transformer model, a multi-to-multi-parameter simulation model is established, with parameter sequences as input sequences and spectral sequences and performance sequences as output sequences. The nonlinear coupling relationship between the multiple variables of the input and output sequences is analyzed, and a multi-head attention mechanism is introduced to perform linear layer transformations on different parameters in the input sequence, and obtain the corresponding query, key, and value values ​​respectively. The attention score is then obtained through the multi-head attention calculation formula.

[0082] The long-term dependency between input and output sequences is analyzed through attention scores, and long-term prediction training of input and output sequences is performed. The prediction optimization of the simulation model is performed using the mean square error as the loss function.

[0083] An LSTM model is constructed in the simulation model. The weight values ​​of the input parameters are set according to the long-term dependencies and the attention scores of different parameters. The input and output sequences are respectively imported into the LSTM model for short-term prediction training. The mean square error is used as the loss function, and the weight value is used as the prediction weight of the input sequence for the output sequence. Through the backpropagation algorithm, the LSTM model parameters are automatically adjusted to reduce the value of the loss function until the loss function converges to the preset value or the number of training iterations is reached.

[0084] It should be noted that the simulation model includes an LSTM model. The weights of the input parameters are proportional to the attention score. The prediction weights represent the contribution of the corresponding input sequence to the output sequence. By adding weights, short-term dependencies and associations can be efficiently and accurately simulated, thereby training and optimizing the LSTM model and applying it to the simulation model. The simulation model includes a Transformer long-term prediction application model and an LSTM short-term prediction application model.

[0085] The calculation formula of multi-head attention is as follows:

[0086] ;

[0087] Represents Query, Key and Value respectively, Indicates the dimension of the key vector.

[0088] It is worth noting that in material production and process optimization, due to the differences in materials under different processes, traditional technologies often rely on large amounts of experimental data and manual adjustments for optimization and evaluation. This results in poor process repeatability and an inability to achieve real-time feedback and rapid adjustment. Although some computational simulation and optimization methods have been developed in some high-end material manufacturing fields, these methods are unable to adapt to complex and changing process requirements due to limitations in computing resources and modeling accuracy. Furthermore, most existing optimization methods rely on manual experience and step-by-step debugging, lacking automated and intelligent adjustment mechanisms. Especially when dealing with multi-variable and multi-objective process optimization, manual adjustments are not only inefficient but also easily limited by the designer's experience and knowledge depth, making it impossible to quickly adapt to changing process requirements. At the same time, existing technologies lack spectral efficiency analysis processes and predictive models for material production, making it difficult to accurately predict material properties in the long and short term.

[0089] Based on this, the present invention obtains and serializes various process parameters through multiple control nodes, and combines the spectral effect sequence data of the material to perform joint prediction training and correlation analysis of multi-dimensional parameters. The prediction training introduces the Transformer model and the multi-head attention mechanism, and realizes the feasibility of long-term prediction in material process production. Furthermore, combined with the LSTM model, the accuracy of short-term prediction in material process production is improved, and a simulation model with complex process adaptive learning is trained to realize twin simulation analysis of material process. Through this simulation model, the final spectral effect of the material is predicted based on the process state parameters, and the process parameters are optimized in combination with the material requirements. Furthermore, through big data literature retrieval of spectral effect data of similar or identical materials, the prediction results of the target material are analyzed for differences, the rationality of the prediction results is evaluated in real time, and the optimization parameters and dynamic optimization simulation model are set.

[0090] The present invention can achieve the following technical effects:

[0091] This invention builds a simulation model based on intelligent digital twins to virtually recreate real-world experiments and production processes, reducing repetitive and time-consuming experimental processes. Through simulation and optimization algorithms, the system can quickly predict results under different process conditions, shortening experimental cycles and reducing costs. This improves the efficiency and reliability of process optimization.

