A cloud computing-based paint production control system

Abstract

The application relates to the technical field of industrial automation and paint production, and particularly relates to a paint production control system based on cloud computing. The paint production control system comprises a raw material characteristic sensing unit, a fluctuation evaluation unit, a production parameter optimization unit and a process closed-loop control unit. The raw material characteristic sensing unit is used for acquiring key physical and chemical parameters of raw materials entering a factory and constructing a raw material feature vector. The fluctuation evaluation unit is used for comparing and analyzing a fluctuation index with a preset trigger threshold value, and generating a formula dynamic adjustment instruction when the fluctuation index is greater than the trigger threshold value. The production parameter optimization unit is used for reversely solving an optimal formula parameter vector and an optimal process parameter vector. The process closed-loop control unit is used for setting the optimal process parameter vector as an initial set point of a production process and adjusting the production process in real time. The model self-evolution unit is used for generating a model correction factor and weighting and retraining a data-driven prediction model. The application realizes adaptive production of non-standard and high-fluctuation raw materials, thereby guaranteeing the high stability of the quality of final products.

Description

Technical Field

[0001] This invention relates to the fields of industrial automation and coating production technology, specifically to a cloud-based coating production control system. Background Technology

[0002] In the coatings manufacturing industry, industrial by-products are often used as raw materials to reduce costs. However, the physicochemical properties of these raw materials, such as chemical composition, particle size distribution, and dynamic viscosity, exhibit natural and significant fluctuations. For example, in the production of environmentally friendly asphalt pavement maintenance coatings, industrial by-products or wastes such as petroleum coke powder and glass fiber powder are often used as functional fillers to reduce costs. However, the key parameters of these by-products, such as carbon content, sulfur content, particle size, and morphology, vary greatly from batch to batch, directly affecting the final coating's resistance to high and low temperatures, rutting resistance, and water stability. Traditional production control methods mainly rely on fixed formulas and process parameters. This static production mode cannot actively perceive and adapt to batch-to-batch differences in incoming raw materials. When the actual characteristics of raw materials deviate from the preset standards, the fixed production parameters cannot be adjusted accordingly, resulting in unstable performance of the final product and difficulty in ensuring consistent quality. Therefore, existing technologies face a core challenge: how to effectively overcome the negative impact of the volatility of non-standard raw materials with unstable physicochemical properties when using them for production, to ensure the continuous stability of coating product quality, and thus improve production flexibility and economic benefits.

[0003] The information disclosed in the background section above is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention discloses a cloud computing-based paint production control system. Specifically, the technical solution of this invention is as follows: A cloud-based paint production control system includes: The raw material characteristic sensing unit is used to acquire key physicochemical parameters of incoming raw materials and construct a raw material feature vector based on these key physicochemical parameters. The volatility assessment unit is used to calculate the volatility index between the raw material characteristic vector and the preset benchmark reference value, and compare the volatility index with the preset trigger threshold. When the volatility index is greater than the trigger threshold, a dynamic adjustment instruction for the formula is generated. The production parameter optimization unit is used to respond to the dynamic adjustment command of the formula and, based on the data-driven prediction model, solve in reverse to obtain the optimal formula parameter vector and the optimal process parameter vector. The process closed-loop control unit is used to set the optimal process parameter vector as the initial set point of the production process and to adjust the production process in real time through a closed-loop feedback control algorithm. The model self-evolution unit is used to calculate the prediction deviation between the actual finished product performance index and the predicted performance value after production is completed, generate the model correction factor, and use the model correction factor to perform weighted retraining of the data-driven prediction model.

[0005] Preferably, the key physicochemical parameters include chemical component concentration, particle size distribution, and dynamic viscosity; the raw material characteristic sensing unit determines the chemical component concentration through near-infrared spectroscopy analysis, determines the particle size distribution through laser particle size analysis, and determines the dynamic viscosity through online viscosity measurement.

