Multi-material color mixing process digital twin simulation platform

Abstract

The application relates to the field of digital twin simulation, and specifically discloses a multi-material color mixing process digital twin simulation platform, which converts a geometric model, material properties and rated process parameters into basic physical simulation data by constructing a multi-material color mixing process digital twin simulation platform, and acquires real-time process parameters by means of an IoT interface to determine a parameter drift vector. At the same time, a lightweight machine learning model is used to process parameter drift characteristics and color distribution characteristics to generate a color difference correction map to dynamically correct color prediction results. In this way, color deviation caused by parameter drift can be dynamically corrected, accurate simulation of the dynamic evolution of the multi-material color mixing process can be realized, the efficiency of color mixing process design and optimization can be improved, time and material costs can be reduced, and the rapid switching demand in the multi-variety small-batch production scene can be met.

Description

Technical Field

[0001] This application relates to the field of digital twin simulation, and more specifically, to a digital twin simulation platform for multi-material color mixing processes. Background Technology

[0002] In modern industrial production, multi-material color mixing processes are widely used in coatings, plastics, inks, and other fields. These processes achieve stable and desired color effects by precisely controlling the mixing ratios and process parameters of different materials. However, in actual production, factors such as equipment aging, fluctuations in ambient temperature and humidity, and batch differences in raw materials can cause unpredictable drifts in key process parameters (such as temperature, flow rate, and stirring rate). These drifts directly alter the core physical properties of the materials, such as viscosity and fluidity, thus affecting the flow patterns and microstructure of the materials within the mixing chamber, ultimately leading to a deviation of the macroscopic color effect from the design target. Due to the lack of precise simulation methods for the dynamic evolution of the color mixing process, traditional process design and optimization rely solely on manual trial and error and experience-based adjustments. This not only consumes significant time and material costs but also struggles to meet the rapid changeover requirements of multi-variety, small-batch production scenarios.

[0003] In existing technologies, the simulation and control of multi-material color mixing processes are mainly based on simplified static physical models and empirical formulas. These models typically assume that process parameters remain stable and ignore the dynamic impact of parameter drift on the rheological properties of materials during production, simplifying the color mixing process into a static mixing calculation under fixed parameters. For example, when simulating material flow, traditional models assume viscosity to be constant and do not consider the nonlinear viscosity fluctuations caused by temperature changes. When predicting color effects, they rely solely on the linear mapping relationship between material ratios and color space, neglecting the complex modulation effect of micro-mixing structures on optical properties. When parameter drift occurs in actual production, static models cannot capture the dynamic causal chain of "parameter change - rheological property change - mixing structure evolution - color deviation," leading to significant deviations between simulation results and actual color mixing effects. This deficiency necessitates frequent physical prototyping and parameter calibration on the production floor, increasing scrap rates and energy costs, and potentially causing batch quality problems due to debugging delays. More importantly, in the face of increasingly complex multi-component material systems and high-precision color requirements, the predictive ability of static models is declining exponentially, making it difficult to meet the core demands of intelligent manufacturing for precise process control.

[0004] Therefore, an optimized digital twin simulation platform for multi-material color mixing processes is needed. Summary of the Invention

[0005] This application is made in order to solve the above-mentioned technical problems.

[0006] According to one aspect of this application, a digital twin simulation platform for multi-material color mixing processes is provided, comprising:

[0007] The input data acquisition module is used to acquire the geometric model, material identification ID, and rated process parameters input by the user.

[0008] The basic physics simulation module is used to perform rapid basic physics simulation on the geometric model based on the material identification ID and the rated process parameters to obtain basic microstructure prediction results and basic color prediction results;

[0009] The real-time process parameter acquisition module is used to acquire real-time process parameters through an IoT interface;

[0010] The deviation prediction module is used to calculate the parameter drift vector between the real-time process parameters and the rated process parameters, and to perform deviation prediction on the basic color prediction results based on the parameter drift vector to obtain a color difference correction map.

[0011] The color appearance correction module is used to apply the color difference correction map to the basic color prediction result to obtain the final color appearance prediction map.

[0012] Compared with existing technologies, this application provides a digital twin simulation platform for multi-material color mixing processes. By constructing such a platform, it transforms geometric models, material properties, and rated process parameters into basic physical simulation data. It then uses an IoT interface to acquire real-time process parameters to determine the parameter drift vector. Simultaneously, a lightweight machine learning model is used to process the parameter drift features and color distribution features, generating a color difference correction map to dynamically correct the color prediction results. In this way, color deviations caused by parameter drift can be dynamically corrected, achieving accurate simulation of the dynamic evolution of the multi-material color mixing process. This improves the efficiency of color mixing process design and optimization, reduces time and material costs, and meets the rapid changeover requirements in multi-variety, small-batch production scenarios. Attached Figure Description

[0013] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0014] Figure 1 This is a block diagram of a digital twin simulation platform for multi-material color mixing processes according to an embodiment of this application.

[0015] Figure 2 This is a schematic diagram of data flow in a digital twin simulation platform for multi-material color mixing processes according to an embodiment of this application.

[0016] Figure 3 This is a block diagram of the basic physical simulation module in the digital twin simulation platform for multi-material color mixing processes according to an embodiment of this application.

[0017] Figure 4 This is a block diagram of the deviation prediction module in the digital twin simulation platform for multi-material color mixing processes according to an embodiment of this application.

[0018] Figure 5 This is a block diagram of the color difference correction unit in the digital twin simulation platform for multi-material color mixing processes according to an embodiment of this application.