[0092] In order to address the problem that traditional material process optimization methods usually rely on a small amount of experimental data and fail to fully explore and utilize the potential of big data analysis, resulting in the value of data not being fully utilized, the present invention is based on a semantic analysis model to conduct real-time mining and analysis of a large amount of historical experimental data, real-time production data and literature data, set a spectral effect feature set based on the target material, and further reasonably evaluate the prediction results, effectively improving the simulation model's prediction and optimization parameter capabilities.

[0093] Through a simulation model based on digital twins, the present invention can provide precise, efficient and intelligent support in the optimization of material processing, significantly overcome various limitations in existing technologies, and promote technological innovation in the field of materials manufacturing.

[0094] According to an embodiment of the present invention, the S5 is specifically:

[0095] Generate text-based material content data based on the target material, perform word segmentation and keyword annotation on the material content data using the TF-IDF algorithm, and generate search tags;

[0096] Deploy the SciBERT semantic analysis model to match and filter relevant content from big data documents using search tags to obtain search content data;

[0097] Based on the search content data, word segmentation and word context association are performed to match preparation parameters, process parameters, spectral characteristics and performance characteristics phrases, and spectral characteristic data and performance characteristic data are screened out. The screened data are integrated to obtain the spectrum-efficiency data set.

[0098] It should be noted that material content data includes the target material's name, function, various parameters, and spectral performance information. Spectral performance information refers to the spectrum and performance information. Word segmentation and word context analysis based on the search content data specifically includes matching information on preparation methods, process parameters, performance indicators, and other dimensions. Big data literature includes historical experimental data, real-time production data, and big data from material literature text.

[0099] The SciBERT used in this paper is a pre-trained language model based on the BERT architecture, which can be optimized for chemical materials science texts. Through domain-specific vocabulary and corpus training, it can effectively improve the performance of NLP tasks in specific fields.

[0100] According to an embodiment of the present invention, the S5 is specifically:

[0101] Set the user simulation process sequence based on the target material production process;

[0102] Constructing a real-time parameter sequence based on the parameters of the current target material real-time production record;

[0103] Import the user simulation process sequence or real-time parameter sequence into the simulation to perform sequence prediction and obtain the output prediction sequence;

[0104] The output prediction sequence is subjected to spectrum-efficacy characteristic data analysis to generate predicted spectrum-efficacy data;

[0105] Multi-dimensional parameter deviation calculations were performed on the predicted spectrum-efficiency data and the spectrum-efficiency data set. The prediction weights of different parameters in the simulation model were introduced to evaluate the deviation. The rationality of the simulation model prediction and the degree of optimization of the process parameters were evaluated through the parameter deviation.

[0106] If the deviation range is within the preset range, the predicted spectrum-efficiency data is used as the target optimization data, multiple optimization parameter sequences are set for prediction analysis, and finally the optimal result is screened out, and the spectrum-efficiency prediction data and process optimization parameter sequence in the optimal result are recorded.

[0107] It should be noted that the deviation range is within the preset range. Specifically, the comprehensive deviation rate is calculated using multi-dimensional parameters combined with weights. If the comprehensive deviation rate is above 60%, the deviation is considered within the reasonable range. Various optimization parameters include preparation equipment parameters, processing parameters, and process environment parameters. The user simulation process sequence is a user-defined parameter sequence input, which is used to simulate the material results of the user-defined process state. The real-time parameter sequence is the real-time parameter information obtained during the production process of the target material. The simulation model is used to perform real-time spectral effect prediction and adjust process parameters.

[0108] According to an embodiment of the present invention, the S6 is specifically:

[0109] Conduct feasibility analysis on target materials according to the process optimization parameter sequence, and set feasible optimization parameters to optimize the production process;

[0110] After process optimization, the result parameter data of material production are recorded, serialized and imported into the simulation model as real data for predictive training and dynamic optimization.

