An intelligent paint matching method, device, equipment and medium based on artificial intelligence

Abstract

This application discloses an intelligent paint mixing method, apparatus, equipment, and medium based on artificial intelligence, relating to the field of artificial intelligence technology. The method includes: screening raw material physical feature vectors from a hybrid architecture database; splicing and fusing features from paint mixing demand data, processed environmental perception data, and raw material physical feature vectors to construct a target feature vector; inputting the target feature vector into a machine learning model to output a predicted paint film performance value; generating an orthogonal experimental table based on the predicted paint film performance value and user-defined optimization objectives; predicting the performance of candidate paint mixing ratios in the orthogonal experimental table; performing range analysis on the performance prediction results; using the range analysis results to perform multi-objective optimization of the candidate paint mixing ratios to obtain the target paint mixing ratio; parsing the target paint mixing ratio to obtain an action instruction sequence; executing the action instruction sequence to complete the paint mixing operation; achieving optimal paint mixing ratio under complex constraints, improving the accuracy and efficiency of paint mixing.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an intelligent paint mixing method, apparatus, equipment and medium based on artificial intelligence. Background Technology

[0002] Currently, existing automatic paint mixing solutions use a high-precision portable spectrometer to scan the target area and obtain digital spectral data of color reflectance. The system software compares the measured spectral data with a large cloud or local color matching database, retrieving several pre-stored ratios with the smallest color difference. Some advanced systems can provide simple adjustment suggestions based on the color difference data, such as "add 0.1% magenta paste." The operator confirms the fine-tuned ratio, and the system drives the automated dispenser to perform metering and mixing. After the mixed paint is dried on the slab, the color difference is measured again. If it is not up to standard, the operator needs to manually modify the mixing parameters based on the new data and personal experience, and then re-execute the metering and mixing. The disadvantages are as follows: weak ability to respond to new demands: when faced with colors outside the database or new performance combinations, the system fails, reverting to an inefficient manual trial-and-error mode; lack of predictive and proactive optimization capabilities: the system can only recall historical ratios and cannot predict the performance of unknown ratios, let alone proactively seek optimization within a large ratio space; lack of adaptive and learning capabilities: the system cannot self-optimize based on environmental changes or historical paint mixing results, resulting in performance stagnation.

[0003] As can be seen from the above, how to adapt to different paint mixing needs, achieve optimal paint mixing ratios under complex constraints, improve the accuracy and efficiency of paint mixing, and enhance resource utilization are problems that need to be solved in this field. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide an intelligent paint mixing method, apparatus, device, and medium based on artificial intelligence, which can adapt to different paint mixing needs, improve the accuracy and efficiency of paint mixing, enhance resource utilization, and achieve optimal paint mixing ratios under complex constraints. The specific solution is as follows: In a first aspect, this application discloses an intelligent paint mixing method based on artificial intelligence, comprising: Acquire user paint mixing requirements data and real-time environmental perception data, process the environmental perception data, and obtain the processed environmental perception data. The raw material materialization feature vectors for paint mixing are selected from the hybrid architecture database. The paint mixing requirement data, the processed environmental perception data, and the raw material materialization feature vectors are then spliced ​​and fused to construct the target feature vector. The target feature vector is input into a pre-trained machine learning model to output a predicted value of the paint film performance corresponding to the target feature vector; the predicted value of the paint film performance includes at least one of color difference, gloss, drying time, and salt spray resistance level, as well as a cost estimate. Based on the predicted paint film performance and the user-defined optimization objectives, an orthogonal experimental table is generated. The performance prediction model is used to predict the performance of each candidate paint mixing ratio in the orthogonal experimental table to obtain the mixing ratio performance prediction results. The mixing ratio performance prediction results are then subjected to range analysis. The range analysis results are used to perform multi-objective optimization on each candidate paint mixing ratio to obtain the target paint mixing ratio. The target paint mixing ratio is analyzed to obtain an action instruction sequence, and the action instruction sequence is executed to complete the paint mixing operation.

[0005] Optionally, the step of acquiring the user's paint mixing requirements data and real-time environmental perception data, and processing the environmental perception data to obtain processed environmental perception data, includes: The user's paint mixing requirements are obtained from the graphical user interface; these requirements include color, gloss, drying time, and salt spray resistance. Real-time environmental sensing data is acquired from sensors using a serial peripheral interface or integrated circuit bus protocol; the sensors include digital temperature and humidity sensors and atmospheric pressure sensors. The environmental perception data is filtered, denoised, and have data stamps added to obtain the processed environmental perception data.

[0006] Optionally, the step of filtering the physical feature vectors of raw materials for paint mixing from the hybrid architecture database includes: A hybrid architecture database is built based on relational databases, time-series databases, and object repositories; The physical feature vectors of raw materials for paint mixing are filtered out from the relational database.

[0007] Optionally, before inputting the target feature vector into the pre-trained machine learning model, the method further includes: Historical paint mixing datasets were selected from the time-series database; The machine learning model was trained, tested, and validated using the five-fold cross-validation method and based on the historical paint mixing dataset to obtain a pre-trained machine learning model; the machine learning model is a random forest regression and classification ensemble model.

[0008] Optionally, the step of performing feature splicing and fusion on the paint mixing requirement data, the processed environmental perception data, and the materialized feature vector of the raw materials to construct a target feature vector includes: Color features, environmental features, and material physical features are extracted from the paint mixing demand data, the processed environmental perception data, and the raw material physical feature vector. Based on the attention weight allocation mechanism, the color features, the environmental features, and the material physical features are spliced ​​and fused to construct the target feature vector.

[0009] Optionally, based on the predicted paint film performance value and the user-defined optimization objective, an orthogonal experimental table is generated. A performance prediction model is used to predict the performance of each candidate paint mixing ratio in the orthogonal experimental table, obtaining the ratio performance prediction results. Range analysis is performed on the ratio performance prediction results, and multi-objective optimization is conducted on each candidate paint mixing ratio using the range analysis results, including: The predicted values ​​of the paint film performance and the optimization targets set by the user are input into the decision-maker, so that the decision-maker can use a preset sampling algorithm to generate an orthogonal experimental table based on the predicted values ​​of the paint film performance and the optimization targets. Convert each candidate paint mixing ratio in the orthogonal experimental table into a mixing ratio feature vector; The performance prediction model is used to predict the performance of the proportion feature vector to obtain the proportion performance prediction result; A range analysis is performed on the predicted performance of the proportions to obtain a range analysis result including the range values; A non-dominated sorting genetic algorithm was used, and multi-objective optimization was performed on each candidate paint formulation based on the range analysis results.

[0010] Optionally, after completing the paint mixing operation, the method further includes: After the paint film dries, the actual performance data of the paint film is collected using an online viscometer, an automatic gloss meter, and an image recognition system. Align the actual performance data of the paint film, the target paint mixing ratio, and the predicted performance value of the paint film corresponding to the target paint mixing ratio to generate a triplet record; The triplet records are stored in the time-series database of the hybrid architecture database.