[0006] Preferably, the volatility assessment unit retrieves the average parameter values ​​of the gold batch raw materials calibrated in historical data and sets them as the benchmark reference value; and uses a weighted Euclidean distance model to quantify the deviation between the raw material feature vector and the benchmark reference value to generate a volatility index.

[0007] Preferably, the production parameter optimization unit sets the acquired target product performance as the optimization target and uses the raw material feature vector as the fixed input of the data-driven prediction model; and adopts a global optimization algorithm to iteratively solve for the formula parameter vector and process parameter vector that minimize the error between the prediction output of the data-driven prediction model and the optimization target, so as to generate the optimal formula parameter vector and the optimal process parameter vector.

[0008] Preferably, the data-driven prediction model is a nonlinear mapping model established by supervised learning training on historical production data, used to reveal the intrinsic relationship between raw material characteristics, formula parameters and process parameters and the performance indicators of the final product.

[0009] Preferably, the process closed-loop control unit performs the following steps: real-time monitoring of process state variables at key nodes of the production line; determining the error between the real-time monitored value of the process state variable and the initial set point; and calculating the output adjustment amount of the production line actuator based on the error using a proportional-integral-derivative control algorithm.

[0010] Preferably, the model self-evolution unit performs the following steps: obtaining the actual finished product performance indicators in the quality inspection process and the predicted performance values ​​output by the data-driven prediction model; calculating the prediction deviation between the actual finished product performance indicators and the predicted performance values; and normalizing the prediction deviation to generate a model correction factor.

[0011] Preferably, the model self-evolution unit is further used to: archive the model correction factor as sample weights together with the production data records of the batch; when the accumulated production data records reach a preset number, trigger the retraining of the data-driven prediction model; the retraining realizes the weighted update of the data-driven prediction model by introducing the sample weights into the loss function.

[0012] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention effectively overcomes the shortcomings of traditional fixed processes that cannot cope with the uncertainty of raw materials by sensing the physicochemical properties of incoming raw materials in real time and dynamically adjusting the production formula and process parameters. It achieves adaptive production of non-standard and highly volatile raw materials, thereby ensuring the high stability of the final product quality.

[0013] 2. This invention uses a data-driven prediction model to solve the problem in reverse, tailoring the optimal production plan for each batch of raw materials, and ensuring that the plan is accurately executed during production through closed-loop feedback control, which greatly improves the accuracy of production control and the achievement rate of product performance.

[0014] 3. This invention establishes a learning mechanism for model self-evolution, which can continuously optimize and iteratively update the core prediction model by comparing actual production results with model predictions, enabling the system to have the ability to adapt and improve itself in order to cope with the long-term and structural changes in raw material supply.

[0015] 4. This invention constructs a complete technical closed loop by integrating raw material sensing, dynamic optimization, precise control, and self-learning, which significantly improves the automation and intelligence level of coating production, making it possible to stably use low-cost, high-fluctuation industrial by-products, thereby enhancing the overall flexibility and economic benefits of production. Attached Figure Description

[0016] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart of the system of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0018] Example 1:

[0019] Please see Figure 1 A cloud-based paint production control system includes: The raw material characteristic sensing unit is used to acquire key physicochemical parameters of incoming raw materials and construct raw material feature vectors based on these key physicochemical parameters. The volatility assessment unit is used to calculate the volatility index between the raw material characteristic vector and the preset benchmark reference value, and compare the volatility index with the preset trigger threshold. When the volatility index is greater than the trigger threshold, a dynamic adjustment instruction for the formula is generated. The production parameter optimization unit is used to respond to the dynamic adjustment command of the formula and, based on the data-driven prediction model, solve in reverse to obtain the optimal formula parameter vector and the optimal process parameter vector. The process closed-loop control unit is used to set the optimal process parameter vector as the initial set point of the production process and to adjust the production process in real time through a closed-loop feedback control algorithm. The model self-evolution unit is used to calculate the prediction deviation between the actual finished product performance index and the predicted performance value after production, generate the model correction factor, and use the model correction factor to perform weighted retraining of the data-driven prediction model; through the above-mentioned technical closed loop, this invention realizes the adaptive production of non-standard and highly volatile raw materials; at the same time, because it can stably use water-based emulsions and low-volatility industrial by-products and supports cold construction processes, it avoids the environmental hazards of organic solvents in the production and use process; A cloud-based coating production control system aims to solve the technical challenge that the inherent volatility of the physicochemical properties of raw materials such as industrial by-products makes it impossible to guarantee the stability of finished product quality using traditional fixed formulas and process parameters. In this embodiment, the system constructs a complete technical closed loop, from raw material arrival sensing, dynamic optimization of production instructions, closed-loop control of the production process to model self-evolution, ensuring adaptive production to non-standard raw materials and continuous stability of product performance. The system includes the following cooperative units: The raw material characteristic sensing unit aims to accurately digitally characterize each batch of raw materials entering the factory. In this embodiment, the unit is deployed at the feeding stage of the production line. It acquires key physicochemical parameters of the raw materials in real time through an integrated sensing system and transforms these multi-dimensional and heterogeneous data into structured digital signals. Based on these parameters, the unit constructs a standardized raw material feature vector for the current batch of raw materials. Here, the raw material feature vector refers to a multi-dimensional vector that can uniquely identify and quantify the characteristics of the batch of raw materials. Its role is to serve as the digital basis and unified input for all subsequent analysis, evaluation, and optimization stages. Its source is obtained by real-time acquisition and processing by the sensing system of this unit. The purpose of the volatility assessment unit is to quantify the degree of deviation between the characteristics of the current batch of raw materials and the ideal state, and to determine whether dynamic adjustment of production parameters needs to be initiated. In this embodiment, the unit receives the raw material feature vector generated by the raw material characteristic sensing unit and compares it with a preset benchmark reference value. A comprehensive volatility index is calculated using a specific mathematical model. This volatility index is a dimensionless indicator that comprehensively quantifies the degree to which the current raw material characteristics deviate from the benchmark. Its purpose is to provide a clear and singular trigger for subsequent decisions. It is derived from the deviation between the current raw material characteristic vector and the benchmark reference value calculated by this unit. When the volatility index is greater than a preset trigger threshold, it means that the volatility of the raw material has exceeded the acceptable range. At this time, the unit will generate and issue a dynamic adjustment instruction for the formula. To ensure the scientific nature of this judgment, the trigger threshold is determined by: performing statistical distribution analysis based on the volatility index of historical golden batches of raw materials and taking the upper bound of its 95% confidence interval, thereby achieving a technical balance between sensitivity and false alarm rate. When the volatility index is less than or equal to the trigger threshold, it indicates that the raw material characteristics are within an acceptable stable range. The system will directly adopt the standard formula parameter vector and process parameter vector corresponding to the historical golden batch, and the process closed-loop control unit will execute subsequent production. The production parameter optimization unit aims to respond to dynamic formula adjustment instructions and tailor an optimal production plan for the current batch of raw materials with specific volatility. In this embodiment, the core of this unit is a data-driven prediction model deployed in the cloud. This data-driven prediction model is a mathematical model that can reveal the complex nonlinear relationship between raw material characteristics, formula parameters, process parameters, and final product performance. Its function is to predict the possible performance output of the product under a specific input combination. It is obtained by machine learning training on massive historical production data. After receiving the instruction, the unit uses the feature vector of the raw materials of the current batch as the fixed input of the model and the target product performance required by national standards or orders as the optimization objective. By executing a global optimization algorithm in the cloud, it reversely solves for the combination of formula parameters and process parameters that minimizes the error between the model's predicted performance and the target performance, thereby generating the optimal formula parameter vector and the optimal process parameter vector. The purpose of the process closed-loop control unit is to ensure that the theoretically optimal process parameters calculated by the optimization unit can be executed accurately and stably in the actual production process. In this embodiment, the unit sets each parameter value in the optimal process parameter vector as the initial set point of the corresponding actuators on the production line, such as heaters and stirring motors. During the production process, the unit monitors the process state variables in real time through sensors deployed at key nodes and compares them with the initial set points to calculate the real-time error. Based on this error, the unit continuously calculates the output adjustment amount through a closed-loop feedback control algorithm such as PID control to dynamically adjust the operating state of each actuator, thereby overcoming on-site environmental disturbances and equipment state fluctuations, and ensuring a high degree of consistency between the actual production trajectory and the optimal set trajectory. The self-evolving model unit aims to enable the entire control system to have the ability to learn and continuously optimize itself in order to adapt to long-term and structural changes that may occur in the supply of raw materials. In this embodiment, after each batch of products is produced, the unit will obtain the actual finished product performance indicators measured in the quality inspection process and compare them with the predicted performance values ​​output by the data-driven prediction model before production, thereby calculating the prediction deviation. This unit generates a model correction factor based on this deviation. This model correction factor is a weighted index that quantifies the accuracy of a single prediction by the model. Its function is to increase the learning weight of inaccurate samples in subsequent model retraining. It is derived from the normalization calculation based on the prediction deviation by this unit. This factor will be archived along with all production data of this batch. When the accumulated data meets the preset trigger conditions, the system will automatically trigger the weighted retraining of the data-driven prediction model, thereby realizing the iterative update of the model and the continuous evolution of its performance. The preset trigger conditions can be set to the accumulated production data records reaching a preset number, or to the average prediction deviation of the model in the most recent N batches exceeding the warning limit, so as to realize a more adaptive retraining trigger mechanism. This invention constructs an intelligent production control system through the collaborative work of the above five units. It can actively sense raw material fluctuations, dynamically optimize production parameters, precisely control the production process, and continuously learn and evolve. It solves the pain point that traditional fixed processes cannot cope with the uncertainty of raw materials. It achieves the overall technical effect of being able to stably produce high-quality, high-performance coatings even when using low-cost, highly volatile industrial by-products as raw materials, and significantly improves the flexibility, stability and economic benefits of production.