[0111] It should be noted that the result parameter data includes multi-dimensional parameter sequences (preparation equipment parameters, processing technology parameters and processing environment parameters, etc.) and spectral effect data, etc.

[0112] According to an embodiment of the present invention, the further embodiment includes:

[0113] The deviation between the actual parameters and the predicted parameters is evaluated based on the result parameter data and the spectral effect prediction data, and the weighted deviation rate is calculated by combining the prediction weights of different parameters;

[0114] The same parameter sequence is selected from the process optimization parameter sequence and the multi-dimensional parameter sequence for correlation analysis. The correlation analysis introduces the grey correlation method for evaluation and obtains the correlation degree.

[0115] The correlation of all parameters is calculated, and based on the correlation, the prediction weight of each parameter is optimized in the simulation model.

[0116] It should be noted that the multi-dimensional parameter sequence represents the sequence of preparation equipment parameters, processing parameters, and processing environment parameters in the actual production process. The process optimization parameter sequence includes multiple parameters, such as preparation equipment parameters, processing parameters, and processing environment parameters, which are used to guide production optimization. Correlation analysis is based on the selection analysis of a certain parameter, such as selecting a certain processing parameter. Theoretically, the larger the prediction weight, the larger the corresponding prediction weight. Therefore, it is possible to establish a corresponding correlation relationship, determine the mismatch between the prediction weight and the correlation degree, and dynamically optimize the prediction weight (increase or decrease the value) to maintain the relationship between the correlation degree and the prediction weight. At the same time, it is possible to optimize the long-term and short-term prediction accuracy of the simulation model and prevent the model from overfitting the prediction.

[0117] Figure 3 A block diagram of an intelligent twin control system based on material process optimization of the present invention is shown.

[0118] The second aspect of the present invention further provides an intelligent twin control system 3 based on material process optimization, which includes: a memory 31 and a processor 32. The memory 31 includes an intelligent twin control program based on material process optimization. When the intelligent twin control program based on material process optimization is executed by the processor 32, the following steps are implemented:

[0119] S1: According to the target material production process, multiple control nodes are set based on the time dimension, and the preparation equipment parameters, processing parameters and processing environment parameters are recorded;

[0120] S2: Sequence the preparation equipment parameters, processing parameters and processing environment parameters to obtain a multi-dimensional parameter sequence;

[0121] S3: Spectral characteristic data and performance data of the target material are measured experimentally, and the spectral characteristic data and performance data are serialized to obtain a spectral sequence and a performance sequence;

[0122] S4: Based on the Transformer model, a multi-to-multi-parameter simulation model is established. The parameter sequence is used as the input sequence, and the spectrum sequence and performance sequence are used as the output sequence to perform multi-parameter interaction modeling. The dependency analysis and prediction training of the input and output sequences are performed through the multi-head attention mechanism.

[0123] S5: Using the semantic model and target materials as search tags, content data is retrieved from big data literature and a spectrum-efficiency dataset is extracted. User simulation sequences or real-time parameter sequences are imported into the simulation model for spectrum-efficiency prediction. The rationality of the prediction results is evaluated in combination with the spectrum-efficiency dataset, and spectrum-efficiency prediction data and process optimization parameter sequences are obtained.

[0124] S6: Optimize the production process of the target material through spectral effect prediction data and process optimization parameter sequence, and dynamically optimize the simulation model based on the actual optimized parameter results as training data.

[0125] It should be noted that the simulation model is specifically a many-to-many parameter prediction neural network based on Transformer.

[0126] According to an embodiment of the present invention, the S1 is specifically:

[0127] Based on the material production process, multiple control time points are set in a production process, and each control time point is used as a control node;

[0128] In a production process, the processing parameters are monitored in real time through equipment sensors and environmental sensors. For each control node, the preparation equipment parameters, processing parameters and processing environment parameters are recorded.

