Intelligent optimization method for mold production parameters based on cloud platform

Abstract

The invention discloses a mold production parameter intelligent optimization method based on a cloud platform, and relates to the technical field of mold manufacturing and industrial intelligent optimization. Comprising the steps that S1, a cloud platform collects equipment operation parameters, material performance parameters and environment state parameters involved in the mold production process from a plurality of heterogeneous data sources in real time; s2, the cloud platform performs standardized preprocessing and multi-dimensional feature fusion on the collected parameters, and constructs a global perception data model covering the whole process of mold production; and S3, the cloud platform constructs a multivariable and strong-coupling mold production dynamic prediction model through a deep learning technology based on the global perception data model. According to the method, equipment operation, material performance, environment state parameters and cloud historical data of multiple heterogeneous data sources are collected in real time through the cloud platform, a global perception data model is constructed through standardized preprocessing and multi-dimensional feature fusion, and the problems that a traditional method is limited by local information and lacks global collaborative analysis are solved.

Description

Technical Field

[0001] This invention relates to the field of mold manufacturing and industrial intelligent optimization technology, specifically to a cloud platform-based intelligent optimization method for mold production parameters. Background Technology

[0002] Mold manufacturing is a core component of the manufacturing industry, and the setting of its processing parameters directly affects product precision, production efficiency, and cost. Traditional mold manufacturing typically relies on engineers' experience or local computer-aided engineering software for parameter optimization, such as simulating key variables like temperature, pressure, and cooling time during the injection molding process using finite element analysis. With the development of industrial internet technology, some companies are beginning to utilize local servers or small data centers to conduct preliminary analysis of production data to adjust parameters such as injection speed and holding pressure curves, thereby achieving quality control and energy consumption optimization goals.

[0003] However, existing optimization methods based on local or distributed computing struggle to address the multivariate, tightly coupled, and dynamically changing parameter adjustment needs in mold production. Because mold production involves real-time interactions of multiple factors such as injection molding machine performance, material properties, and environmental temperature and humidity, traditional optimization models often rely solely on static historical data or local information from a single production line, lacking the ability to perceive and collaboratively analyze the overall production status. This results in insufficient adaptability of parameter optimization results under complex operating conditions, an inability to respond in real-time to sudden fluctuations or cross-equipment collaborative needs, and consequently, problems such as unstable mold forming quality, high energy consumption, and increased potential equipment wear and tear. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a cloud-based intelligent optimization method for mold production parameters. This method solves the problems of existing technologies, which struggle to cope with the multivariate and strongly coupled dynamic changes in mold production, lack of global awareness, and consequently lead to unstable molding quality, high energy consumption, and significant equipment wear.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a cloud platform-based intelligent optimization method for mold production parameters, comprising: S1. The cloud platform collects equipment operating parameters, material performance parameters, and environmental status parameters involved in the mold production process from multiple heterogeneous data sources in real time. S2, the cloud platform performs standardized preprocessing and multi-dimensional feature fusion on the collected parameters to build a global perception data model covering the entire mold production process; S3 and the cloud platform are based on a global perception data model and use deep learning technology to build a multivariate, strongly coupled dynamic prediction model for mold production. The prediction model can evaluate the mold forming quality, predict energy consumption trends, and identify potential equipment risks in real time. S4. The cloud platform utilizes a multi-objective intelligent optimization algorithm to calculate and generate the optimal mold production parameter optimization scheme in real time based on the evaluation results output by the dynamic prediction model and the preset production optimization objectives. The S5 cloud platform distributes the optimal mold production parameter optimization scheme to the corresponding mold production execution equipment, driving the mold production execution equipment to achieve adaptive and dynamic adjustment of production parameters.

[0006] Furthermore, the cloud platform in S1 collects equipment operating parameters, material performance parameters, and environmental status parameters involved in the mold production process from multiple heterogeneous data sources in real time, specifically including: The cloud platform acquires real-time data on mold cavity temperature, injection pressure, holding pressure, cooling time, injection speed, screw position, clamping force, raw material batch number, material melt index, ambient temperature, ambient humidity, and product defect rate from injection molding machines, mold temperature sensors, mold pressure sensors, coolant flow meters, material feeding systems, ambient temperature and humidity sensors, and machine vision inspection systems via industrial communication interfaces.