[0019] Figure 6 This is a block diagram of a feature modeling secondary subunit in a digital twin simulation platform for multi-material color mixing processes according to an embodiment of this application. Detailed Implementation

[0020] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0021] It should be understood that the steps described in the method embodiments of this disclosure may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this disclosure is not limited in this respect.

[0022] This application is made in response to the problems mentioned above in the background art. Figure 1 This is a block diagram of a digital twin simulation platform for multi-material color mixing processes according to an embodiment of this application. Figure 2 This is a schematic diagram of the data flow in a digital twin simulation platform for a multi-material color mixing process according to an embodiment of this application. Specifically, as shown... Figure 1 and Figure 2As shown, the multi-material color mixing process digital twin simulation platform 100 according to an embodiment of this application includes: an input data acquisition module 110, used to acquire a geometric model, material identifier ID, and rated process parameters input by a user; a basic physical simulation module 120, used to perform rapid basic physical simulation on the geometric model based on the material identifier ID and the rated process parameters to obtain basic microstructure prediction results and basic color prediction results; a real-time process parameter acquisition module 130, used to acquire real-time process parameters through an IoT interface; a deviation prediction module 140, used to calculate the parameter drift vector between the real-time process parameters and the rated process parameters, and perform deviation prediction on the basic color prediction results based on the parameter drift vector to obtain a color difference correction map; and a color appearance correction module 150, used to apply the color difference correction map to the basic color prediction results to obtain a final color appearance prediction map.

[0023] Specifically, the input data acquisition module 110 is used to acquire the geometric model, material identifier ID, and rated process parameters input by the user. In particular, in one example of this application, the material identifier ID includes material identifier IDs for two or more materials. It should be understood that since the digital twin simulation of multi-material color mixing processes requires a digital representation of physical objects, the geometric model is a digital mapping of the spatial structure of the color mixing equipment, the material identifier ID is a unique index for retrieving material physical properties, and the rated process parameters are a digital definition of standard operating conditions, this application provides standardized initial data input for each functional module of the platform by acquiring the geometric model, material identifier ID, and rated process parameters. The geometric model clarifies the spatial boundaries of material mixing, the material identifier ID is used to associate rheological parameters with optical parameters and other physical property data, and the rated process parameters are used to set the baseline conditions for simulation. This allows subsequent modules to perform physical simulation and deviation analysis based on a unified data source, avoiding simulation link breaks due to data loss.

[0024] In a specific example of this application, the operational mission requirement acquisition module 110 is implemented as follows: First, a user input interface needs to be constructed. This interface supports the visual uploading of geometric models, allows users to select 3D model files in standard formats such as STL and STEP, and provides a drop-down menu for material identifiers (IDs). The menu data is linked to a material database, allowing selection of two or more material identifiers. Input fields for rated process parameters are also set, including input boxes for numerical parameters such as temperature and flow rate ratio. After receiving the input data, the system initiates a data verification mechanism to check the integrity of the geometric model, verifying the existence of the material identifiers in the database, and verifying whether the rated process parameters are within a preset reasonable range, such as the temperature not exceeding the material tolerance threshold. If data anomalies are found, a prompt is issued to the user requesting correction. After successful verification, the system converts the geometric model into a format suitable for coarse-mesh CFD simulation, retrieves rheological and optical parameters in batches from the database based on the material identifiers, and stores the rated process parameters, material properties, and geometric model in a structured association. For multi-material scenarios, the system automatically identifies the material types, ensures correct matching of attribute data through hash indexing, and finally packages and transmits the processed data to the subsequent basic physical simulation module, providing a standardized input dataset for subsequent simulations.

[0025] Specifically, the basic physical simulation module 120 is used to perform rapid basic physical simulation on the geometric model based on the material identifier ID and the rated process parameters to obtain basic microstructure prediction results and basic color prediction results. It should be understood that since the microstructure of a material directly determines the scattering and absorption behavior of light, thus affecting color appearance, this application couples material properties with process parameters through physical simulation to reveal the intrinsic relationship between microstructure and color, providing comparable initial benchmark data for the subsequent analysis and processing of the deviation prediction module.

[0026] In a specific example of this application, Figure 3 This is a block diagram of the basic physical simulation module in the digital twin simulation platform for multi-material color mixing processes according to an embodiment of this application. Figure 3 As shown, the basic physical simulation module 120 includes: a material property loading unit 121, used to load physical property data of selected materials from a material database based on the material identifier ID; a CFD simulation unit 122, used to obtain the basic microstructure prediction result based on the physical property data and the rated process parameters and through coarse-grid CFD simulation; and a basic color prediction unit 123, used to process the basic microstructure prediction result based on the Kubelka-Monk optical model to obtain the basic color prediction result.

[0027] Specifically, the material property loading unit 121 is used to load the physical property data of the selected material from the material database based on the material identifier ID. In particular, in one example of this application, the physical property data includes rheological parameters and optical parameters, and the rated process parameters include temperature and flow rate ratio. It should be understood that the rheological parameters of a material directly determine its flow characteristics and microstructure formation during the mixing process, while optical parameters form the physical basis for color prediction. Temperature and flow rate ratio, as key process parameters, significantly affect the material's physical properties and mixing effect. Therefore, this application loads the physical property data of the selected material from the material database, thereby shifting the color mixing simulation from experience-driven to data-driven, providing standardized and accurate initial data support for subsequent modules of the multi-material color mixing process digital twin simulation platform.