[0020] Example 2:

[0021] The key physicochemical parameters in Example 2 include chemical component concentration, particle size distribution, and dynamic viscosity; the raw material characteristic sensing unit determines the chemical component concentration through near-infrared spectroscopy analysis, determines the particle size distribution through laser particle size analysis, and determines the dynamic viscosity through online viscosity measurement. Based on Example 1, this embodiment specifies the implementation method of the raw material characteristic sensing unit; its purpose is to ensure that the constructed raw material feature vector can comprehensively and accurately reflect the inherent characteristics of the raw material through standardized and high-precision measurement methods. Specifically, the key physicochemical parameters in this embodiment are clearly defined as three core dimensions: chemical composition concentration, particle size distribution, and dynamic viscosity. These three parameters were chosen because they jointly determine the reaction kinetics of the coating production process and the core performance of the final product from three levels: chemical composition, physical morphology, and rheological properties. Specifically, in the production scenario of environmentally friendly asphalt pavement maintenance coatings, the raw material characteristic sensing unit can measure the carbon content, sulfur content, and volatile matter of petroleum coke powder, which is the main filler, as well as the particle size and density of glass fiber powder, providing accurate input vectors for subsequent dynamic optimization of the formulation. To acquire these parameters, the raw material characteristic sensing unit integrates and performs the following specialized measurement techniques: The concentration of chemical components was determined using near-infrared spectroscopy (NIR) analysis. NIR spectroscopy is a technique that utilizes information from the near-infrared spectral region for qualitative and quantitative analysis. Its function in this invention is to rapidly and non-destructively determine the concentrations of key chemical components such as carbon, sulfur, and ash in incoming raw materials, such as petroleum coke. The data is obtained from the chemical component concentration data collected and analyzed by a near-infrared spectrometer deployed at the feed inlet. The particle size distribution is determined by laser particle size analysis. Laser particle size analysis is a method for measuring particle size distribution based on the principles of light diffraction and scattering. Its function is to accurately quantify the particle size of raw materials, which directly affects their dispersibility and reactivity in coating systems. This embodiment obtains key distribution nodes, such as d10, d50, and d90, using this method. Dynamic viscosity is determined through online viscosity measurement. Online viscosity measurement is a technique that directly measures viscosity in pipes or containers where materials flow. Its function is to obtain the rheological properties of raw materials in real time at specific shear rates. These properties are crucial for setting subsequent process parameters such as mixing and pumping. This embodiment obtains dynamic viscosity using an online viscometer. .