[0011] Secondly, this application discloses an intelligent paint mixing device based on artificial intelligence, comprising: The processing module is used to acquire user paint mixing requirements data and real-time environmental perception data, process the environmental perception data, and obtain the processed environmental perception data. The feature splicing and fusion module is used to filter out the materialized feature vectors of raw materials for paint mixing from the hybrid architecture database, and to splice and fuse the paint mixing requirement data, the processed environmental perception data, and the materialized feature vectors of raw materials to construct a target feature vector. The paint film performance prediction module is used to input the target feature vector into a pre-trained machine learning model to output a paint film performance prediction value corresponding to the target feature vector; the paint film performance prediction value includes at least one of color difference, gloss, drying time, and salt spray resistance level, as well as a cost estimate. The multi-objective optimization module is used to generate an orthogonal experimental table based on the predicted value of the paint film performance and the optimization objectives set by the user, use a performance prediction model to predict the performance of each candidate paint ratio in the orthogonal experimental table, obtain the ratio performance prediction result, perform range analysis on the ratio performance prediction result, and use the range analysis result to perform multi-objective optimization on each candidate paint ratio to obtain the target paint ratio. The parsing module is used to parse the target paint mixing ratio, obtain an action instruction sequence, and execute the action instruction sequence to complete the paint mixing operation.

[0012] Thirdly, this application discloses an electronic device, including: Memory, used to store computer programs; A processor is used to execute the computer program to implement the aforementioned artificial intelligence-based intelligent paint mixing method.

[0013] Fourthly, this application discloses a computer storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the aforementioned disclosed intelligent paint mixing method based on artificial intelligence.

[0014] As can be seen, this application provides an intelligent paint mixing method based on artificial intelligence, including acquiring user paint mixing demand data and real-time environmental perception data; processing the environmental perception data to obtain processed environmental perception data, eliminating random errors and dimensional differences in the acquisition process, and ensuring the authenticity and consistency of the data; selecting the physical feature vectors of raw materials for paint mixing from a hybrid architecture database; performing feature splicing and fusion on the paint mixing demand data, processed environmental perception data, and physical feature vectors of raw materials to construct a target feature vector, enabling the machine learning model to make performance predictions based on comprehensive information, improving the accuracy and generalization ability of the prediction results; and inputting the target feature vector into a pre-trained machine learning model to output a paint film performance prediction corresponding to the target feature vector. The system provides a comprehensive and rapid prediction of paint film performance, including at least one of color difference, gloss, drying time, and salt spray resistance rating, as well as a cost estimate. Based on the predicted paint film performance and user-defined optimization objectives, an orthogonal experimental table is generated. A performance prediction model is then used to predict the performance of each candidate paint mixing ratio in the orthogonal experimental table, yielding the ratio performance prediction results. Range analysis is performed on the ratio performance prediction results, and the results are used to perform multi-objective optimization on each candidate paint mixing ratio to obtain the target paint mixing ratio, achieving optimal paint mixing ratio under complex constraints. The target paint mixing ratio is then analyzed to obtain an action instruction sequence. This sequence is executed to complete the paint mixing operation, adapting to different paint mixing needs, improving the accuracy and efficiency of paint mixing, and enhancing resource utilization. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0016] Figure 1 This is a flowchart of an intelligent paint mixing method based on artificial intelligence disclosed in this application; Figure 2 This is an overall architecture diagram of an intelligent paint mixing system disclosed in this application; Figure 3 This is a flowchart of the workflow of an AI prediction engine disclosed in this application; Figure 4 This application discloses a specific flowchart for implementing intelligent paint mixing; Figure 5 This is a schematic diagram of the structure of an intelligent paint mixing device based on artificial intelligence disclosed in this application; Figure 6 This application provides a structural diagram of an electronic device. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Currently, existing automatic paint mixing solutions use a high-precision portable spectrometer to scan the target area and obtain digital spectral data of color reflectance. The system software compares the measured spectral data with a large cloud or local color matching database, retrieving several pre-stored ratios with the smallest color difference. Some advanced systems can provide simple adjustment suggestions based on the color difference data, such as "add 0.1% magenta paste." The operator confirms the fine-tuned ratio, and the system drives the automated dispenser to perform metering and mixing. After the mixed paint is dried on the slab, the color difference is measured again. If it is not up to standard, the operator needs to manually modify the mixing parameters based on the new data and personal experience, and then re-execute the metering and mixing. The disadvantages are as follows: weak ability to respond to new demands: when faced with colors outside the database or new performance combinations, the system fails, reverting to an inefficient manual trial-and-error mode; lack of predictive and proactive optimization capabilities: the system can only recall historical ratios and cannot predict the performance of unknown ratios, let alone proactively seek optimization within a large ratio space; lack of adaptive and learning capabilities: the system cannot self-optimize based on environmental changes or historical paint mixing results, resulting in performance stagnation. As can be seen from the above, how to adapt to different paint mixing needs, achieve optimal paint mixing ratios under complex constraints, improve the accuracy and efficiency of paint mixing, and enhance resource utilization are problems that need to be solved in this field.

[0019] The technical differences between this application and the prior art are shown in Table 1: Table 1 Technical Differences

[0020] See Figure 1 As shown in the figure, this invention discloses an intelligent paint mixing method based on artificial intelligence, which may specifically include: Step S11: Obtain the user's paint mixing requirements data and real-time environmental perception data, process the environmental perception data, and obtain the processed environmental perception data.

[0021] In this embodiment, the user's paint mixing requirements data are obtained from the graphical user interface; the paint mixing requirements data includes color, gloss, drying time requirements, and salt spray resistance requirements; real-time environmental perception data is obtained from sensors using a serial peripheral interface or integrated circuit bus protocol; the sensors include digital temperature and humidity sensors and atmospheric pressure sensors; the environmental perception data is filtered and noise reduced and data stamps are added to obtain the processed environmental perception data.

[0022] This application proposes an intelligent paint mixing system, the overall architecture of which is as follows: Figure 2 As shown, this system is not a single device, but a complete technological system integrating perception, decision-making, execution, and learning. It consists of six major functional modules: user interaction module, environmental sensing module, data and knowledge base module, AI (Artificial Intelligence) prediction engine, automated execution module, and feedback and learning module.

[0023] This step is implemented through a user interaction module and an environmental sensing module. Regarding the user interaction module: Function: Transforms users' unstructured paint mixing requirements into paint mixing requirement data that the system can process; Hardware: Includes industrial touchscreen, keyboard, barcode scanner, and optional high-precision spectrophotometer; Software: Provides a graphical user interface based on Web (World Wide Web) or client-side technologies such as HTML5 (HyperText Markup Language 5) and QT (Qt Toolkit). Core software functions include: Structured form for requirements: Provides drop-down menus, sliders, input boxes, and other controls, allowing users to select or input indicators such as "color code," "gloss (60° GU value)," "drying time requirement," and "salt spray resistance requirement." Visual color capture: If connected to a spectrophotometer, it acquires L, a, b color space data via an application programming interface (API). If using images, it calls the built-in open-source computer vision library for color region recognition and average color calculation.

[0024] Data encapsulation and transmission: The collected structured data (such as "color_lab": [85, 2, -5], "gloss": 90, "dry_time": "<30min") is encapsulated into a JSON (JavaScript Object Notation) format data packet and sent to the AI ​​prediction engine through the application programming interface.

[0025] For the environmental sensing module: Function: Real-time sensing of the paint mixing workshop environment to provide parameters for dynamic mixing ratio optimization; Hardware: Digital temperature and humidity sensors and atmospheric pressure sensors are deployed in the paint mixing room. The sensors communicate with the data acquisition terminal using a serial peripheral interface or integrated circuit bus protocol.