[0028] In a specific example of this application, the material property loading unit 121 is implemented as follows: The material database adopts a three-layer architecture design. The basic information layer queries the material base table based on the material identifier ID to obtain basic data. The rheological parameter layer extracts parameters such as viscosity-temperature curves by associating the rheological property table with the material identifier ID. The optical parameter layer accesses the optical property table to load parameters required by the Kubelka-Monk model, such as the absorption coefficient K and scattering coefficient S. For multi-material scenarios, the system simultaneously obtains multiple material properties through parallel query technology and stores them in a hash table to ensure a one-to-one correspondence between properties and materials. The loaded physical property data needs to undergo format conversion and structuring processing. Rheological parameters are converted into numerical matrices in CSV format, optical parameters are segmented by wavelength and organized into JSON arrays, and multiple material properties are expanded through dimensionality to form a three-dimensional tensor to adapt to the mesh calculation requirements of subsequent CFD simulations. If a database query times out or returns a null value, the system automatically triggers a backup data source and records an exception log. At the same time, it enables an attribute caching mechanism to store commonly used material attributes in a Redis cache. When loading repeatedly, the data is read directly from the cache to improve loading efficiency. Finally, the processed attribute data is pushed to the basic physical simulation module through a message queue to complete the association with the geometric model and process parameters.

[0029] Specifically, the CFD simulation unit 122 is used to obtain the basic microstructure prediction results based on the physical property data and the rated process parameters through coarse-grid CFD simulation. It should be understood that coarse-grid CFD simulation can perform a basic simulation of the material mixing process based on physical property data and rated process parameters while reducing computational complexity. Specifically, by combining material physical property data and rated process parameters, coarse-grid CFD simulation is used to quickly construct a basic prediction of the flow pattern and distribution state of the material within the mixing cavity, obtaining results that reflect the microstructure of the material mixing. This CFD simulation avoids the computational burden of full physical simulation while preserving the interpretability of the physical model, enabling the system to quickly output valuable benchmark predictions during real-time operation, providing an effective starting point for subsequent dynamic correction.

[0030] In a specific example of this application, the CFD simulation unit 122 is implemented as follows: First, the simulation environment needs to be initialized based on the physical property data of the selected material and the rated process parameters. Specifically, an appropriate set of fluid dynamics equations is set, such as the Navier-Stokes equations, i.e. , ,in, For fluid density, For time, For divergence calculation, It is a velocity vector. This involves calculating partial derivatives and considering mass transfer phenomena in multiphase systems, with particular attention to the interactions between different materials and their impact on mixing. During this process, precise boundary conditions, such as inlet velocity distribution and no-slip conditions on the walls, need to be determined based on rated process parameters. Subsequently, a coarse-grid CFD simulation is executed, iteratively solving a series of partial differential equations to capture the flow patterns and distribution states of the fluid within the mixing chamber. The data generated during the simulation includes information on the flow field, pressure field, and velocity field, providing a detailed description of how materials flow and mix within the mixing chamber. In-depth analysis of the simulation results yields predictions of the fundamental microstructure reflecting the microstructure of the material mixing process. Here, the fundamental microstructure predictions refer to the microstructure of color and texture characteristics at the macroscale, directly affecting the color performance of the final product. This approach achieves accurate simulation of the dynamic evolution of the multi-material color mixing process, improving the efficiency of color mixing process design and optimization, and reducing time and material costs.

[0031] Specifically, the basic color prediction unit 123 is used to process the basic microstructure prediction results based on the Kubelka-Monk optical model to obtain the basic color prediction results. It should be understood that the Kubelka-Monk optical model is a mathematical model used to describe the absorption and scattering behavior of light when propagating in a multi-material hybrid system. Its core is to establish a mapping relationship between the microscopic material distribution and macroscopic optical properties by quantifying the absorption coefficient K and scattering coefficient S of the material. Specifically, this application uses the basic microstructure prediction results output by the CFD simulation unit as input, and calculates the absorption and scattering process of light in the hybrid material based on the K and S parameters pre-stored in the material database. Then, the microstructure information is converted into the luminance component, red-green color component, and yellow-blue color component values ​​in the CIELAB color space, thereby generating the basic color prediction results and laying the optical theoretical foundation for subsequent dynamic correction of color deviation.

[0032] In a specific example of this application, the basic color prediction unit 123 is implemented as follows: After obtaining the basic microstructure prediction results from the coarse-grid CFD simulation, the system first loads the optical parameters of the selected materials from the material database, including the absorption coefficient K and scattering coefficient S required by the Kubelka-Monk model. These parameters correspond one-to-one with the material identifier ID and have been pre-calibrated according to the material's physical properties. Next, the spatial distribution data of the materials in the basic microstructure prediction results, such as the volume fraction and layered distribution characteristics of different materials at various locations in the mixing cavity, are converted into a format suitable for optical model input, typically represented as a material composition matrix within a three-dimensional mesh cell. Then, based on the basic equations of the Kubelka-Monk model, i.e. ,in The absorption coefficient is... The scattering coefficient is... For diffuse reflectance, optical properties are calculated for each grid cell. For multi-layered or mixed materials, the absorption and scattering effects of light within the cell are calculated based on the absorption and scattering coefficients. The optical effects of adjacent cells are accumulated through iteration or integration to simulate the light transmission process in the mixed material. During this process, the influence of the microstructure characteristics of the material distribution on light scattering needs to be considered, and the model parameters are adaptively adjusted. After calculation, the optical properties of each grid cell are converted into the luminance, red-green, and yellow-blue color components in the CIELAB color space, forming a basic color prediction matrix corresponding to the geometric model space. Finally, the basic color prediction matrix is ​​mapped to the user-input geometric model to generate a visualized basic color prediction result.