[0007] Furthermore, the cloud platform in S2 performs standardized preprocessing and multi-dimensional feature fusion on the collected parameters, specifically including: The cloud platform performs data cleaning on the collected parameters, including removing duplicates, handling missing data, and correcting abnormal data. The cloud platform normalizes and standardizes the cleaned parameters to eliminate the influence of data with different dimensions; The cloud platform performs time-series alignment of parameters from different sources and timestamps, and uses principal component analysis or independent component analysis to perform feature dimensionality reduction, extracting an integrated feature vector that represents the core state of mold production.

[0008] Furthermore, the cloud platform in S3, based on a globally perceptive data model, constructs a multivariate, strongly coupled dynamic prediction model for mold production using deep learning technology, specifically including: The cloud platform uses historical production batch data, known equipment parameter configurations, and corresponding product quality inspection reports to construct a labeled dataset; The cloud platform uses convolutional neural network or recurrent neural network structure to train integrated feature vectors and learn the nonlinear mapping relationship between mold production parameters and molding quality, energy consumption and equipment status. The cloud platform optimizes the performance of the prediction model through cross-validation and hyperparameter tuning, ensuring its accurate perception of dynamic changes in mold production.

[0009] Furthermore, the cloud platform in S4 utilizes a multi-objective intelligent optimization algorithm to calculate and generate the optimal mold production parameter optimization scheme in real time based on the evaluation results output by the dynamic prediction model and the preset production optimization objectives, specifically including: The cloud platform uses the predicted values ​​of mold forming quality, energy consumption trend, and equipment risk index output by the prediction model as inputs to the optimization algorithm; The cloud platform is configured with a multi-objective optimization function that includes maximizing product qualification rate, minimizing unit product energy consumption, minimizing equipment wear rate, and shortening production cycle. The cloud platform employs reinforcement learning algorithms based on advanced search strategies such as particle swarm optimization, genetic algorithms, or Bayesian optimization to dynamically search for the optimal combination of injection speed curve, holding pressure curve, mold heating / cooling temperature curve, and cooling time, while satisfying equipment physical constraints and material process constraints.

[0010] Furthermore, the cloud platform in S5 distributes the optimal mold production parameter optimization scheme to the corresponding mold production execution equipment, driving the mold production execution equipment to achieve adaptive and dynamic adjustment of production parameters, specifically including: The cloud platform converts mold production parameter optimization schemes into executable control commands through an industrial communication interface based on the OPCUA protocol; The cloud platform transmits control commands in real time to the controllers of mold production execution equipment such as injection molding machines, mold temperature controllers, and cooling machines; The controller of the mold production execution equipment receives control commands and automatically adjusts the parameters of each stage of the injection molding process according to the command requirements, so as to achieve precise closed-loop control of the production process.

[0011] Furthermore, the data source in S1 further includes a cloud-based historical database, specifically including: The cloud platform obtains long-term production data for different mold types, material grades, and product batches from a historical database in the cloud. The cloud platform integrates long-term production data with real-time collected data to support offline training and periodic updates of the mold production dynamic prediction model.

[0012] Furthermore, the prediction model possesses self-learning and adaptive capabilities, specifically including: The cloud platform periodically or when it detects significant deviations in the production process triggers online retraining of the prediction model; Online retraining uses the latest real-time production and feedback data to iteratively update the parameters of the prediction model to adapt to dynamic changes in the production environment and equipment aging.

[0013] Furthermore, the search strategy of the intelligent optimization algorithm specifically includes: Before generating parameter optimization schemes, the algorithm first uses sensitivity analysis to determine the key parameters that have the greatest impact on the optimization objective; During the search process, the algorithm uses historical optimization experience as prior knowledge to accelerate the convergence of the optimal solution; The algorithm weighs multiple optimization objectives to generate a set of Pareto optimal solutions for production decision-makers to choose from.