[0129] It should be noted that the preparation equipment parameters include the equipment model and specifications, equipment operating time, temperature control accuracy, pressure control accuracy, and time control accuracy; the processing technology parameters include heating time, cooling time, reaction time, atmosphere, equipment speed and rotation speed, temperature, pressure, flow rate and other parameters; the processing process environmental parameters include ambient temperature and humidity, cleanliness, vibration and noise parameters, etc.

[0130] According to an embodiment of the present invention, S2 specifically includes:

[0131] Based on the time dimension, each parameter in the preparation equipment parameters, processing technology parameters and processing environment parameters is serialized, the sequence point corresponds to each control node, and a multi-dimensional parameter sequence is obtained.

[0132] It should be noted that the parameter sequence includes multiple parameters, which is specifically related to the number of parameter types.

[0133] According to an embodiment of the present invention, S3 specifically includes:

[0134] Through material experiments, test nodes are set and the spectrum of the target material is measured. Multi-segment layered information is extracted from the spectrum, and the spectrum information is stored as characteristic values ​​to obtain spectral characteristic data.

[0135] Spectral information includes pixel spacing, wavelength response, reflectivity, transmittance, and absorptivity of electromagnetic waves of different wavelengths;

[0136] Determine the performance status of the target material based on the test node, and serialize the obtained performance information to obtain a performance sequence;

[0137] Performance information includes material strength, electrical conductivity, and heat resistance.

[0138] It should be noted that the test node is specifically a test cycle or test time point. The corresponding characteristic data is serialized based on the test node, and the obtained spectral sequence and performance sequence include multi-dimensional spectral effect parameters. The spectral effect parameters can be correlated with the multi-dimensional parameter sequence, sequence feature learning and corresponding multi-parameter prediction. The target material can be measured based on different production process steps or based on the completion status, depending on the specific production process optimization requirements. Material experiments can be spectrally measured based on a step-by-step Fourier interferometer.

[0139] According to an embodiment of the present invention, the S4 is specifically:

[0140] Based on the Transformer model, a multi-to-multi-parameter simulation model is established, with parameter sequences as input sequences and spectral sequences and performance sequences as output sequences. The nonlinear coupling relationship between the multiple variables of the input and output sequences is analyzed, and a multi-head attention mechanism is introduced to perform linear layer transformations on different parameters in the input sequence, and obtain the corresponding query, key, and value values ​​respectively. The attention score is then obtained through the multi-head attention calculation formula.

[0141] The long-term dependency between input and output sequences is analyzed through attention scores, and long-term prediction training of input and output sequences is performed. The prediction optimization of the simulation model is performed using the mean square error as the loss function.

[0142] An LSTM model is constructed in the simulation model. The weight values ​​of the input parameters are set according to the long-term dependencies and the attention scores of different parameters. The input and output sequences are respectively imported into the LSTM model for short-term prediction training. The mean square error is used as the loss function, and the weight value is used as the prediction weight of the input sequence for the output sequence. Through the backpropagation algorithm, the LSTM model parameters are automatically adjusted to reduce the value of the loss function until the loss function converges to the preset value or the number of training iterations is reached.

[0143] It should be noted that the simulation model includes an LSTM model. The weights of the input parameters are proportional to the attention score. The prediction weights represent the contribution of the corresponding input sequence to the output sequence. By adding weights, short-term dependencies and associations can be efficiently and accurately simulated, thereby training and optimizing the LSTM model and applying it to the simulation model. The simulation model includes a Transformer long-term prediction application model and an LSTM short-term prediction application model.

[0144] The calculation formula of multi-head attention is as follows:

[0145] ;

[0146] Represents Query, Key and Value respectively, Indicates the dimension of the key vector.