[0026] Software: A lightweight agent runs on a data acquisition terminal (such as a Raspberry Pi) to poll environmental perception data from sensors at a fixed frequency (e.g., once per second) and performs filtering and noise reduction processing. The filtered and noise-reduced environmental perception data is timestamped and pushed in real time to a designated topic in the AI ​​prediction engine via a message queue telemetry transmission protocol.

[0027] Step S12: Select the materialized feature vectors of paint mixing raw materials from the hybrid architecture database, and perform feature splicing and fusion on the paint mixing requirement data, the processed environmental perception data, and the materialized feature vectors of raw materials to construct the target feature vector.

[0028] In this embodiment, a hybrid architecture database is constructed based on a relational database, a time-series database, and an object repository. The physical feature vectors of raw materials for paint mixing are selected from the relational database. Color features, environmental features, and material physical features are extracted from the paint mixing demand data, the processed environmental perception data, and the physical feature vectors of raw materials. Based on an attention weight allocation mechanism, the color features, environmental features, and material physical features are spliced ​​and fused to construct a target feature vector.

[0029] This step is implemented through the data and knowledge base module and the AI ​​prediction engine. Regarding the data and knowledge base module: Features: Securely store and efficiently manage all relevant data, and provide high-speed data services for the AI ​​prediction engine; Storage architecture: A hybrid architecture combining relational database, time-series database, and object storage is adopted; Relational database: stores structured master data of proportions and a library of physical and chemical properties of raw materials; the master data of proportions includes, but is not limited to, proportion ID (Identification, serial number), name of each component, and theoretical addition amount; the library of physical and chemical properties of raw materials includes, but is not limited to, the coloring intensity of color pastes and the hydroxyl value / glass transition temperature of resins.

[0030] Time-series database: Specifically designed for efficient storage and management of massive amounts of timestamped process data, such as environmental data, execution parameters, and feedback results for each paint mixing session.

[0031] Object storage: Stores unstructured data, such as color spectrum curve files and images of paint film performance test reports.

[0032] Knowledge Management and Services: Deploy database management systems and data warehouse technologies to provide efficient querying. Provide data access interfaces for the AI ​​prediction engine through a representational state transfer application programming interface, enabling secure and controllable data retrieval.

[0033] For AI prediction engines: Function: Receives input, predicts performance through machine learning models, and runs optimization algorithms to obtain the optimal ratio.

[0034] Hardware: Deployed on local servers or in the cloud, with a graphics processor as the core computing unit to accelerate model training and prediction.

[0035] Software Architecture and Workflow: The workflow of an AI prediction engine is as follows Figure 3 As shown, it operates in a pipeline manner, consisting of three stages: feature engineering, performance prediction, and optimization decision-making.

[0036] This step is the feature engineering stage: The system performs feature concatenation and fusion on user paint mixing requirements data, processed environmental perception data, and raw material materialization feature vectors retrieved from a hybrid architecture database to construct a complete target feature vector. Specifically: The AI ​​prediction engine receives paint mixing demand data from users and the environment, as well as processed environmental perception data. It also retrieves relevant raw material materialization feature vectors from the hybrid architecture database. Through feature fusion processing, the three are combined into a high-dimensional complete feature vector that can comprehensively describe the current paint mixing task, preparing for subsequent predictions.

[0037] Inputs: Paint mixing requirement data from the user interaction module (including but not limited to color L, a, b values, gloss, and drying time), environmental perception data processed by the environmental sensing module (including but not limited to temperature and humidity), and raw material physical feature vectors retrieved from the hybrid architecture database.

[0038] Processing: The AI ​​prediction engine extracts the corresponding physicochemical feature vectors from the hybrid architecture database based on the resin and pigment types in the requirements. For example, it converts "Resin A" into [hydroxyl value: 120, Tg: 45, solid content: 60%]. The paint mixing requirements data, the processed environmental perception data, and the raw material physicochemical feature vectors are then concatenated to form a complete, high-dimensional target feature vector. This is a crucial data preprocessing step for the model to work accurately.

[0039] Feature extraction and fusion: Color characteristics: The L, a, b values ​​output by the spectrophotometer are used directly, or calculated using the RGB (Red Green Blue) to Lab formula.

[0040] Environmental characteristics: Temperature and humidity were standardized and their deviations from standard conditions (23℃, 50%RH) were calculated.

[0041] Material physicochemical characteristics: Hydroxyl value, glass transition temperature, solid content of resin, coloring intensity, particle size distribution, weather resistance grade, etc. of pigment are extracted from the hybrid architecture database to constitute the material physicochemical characteristics.

[0042] Feature fusion method: The concatenation + attention weight allocation mechanism is adopted. First, the three types of features are concatenated into a high-dimensional vector. Then, a small neural network (attention layer) is used to calculate the weight of each feature dimension to the current task, and weighted fusion is performed to form the final target feature vector F.

[0043] Step S13: Input the target feature vector into a pre-trained machine learning model to output a predicted value of the paint film performance corresponding to the target feature vector; the predicted value of the paint film performance includes at least one of color difference, gloss, drying time, salt spray resistance level, and cost estimate.

[0044] In this embodiment, before inputting the target feature vector into the pre-trained machine learning model, the method further includes: selecting a historical paint mixing dataset from the time series database; using five-fold cross-validation and training, testing, and validating the machine learning model based on the historical paint mixing dataset to obtain a pre-trained machine learning model; the machine learning model is a random forest regression and classification ensemble model.

[0045] This step is the performance prediction stage in the AI ​​prediction engine: First, the target feature vector F is input into the machine learning model. The machine learning model quickly calculates multiple performance indicators (such as color difference, gloss, durability, etc.) corresponding to the ratio in the virtual space, and outputs performance prediction results including paint film performance prediction values. The machine learning model includes, but is not limited to, random forest regression and classification ensemble models.

[0046] Model: The core is a pre-trained random forest regression and classification ensemble model. Random forests consist of multiple decision trees, which can effectively handle high-dimensional nonlinear relationships and have strong resistance to overfitting.

[0047] Operation: The target feature vector is input into the model. Multiple decision trees within the model make judgments, and finally, through a voting or averaging mechanism, the model outputs predicted values ​​for multiple properties of the final paint film, i.e., paint film performance prediction values. These multiple properties include, but are not limited to, predicted color difference ΔE, predicted gloss, and predicted weather resistance rating. These predictions are made in a virtual space without the need for actual paint mixing.

[0048] Model structure: A random forest regression model is used, consisting of 200 decision trees, each with a maximum depth of 15. The model input is the target feature vector F, and the output is the predicted values ​​of the paint film performance, including: color difference ΔE, gloss, drying time, salt spray resistance level, and cost estimate.

[0049] The salt spray resistance rating is based on a scale of 1 to 5 (5 being the best), and the estimated cost is automatically calculated based on the unit price of each component and the predicted usage.

[0050] The training process of the model is as follows: Training data: Historical paint mixing dataset from a time-series database, totaling approximately 5000 entries. Each entry contains a complete target feature vector F and a measured performance label.

[0051] Data labeling: Performance labels are obtained through standard laboratory testing (e.g., ΔE measured by a colorimeter, GU value measured by a gloss meter).