[0022] Example 3: The volatility assessment unit retrieves the average parameter values ​​of the gold batch raw materials calibrated in historical data and sets them as the benchmark reference value; and uses a weighted Euclidean distance model to quantify the deviation between the raw material feature vector and the benchmark reference value to generate a volatility index. Based on Example 1, this embodiment optimizes the specific algorithm logic of the volatility assessment unit; its purpose is to establish a more scientific and robust quantitative assessment model to accurately determine the degree of volatility of raw materials. In this embodiment, the volatility assessment unit performs the following steps to generate a volatility index: First, this unit retrieves the average parameter values ​​of the raw materials for the identified gold batches from the historical production database and sets them as the benchmark reference values. The "golden batch" refers to those raw material batches that have historically been able to consistently produce products of the highest quality. Its purpose is to provide a practically validated and optimal reference standard for the evaluation system. It is derived from historical data through data mining and screening. This method ensures the objectivity and practicality of the benchmark value and avoids the risk of relying on theoretical design values ​​and deviating from actual production. Secondly, after obtaining the feature vector of the current batch of raw materials. and reference vector This unit uses a weighted Euclidean distance model to quantify the deviation between the two; to accurately assess the degree to which the current raw material characteristics deviate from the ideal state, a volatility index is introduced. The calculation method is as follows: in, For the first The volatility index of batch raw materials is a dimensionless comprehensive deviation measure calculated by this formula; The first is provided by the raw material characteristic sensing unit. The first batch of raw materials Real-time measured values ​​of each characteristic parameter; The corresponding benchmark reference value obtained from historical gold batch data is the first Each parameter value; It is the total dimension of the raw material feature vector, that is, the total number of key physicochemical parameters; It is the first The weighting coefficients of each parameter, which are dimensionless parameters; It is a very small positive constant to ensure computational stability, for example... Used to avoid benchmark reference values When the result is zero or close to zero, the denominator is zero, thus ensuring the robustness of the calculation; To clarify The calibration process must be clearly stated that the coefficients are not pre-set, but determined through multiple regression analysis on an independent 'calibration dataset'; this calibration dataset consists of the feature vectors of raw materials from historical production and their corresponding actual finished product performance index vectors. Composition; This analysis quantifies the sensitivity of each raw material parameter to the performance of the final product, and normalizes this sensitivity before using it as a weighting coefficient. This allows the weights of each parameter to independently reflect their impact on product performance.

[0023] Example 4:

[0024] The production parameter optimization unit sets the acquired target product performance as the optimization objective and uses the raw material feature vector as the fixed input of the data-driven prediction model. It then uses a global optimization algorithm to iteratively solve for the formula parameter vector and process parameter vector that minimize the error between the prediction output of the data-driven prediction model and the optimization objective, in order to generate the optimal formula parameter vector and the optimal process parameter vector. Data-driven prediction models are nonlinear mapping models established through supervised learning training on historical production data. They are used to reveal the intrinsic relationship between raw material characteristics, formula parameters, and process parameters and the performance indicators of the final product. Based on Example 1, this embodiment elaborates on the working mechanism of the production parameter optimization unit and the data-driven prediction model on which it depends; its underlying logic is to transform the passive production adaptation process into an active, model-predictive reverse optimization process. Data-driven prediction models are the core foundation for achieving optimization. In this embodiment, the model is a nonlinear mapping relationship model established by supervised learning training on massive historical production data, such as Radial Basis Function Network (RBFN) or Gaussian Process Regression (GPR). Supervised learning training is a machine learning method that provides the model with a dataset containing input features and corresponding expected outputs, i.e., labels, allowing the model to learn the mapping relationship between them. In this invention, raw material feature vectors, formula parameter vectors, and process parameter vectors from historical data are used as input features, and the corresponding actual finished product performance indicators are used as labels; the model's role is to deeply reveal the characteristics of raw materials. Formula parameters With process parameters How do these three types of inputs collectively affect the final product performance metrics? This complex internal connection; the model can be abstractly expressed as a functional relationship. in, This model The calculated output is a vector of predicted performance values ​​for the product. It is the feature vector of the raw materials in the current batch, which is used as a known input during the optimization solution; and These are the vectors of formulation parameters and process parameters to be solved, generated iteratively by the optimization algorithm as optimization variables; in the optimization scenario of this type of environmentally friendly asphalt pavement maintenance coating, the target product performance... Specifically, it can be defined as the road performance of the finished coating, such as texture depth TD, pavement permeability coefficient Cw, and friction coefficient pendulum value BPN; while the formulation parameter vector serves as the optimization variable. This directly corresponds to the mass ratio of each component, such as the proportions of SBR modified emulsified asphalt, petroleum coke powder, glass fiber powder, and C5 / C9 petroleum resin; process parameter vector. This corresponds to key process parameters such as stirring time and settling temperature; the production parameter optimization unit then performs a reverse optimization task based on the above model; the process is as follows: The target product performance required by national standards or customer orders is then optimized. The ultimate goal of this optimization is set as the raw material feature vector of the current batch obtained by the raw material characteristic sensing unit. As a prediction model Given a known, fixed input parameter, a global optimization algorithm, such as Particle Swarm Optimization (PSO), is used to iteratively search the formula parameter vector within the feasible region of the preset formula and process parameters. This feasible region is a hard constraint boundary determined based on equipment physical limits, safety production regulations, and historical experience data. and process parameter vector The combination; Global optimization algorithms are a class of heuristic algorithms designed to find the global optimum rather than a local optimum for complex problems. Their purpose is to avoid getting trapped in local optima during the optimization process, thereby maximizing the chances of finding the theoretically optimal combination of production parameters. In each iteration, the algorithm selects a set of candidate... With fixed Input together into the prediction model In the process, the prediction performance is obtained. The goal of the algorithm is to find a solution that satisfies the objective function. The solution is to find the formulation and process parameters that minimize the error between the model's predicted output and the optimization objective, such as the L2 norm. When the algorithm converges or reaches the preset number of iterations, the optimal solution it finds is the optimal recipe parameter vector. and optimal process parameter vector If, within the feasible region, the algorithm cannot find a solution that makes the error between the model's predicted output and the optimization objective less than a preset acceptable threshold, the system will determine that the current raw materials cannot achieve the target performance while meeting cost and safety constraints, and will issue an alarm to the operator, suggesting adjustments to the target performance or isolation of the batch of raw materials. The preset acceptable threshold is determined based on the tolerance range of various performance indicators of the product in national standards or customer orders. For example, it can be set to 80% of the allowable tolerance range of the target performance indicator to ensure that the optimization results have a sufficiently high degree of acceptance in engineering practice. This invention can make maximum use of industrial by-products with poor characteristics, and significantly reduce dependence on expensive raw materials while ensuring that the final product performance meets the standards. To ensure the reliability of the model's predictions, the system also includes an input verification module. This module verifies the input vector provided by the raw material characteristic sensing unit. When the value of one or more dimensions exceeds the distribution range of historical training data, the system will trigger an alert and initiate a conservative, validated baseline production scheme to prevent unreliable prediction model output from causing production anomalies.