[0147] It is worth noting that in material production and process optimization, due to the differences in materials under different processes, traditional technologies often rely on large amounts of experimental data and manual adjustments for optimization and evaluation. This results in poor process repeatability and an inability to achieve real-time feedback and rapid adjustment. Although some computational simulation and optimization methods have been developed in some high-end material manufacturing fields, these methods are unable to adapt to complex and changing process requirements due to limitations in computing resources and modeling accuracy. Furthermore, most existing optimization methods rely on manual experience and step-by-step debugging, lacking automated and intelligent adjustment mechanisms. Especially when dealing with multi-variable and multi-objective process optimization, manual adjustments are not only inefficient but also easily limited by the designer's experience and knowledge depth, making it impossible to quickly adapt to changing process requirements. At the same time, existing technologies lack spectral efficiency analysis processes and predictive models for material production, making it difficult to accurately predict material properties in the long and short term.

[0148] Based on this, the present invention obtains and serializes various process parameters through multiple control nodes, and combines the spectral effect sequence data of the material to perform joint prediction training and correlation analysis of multi-dimensional parameters. The prediction training introduces the Transformer model and the multi-head attention mechanism, and realizes the feasibility of long-term prediction in material process production. Furthermore, combined with the LSTM model, the accuracy of short-term prediction in material process production is improved, and a simulation model with complex process adaptive learning is trained to realize twin simulation analysis of material process. Through this simulation model, the final spectral effect of the material is predicted based on the process state parameters, and the process parameters are optimized in combination with the material requirements. Furthermore, through big data literature retrieval of spectral effect data of similar or identical materials, the prediction results of the target material are analyzed for differences, the rationality of the prediction results is evaluated in real time, and the optimization parameters and dynamic optimization simulation model are set.

[0149] The present invention can achieve the following technical effects:

[0150] This invention builds a simulation model based on intelligent digital twins to virtually recreate real-world experiments and production processes, reducing repetitive and time-consuming experimental processes. Through simulation and optimization algorithms, the system can quickly predict results under different process conditions, shortening experimental cycles and reducing costs. This improves the efficiency and reliability of process optimization.

[0151] In order to address the problem that traditional material process optimization methods usually rely on a small amount of experimental data and fail to fully explore and utilize the potential of big data analysis, resulting in the value of data not being fully utilized, the present invention is based on a semantic analysis model to conduct real-time mining and analysis of a large amount of historical experimental data, real-time production data and literature data, set a spectral effect feature set based on the target material, and further reasonably evaluate the prediction results, effectively improving the simulation model's prediction and optimization parameter capabilities.

[0152] Through a simulation model based on digital twins, the present invention can provide precise, efficient and intelligent support in the optimization of material processing, significantly overcome various limitations in existing technologies, and promote technological innovation in the field of materials manufacturing.

[0153] According to an embodiment of the present invention, the S5 is specifically:

[0154] Generate text-based material content data based on the target material, perform word segmentation and keyword annotation on the material content data using the TF-IDF algorithm, and generate search tags;

[0155] Deploy the SciBERT semantic analysis model to match and filter relevant content from big data documents using search tags to obtain search content data;

[0156] Based on the search content data, word segmentation and word context association are performed to match preparation parameters, process parameters, spectral characteristics and performance characteristics phrases, and spectral characteristic data and performance characteristic data are screened out. The screened data are integrated to obtain the spectrum-efficiency data set.

[0157] It should be noted that material content data includes the target material's name, function, various parameters, and spectral performance information. Spectral performance information refers to the spectrum and performance information. Word segmentation and word context analysis based on the search content data specifically includes matching information on preparation methods, process parameters, performance indicators, and other dimensions. Big data literature includes historical experimental data, real-time production data, and big data from material literature text.

[0158] The SciBERT used in this paper is a pre-trained language model based on the BERT architecture, which can be optimized for chemical materials science texts. Through domain-specific vocabulary and corpus training, it can effectively improve the performance of NLP tasks in specific fields.