[0052] Training period: Five-fold cross-validation was used for validation, and the training time was about 2 hours, including GPU (Graphics Processing Unit) acceleration.

[0053] Prediction process: The target feature vector F is input into the pre-trained random forest model. Each decision tree outputs a prediction value independently. Finally, the average of all trees is taken as the final prediction result.

[0054] Step S14: Based on the predicted paint film performance value and the optimization target set by the user, generate an orthogonal experimental table, use the performance prediction model to predict the performance of each paint mixing candidate ratio in the orthogonal experimental table, obtain the ratio performance prediction result, perform range analysis on the ratio performance prediction result, and use the range analysis result to perform multi-objective optimization on each paint mixing candidate ratio to obtain the target paint mixing ratio.

[0055] In this embodiment, the predicted paint film performance value and the user-defined optimization objective are input into the decision-maker, so that the decision-maker can generate an orthogonal experimental table based on the predicted paint film performance value and the optimization objective using a preset sampling algorithm; each paint mixing candidate ratio in the orthogonal experimental table is converted into a ratio feature vector; the performance prediction model is used to predict the performance of the ratio feature vector to obtain the ratio performance prediction result; range analysis is performed on the ratio performance prediction result to obtain the range analysis result including the range value; a non-dominated sorting genetic algorithm is used to perform multi-objective optimization on each paint mixing candidate ratio based on the range analysis result to obtain the target paint mixing ratio.

[0056] This step is the optimization decision-making stage in the AI ​​prediction engine: By combining the predicted values ​​of paint film performance with the user-defined optimization objectives, a virtual orthogonal experimental table is generated, and optimization algorithms such as "range analysis" are executed to quickly locate and output one or more optimal target paint mixing ratios in a vast virtual mixing space.

[0057] The predicted coating performance values, along with the user-defined optimization objectives, are fed into the decision-maker. The decision-maker employs efficient optimization strategies (such as combining virtual orthogonal experimental tables for simulation calculations and range analysis) to automatically find the optimal ratio within a vast mixing space, ultimately outputting the recommended optimal ratio and the ratio performance prediction results for each of its various properties.

[0058] Input: Predicted film performance values, user-defined optimization objectives (such as "lowest cost" and "fastest drying") and constraints (such as "ΔE must be <1.5" and "gloss >85GU").

[0059] The "prediction model-based sequence optimization" method is adopted. First, a batch of representative paint mixing candidate ratios are intelligently generated in the huge ratio combination space using a preset sampling algorithm to generate an orthogonal experimental table. The preset sampling algorithm includes, but is not limited to, the Latin hypercube sampling algorithm. Then, each candidate paint mixing ratio in the orthogonal experimental table is converted into a ratio feature vector; Then, the performance prediction model is called to predict the performance of the proportion feature vectors in batches, and the proportion performance prediction results are obtained. Finally, a multi-objective optimization algorithm is used to screen and iteratively evolve the candidate paint formulations. This algorithm simulates the biological evolution process, and through operations such as selection, crossover, and mutation, it can approximate the Pareto optimal solution set that satisfies all constraints within a few generations, thus obtaining the target paint formulation.

[0060] To address small sample scenarios, the engine incorporates an "orthogonal design" strategy, automatically generating orthogonal experimental tables. It selects only a few of the most representative virtual ratios for prediction and then uses range analysis to quickly identify key influencing factors and optimization directions, greatly improving optimization efficiency.

[0061] Virtual orthogonal experiments and dynamic optimization: Dynamic generation of orthogonal experimental tables: The system automatically generates an L9(34) or L18 orthogonal experimental table based on the number of components involved in the current task (e.g., 5 base paints) and the number of levels to be investigated (e.g., 3 addition ratio levels for each paint). This table is not used for physical experiments, but rather serves as a virtual proportioning sampling point.

[0062] Batch virtual prediction: The candidate paint mixing schemes corresponding to each row of the orthogonal experimental table are transformed into a mixing feature vector, and then input into the performance prediction model to obtain the mixing performance prediction results.

[0063] Range analysis optimization: Perform range analysis on the prediction results to calculate the degree of influence of each component (factor) on each performance index (such as ΔE, cost) (range R value), thereby identifying key influencing factors and their optimal level combination.

[0064] Multi-objective collaborative decision-making: The system employs a non-dominated sorting genetic algorithm with an elitist strategy to optimize each candidate paint formulation based on range analysis results. Using the aforementioned optimal combination as the initial population, and with ΔE, cost, and drying time as optimization objectives, the system iterates and evolves for approximately 100 generations in a virtual space, ultimately outputting a set of Pareto optimal solutions. Users can then select the final target paint formulation from this set according to their preferences.

[0065] Environmentally Adaptive Weight Adjustment: The target weights in the optimization algorithm can be dynamically adjusted based on environmental data. For example, when the ambient humidity is >70%, the system automatically increases the weight of the "drying time" target to prioritize ensuring workability.

[0066] Output: One or more optimal target paint mixing ratios, along with predicted values ​​for their various performance characteristics, cost estimates, and confidence intervals.

[0067] In summary, the internal workflow of an AI prediction engine consists of three stages: Phase 1: Feature Engineering Input: Receive paint mixing requirement data (such as color Lab value, gloss requirement) from the user interaction module, environmental perception data (temperature, humidity) from the environmental sensing module, and raw material physical and chemical feature vectors (such as resin hydroxyl value, glass transition temperature, color intensity of pigment) retrieved from the hybrid architecture database. Processing: The environmental perception data is standardized to eliminate the influence of dimensions. Then, all features are concatenated into a high-dimensional vector. To further enhance feature representation, an attention mechanism is introduced. A small neural network calculates the weights of each feature dimension and performs weighted fusion to generate the final high-dimensional target feature vector F. This vector comprehensively encodes the integrated information of the current paint mixing task.

[0068] Phase Two: Performance Prediction

[0069] Input: The target feature vector F output from the feature engineering stage.

[0070] Processing: The target feature vector F is input into a pre-trained machine learning model (such as a random forest model containing 200 decision trees). This model has been trained on historical "ratio-performance" data and is able to learn the complex nonlinear relationships between materials, environment, requirements, and the final paint film performance. The model performs forward computation on the input target feature vector F, with each decision tree outputting a predicted value. Finally, through ensemble (averaging) methods, it outputs multiple predicted paint film performance values ​​corresponding to the virtual ratio, including color difference ΔE, gloss, drying time, weather resistance rating, cost estimation, etc.

[0071] Phase 3: Optimizing Decisions

[0072] Input: Multiple predicted values ​​of coating performance output from the performance prediction stage, as well as user-defined optimization objectives (such as "lowest cost" and "fastest drying") and hard constraints (such as "ΔE<1.5" and "gloss>85GU").

[0073] Processing: Virtual Orthogonal Experiment: The system dynamically generates an orthogonal experimental table (e.g., L9) based on the types of components involved in the current task and the levels to be investigated. The orthogonal experimental table defines a small number of highly representative virtual combination ratios.

[0074] Batch Virtual Prediction: The candidate paint ratios corresponding to each row in the orthogonal experiment table are converted into ratio feature vectors. The performance prediction model is then called to perform batch predictions, and the performance prediction results of each ratio feature vector are obtained.