[0014] Furthermore, the adaptive, dynamic adjustment further includes: After the mold production parameters are adjusted, the cloud platform continuously monitors the actual operation effect and product quality of the mold production process; The cloud platform uses actual operational results and product quality data as feedback signals to verify the effectiveness of parameter optimization schemes in real time. When the cloud platform detects a deviation between the production results and the optimization goals, it immediately starts a new round of parameter optimization cycle to ensure continuous optimization of the mold production process.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention collects data from multiple heterogeneous data sources in real time through a cloud platform, including equipment operation, material performance, environmental status parameters, and historical data from the cloud. After standardized preprocessing and multi-dimensional feature fusion, a globally perceptive data model is constructed, overcoming the limitations of traditional methods in terms of local information and lack of global collaborative analysis. A dynamic prediction model is then built using deep learning to accurately assess molding quality, predict energy consumption trends, and assess equipment risks, addressing the highly coupled nature of multi-variable mold production. Combined with a multi-objective intelligent optimization algorithm, optimal parameter schemes are generated to balance product qualification rate, energy consumption, equipment wear, and production cycle. Finally, instructions are issued via the OPCUA protocol with closed-loop feedback adjustments to respond to production fluctuations in real time, solving the problems of unstable molding quality, high energy consumption, and significant equipment wear caused by insufficient adaptability in traditional optimization methods, thus achieving continuous optimization of mold production. Attached Figure Description

[0016] Figure 1 This is a flowchart of the method of the present invention. 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] Please see Figure 1 This invention provides a cloud-based intelligent optimization method for mold production parameters, comprising: S1. The cloud platform collects equipment operating parameters, material performance parameters, and environmental status parameters involved in the mold production process from multiple heterogeneous data sources in real time. S2, the cloud platform performs standardized preprocessing and multi-dimensional feature fusion on the collected parameters to build a global perception data model covering the entire mold production process; S3 and the cloud platform are based on a global perception data model and use deep learning technology to build a multivariate, strongly coupled dynamic prediction model for mold production. The prediction model can evaluate the mold forming quality, predict energy consumption trends, and identify potential equipment risks in real time. S4. The cloud platform utilizes a multi-objective intelligent optimization algorithm to calculate and generate the optimal mold production parameter optimization scheme in real time based on the evaluation results output by the dynamic prediction model and the preset production optimization objectives. The S5 cloud platform distributes the optimal mold production parameter optimization scheme to the corresponding mold production execution equipment, driving the mold production execution equipment to achieve adaptive and dynamic adjustment of production parameters.

[0019] Specifically, Phase S1 involves real-time acquisition of multi-source parameters. The cloud platform first establishes a heterogeneous data acquisition network covering the entire mold production scenario, connecting various data sources through industrial communication interfaces. These sources include injection molding machines, mold temperature sensors, mold pressure sensors, coolant flow meters, material feeding systems, environmental temperature and humidity sensors, and machine vision inspection systems on the production site. It also accesses a cloud-based historical database to obtain long-term production data. The acquisition process employs a real-time transmission mechanism, specifically acquiring and uploading key parameters every 50 milliseconds. This ensures the simultaneous acquisition of equipment operating parameters such as mold cavity temperature, injection pressure, holding pressure, cooling time, injection speed, screw position, and clamping force; material performance parameters such as raw material batch number and melt flow index; and environmental status and quality feedback parameters such as ambient temperature, ambient humidity, and product defect rate. This multi-source acquisition method breaks down the data silos of traditional single production lines or local systems, laying the foundation for comprehensive analysis.