[0159] According to an embodiment of the present invention, the S5 is specifically:

[0160] Set the user simulation process sequence based on the target material production process;

[0161] Constructing a real-time parameter sequence based on the parameters of the current target material real-time production record;

[0162] Import the user simulation process sequence or real-time parameter sequence into the simulation to perform sequence prediction and obtain the output prediction sequence;

[0163] The output prediction sequence is subjected to spectrum-efficacy characteristic data analysis to generate predicted spectrum-efficacy data;

[0164] Multi-dimensional parameter deviation calculations were performed on the predicted spectrum-efficiency data and the spectrum-efficiency data set. The prediction weights of different parameters in the simulation model were introduced to evaluate the deviation. The rationality of the simulation model prediction and the degree of optimization of the process parameters were evaluated through the parameter deviation.

[0165] If the deviation range is within the preset range, the predicted spectrum-efficiency data is used as the target optimization data, multiple optimization parameter sequences are set for prediction analysis, and finally the optimal result is screened out, and the spectrum-efficiency prediction data and process optimization parameter sequence in the optimal result are recorded.

[0166] It should be noted that the deviation range is within the preset range. Specifically, the comprehensive deviation rate is calculated using multi-dimensional parameters combined with weights. If the comprehensive deviation rate is above 60%, the deviation is considered within the reasonable range. Various optimization parameters include preparation equipment parameters, processing parameters, and process environment parameters. The user simulation process sequence is a user-defined parameter sequence input, which is used to simulate the material results of the user-defined process state. The real-time parameter sequence is the real-time parameter information obtained during the production process of the target material. The simulation model is used to perform real-time spectral effect prediction and adjust process parameters.

[0167] According to an embodiment of the present invention, the S6 is specifically:

[0168] Conduct feasibility analysis on target materials according to the process optimization parameter sequence, and set feasible optimization parameters to optimize the production process;

[0169] After process optimization, the result parameter data of material production are recorded, serialized and imported into the simulation model as real data for predictive training and dynamic optimization.

[0170] It should be noted that the result parameter data includes multi-dimensional parameter sequences (preparation equipment parameters, processing technology parameters and processing environment parameters, etc.) and spectral effect data, etc.

[0171] The third aspect of the present invention also provides a computer-readable storage medium, which includes an intelligent twin control program based on material process optimization. When the intelligent twin control program based on material process optimization is executed by a processor, the steps of the intelligent twin control method based on material process optimization as described in any one of the above items are implemented.

[0172] The present invention discloses an intelligent twin control method and system based on material process optimization. A multi-node control mechanism is set along the time axis to collect equipment parameters, process parameters and environmental parameters and serialize them into a multi-dimensional input sequence; the spectral characteristics and performance indicators of the material are obtained synchronously to construct a spectrum-performance output sequence; a multi-to-multi-parameter simulation model is established based on the Transformer architecture, and the dependency relationship between parameters is analyzed using a multi-head attention mechanism; the spectrum-efficiency data set is extracted from the literature in combination with the semantic model, and the real-time / simulation parameters are predicted and analyzed and the parameter settings are optimized. Finally, the process adjustment is guided by the prediction results and parameters, and a closed-loop feedback iterative optimization simulation model is formed. The present invention realizes dynamic modeling, accurate prediction and control analysis of material processes, significantly improves process optimization efficiency and material performance prediction accuracy, and reduces the cost of R&D trial and error.

[0173] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0174] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed across multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0175] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0176] Those skilled in the art will appreciate that all or part of the steps of the above-mentioned method embodiments may be implemented by hardware associated with program instructions, and the aforementioned program may be stored in a computer-readable storage medium. When the program is executed, the program executes the steps of the above-mentioned method embodiments. The aforementioned storage medium includes various media that can store program codes, such as mobile storage devices, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0177] Alternatively, if the integrated units described above are implemented as software modules and sold or used as standalone products, they can also be stored on a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of the present invention, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product, stored on a storage medium, includes instructions for enabling a computer device (such as a personal computer, server, or network device) to execute all or part of the methods described in various embodiments of the present invention. The aforementioned storage media include various media capable of storing program code, such as removable storage devices, ROM, RAM, magnetic disks, or optical disks.