[0075] Range analysis guidance: Perform range analysis on the performance prediction results of the formulation to calculate the degree of influence of each component (factor) on each performance index (range R value), thereby quickly identifying key influencing factors and their general optimization direction.

[0076] Multi-objective evolutionary optimization: Guided by the aforementioned analysis results, a multi-objective optimization algorithm (such as NSGA-II) is used to perform iterative searches in a larger matching space. The algorithm uses a predictive model as a performance evaluator, simulates the selection, crossover, and mutation processes of biological evolution, and approaches the Pareto optimal front for multiple objectives such as color, performance, and cost while satisfying all constraints.

[0077] Environmental adaptive adjustment: During the optimization process, the algorithm will dynamically adjust the weight priority of each target based on real-time environmental data (such as high humidity).

[0078] Output: The final output includes one or more Pareto optimal target paint mixing ratios, as well as a detailed performance prediction report and cost estimate for each scheme.

[0079] Step S15: Analyze the target paint mixing ratio to obtain an action instruction sequence, and execute the action instruction sequence to complete the paint mixing operation.

[0080] This step is implemented through an automated execution module. Regarding the automated execution module: Function: It can convert one or more optimal target paint mixing ratios output by the AI ​​prediction engine, as well as their predicted performance values, cost estimates and confidence intervals, into a sequence of action instructions without loss and with high accuracy.

[0081] Hardware: Automatic paint mixing machine based on a programmable logic controller (PLC). Includes: a high-precision mass flow meter or a plunger pump driven by a servo motor (metering accuracy ±0.1%), an explosion-proof stirring motor, a PLC, a pneumatic valve assembly, and a cleaning circuit.

[0082] Mixing ratio analysis and instruction issuance: Receive the target paint mixing ratio from the AI ​​prediction engine, and the programmable logic controller will analyze it into a sequence of action instructions for each channel actuator (such as "Pump A is turned on, target flow rate XX grams").

[0083] Closed-loop feedback control: The flow meter feeds back the actual flow data to the programmable logic controller (PLC) in real time. The PLC compares the data with the target value and dynamically adjusts the pump speed or valve opening through a proportional-integral-derivative (PI-DE) control algorithm to ensure the accuracy of the flow rate.

[0084] Sequential control and safety interlock: Strictly follow the sequential logic of "solvent rinsing → metering addition → stirring and mixing → discharge", with sensor interlocks between each step to prevent misoperation.

[0085] In this embodiment, after the paint mixing operation is completed, the method further includes: after the paint film dries, collecting actual performance data of the paint film using an online viscometer, an automatic gloss meter, and an image recognition system; aligning the actual performance data of the paint film, the target paint mixing ratio, and the predicted performance value of the paint film corresponding to the target paint mixing ratio to generate a triplet record; and storing the triplet record in the time-series database of the hybrid architecture database.

[0086] This step is achieved through the feedback and learning module, whose function is to form a closed loop, enabling the system to continuously evolve.

[0087] Data Acquisition: Using online sensors, including but not limited to online viscometers, automatic gloss meters, and image recognition systems, the actual performance data of the paint film is automatically collected after the paint is mixed or after the paint film has dried.

[0088] Data alignment and storage: The actual result of this paint mixing is precisely aligned with the prediction value of the AI ​​prediction engine and the input target paint mixing ratio to form a complete "input-prediction-output" triple record, which is then stored in the time series database of the hybrid architecture database.

[0089] Incremental Model Learning: The system is configured with a scheduled task (e.g., every Sunday morning) to initiate the model retraining process. This process automatically extracts new data from the knowledge base and performs incremental learning or fine-tuning on the existing random forest model, allowing the model's predictive ability to continuously evolve as more data is used. All model versions are managed and archived for rollback when necessary.

[0090] Model version management and rollback mechanism: The system manages the random forest model in the AI ​​prediction engine through versioning. A new version (e.g., v2.1.3) is generated after each incremental training.

[0091] All model versions and their corresponding training datasets are archived in a hybrid architecture database.

[0092] If the accuracy of the new version model on the validation set drops by more than 5%, the system will automatically trigger an alarm and can roll back to the previous stable version with one click to ensure the stability of the production environment.

[0093] The feedback and learning modules in this application include a model version manager and a rollback controller, which ensure system robustness.

[0094] In summary, the modules in the intelligent paint mixing system exchange data through standard communication protocols, including but not limited to MQTT (Message Queuing Telemetry Transport), HTTP (Hypertext Transfer Protocol) / HTTPS (Hypertext Transfer Protocol Secure). The working principle is as follows: User interaction module: Serving as the system input port, this module provides a graphical user interface (such as a touchscreen) that allows operators to input structured requirements such as color codes and performance indicators (gloss, drying time, etc.). This module can also connect to hardware such as barcode scanners and high-precision spectrophotometers to automatically acquire color data (L, a, b values).

[0095] Environmental sensing module: Temperature and humidity sensors, atmospheric pressure sensors, etc., deployed at the paint mixing site collect environmental perception data in real time through data acquisition terminals (such as Raspberry Pi), and push the filtered data to the system to provide real-time parameters for dynamic ratio optimization.

[0096] Data and Knowledge Base Module: Serving as the system's memory unit, this module employs a hybrid storage architecture. A relational database stores master data on proportions and the physicochemical properties of raw materials (such as resin hydroxyl value and pigment coloring intensity); a time-series database stores time-stamped process data; and an object database stores unstructured data such as spectral curves. This module provides data services to other modules via an API (Application Programming Interface).

[0097] The AI ​​prediction engine, acting as the system's decision-making brain, is its core. It receives data from users and the environment, extracts materialized feature vectors of raw materials from a hybrid architecture database, constructs high-dimensional feature vectors through feature engineering, inputs them into a pre-trained machine learning model (such as a random forest) for performance prediction, and then combines optimization algorithms (such as virtual orthogonal experiments and multi-objective optimization algorithms) to output the optimal target paint mixing ratio.

[0098] Automated execution module: Serving as the system's limbs, this is an automatic paint mixing machine controlled by a PLC (Programmable Logic Controller). It receives mixing instructions from an AI prediction engine, driving high-precision metering pumps, agitators, and other actuators. Through PID (Proportional-Integral-Derivative) closed-loop control, it precisely completes the metering, addition, and mixing of each component, producing a solid paint coating.

[0099] Feedback and Learning Module: Serving as the system's evolutionary organ, this module collects actual performance data of the manufactured coatings using online sensors (such as colorimeters and viscometers), aligns this data with predicted values, and stores it in a knowledge base. The system periodically triggers incremental learning to update the AI ​​model and manages model versions, achieving closed-loop self-evolution.