[0020] Phase S2 involves data preprocessing and the construction of a global perception data model. The cloud platform first performs standardized preprocessing on the collected multi-source parameters, starting with data cleaning. During data cleaning, duplicate data is removed by comparing data collection timestamps with duplicate identifiers; for missing data, linear interpolation or mean imputation is used; for outlier data, the 3σ principle is applied for correction, which considers data exceeding the mean ± 3 standard deviations as outliers. Next, normalization is performed to eliminate the influence of different units. For example, for parameters such as injection pressure in MPa and cooling time in seconds, the min-max normalization formula is used: ; in These are the normalized parameter values. The original value of the parameter. This is the minimum value of the parameter in the historical data. This parameter represents the maximum value in historical data. Subsequently, parameters from different sources and timestamps are aligned according to the production cycle to ensure a one-to-one correspondence between equipment, material, and environmental parameters at the same time point. Finally, principal component analysis or independent component analysis is used for feature dimensionality reduction to extract integrated feature vectors representing the core state of mold production. For example, eight core features are extracted from 20 original parameters to construct a globally perceptual data model covering the entire mold production process, enabling overall control of the production status.

[0021] Phase S3 involves building a dynamic prediction model. The cloud platform uses the integrated feature vectors from the globally perceptual data model as a foundation, combined with deep learning technology to construct a multivariate, strongly coupled prediction model. First, a labeled dataset is built, containing integrated feature vectors from historical production batches, corresponding batch equipment parameter configurations, product quality inspection reports, energy consumption records, and equipment status records. Equipment parameter configurations include injection speed settings and holding pressure settings; product quality inspection reports cover dimensional accuracy and surface defect conditions; energy consumption records include power consumption per unit product; and equipment status records involve equipment vibration values ​​and component temperatures. Next, a suitable deep learning network structure is selected, such as using a recurrent neural network (LSTM) to process time-series production parameters, or a convolutional neural network (CNN) to extract spatial correlation features between parameters. The network is trained using a backpropagation algorithm to learn the nonlinear mapping relationship between mold production parameters and molding quality, energy consumption, and equipment status. During training, a 5-fold cross-validation method was used to divide the training and validation sets. At the same time, a grid search method was used to optimize hyperparameters, such as adjusting the number of hidden layer neurons and learning rate of the LSTM, to ensure that the model can accurately perceive the dynamic changes in mold production, achieve real-time evaluation of molding quality, accurate prediction of energy consumption trends, and effective identification of potential equipment risks.

[0022] Phase S4 involves multi-objective intelligent optimization. The cloud platform uses the molding quality predictions (such as predicted pass rate), energy consumption trend predictions (such as predicted unit product energy consumption), and equipment risk indices (such as predicted equipment wear rate) output by the dynamic prediction model as inputs to the optimization algorithm, setting a multi-objective optimization function. Since the dimensions of each optimization objective are different, each objective value is first normalized, and then a weighted summation objective function is constructed. ; in To optimize the objective function value, This represents the normalized product qualification rate, with a maximum value of 1. This represents the normalized energy consumption per unit of product, with a minimum value of 0. The normalized equipment wear rate has a minimum value of 0. The normalized production cycle has a minimum value of 0. The weights for each objective are denoted by , and their sum is 1. The optimization process employs a reinforcement learning algorithm based on particle swarm optimization, simultaneously satisfying both equipment physical constraints and material process constraints. Equipment physical constraints include, for example, an injection pressure not exceeding 180 MPa, while material process constraints include, for example, maintaining the material melt temperature between 180-220℃. The algorithm dynamically searches for the optimal parameter combinations, such as generating optimal injection speed curves, holding pressure curves, mold heating / cooling temperature curves, and cooling time, thus forming the optimal production parameter optimization scheme.

[0023] Phase S5 involves parameter distribution and adaptive adjustment. The cloud platform, through an industrial communication interface based on the OPCUA protocol, converts the optimal parameter optimization scheme into executable control commands. For example, it converts the injection speed curve into pulse signal commands that the injection molding machine controller can recognize. These control commands are then transmitted in real-time to the corresponding production execution equipment controllers, including injection molding machines, mold temperature controllers, and cooling systems. Upon receiving the commands, the equipment controllers automatically adjust parameters at each stage of the production process. For instance, the mold temperature controller adjusts the mold temperature from the current 80℃ to 85℃ according to the command, and the cooling system adjusts the coolant flow rate to match the cooling time requirements. Simultaneously, the cloud platform continuously monitors equipment operating parameters and product quality, comparing actual data with the optimization targets. If a deviation is detected, a new optimization cycle is initiated, achieving adaptive dynamic adjustment of production parameters.