[0033] Specifically, the real-time process parameter acquisition module 130 is used to acquire real-time process parameters through an IoT interface. Specifically, it collects process parameters from the production site in real time via the IoT interface, establishing a dynamic data connection between the physical world and the digital twin model. That is, the IoT interface allows for real-time and accurate acquisition of process parameters during production, providing the system with high-frequency dynamic data input and ensuring that the accuracy of parameter drift detection reaches the actual sampling frequency of the production site. This real-time data accurately reflects the impact of equipment operating status and environmental changes on process parameters, providing a reliable data source for subsequent deviation prediction.

[0034] In a specific example of this application, the real-time process parameter acquisition module 130 is implemented as follows: First, IoT sensors, such as temperature sensors and flow sensors, are deployed at key process parameter monitoring points of the production equipment. The sensor type must be selected according to the characteristics of the process parameters, such as thermocouple temperature sensors or electromagnetic flow meters, and the sampling frequency must be ensured to be no lower than the characteristic frequency of parameter drift, such as 10 samples per second. Second, an edge computing gateway is deployed in the workshop to establish a connection with the sensors through an industrial communication protocol, and to complete the acquisition and preliminary preprocessing of real-time data, such as removing impulse noise, standardizing data format, and converting the analog or digital signals output by the sensors into process parameter data in a unified format, such as temperature values ​​and flow rates. Next, an IoT interface adaptation module is developed in the digital twin simulation platform. This module establishes a bidirectional communication link with the edge computing gateway based on a standardized interface protocol, sets the real-time priority of data transmission, such as prioritizing real-time parameter data over historical data transmission, and encrypts and verifies the integrity of the transmitted data to ensure that the parameter data is not lost or tampered with during transmission. Then, the interface module stores the acquired real-time process parameters in the platform's time-series database, indexed by timestamps, such as including the acquisition time for each data entry. Simultaneously, it pushes the latest parameter data to the drift feature extraction module. During this process, a data buffering mechanism, such as a circular buffer, needs to be implemented to temporarily store untransmitted data when the network is interrupted, ensuring data continuity once the network is restored. Furthermore, the system needs to perform outlier detection on real-time parameters. When a parameter value is detected to exceed the rated range by ±20%, an early warning is triggered, prompting on-site personnel to check the equipment status and ensure the validity of the acquired data. Ultimately, through these steps, seamless acquisition of real-time process parameters from production site sensors to the digital twin platform is achieved, providing dynamic data support for parameter drift analysis and color prediction correction.

[0035] Specifically, the deviation prediction module 140 is used to calculate the parameter drift vector between the real-time process parameters and the rated process parameters, and to perform deviation prediction on the basic color prediction result based on the parameter drift vector to obtain a color difference correction map. Specifically, this application calculates the deviation between the real-time process parameters and the rated process parameters at each predetermined time point, arranges the deviations according to the time dimension to obtain the parameter drift vector, and combines the spatial distribution characteristics of the basic color prediction with a lightweight machine learning model to learn the mapping law between parameter drift patterns and color errors, thereby generating a color difference correction map corresponding to the basic color prediction space. This map is used to subsequently correct the basic color prediction results point by point, enabling the color prediction to adapt to parameter fluctuations in the production environment.

[0036] In a specific example of this application, Figure 4 This is a block diagram of the deviation prediction module in a digital twin simulation platform for multi-material color mixing processes according to an embodiment of this application. Figure 4 As shown, the deviation prediction module 140 includes: a color distribution feature extraction unit 141, used to extract color distribution features from the basic color prediction result to obtain a basic color space distribution feature map; a drift time series feature extraction unit 142, used to extract a parameter drift time series feature encoding vector from the parameter drift vector; and a color difference correction unit 143, used to input the parameter drift time series feature encoding vector and the basic color space distribution feature map into a pre-trained lightweight machine learning model to obtain the color difference correction map.

[0037] Specifically, the color distribution feature extraction unit 141 is used to extract color distribution features from the basic color prediction results to obtain a basic color space distribution feature map. That is, by systematically extracting color distribution features, a basic color space distribution feature map corresponding to the physical model prediction results is constructed. This map contains spatial feature information such as color gradient and local variance, serving as a key basis for the machine learning model to analyze the impact of parameter drift. This allows the model to more accurately learn the mapping relationship between parameter drift and color error, laying a feature foundation for generating a color difference correction map, thereby improving the accuracy and reliability of color deviation prediction.

[0038] In a specific example of this application, the color distribution feature extraction unit 141 is implemented as follows: First, the basic color prediction result is converted from the RGB color space to the CIELAB color space, because the CIELAB space better matches the human eye's perception of color differences, facilitating subsequent color difference calculation and feature extraction. Next, the converted color data is decomposed into channels to obtain three component data: luminance component, red-green component, and yellow-blue component. Then, for each channel, a sliding window technique is used to calculate the color gradient value of the local region, thereby representing the rate of color change in space. Simultaneously, the variance of the local region's color values ​​is calculated to reflect the uniformity of the color distribution. Afterwards, a global mean pooling operation is performed to obtain the global color feature vector for each channel, summarizing the overall color distribution trend. Then, the local gradient, variance features, and global features are fused to form a multi-scale color space distribution feature. To adapt to the input requirements of subsequent machine learning models, the fused features need to be normalized and reconstructed into a color feature map with the same dimension as the basic color prediction result according to spatial location information. In this process, the shallow feature extraction module in the convolutional neural network will be used to further abstract and reduce the dimensionality of the color feature map, retain the key features that best reflect the differences in color distribution, and finally obtain the basic color space distribution feature map, which provides high-quality feature data for the joint input of the parameter drift vector and the feature map and the bias prediction.