[0100] The specific process for achieving intelligent paint mixing is as follows: Figure 4As shown, the process is as follows: Step 1, Receiving Requirements: The user inputs or selects the target requirements for this paint mixing task through the interactive interface, including but not limited to the target color (obtainable through color codes, color cards, or spectrophotometers), key performance indicators (such as gloss ≥90 GU, drying time ≤30 minutes), and the selected raw material system (such as resin type, color paste series). This step receives the user's paint mixing requirement data. Step 2, Environmental Perception: The system uses environmental sensors to perceive environmental data in real time, including but not limited to temperature, humidity, and atmospheric pressure. The data is then processed and timestamped to obtain processed environmental perception data. Step 3, AI Prediction and Optimization: The AI ​​intelligent prediction engine, based on the input paint mixing requirement data and the processed environmental perception data, combined with the material physical feature vectors of raw materials in the hybrid architecture database, performs performance prediction and multi-objective optimization, outputting the results. The optimal target paint mixing ratio is determined, and the entire process is completed on the server / cloud within minutes, without the need for physical trial and error. The fourth step is automated execution: the automated execution module receives the target paint mixing ratio, the programmable logic controller parses the instructions, and drives high-precision metering pumps, mass flow meters, and other actuators to accurately measure and add each component of the base paint according to the set ratio. Then, the mixer is started for uniform mixing, producing a physical coating and accurately completing the physical mixing of the paint. The fifth step is verification and feedback: the system collects actual performance data of the finished paint or film through online sensors or manual input, and feeds the results back to the hybrid architecture database. The system periodically (or based on the amount of accumulated data) initiates an incremental learning process, using new data to fine-tune and optimize the AI ​​prediction model, updating the model version, thereby enabling the system to make more accurate predictions in the next task and achieving continuous evolution.

[0101] Taking the customized R&D and production material preparation of the topcoat for a large drone as an example, the specific steps to achieve intelligent paint mixing are as follows: (1) Requirements Input and Data Awareness: The coating process engineer inputs the task objective into the system interface, such as "Preparing the skin topcoat for XX type UAV". Paint mixing requirements data include: Color: such as a specific gray label; Core performance characteristics include: UV aging resistance ≥2000 hours, high and low temperature cycling resistance (-55℃ to +70℃), and surface resistivity meeting electromagnetic compatibility requirements. Application: If applicable to robotic spraying, the application period is >4 hours.

[0102] (2) The system reads real-time environmental sensing data of the paint preparation workshop, such as temperature 23℃ and humidity 50%.

[0103] (3) AI Prediction and Optimization: The AI ​​prediction engine is activated. First, it retrieves the physical and chemical feature vectors of raw materials such as resin base materials (e.g., high-performance fluorocarbon resin), special pigments (e.g., radar wave absorbing pigments), and functional additives for drone coatings from the hybrid architecture database and integrates them with the above-mentioned paint mixing requirements data and processed environmental perception data. Subsequently, the machine learning model performs performance prediction in a virtual mixing space of tens of thousands. The optimization decision module takes "comprehensive performance compliance" as the primary constraint and "raw material cost" and "viscosity stability" as optimization objectives, and uses virtual orthogonal design for rapid screening. Finally, it outputs the optimal target paint mixing ratio and predicted performance report within a few minutes: such as fluorocarbon resin A: 60 parts, anti-rust pigment B: 15 parts, directional alignment additive C: 1.5 parts, special solvent D: 23.5 parts, etc., and predicts its salt spray resistance > 3000h and application viscosity 45 seconds (Ford Cup 4).

[0104] (4) Precise execution and sample preparation: The sequence of action instructions corresponding to the target paint mixing ratio is sent to the automated paint mixing station in the workshop. The high-precision metering pump group adds the paint sequentially according to the ratio, preparing 20 kg batches of topcoat base material. The preparation process data is recorded throughout the entire process.

[0105] (5) Verification, Feedback, and Model Evolution: Test panels were prepared for this batch of coatings according to standards, and verification tests such as accelerated aging and electrochemical impedance were conducted. Two weeks later, the measured data (e.g., UV resistance for 2050 hours, surface resistivity) were collected. The data is then sent back to the system. The system uses this complete "demand-ratio-measured result" data pair as a high-value sample, triggering incremental learning of the AI ​​model to optimize the prediction accuracy of special coatings for drones in the future.

[0106] Example of experimental verification data (based on simulation and pilot data): Efficiency Improvement: In the task of mixing 10 new colors, the traditional trial and error method requires an average of 8.3 physical experiments and takes about 4.5 hours; this application, through virtual orthogonal optimization, requires an average of only 1.2 physical verifications and takes about 25 minutes, improving efficiency by about 90%.

[0107] Prediction accuracy: The root mean square error of the random forest model in predicting color difference ΔE on the independent test set is 0.42, and the mean absolute percentage error in predicting gloss is 3.1%, which meets the requirements of industrial-grade accuracy.

[0108] Small sample adaptability: In a new material system with only 30 historical data points, the system can still find a suitable ratio within 3 iterations by combining transfer learning and virtual orthogonal experiments, breaking through the dependence of traditional data-driven models on small samples.

[0109] Environmental Adaptive Effect: In environments where humidity increases from 50% to 80%, the system automatically adjusts and optimizes the weights, reducing the average predicted drying time of the output mix by 18%, significantly improving construction adaptability.

[0110] The innovation of this application lies in combining machine learning performance prediction with orthogonal design-based optimization decision-making in the field of paint mixing, forming a complete new method. Its core process is: feature construction → AI model prediction of virtual mixing performance → efficient optimization using orthogonal design and other strategies in the prediction results → outputting the optimal target paint mixing ratio. To solve the "small sample" problem in on-site paint mixing, it proposes generating an orthogonal experimental table. This involves no or very few physical experiments, but instead, the AI ​​model performs performance predictions at mixing points arranged in a virtual orthogonal table. Range analysis is then performed on the prediction results to quickly identify key influencing factors and the optimal level combination. A closed-loop data flow is established: demand input → AI decision → automatic execution → effect feedback → model update. Each actual paint mixing result is used as labeled training data and fed back into the system. The system is used for incremental learning, driving the continuous evolution of the AI ​​model. The modules in the intelligent paint mixing system are not simply stacked together, but rather form an organic whole capable of completing the entire intelligent behavior of "perception-decision-execution-learning" through specific data interfaces and logical relationships. Furthermore, the modules establish connection relationships and data interaction protocols to achieve the aforementioned intelligent closed loop. Within the AI ​​prediction engine, the core role of the feature engineering unit is clearly defined: it does not simply transmit raw material ratios, but actively integrates user needs, environmental data, and raw material physical feature vectors extracted from the knowledge base (such as hydroxyl value, Tg, and tinting strength) to construct high-dimensional features. These features are then input into a specific machine learning model (such as random forest) for prediction. The target paint mixing ratio output by the system is not isolated data, but is bound to optimization goals (such as "lowest cost"). When the automated execution module receives the target paint mixing ratio, it can adjust execution parameters (such as stirring speed and curing time) to ensure that the final product not only has the correct composition but also its performance approaches the predicted optimal point.