[0024] In this embodiment, the cloud platform in S1 collects equipment operating parameters, material performance parameters, and environmental status parameters involved in the mold production process from multiple heterogeneous data sources in real time, specifically including: The cloud platform acquires real-time data on mold cavity temperature, injection pressure, holding pressure, cooling time, injection speed, screw position, clamping force, raw material batch number, material melt index, ambient temperature, ambient humidity, and product defect rate from injection molding machines, mold temperature sensors, mold pressure sensors, coolant flow meters, material feeding systems, ambient temperature and humidity sensors, and machine vision inspection systems via industrial communication interfaces.

[0025] Specifically, the process of the cloud platform collecting parameters from multiple heterogeneous data sources in real time relies on standardized industrial communication interfaces and a comprehensive data collection range, as detailed below: First, determine the type of industrial communication interface. The cloud platform selects the appropriate interface for data transmission based on the communication protocols of the equipment on the production site. For example, for injection molding machines that support the industrial Ethernet protocol, an RJ45 Ethernet interface is used to establish a connection; for sensors distributed in different areas of the workshop, such as environmental temperature and humidity sensors, data is uploaded via 4G or 5G wireless communication modules, ensuring real-time communication between various data sources and the cloud platform.

[0026] Secondly, the data sources and specific content of each parameter are clarified. Mold cavity temperature and mold surface temperature are collected from a mold temperature sensor, with the mold cavity temperature measurement range typically 0-300℃; injection pressure, holding pressure, and internal mold cavity pressure are collected from a mold pressure sensor, with the injection pressure measurement range typically 0-250MPa; instantaneous and cumulative coolant flow rates are collected from a coolant flow meter, with the instantaneous flow rate measurement range typically 0-50L / min; injection speed, screw position, clamping force, and injection cycle time are collected from the injection molding machine's control system, with the injection speed measurement range typically 0-150mm / s and the screw position measurement range typically... The clamping force is typically measured in the range of 0-300mm, and the clamping force is typically measured in the range of 0-5000kN. The batch number and melt flow index of the raw materials are collected from the feeding system. The batch number is used to trace the source of the materials, and the melt flow index reflects the fluidity of the materials and is usually expressed in g / 10min. The ambient temperature and humidity in the production workshop are collected from the ambient temperature and humidity sensors. The ambient temperature is typically measured in the range of 0-40℃, and the ambient humidity is typically measured in the range of 20% to 80%RH. The product defect rate is collected from the machine vision inspection system by taking pictures of the products, identifying surface scratches, bubbles, and other defects, and calculating their percentage.

[0027] A synchronous sampling mechanism is used during the data acquisition process. Specifically, all parameters are collected at 50-millisecond intervals to ensure that the parameters at the same time point can correspond accurately. This provides a time-consistent data foundation for the subsequent construction of the global perception data model and avoids analysis errors caused by asynchronous parameter acquisition.

[0028] In this embodiment, the cloud platform in S2 performs standardized preprocessing and multi-dimensional feature fusion on the collected parameters, specifically including: The cloud platform performs data cleaning on the collected parameters, including removing duplicates, handling missing data, and correcting abnormal data. The cloud platform normalizes and standardizes the cleaned parameters to eliminate the influence of data with different dimensions; The cloud platform performs time-series alignment of parameters from different sources and timestamps, and uses principal component analysis or independent component analysis to perform feature dimensionality reduction, extracting an integrated feature vector that represents the core state of mold production.