[0178] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. An intelligent twin control method based on material process optimization, characterized in that: include: S1: According to the target material production process, multiple control nodes are set based on the time dimension, and the preparation equipment parameters, processing parameters and processing environment parameters are recorded; S2: Sequence the preparation equipment parameters, processing parameters and processing environment parameters to obtain a multi-dimensional parameter sequence; S3: Spectral characteristic data and performance data of the target material are measured experimentally, and the spectral characteristic data and performance data are serialized to obtain a spectral sequence and a performance sequence; S4: Based on the Transformer model, a multi-to-multi-parameter simulation model is established. The parameter sequence is used as the input sequence, and the spectrum sequence and performance sequence are used as the output sequence to perform multi-parameter interaction modeling. The dependency analysis and prediction training of the input and output sequences are performed through the multi-head attention mechanism. S5: Using the semantic model and target materials as search tags, content data is retrieved from big data literature and a spectrum-efficiency dataset is extracted. User simulation sequences or real-time parameter sequences are imported into the simulation model for spectrum-efficiency prediction. The rationality of the prediction results is evaluated in combination with the spectrum-efficiency dataset, and spectrum-efficiency prediction data and process optimization parameter sequences are obtained. S6: Optimize the production process of the target material through spectral effect prediction data and process optimization parameter sequence, and dynamically optimize the simulation model based on the actual optimized parameter results as training data.

2. The intelligent twin control method based on material process optimization according to claim 1 is characterized in that: Said S1 is specifically: Based on the material production process, multiple control time points are set in a production process, and each control time point is used as a control node; In a production process, the processing parameters are monitored in real time through equipment sensors and environmental sensors. For each control node, the preparation equipment parameters, processing parameters and processing environment parameters are recorded.

3. The intelligent twin control method based on material process optimization according to claim 1 is characterized in that: The S2 is specifically: Based on the time dimension, each parameter in the preparation equipment parameters, processing technology parameters and processing environment parameters is serialized, the sequence point corresponds to each control node, and a multi-dimensional parameter sequence is obtained.

4. The intelligent twin control method based on material process optimization according to claim 1 is characterized in that: The S3 is specifically: Through material experiments, test nodes are set and the spectrum of the target material is measured. Multi-segment layered information is extracted from the spectrum, and the spectrum information is stored as characteristic values ​​to obtain spectral characteristic data. Spectral information includes pixel spacing, wavelength response, reflectivity, transmittance, and absorptivity of electromagnetic waves of different wavelengths; Determine the performance status of the target material based on the test node, and serialize the obtained performance information to obtain a performance sequence; Performance information includes material strength, electrical conductivity, and heat resistance.

5. The intelligent twin control method based on material process optimization according to claim 1 is characterized in that: The S4 is specifically: Based on the Transformer model, a multi-to-multi-parameter simulation model is established, with parameter sequences as input sequences and spectral sequences and performance sequences as output sequences. The nonlinear coupling relationship between the multiple variables of the input and output sequences is analyzed, and a multi-head attention mechanism is introduced to perform linear layer transformations on different parameters in the input sequence, and obtain the corresponding query, key, and value values ​​respectively. The attention score is then obtained through the multi-head attention calculation formula. The long-term dependency between input and output sequences is analyzed through attention scores, and long-term prediction training of input and output sequences is performed. The prediction optimization of the simulation model is performed using the mean square error as the loss function. An LSTM model is constructed in the simulation model. The weight values ​​of the input parameters are set according to the long-term dependencies and the attention scores of different parameters. The input and output sequences are respectively imported into the LSTM model for short-term prediction training. The mean square error is used as the loss function, and the weight value is used as the prediction weight of the input sequence for the output sequence. Through the backpropagation algorithm, the LSTM model parameters are automatically adjusted to reduce the value of the loss function until the loss function converges to the preset value or the number of training iterations is reached.