[0111] The beneficial effects of this application are as follows: It fundamentally upgrades paint mixing technology from "experience-driven, open-loop retrieval" to "data-driven, closed-loop decision-making"; it establishes a paint mixing decision-making kernel with predictive and creative capabilities. To overcome the core shortcoming of existing technologies that "can only retrieve, not create," this invention constructs a performance prediction model based on machine learning (such as random forests and neural networks). This model can directly predict the comprehensive performance of any virtual ratio based on the input raw material characteristics, environmental parameters, and performance requirements, thereby gaining the ability to actively generate and screen ratios in a vast ratio space; it enables rapid, one-time, and accurate paint mixing under new demands. Addressing the inefficiency of existing technologies in responding to new demands, the purpose of this invention is to use the aforementioned AI prediction model, combined with efficient optimization algorithms (such as multi-objective optimization and orthogonal design), to conduct massive computational experiments in the digital space to quickly locate the globally optimal or suboptimal ratio. This shortens the paint mixing cycle from the "days / hours" level relying on physical trial and error to the "minutes" level primarily based on computation, ultimately avoiding physical trial and error altogether; and it achieves automatic multi-objective optimization under complex constraints. Addressing the shortcoming of existing technologies with a single optimization objective, this invention integrates a multi-objective optimization decision-making module into the system. This module uses multiple indicators such as color matching, durability, cost, and workability as common constraints or optimization goals, and automatically weighs and optimizes based on AI predictions. Thus, in a single output of paint mixing ratios, it simultaneously satisfies color and multiple performance indicators, reducing overall cost and improving resource utilization. Addressing the open-loop and rigid shortcomings of existing technologies, this invention designs a system architecture with a data feedback loop. Each actual paint mixing result (whether successful or unsuccessful) serves as new training data for continuously updating and optimizing the AI ​​prediction model. This enables the system to continuously improve and adapt during use, reducing reliance on manual maintenance and becoming increasingly accurate and reliable over time.

[0112] This invention changes the core of paint mixing from finding the ratio to calculating the ratio, and gives the system the vitality of learning and evolution, thereby solving the fundamental technical shortcomings of existing technologies in terms of efficiency, cost, adaptability and sustainability.

[0113] This application provides an intelligent paint mixing method based on artificial intelligence, including acquiring user paint mixing demand data and real-time environmental perception data; processing the environmental perception data to obtain processed environmental perception data, eliminating random errors and dimensional differences in the acquisition process, and ensuring the authenticity and consistency of the data; selecting physical feature vectors of raw materials for paint mixing from a hybrid architecture database; and performing feature splicing and fusion on the paint mixing demand data, processed environmental perception data, and physical feature vectors of raw materials to construct a target feature vector, enabling the machine learning model to make performance predictions based on comprehensive information, thereby improving the accuracy and generalization ability of the prediction results; and inputting the target feature vector into a pre-trained machine learning model to output the predicted paint film performance value corresponding to the target feature vector. The system predicts paint film performance values, including at least one of color difference, gloss, drying time, and salt spray resistance level, as well as a cost estimate, enabling rapid and comprehensive prediction of paint film performance. Based on the predicted paint film performance values ​​and user-defined optimization objectives, an orthogonal experimental table is generated. The performance prediction model is then used to predict the performance of each candidate paint mixing ratio in the orthogonal experimental table, yielding the ratio performance prediction results. Range analysis is performed on the ratio performance prediction results, and the range analysis results are used to perform multi-objective optimization on each candidate paint mixing ratio to obtain the target paint mixing ratio, achieving optimal paint mixing ratio under complex constraints. The target paint mixing ratio is analyzed to obtain an action instruction sequence, which is then executed to complete the paint mixing operation, adapting to different paint mixing needs, improving the accuracy and efficiency of paint mixing, and enhancing resource utilization.

[0114] See Figure 5 As shown in the figure, an embodiment of the present invention discloses an intelligent paint mixing device based on artificial intelligence, which may specifically include: Processing module 11 is used to acquire user paint mixing requirements data and real-time environmental perception data, process the environmental perception data, and obtain the processed environmental perception data. The feature splicing and fusion module 12 is used to filter the materialized feature vectors of paint mixing raw materials from the hybrid architecture database, and to splice and fuse the paint mixing requirement data, the processed environmental perception data and the materialized feature vectors of raw materials to construct a target feature vector. The paint film performance prediction module 13 is used to input the target feature vector into a pre-trained machine learning model to output a paint film performance prediction value corresponding to the target feature vector; the paint film performance prediction value includes at least one of color difference, gloss, drying time, and salt spray resistance level, as well as a cost estimate. The multi-objective optimization module 14 is used to generate an orthogonal experimental table based on the predicted value of the paint film performance and the optimization objective set by the user, use a performance prediction model to predict the performance of each paint mixing candidate ratio in the orthogonal experimental table, obtain the ratio performance prediction result, perform range analysis on the ratio performance prediction result, and use the range analysis result to perform multi-objective optimization on each paint mixing candidate ratio to obtain the target paint mixing ratio. The parsing module 15 is used to parse the target paint mixing ratio, obtain an action instruction sequence, and execute the action instruction sequence to complete the paint mixing operation.

[0115] In some specific embodiments, the processing module 11 may specifically include: The paint mixing requirement data acquisition module is used to acquire the user's paint mixing requirement data from the graphical user interface; the paint mixing requirement data includes color, gloss, drying time requirements, and salt spray resistance requirements; An environmental sensing data acquisition module is used to acquire real-time environmental sensing data from sensors using a serial peripheral interface or integrated circuit bus protocol; the sensors include a digital temperature and humidity sensor and an atmospheric pressure sensor. The filtering, noise reduction, and data stamping module is used to filter, reduce noise, and add data stamps to the environmental sensing data to obtain the processed environmental sensing data.

[0116] In some specific embodiments, the feature splicing and fusion module 12 may specifically include: Hybrid architecture database building module, used to build hybrid architecture databases based on relational databases, time-series databases, and object repositories; The raw material physical feature vector filtering module is used to filter out the physical feature vectors of raw materials for paint mixing from the relational database.

[0117] In some specific embodiments, the paint film performance prediction module 13 may specifically include: The historical paint mixing dataset filtering module is used to filter historical paint mixing datasets from the time-series database; The training, testing, and validation modules are used to train, test, and validate the machine learning model based on the historical paint mixing dataset using the five-fold cross-validation method to obtain a pre-trained machine learning model; the machine learning model is a random forest regression and classification ensemble model.

[0118] In some specific embodiments, the feature splicing and fusion module 12 may specifically include: The extraction module is used to extract color features, environmental features, and material physical features from the paint mixing requirement data, the processed environmental perception data, and the raw material physical feature vector. The target feature vector construction module is used to perform feature splicing and fusion on the color features, the environmental features and the material physical features based on the attention weight allocation mechanism to construct the target feature vector.

[0119] In some specific embodiments, the multi-objective optimization module 14 may specifically include: The orthogonal experimental table generation module is used to input the predicted value of the paint film performance and the optimization target set by the user into the decision-maker, so that the decision-maker can use a preset sampling algorithm to generate an orthogonal experimental table based on the predicted value of the paint film performance and the optimization target. The conversion module is used to convert each candidate paint mixing ratio in the orthogonal experimental table into a mixing ratio feature vector; The performance prediction module is used to perform performance prediction on the ratio feature vector using a performance prediction model to obtain the ratio performance prediction result. The range analysis module is used to perform range analysis on the predicted performance of the proportioning ratio to obtain range analysis results including the range values. The multi-objective optimization module is used to perform multi-objective optimization of each paint candidate ratio based on the range analysis results using a non-dominated sorting genetic algorithm.

[0120] In some specific embodiments, the parsing module 15 may specifically include: The actual performance data acquisition module for paint film is used to collect actual performance data of paint film after it dries using an online viscometer, an automatic gloss meter, and an image recognition system. The alignment module is used to align the actual performance data of the paint film, the target paint mixing ratio, and the predicted performance value of the paint film corresponding to the target paint mixing ratio to generate a triplet record. A record storage module is used to store the triplet records into a time-series database within a hybrid architecture database.