[0039] Specifically, the drift temporal feature extraction unit 142 is used to extract a parameter drift temporal feature encoding vector from the parameter drift vector. In particular, in a specific example of this application, the parameter drift vector is processed by a temporal correlation feature extractor based on an LSTM model to obtain the parameter drift temporal feature encoding vector. That is, by extracting the temporal features of the parameter drift vector, a parameter drift temporal feature encoding vector containing dynamic information in the time dimension is constructed. This vector can characterize the rate of change, periodic fluctuations, and anomalous jumps of parameter drift in the time series, providing key temporal dimension input support for subsequent machine learning models to analyze the impact of dynamic drift on color prediction. This enables the model to more accurately capture the temporal correlation of parameter drift, improve the timeliness and accuracy of color deviation prediction in dynamic drift scenarios, and provide temporal dimension feature basis for subsequent generation of color difference correction maps.

[0040] In a specific example of this application, the drift time-series feature extraction unit 142 is implemented as follows: First, the parameter drift vectors are arranged into a time-series sequence according to the time dimension, and the dimensional differences are eliminated through normalization processing, and the sequence is divided into fixed-length segment segments. Second, a time-series correlation feature extractor based on the LSTM model is constructed. The LSTM model is a special type of recurrent neural network, whose core structure includes an input gate, a forget gate, and an output gate. By memorizing long-term dependent information through cell state memory, it can effectively capture the trend, periodicity, and other time-series features in parameter drift. For example, for the slow drift or periodic fluctuation of process parameters such as temperature and flow rate, LSTM can selectively retain key information and filter noise through a gating mechanism, thereby accurately extracting the time-series features of parameter drift. Then, the LSTM model is trained using historical parameter drift data, with the sequence segments of the parameter drift vector as input and the system state such as actual color deviation as output labels. During training, the model learns the mapping relationship between the parameter drift time sequence and color deviation, and adjusts the weights to improve the feature extraction accuracy. Finally, the parameter drift vector is input into the trained LSTM model, which processes the time series data layer by layer. The last hidden state is the parameter drift time series feature encoding vector.

[0041] Specifically, the color difference correction unit 143 is used to input the parameter drift temporal feature encoding vector and the basic color space distribution feature map into a pre-trained lightweight machine learning model to obtain the color difference correction map. In particular, this application utilizes the fitting ability of machine learning models to dynamic features, learns the mapping law between parameter drift patterns and color errors through the model, and generates a color difference correction map corresponding to the basic color prediction space. This achieves accurate prediction and dynamic compensation for color deviations caused by parameter drift, enabling the color prediction results to adapt to parameter fluctuations in the production environment.

[0042] In a specific example of this application, Figure 5 This is a block diagram of the color difference correction unit in a digital twin simulation platform for multi-material color mixing processes according to an embodiment of this application. Figure 5 As shown, the color difference correction unit 143 includes: a feature modeling secondary subunit 1431, used to input the parameter drift temporal feature encoding vector and the basic color space distribution feature map into the encoder of the pre-trained lightweight machine learning model to obtain a color space distribution shift latent modeling feature map; and a feature decoding secondary subunit 1432, used to input the color space distribution shift latent modeling feature map into the decoder of the pre-trained lightweight machine learning model to obtain the color difference correction map.

[0043] More specifically, the feature modeling secondary subunit 1431 is used to input the parameter drift temporal feature encoding vector and the basic color space distribution feature map into the encoder of the pre-trained lightweight machine learning model to obtain a color space distribution shift latent modeling feature map. In a specific example of this application, Figure 6 This is a block diagram of a secondary subunit for feature modeling in a digital twin simulation platform for multi-material color mixing processes according to an embodiment of this application. Figure 6 As shown, the feature modeling secondary subunit 1431 includes: a global mean pooling tertiary subunit 14311, used to perform global mean pooling on the basic color space distribution feature map to obtain a basic color space distribution global pooling encoding vector; a dynamic interaction analysis tertiary subunit 14312, used to perform global dynamic interaction analysis on the parameter drift temporal feature encoding vector and the basic color space distribution global pooling encoding vector to obtain a color space distribution full-dimensional dynamic perception encoding matrix; and a nonlinear activation tertiary subunit 14313, used to perform nonlinear activation on the color space distribution full-dimensional dynamic perception encoding matrix and then apply it to the basic color space distribution feature map to obtain the color space distribution offset implicit modeling feature map.

[0044] More specifically, the global mean pooling three-level subunit 14311 is expressed by the formula:

[0045]

[0046] in, and These are the height and width of the basic color space distribution feature map, respectively. It is the basic color space distribution feature map Location feature value, It is a global pooling encoding vector based on the color space distribution.

[0047] In other words, by using global mean pooling, the high-dimensional feature data of the basic color space distribution feature map is compressed into a low-dimensional global pooling encoding vector of the basic color space distribution. This achieves feature dimensionality reduction and parameter regularization, effectively reducing feature dimensionality, computational complexity, and enhancing the model's robustness to color space distribution translation changes. It also avoids overfitting and ensures that when dynamically interacting with parameter drift time-series features, the model can establish a macroscopic correlation between parameter changes and color distribution deviations, providing crucial global feature support for generating accurate color difference correction maps.