6. The intelligent twin control method based on material process optimization according to claim 1 is characterized in that: The S5 is specifically: Generate text-based material content data based on the target material, perform word segmentation and keyword annotation on the material content data using the TF-IDF algorithm, and generate search tags; Deploy the SciBERT semantic analysis model to match and filter relevant content from big data documents using search tags to obtain search content data; Based on the search content data, word segmentation and word context association are performed to match preparation parameters, process parameters, spectral characteristics and performance characteristics phrases, and spectral characteristic data and performance characteristic data are screened out. The screened data are integrated to obtain the spectrum-efficiency data set.

7. The intelligent twin control method based on material process optimization according to claim 6 is characterized in that: The S5 is specifically: Set the user simulation process sequence based on the target material production process; Constructing a real-time parameter sequence based on the parameters of the current target material real-time production record; Import the user simulation process sequence or real-time parameter sequence into the simulation to perform sequence prediction and obtain the output prediction sequence; The output prediction sequence is subjected to spectrum-efficacy characteristic data analysis to generate predicted spectrum-efficacy data; Multi-dimensional parameter deviation calculations were performed on the predicted spectrum-efficiency data and the spectrum-efficiency data set. The prediction weights of different parameters in the simulation model were introduced to evaluate the deviation. The rationality of the simulation model prediction and the degree of optimization of the process parameters were evaluated through the parameter deviation. If the deviation range is within the preset range, the predicted spectrum-efficiency data is used as the target optimization data, multiple optimization parameter sequences are set for prediction analysis, and finally the optimal result is screened out, and the spectrum-efficiency prediction data and process optimization parameter sequence in the optimal result are recorded.

8. The intelligent twin control method based on material process optimization according to claim 1 is characterized in that: Said S6 is specifically: Conduct feasibility analysis on target materials according to the process optimization parameter sequence, and set feasible optimization parameters to optimize the production process; After process optimization, the result parameter data of material production are recorded, serialized and imported into the simulation model as real data for predictive training and dynamic optimization.

9. An intelligent twin control system based on material process optimization, characterized in that: The system includes: a memory and a processor. The memory includes an intelligent twin control program based on material process optimization. When the intelligent twin control program based on material process optimization is executed by the processor, the following steps are implemented: S1: According to the target material production process, multiple control nodes are set based on the time dimension, and the preparation equipment parameters, processing parameters and processing environment parameters are recorded; S2: Sequence the preparation equipment parameters, processing parameters and processing environment parameters to obtain a multi-dimensional parameter sequence; S3: Spectral characteristic data and performance data of the target material are measured experimentally, and the spectral characteristic data and performance data are serialized to obtain a spectral sequence and a performance sequence; S4: Based on the Transformer model, a multi-to-multi-parameter simulation model is established. The parameter sequence is used as the input sequence, and the spectrum sequence and performance sequence are used as the output sequence to perform multi-parameter interaction modeling. The dependency analysis and prediction training of the input and output sequences are performed through the multi-head attention mechanism. S5: Using the semantic model and target materials as search tags, content data is retrieved from big data literature and a spectrum-efficiency dataset is extracted. User simulation sequences or real-time parameter sequences are imported into the simulation model for spectrum-efficiency prediction. The rationality of the prediction results is evaluated in combination with the spectrum-efficiency dataset, and spectrum-efficiency prediction data and process optimization parameter sequences are obtained. S6: Optimize the production process of the target material through spectral effect prediction data and process optimization parameter sequence, and dynamically optimize the simulation model based on the actual optimized parameter results as training data.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium includes an intelligent twin control program based on material process optimization. When the intelligent twin control program based on material process optimization is executed by the processor, the steps of the intelligent twin control method based on material process optimization as described in any one of claims 1 to 8 are implemented.

Images