[0029] Specifically, the cloud platform needs to perform standardized preprocessing of collected parameters and multi-dimensional feature fusion in an orderly manner step by step to ensure data quality and feature validity, as follows: The first step is data cleaning. The cloud platform first removes duplicate data from the collected raw parameters by comparing the collection timestamp, device number, and parameter value of each data point to delete completely duplicate records. For missing data, an appropriate imputation method is selected based on the parameter type. For example, for linearly varying data such as injection speed, linear interpolation is used to imput missing values. Specifically, the slope is calculated by using two adjacent valid data points before and after the missing value to deduce the missing value. For data with small random fluctuations such as ambient temperature, mean imputation is used. Specifically, the average value of the parameter over the next 10 sampling periods is used to imput missing values. For outlier data, the 3σ principle is used for correction. The mean value of the parameter in the historical 30-day data collection is first calculated. with standard deviation If a certain sample value exceeds If the range is exceeded, it is considered abnormal data and is used... Replace the outlier value with a new one to avoid interference from abnormal data in subsequent analysis.

[0030] The second step is normalization and standardization. Since the dimensions of different parameters differ significantly—for example, injection pressure is in MPa and cooling time is in seconds—the influence of these dimensions needs to be eliminated through processing. This implementation method uses the min-max normalization method to calculate the normalized value for each parameter, as shown in the following formula: ; in The normalized parameter value, with a range of [0,1]; These are the original sampled values ​​of the parameters; This is the minimum value of the parameter in the data collected over the past 6 months; This is the maximum value of this parameter in the data collected over the past 6 months. For example, the original value of the injection pressure for a certain batch is 120 MPa. It is 50 MPa. If the pressure is 200 MPa, then the normalized value is Ensure that all parameters are on the same order of magnitude.

[0031] The third step is time alignment and feature dimensionality reduction. First, the time granularity for time alignment is determined based on the production cycle. For example, using a complete mold production cycle as a unit, parameters collected at different timestamps within each cycle are uniformly interpolated to the same time node, achieving time synchronization of parameters from different sources. Then, Principal Component Analysis (PCA) is used for feature dimensionality reduction. First, the covariance matrix of all normalized parameters is calculated, then the eigenvalues ​​and eigenvectors of the covariance matrix are solved. Eigenvectors with eigenvalues ​​greater than 1 are selected as principal components, and the corresponding integrated feature vectors are extracted. For example, from 10 parameters—mold cavity temperature, injection pressure, holding pressure, cooling time, injection speed, screw position, clamping force, material melt index, ambient temperature, and ambient humidity—4 principal components are extracted using PCA to form an integrated feature vector, reducing data redundancy while retaining core information about the production status.

[0032] In this embodiment, the cloud platform in S3 is based on a globally perceptive data model and uses deep learning technology to construct a multivariate, strongly coupled dynamic prediction model for mold production, specifically including: The cloud platform uses historical production batch data, known equipment parameter configurations, and corresponding product quality inspection reports to construct a labeled dataset; The cloud platform uses convolutional neural network or recurrent neural network structure to train integrated feature vectors and learn the nonlinear mapping relationship between mold production parameters and molding quality, energy consumption and equipment status. The cloud platform optimizes the performance of the prediction model through cross-validation and hyperparameter tuning, ensuring its accurate perception of dynamic changes in mold production.

[0033] Specifically, the process of building a dynamic prediction model based on a globally aware data model in a cloud platform requires combining the construction of labeled datasets, deep learning network training, and model optimization, as detailed below: First, a labeled dataset is constructed. The cloud platform extracts integrated feature vectors from the past 12 months from the global perception data model, and simultaneously associates them with corresponding equipment parameter configurations, product quality inspection reports, energy consumption records, and equipment status records to form the labeled dataset. The equipment parameter configurations include the injection speed settings, holding pressure settings, mold temperature settings, and cooling time settings of the injection molding machine; the product quality inspection reports include the dimensional accuracy of the product, such as deviations in critical dimensions, and surface defect types such as scratches, bubbles, and defect rates; the energy consumption records include the total power consumption for each batch of production and the power consumption per unit product; and the equipment status records include the vibration value of the injection molding machine screw, the temperature of the heating coil, and the pressure of the hydraulic system. The dataset is divided into a training set and a validation set in a 7:3 ratio for model training and performance validation.