[0025] Example 5:

[0026] The process closed-loop control unit performs the following steps: real-time monitoring of process state variables at key nodes of the production line; determination of the error between the real-time monitored value of the process state variable and the initial set point; and calculation of the output adjustment of the production line actuator based on the error using a proportional-integral-derivative control algorithm. This embodiment, based on Embodiment 1, defines in detail the specific execution steps of the process closed-loop control unit; its purpose is to transform the theoretically optimal process parameter vector generated in the previous step. It accurately translates into actual production line operations, effectively suppressing various interferences in the production process; The process closed-loop control unit strictly executes the following steps to form a continuously operating feedback control loop: the system will use the optimal formula parameter vector output by the production parameter optimization unit... Send to the automated batching system to complete the precise feeding and premixing of the current batch; vectorize the optimal process parameters. The various parameter values, such as target temperature and target speed, serve as the initial setpoints for the controllers of each key actuator on the production line. ; Deploy corresponding sensors at key nodes in the production line, such as mixing tanks and reaction vessels, to monitor process state variables. High-frequency, real-time monitoring is required; process state variables refer to key physical or chemical quantities that need to be precisely controlled during the production process, such as the real-time viscosity of a mixture. Real-time temperature Its function is to reflect the actual state of the production process in real time; The control system continuously transmits real-time monitoring values. and the corresponding initial setpoint Compare the two and calculate the error between them. Based on the calculated error The mature proportional-integral-derivative PID control algorithm in the field of industrial control is used to calculate the output adjustment of the production line actuators. The proportional-integral-derivative (PID) control algorithm is a classic feedback control algorithm that combines proportional, integral, and derivative control laws. Its function is to quickly, stably, and without steady-state error eliminate control errors. Its calculation formula is: in, , , These represent the proportional, integral, and derivative gain coefficients in a PID controller, respectively, and their specific dimensions depend on the error signal. and output adjustment amount The physical units must be consistent across the left and right sides of the formula; for example, if the error unit is °C and the output is a dimensionless percentage, then... The dimensions are , The dimensions are , The dimensions are These coefficients are not set arbitrarily, but are systematically determined by conducting step response tests on the production process and combining mature engineering tuning methods such as Ziegler-Nichols, in order to ensure the speed and stability of the control loop.

[0027] Example 6: The model self-evolution unit performs the following steps: obtaining the actual finished product performance indicators in the quality inspection process and the predicted performance values ​​output by the data-driven prediction model; calculating the prediction deviation between the actual finished product performance indicators and the predicted performance values; and normalizing the prediction deviation to generate a model correction factor. The model self-evolution unit is further used to: use the model correction factor as sample weights and archive it together with the production data records of this batch; when the accumulated production data records reach a preset number, it triggers the retraining of the data-driven prediction model; the retraining realizes the weighted update of the data-driven prediction model by introducing the sample weights into the loss function. Based on Example 1, this embodiment deepens and refines the execution steps and internal logic of the model's self-evolving unit; its purpose is to construct a complete, self-improving learning loop to ensure the long-term accuracy of the data-driven prediction model and its adaptability to environmental changes. The execution process of the model self-evolutionary unit is divided into two closely connected stages: the generation of the model correction factor and the weight-based model retraining. To further clarify, after each batch of production is completed, this unit performs the following steps to generate model correction factors: obtaining the actual finished product performance index vector of that batch of products from the quality inspection stage. Simultaneously, it retrieves the predicted performance value vector generated by the data-driven prediction model for this batch before production. Calculate the difference between these two vectors, i.e., the prediction bias. The prediction bias is normalized to generate a dimensionless model correction factor. The calculation of the model correction factor incorporates the concept of weighted learning, aiming to transform the prediction error into an effective quantified weight to guide model updates; its calculation formula is as follows: in, It is a dimensionless model correction factor calculated for the kth production batch. Its value range is between [0,1). The larger the value, the larger the model prediction error. It is the L2 norm used to calculate the size of a vector; It is a positive, dimensionless learning rate hyperparameter used to balance the magnitude and stability of model updates. Its value is selected based on independent calibration datasets and optimized through cross-validation. It is a very small normal number, such as This is used to avoid division by zero errors when the measured performance is zero, so as to ensure the robustness of the calculation; Based on the above factors, the model self-evolutionary unit further utilizes these factors to achieve weighted updates of the model: the calculated model correction factor... As the sample weight of the production data records for this batch, it is related to the complete production data records for this batch. The data is archived together and stored in the historical database; when a newly archived data record meets a preset trigger condition, the data-driven prediction model is automatically triggered. The retraining procedure; during the retraining process, the system adjusts the sample weights, i.e., the model correction factors. This is introduced into the loss function of a machine learning model; the loss function is a function used in machine learning to measure the difference between the model's predictions and the true values, and the goal of model training is to minimize this function; by introducing weights, we can... Samples with larger values, i.e., those with larger previous prediction errors, have a greater impact on the updating of model parameters.