[0048] More specifically, the dynamic interactive analysis three-level subunit 14312 is expressed by the formula:

[0049]

[0050]

[0051] in, for Activation function and Let be the projection matrix. The color space distribution is a dynamic weighted perception matrix. The color space distribution is a dynamic weighted perception matrix. For positional product, For vector multiplication, The parameter drift time-series feature encoding vector, A trainable bias vector for full-dimensional perception of color space distribution. It is a full-dimensional dynamic perception encoding matrix for color space distribution.

[0052] In other words, through global dynamic interactive analysis, the parameter drift temporal feature encoding vector and the basic color space distribution global pooling encoding vector are dynamically correlated across all dimensions, thereby capturing the potential influence patterns of all dimensions between the two, generating a full-dimensional dynamic perception encoding matrix of color space distribution that can finely encode the interdependence between parameter drift features and color distribution features. This provides key interactive feature representations for subsequent modulation of the basic color space distribution feature map based on this matrix to generate a color difference correction map, realizing deep fusion and dynamic mapping of cross-modal features.

[0053] Preferably, here, the color space distribution dynamic weight perception matrix As a time-series feature encoding vector of parameter drift and the global pooling encoding vector of the basic color space distribution Based on projection matrix and The preset dimensions of implicit association modeling are further used to analyze different dimensions. The vectors are dynamically and perceptually encoded across all dimensions, thus requiring a dynamic weighted perceptual matrix to be distributed in the color space. and projection matrix and It can achieve intrinsic convergence, thereby avoiding dimensionality compatibility conflicts and improving the encoding matrix. The interactive representation of the encoding effect.

[0054] Specifically, firstly regarding the projection matrix and and the dynamic weight perception matrix of color space distribution Calculate different multi-order statistical invariants:

[0055]

[0056]

[0057] in, Let F be the norm of the matrix. , , and These are different multi-order statistical invariants. This is the inverse of the matrix. Therefore, , and , The fractional spectral characterization theorem is satisfied at different levels of operator action.

[0058] Then, based on the effects of fractional-order invariant space dimensions at different orders, the dynamic weight perception matrix of color space distribution is obtained by using dimensional fiberization based on statistical angles. Relative to the projection matrix and The tensor fiber bundle constraint is:

[0059]

[0060]

[0061] in, and This is the coupling correction matrix.

[0062] In other words, fractional invariants of different orders are regarded as dimensional fibrils at different levels of spatial dimension to ensure that the asymptotic convergence between matrices has a single canonical configuration in terms of dimensional correlation. For example, it can be simply understood as achieving canonical uniqueness through operator noncommutativity.

[0063] Thus, based on and The coupling correction matrix is ​​used to correct the color space distribution dynamic weight perception matrix. ,Right now:

[0064]

[0065] in, The dynamic weighted perception matrix for the corrected color space distribution.

[0066] This is achieved through a dynamic weighted perception matrix distributed in the color space. and projection matrix and The convergence of the intrinsic values ​​between them achieves dimensional compatibility consensus, thereby improving the full-dimensional dynamic perception coding matrix of color space distribution. The interactive representation of the encoding effect. Finally, a global dynamic interactive analysis is performed on the parameter drift temporal feature encoding vector and the basic color space distribution global pooling encoding vector using the corrected color space distribution dynamic weight perception matrix:

[0067]

[0068] in, To correct the color space distribution, a full-dimensional dynamic sensing encoding matrix is ​​used.

[0069] More specifically, the nonlinear activation third-level subunit 14313 includes: performing nonlinear activation on the color space distribution full-dimensional dynamic perception coding matrix to obtain an activated color space distribution full-dimensional dynamic perception coding matrix; and performing positional dot product between the activated color space distribution full-dimensional dynamic perception coding matrix and each of the basic color space distribution feature matrices along the channel dimension in the basic color space distribution feature map, so as to apply the activated color space distribution full-dimensional dynamic perception coding matrix to the basic color space distribution feature map to obtain the color space distribution offset implicit modeling feature map, expressed by the formula:

[0070]

[0071] in, for Activation function The activated color space distribution is a full-dimensional dynamic perceptual coding matrix. This is a positional dot product along the channel dimension. Based on the basic color space distribution feature map The color space distribution offset is implicitly modeled as a feature map.

[0072] In other words, by using the activated color space distribution full-dimensional dynamic perception encoding matrix as adaptive weights, the basic color space distribution feature matrices of the basic color space distribution feature map are modulated by positional dot product along the channel dimension. This realizes the dynamic guidance of parameter drift temporal features on the basic color distribution features, so that the generated color space distribution offset implicit modeling feature map not only integrates parameter drift temporal information and color space distribution information, but also highlights the color feature regions that are significantly affected by parameter drift through weight allocation, suppresses irrelevant feature interference, and provides more targeted feature support for accurately predicting color difference and generating correction maps, thereby improving the model's ability to capture and model color deviation caused by parameter drift.