[0034] Next, a deep learning network structure is selected and trained. Based on the temporal characteristics of the parameters, this implementation chooses a Long Short-Term Memory (LSTM) network to construct the prediction model. LSTM networks can effectively capture long-term dependencies in temporal data, making them suitable for dynamic processes like mold production. The input layer of the LSTM network is an integrated feature vector, such as a 4-dimensional feature vector. It has two hidden layers with 64 neurons per layer. The output layer has three branches, outputting predicted molding quality values ​​(e.g., defect rate), predicted energy consumption trends (e.g., unit product energy consumption for the next hour), and equipment risk indices (e.g., equipment wear risk based on vibration values). The training process uses the Adam optimizer, and the loss function is the mean squared error (MSE), as shown in the following formula: ; in The loss value. This represents the number of samples in the training set. For the first The actual labeled value for each sample, such as the actual defect rate. For the model to the first The predicted value for each sample. The number of training iterations is set to 100 rounds. After each round, the loss value of the validation set is calculated. If the loss value of the validation set does not decrease for 10 consecutive rounds, training is stopped to avoid model overfitting.

[0035] Finally, model optimization was performed. A 5-fold cross-validation method was used to further optimize model performance. The labeled dataset was randomly divided into 5 non-overlapping subsets. Four subsets were used as the training set and one subset as the validation set in each iteration, repeated 5 times. The average validation loss was then calculated to ensure model stability under different data distributions. Simultaneously, hyperparameter tuning was performed using a grid search method. The candidate ranges for the number of hidden layer neurons were set to 32-128, the learning rate to 0.001-0.01, and the batch size to 16-64. The hyperparameter combination that minimized the validation loss was selected as the final model parameters.

[0036] In this embodiment, the cloud platform in S4 utilizes a multi-objective intelligent optimization algorithm to calculate and generate the optimal mold production parameter optimization scheme in real time based on the evaluation results output by the dynamic prediction model and the preset production optimization objectives. Specifically, this includes: The cloud platform uses the predicted values ​​of mold forming quality, energy consumption trend, and equipment risk index output by the prediction model as inputs to the optimization algorithm; The cloud platform is configured with a multi-objective optimization function that includes maximizing product qualification rate, minimizing unit product energy consumption, minimizing equipment wear rate, and shortening production cycle. The cloud platform employs reinforcement learning algorithms based on advanced search strategies such as particle swarm optimization, genetic algorithms, or Bayesian optimization to dynamically search for the optimal combination of injection speed curve, holding pressure curve, mold heating / cooling temperature curve, and cooling time, while satisfying equipment physical constraints and material process constraints.

[0037] Specifically, the process by which a cloud platform generates the optimal parameter scheme using a multi-objective intelligent optimization algorithm requires clearly defining the input conditions, optimization objectives, algorithm selection, and constraints, as detailed below: First, the input data for the optimization algorithm is determined. The cloud platform uses three core results output by the dynamic prediction model as input: the predicted molding quality, the predicted energy consumption trend, and the equipment risk index. The predicted molding quality is expressed as the predicted product pass rate, ranging from 0% to 100%; the predicted energy consumption trend is expressed as the predicted energy consumption per unit of product, in kW·h / piece; and the equipment risk index is expressed as the predicted equipment wear rate, ranging from 0% to 10%. Simultaneously, basic production parameters, such as raw material type and product model, are acquired to provide foundational information for optimization.

[0038] Next, a multi-objective optimization function and weights are defined. Since the optimization objectives include maximizing product yield, minimizing unit product energy consumption, minimizing equipment wear rate, and shortening production cycle, and these objectives have different dimensions, each objective value is first normalized. After normalization, the product yield rate... ,in The actual pass rate; normalized unit product energy consumption. ,in Actual energy consumption This represents the lowest energy consumption in history. This represents the highest energy consumption in history; after normalizing the equipment wear rate. ,in This represents the actual wear rate; after production cycle normalization. ,in For the actual period, This is the shortest period in history. This represents the longest historical period. A weighted multi-objective optimization function is then constructed: ; The weighting coefficients are set according to production needs. For example, when production focuses on product quality, the pass rate is weighted at 0.4. This weight is higher than other objectives to ensure that the optimization direction is consistent with actual needs.