[0028] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention; any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0029] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A cloud-based paint production control system, characterized in that, include: The raw material characteristic sensing unit is used to acquire key physicochemical parameters of incoming raw materials and construct a raw material feature vector based on these key physicochemical parameters. The volatility assessment unit is used to calculate the volatility index between the raw material characteristic vector and the preset benchmark reference value, and compare the volatility index with the preset trigger threshold. When the volatility index is greater than the trigger threshold, a dynamic adjustment instruction for the formula is generated. The production parameter optimization unit is used to respond to the dynamic adjustment command of the formula and, based on the data-driven prediction model, solve in reverse to obtain the optimal formula parameter vector and the optimal process parameter vector. The process closed-loop control unit is used to set the optimal process parameter vector as the initial set point of the production process and to adjust the production process in real time through a closed-loop feedback control algorithm. The model self-evolution unit is used to calculate the prediction deviation between the actual finished product performance index and the predicted performance value after production is completed, generate the model correction factor, and use the model correction factor to perform weighted retraining of the data-driven prediction model.

2. The cloud-based paint production control system according to claim 1, characterized in that, The key physicochemical parameters include chemical component concentration, particle size distribution, and dynamic viscosity. The raw material characteristic sensing unit determines the chemical component concentration through near-infrared spectroscopy analysis, determines the particle size distribution through laser particle size analysis, and determines the dynamic viscosity through online viscosity measurement.

3. The cloud-based paint production control system according to claim 1, characterized in that, The volatility assessment unit retrieves the average parameter values ​​of the gold batch raw materials calibrated in historical data and sets them as the benchmark reference value; and uses a weighted Euclidean distance model to quantify the deviation between the raw material feature vector and the benchmark reference value to generate a volatility index.

4. The cloud-based paint production control system according to claim 1, characterized in that, The production parameter optimization unit sets the acquired target product performance as the optimization objective and uses the raw material feature vector as the fixed input of the data-driven prediction model. It then uses a global optimization algorithm to iteratively solve for the formula parameter vector and process parameter vector that minimize the error between the prediction output of the data-driven prediction model and the optimization objective, thereby generating the optimal formula parameter vector and the optimal process parameter vector.

5. A cloud-based paint production control system according to claim 4, characterized in that, The data-driven prediction model is a nonlinear mapping model established through supervised learning training of historical production data. It is used to reveal the intrinsic relationship between raw material characteristics, formula parameters, and process parameters and the performance indicators of the final product.

6. The cloud-based paint production control system according to claim 1, characterized in that, The process closed-loop control unit performs the following steps: real-time monitoring of process state variables at key nodes of the production line; determining the error between the real-time monitored value of the process state variable and the initial set point; and, based on the error, calculating the output adjustment amount of the production line actuator using a proportional-integral-derivative control algorithm.

7. A cloud-based paint production control system according to claim 1, characterized in that, The model self-evolution unit performs the following steps: obtaining the actual finished product performance indicators in the quality inspection process and the predicted performance values ​​output by the data-driven prediction model; calculating the prediction deviation between the actual finished product performance indicators and the predicted performance values; The prediction bias is then normalized to generate a model correction factor.

8. A cloud-based paint production control system according to claim 7, characterized in that, The model self-evolution unit is further used to: use the model correction factor as sample weights and archive it together with the production data records of the batch; when the accumulated production data records reach a preset number, trigger the retraining of the data-driven prediction model; the retraining realizes the weighted update of the data-driven prediction model by introducing the sample weights into the loss function.

Images