[0121] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the artificial intelligence-based intelligent paint mixing method disclosed in any of the foregoing embodiments.

[0122] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0123] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored on it include operating system 221, computer program 222 and data 223, etc., and the storage method can be temporary storage or permanent storage.

[0124] The operating system 221 manages and controls the various hardware devices on the electronic device 20 and the computer program 222 to enable the processor 21 to perform calculations and processing on the data 223 in the memory 22. The operating system 221 can be Windows, Unix, Linux, etc. The computer program 222, in addition to including a computer program capable of performing the AI-based intelligent paint mixing method executed by the electronic device 20 as disclosed in any of the foregoing embodiments, may further include computer programs capable of performing other specific tasks. The data 223 may include data received by the AI-based intelligent paint mixing device from external devices, as well as data collected by its own input / output interface 25.

[0125] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0126] Furthermore, this application also discloses a computer-readable storage medium storing a computer program. When the computer program is loaded and executed by a processor, it implements the steps of the artificial intelligence-based intelligent paint mixing method disclosed in any of the foregoing embodiments.

[0127] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0128] The above provides a detailed description of an intelligent paint mixing method, apparatus, device, and storage medium based on artificial intelligence provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. An intelligent paint mixing method based on artificial intelligence, characterized in that, include: Acquire user paint mixing requirements data and real-time environmental perception data, process the environmental perception data, and obtain the processed environmental perception data. The raw material materialization feature vectors for paint mixing are selected from the hybrid architecture database. The paint mixing requirement data, the processed environmental perception data, and the raw material materialization feature vectors are then spliced ​​and fused to construct the target feature vector. The target feature vector is input into a pre-trained machine learning model to output a predicted value of the paint film performance corresponding to the target feature vector; the predicted value of the paint film performance includes at least one of color difference, gloss, drying time, and salt spray resistance level, as well as a cost estimate. Based on the predicted paint film performance and the user-defined optimization objectives, an orthogonal experimental table is generated. The performance prediction model is used to predict the performance of each candidate paint mixing ratio in the orthogonal experimental table to obtain the mixing ratio performance prediction results. The mixing ratio performance prediction results are then subjected to range analysis. The range analysis results are used to perform multi-objective optimization on each candidate paint mixing ratio to obtain the target paint mixing ratio. The target paint mixing ratio is analyzed to obtain an action instruction sequence, and the action instruction sequence is executed to complete the paint mixing operation.

2. The intelligent paint mixing method based on artificial intelligence according to claim 1, characterized in that, The process of acquiring user paint mixing requirements data and real-time environmental perception data, and processing the environmental perception data to obtain processed environmental perception data includes: The user's paint mixing requirements are obtained from the graphical user interface; these requirements include color, gloss, drying time, and salt spray resistance. Real-time environmental sensing data is acquired from sensors using a serial peripheral interface or integrated circuit bus protocol; the sensors include digital temperature and humidity sensors and atmospheric pressure sensors. The environmental perception data is filtered, denoised, and have data stamps added to obtain the processed environmental perception data.

3. The intelligent paint mixing method based on artificial intelligence according to claim 1, characterized in that, The step of filtering the physical feature vectors of raw materials for paint mixing from the hybrid architecture database includes: A hybrid architecture database is built based on relational databases, time-series databases, and object repositories; The physical feature vectors of raw materials for paint mixing are filtered out from the relational database.

4. The intelligent paint mixing method based on artificial intelligence according to claim 3, characterized in that, Before inputting the target feature vector into the pre-trained machine learning model, the method further includes: Historical paint mixing datasets were selected from the time-series database; The machine learning model was trained, tested, and validated using the five-fold cross-validation method and based on the historical paint mixing dataset to obtain a pre-trained machine learning model; the machine learning model is a random forest regression and classification ensemble model.

5. The intelligent paint mixing method based on artificial intelligence according to claim 1, characterized in that, The process of concatenating and fusing the paint mixing requirement data, the processed environmental perception data, and the materialized feature vector of the raw materials to construct a target feature vector includes: Color features, environmental features, and material physical features are extracted from the paint mixing demand data, the processed environmental perception data, and the raw material physical feature vector. Based on the attention weight allocation mechanism, the color features, the environmental features, and the material physical features are spliced ​​and fused to construct the target feature vector.

6. The intelligent paint mixing method based on artificial intelligence according to claim 1, characterized in that, Based on the predicted paint film performance values ​​and the user-defined optimization objectives, an orthogonal experimental table is generated. A performance prediction model is used to predict the performance of each candidate paint blending ratio in the orthogonal experimental table, yielding the ratio performance prediction results. Range analysis is then performed on the ratio performance prediction results, and the range analysis results are used to perform multi-objective optimization of each candidate paint blending ratio, including: The predicted values ​​of the paint film performance and the optimization targets set by the user are input into the decision-maker, so that the decision-maker can use a preset sampling algorithm to generate an orthogonal experimental table based on the predicted values ​​of the paint film performance and the optimization targets. Convert each candidate paint mixing ratio in the orthogonal experimental table into a mixing ratio feature vector; The performance prediction model is used to predict the performance of the proportion feature vector to obtain the proportion performance prediction result; A range analysis is performed on the predicted performance of the proportions to obtain range analysis results including the range values; A non-dominated sorting genetic algorithm was used, and multi-objective optimization was performed on each candidate paint formulation based on the range analysis results.

7. The intelligent paint mixing method based on artificial intelligence according to any one of claims 1 to 6, characterized in that, After the paint mixing process is completed, the following is also included: After the paint film dries, the actual performance data of the paint film is collected using an online viscometer, an automatic gloss meter, and an image recognition system. Align the actual performance data of the paint film, the target paint mixing ratio, and the predicted performance value of the paint film corresponding to the target paint mixing ratio to generate a triplet record; The triplet records are stored in the time-series database of the hybrid architecture database.

8. An intelligent paint mixing device based on artificial intelligence, characterized in that, include: The processing module is used to acquire user paint mixing requirements data and real-time environmental perception data, process the environmental perception data, and obtain the processed environmental perception data. The feature splicing and fusion module is used to filter out the materialized feature vectors of raw materials for paint mixing from the hybrid architecture database, and to splice and fuse the paint mixing requirement data, the processed environmental perception data, and the materialized feature vectors of raw materials to construct a target feature vector. The paint film performance prediction module is used to input the target feature vector into a pre-trained machine learning model to output a paint film performance prediction value corresponding to the target feature vector; the paint film performance prediction value includes at least one of color difference, gloss, drying time, and salt spray resistance level, as well as a cost estimate. The multi-objective optimization module is used to generate an orthogonal experimental table based on the predicted value of the paint film performance and the optimization objectives set by the user, use a performance prediction model to predict the performance of each candidate paint ratio in the orthogonal experimental table, obtain the ratio performance prediction result, perform range analysis on the ratio performance prediction result, and use the range analysis result to perform multi-objective optimization on each candidate paint ratio to obtain the target paint ratio. The parsing module is used to parse the target paint mixing ratio, obtain an action instruction sequence, and execute the action instruction sequence to complete the paint mixing operation.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the artificial intelligence-based intelligent paint mixing method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store computer programs; wherein, when the computer programs are executed by a processor, they implement the intelligent paint mixing method based on artificial intelligence as described in any one of claims 1 to 7.

Images