[0073] More specifically, the feature modeling secondary subunit 1432 is used to input the color space distribution shift implicit modeling feature map into the decoder of the pre-trained lightweight machine learning model to obtain the color difference correction map. That is, using the decoder of the pre-trained lightweight machine learning model, the color space distribution shift implicit modeling feature map, which integrates parameter drift temporal features and color space distribution features, is converted into a color difference correction map that can be directly used to correct the basic color prediction results. Specifically, the color space distribution shift implicit modeling feature map has formed an abstract feature representation containing the influence of parameter drift on each dimension of the color space through cross-modal dynamic interactive analysis of the encoder. However, this representation needs to be restored to a spatial distribution matrix of the same dimension as the basic color prediction map through structural mapping of the decoder, so as to provide an operable pixel-by-pixel correction signal for subsequent color appearance correction. On the one hand, the decoder uses the mapping law between parameter drift and color deviation learned by the pre-trained model to enable the color difference correction map to accurately reflect the color shift distribution under different parameter drift modes, such as the overall brightness shift caused by temperature drift and the local hue deviation caused by flow fluctuation. On the other hand, the lightweight design ensures the real-time nature of the decoding process, enabling the calibration map to be dynamically adjusted as the real-time process parameters are updated, thus achieving online compensation for color deviations.

[0074] In a specific example of this application, the training process of the lightweight machine learning model is as follows: First, a multi-source heterogeneous training dataset is constructed. By combining historical production data collection with physical simulation, stepwise drifts, periodic fluctuations, or random disturbances of parameters such as temperature and flow rate are introduced on the basis of rated process parameters. High-precision CFD simulation and Kubelka-Monk optical model are used to generate basic color prediction results. At the same time, the real color output is obtained through actual color mixing experiments or spectral measurements. The CIE2000 color difference ΔE is calculated to form a parameter drift vector - basic color prediction map - actual color difference. Figure 3 The tuple sample data covers typical operating conditions such as equipment aging and environmental changes to ensure that the drift range is consistent with industrial reality.

[0075] Secondly, feature engineering and data preprocessing are performed. The parameter drift vector is arranged into a time series according to the time dimension. Time series statistics such as mean, variance, and frequency domain features are extracted to generate time series feature encoding vectors. Spatial distribution features such as color gradient and local variance are extracted from the basic color prediction map to construct a feature map. The actual color difference map is converted into a ΔE correction map of the same dimension as a supervision label. At the same time, data augmentation techniques such as adding Gaussian noise and random flipping and translation are used to expand the sample diversity.

[0076] The model architecture adopts a lightweight "encoder-decoder" design. The encoder includes a full-dimensional dynamic perception cross-modal module. First, the global mean pooling of the basic color feature map is used to obtain the global encoding vector. Then, the parameter drift temporal encoding vector and the global encoding vector are mapped through the dimension of the projection matrix. The dynamic weight perception matrix is ​​calculated by the Softmax function and multiplied by the trainable bias vector to generate the full-dimensional dynamic perception encoding matrix. After Sigmoid activation, the basic color feature map is weighted and modulated to achieve cross-modal fusion. The decoder restores the fused feature map to the input size and outputs a ΔE correction map through deconvolution.

[0077] The training process employs end-to-end supervised learning, using the mean square error between the predicted color difference map and the actual ΔE correction map as the loss function. It combines the CIELAB color space to perceive color difference weights, uses the Adam optimizer to adaptively adjust the learning rate, and employs early stopping to avoid overfitting. At the same time, it introduces multi-order statistical invariant constraints, calculates the multi-order statistical invariants of the projection matrix and the dynamic weight matrix, and constructs a tensor fiber bundle constraint matrix to perform intrinsic convergence correction on the dynamic weight perception matrix, ensuring the dimensionality consistency of cross-modal feature mapping.

[0078] Finally, the model is validated using industrial field data. The consistency between the predicted color difference map and the actual measurement results is compared, and the model inference speed is calculated to ensure that the model balances accuracy and real-time performance. Ultimately, a lightweight model with controllable parameters is embedded in the digital twin platform to achieve real-time correction.

[0079] Specifically, the color appearance correction module 150 is used to apply the color difference correction map to the basic color prediction result to obtain the final color appearance prediction map. That is, by merging the color difference correction map with the basic color prediction result, color errors caused by real-time process parameter drift are dynamically compensated, enabling the generated final color appearance prediction map to accurately reflect the true color effect of material mixing in actual production. This provides a reliable visual reference for process design and optimization, facilitates rapid color adjustment in multi-variety, small-batch production scenarios, significantly improves the design efficiency of color mixing processes, reduces material and time costs, and meets the needs of intelligent manufacturing for precise color control.

[0080] In a specific example of this application, the color appearance correction module 150 is implemented as follows: First, the color difference correction image is decomposed into channels to extract the deviation values ​​of brightness, hue, and saturation to form a correction matrix. Then, a weighted superposition algorithm is used to multiply the correction matrix with the basic color prediction result by channel, adjusting the color value of each pixel. The algorithm supports user-defined correction weights to adapt to the color sensitivity of different materials. After correction, the color result is converted into a visual image, outputting color difference evaluation indicators for process engineers to compare with design targets. It also supports interactive parameter adjustment and real-time preview of optimization effects. Through pixel-level correction, it dynamically compensates for color deviations caused by parameter drift, making the simulation results closer to the actual production effect.

[0081] In summary, the multi-material color mixing process digital twin simulation platform 100 based on the embodiments of this application is explained. By constructing a multi-material color mixing process digital twin simulation platform, it transforms geometric models, material properties, and rated process parameters into basic physical simulation data, and acquires real-time process parameters through an IoT interface to determine the parameter drift vector. Simultaneously, a lightweight machine learning model is used to process the parameter drift features and color distribution features, generating a color difference correction map to dynamically correct the color prediction results. In this way, color deviations caused by parameter drift can be dynamically corrected, achieving accurate simulation of the dynamic evolution of the multi-material color mixing process. This improves the efficiency of color mixing process design and optimization, reduces time and material costs, and meets the rapid changeover requirements in multi-variety, small-batch production scenarios.

[0082] As described above, the multi-material color mixing process digital twin simulation platform 100 according to the embodiments of this application can be implemented in various wireless terminals, such as servers with combat plan generation algorithms based on situation assessment. In one possible implementation, the multi-material color mixing process digital twin simulation platform 100 according to the embodiments of this application can be integrated into the wireless terminal as a software module and / or hardware module. For example, the multi-material color mixing process digital twin simulation platform 100 can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the multi-material color mixing process digital twin simulation platform 100 can also be one of many hardware modules of the wireless terminal.

[0083] Alternatively, in another example, the multi-material color mixing process digital twin simulation platform 100 and the wireless terminal can also be separate devices, and the multi-material color mixing process digital twin simulation platform 100 can be connected to the wireless terminal via wired and / or wireless networks, and transmit interactive information in accordance with an agreed data format.

[0084] Various implementations of this disclosure have been described above. The foregoing description is exemplary and not exhaustive. Furthermore, it is not limited to the disclosed implementations, and many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described implementations.

Claims

1. A digital twin simulation platform for multi-material color mixing processes, characterized in that, include: The input data acquisition module is used to acquire the geometric model, material identification ID, and rated process parameters input by the user. The basic physics simulation module is used to perform rapid basic physics simulation on the geometric model based on the material identification ID and the rated process parameters to obtain basic microstructure prediction results and basic color prediction results; The real-time process parameter acquisition module is used to acquire real-time process parameters through an IoT interface; The deviation prediction module is used to calculate the parameter drift vector between the real-time process parameters and the rated process parameters, and to perform deviation prediction on the basic color prediction results based on the parameter drift vector to obtain a color difference correction map. The color appearance correction module is used to apply the color difference correction map to the basic color prediction result to obtain the final color appearance prediction map. The deviation prediction module includes: The color distribution feature extraction unit is used to extract color distribution features from the basic color prediction results to obtain a basic color space distribution feature map. A drift time series feature extraction unit is used to extract a parameter drift time series feature encoding vector from the parameter drift vector; The color difference correction unit is used to input the parameter drift temporal feature encoding vector and the basic color space distribution feature map into a pre-trained lightweight machine learning model to obtain the color difference correction map.

2. The digital twin simulation platform for multi-material color mixing process according to claim 1, characterized in that, The material identification ID includes material identification IDs for two or more materials.

3. The digital twin simulation platform for multi-material color mixing process according to claim 2, characterized in that, The basic physics simulation module includes: A material property loading unit is used to load the physical property data of the selected material from the material database based on the material identifier ID; The CFD simulation unit is used to obtain the prediction results of the basic microstructure based on the physical property data and the rated process parameters and through coarse-grid CFD simulation. The basic color prediction unit is used to process the basic microstructure prediction results based on the Kubelka-Monk optical model to obtain the basic color prediction results.

4. The digital twin simulation platform for multi-material color mixing process according to claim 3, characterized in that, The physical property data includes rheological parameters and optical parameters, and the rated process parameters include temperature and flow rate ratio.

5. The digital twin simulation platform for multi-material color mixing process according to claim 1, characterized in that, The deviation prediction module is used to: calculate the deviation between the real-time process parameters and the rated process parameters at each predetermined time point, and arrange the deviations according to the time dimension to obtain the parameter drift vector.

6. The digital twin simulation platform for multi-material color mixing process according to claim 5, characterized in that, The color difference correction unit includes: The feature modeling secondary subunit is used to input the parameter drift temporal feature encoding vector and the basic color space distribution feature map into the encoder of the pre-trained lightweight machine learning model to obtain the color space distribution offset latent modeling feature map. The feature decoding secondary subunit is used to input the color space distribution offset latent modeling feature map into the decoder of the pre-trained lightweight machine learning model to obtain the color difference correction map.

7. The digital twin simulation platform for multi-material color mixing process according to claim 6, characterized in that, The feature modeling secondary subunit includes: A three-level subunit for global mean pooling is used to perform global mean pooling on the basic color space distribution feature map to obtain the basic color space distribution global pooling encoding vector. The dynamic interactive analysis three-level sub-unit is used to perform global dynamic interactive analysis on the parameter drift time-series feature encoding vector and the basic color space distribution global pooling encoding vector to obtain the full-dimensional dynamic perception encoding matrix of the color space distribution. The nonlinear activation third-level sub-unit is used to apply the nonlinear activation of the full-dimensional dynamic perception coding matrix of the color space distribution to the basic color space distribution feature map to obtain the color space distribution offset implicit modeling feature map.

8. The digital twin simulation platform for multi-material color mixing process according to claim 7, characterized in that, The nonlinear activation three-level subunit is used for: The color space distribution full-dimensional dynamic sensing coding matrix is ​​non-linearly activated to obtain the activated color space distribution full-dimensional dynamic sensing coding matrix. The activated color space distribution full-dimensional dynamic perception encoding matrix and each basic color space distribution feature matrix along the channel dimension in the basic color space distribution feature map are multiplied by position to apply the activated color space distribution full-dimensional dynamic perception encoding matrix to the basic color space distribution feature map to obtain the color space distribution offset implicit modeling feature map.

Images