[0039] Next, an optimization algorithm is selected and a search strategy is set. This implementation uses a reinforcement learning algorithm based on particle swarm optimization (PSO). The initial particle swarm size is set to 50, and each particle represents a set of parameters including injection speed, holding pressure, mold temperature, and cooling time. The particle position update formula is as follows: ; ; in For the first The particle in the first The speed update value on the dimension parameter, The inertia weight is set to 0.7. The learning factor is set to 2 for all of them. A random number between 0 and 1 For the first The historical best position of each particle This represents the globally optimal position for the entire particle swarm. For the first The particle in the first Current position on the dimension parameter This is the updated position.

[0040] Finally, the constraints in the optimization process are clearly defined. Equipment physical constraints include injection pressure not exceeding the equipment's rated pressure, e.g., 180 MPa; and clamping force not exceeding the equipment's maximum clamping force, e.g., 5000 kN. Material process constraints include maintaining the material's melt temperature within the required process range, e.g., 180-220℃ for PP material; and cooling time not less than the minimum time required for material solidification, e.g., 5 seconds. During the search process, if the parameter combination corresponding to a particle violates the constraints, its fitness value is set to the maximum value to prevent that combination from being selected as the optimal solution.

[0041] In this embodiment, the cloud platform in S5 sends the optimal mold production parameter optimization scheme to the corresponding mold production execution equipment, driving the mold production execution equipment to achieve adaptive and dynamic adjustment of production parameters, specifically including: The cloud platform converts mold production parameter optimization schemes into executable control commands through an industrial communication interface based on the OPCUA protocol; The cloud platform transmits control commands in real time to the controllers of mold production execution equipment such as injection molding machines, mold temperature controllers, and cooling machines; The controller of the mold production execution equipment receives control commands and automatically adjusts the parameters of each stage of the injection molding process according to the command requirements, so as to achieve precise closed-loop control of the production process.

[0042] Specifically, the process by which the cloud platform distributes the optimal parameter scheme to the production execution equipment and achieves adaptive adjustments relies on standardized communication protocols and closed-loop control mechanisms, as detailed below: First, the parameter scheme is converted into control commands. The cloud platform generates executable control commands based on the parameters in the optimal parameter scheme and the control protocols of different production equipment. For example, for the injection speed parameter of an injection molding machine, if the optimal scheme is "increase from 50mm / s to 80mm / s in the first 3 seconds, then maintain at 80mm / s", it is converted into pulse signal commands that the injection molding machine controller can recognize. Each pulse corresponds to the displacement of the screw, and the speed is dynamically adjusted by controlling the pulse frequency. For the temperature parameter of a mold temperature controller, if the optimal scheme is "maintain the mold cavity temperature at 85℃", it is converted into a PID control command for the mold temperature controller, specifying the target temperature and adjustment coefficient.

[0043] Secondly, command transmission is achieved through the OPCUA protocol. The cloud platform establishes an industrial communication link based on the OPCUA protocol, which supports standardized communication across vendors and devices, ensuring the reliability and real-time performance of command transmission. For example, the cloud platform's communication module connects to the industrial switch in the workshop via Ethernet, and the switch then connects to the controllers of the injection molding machine, mold temperature controller, and cooling machine, with command transmission latency controlled within 100 milliseconds to avoid parameter adjustment delays caused by transmission latency.

[0044] Next, the equipment controller performs parameter adjustments. Upon receiving control commands, the controllers of each production line automatically adjust the parameters for their respective stages. The injection molding machine controller adjusts the screw's speed and pressure according to the commands, achieving dynamic control of injection speed and holding pressure; the mold temperature controller stabilizes the mold temperature at the target value by adjusting the heating element power and coolant flow rate; and the cooling system controller adjusts the coolant circulation pump speed to match the cooling time requirements. During the adjustment process, the equipment controller provides real-time feedback of current parameter values ​​to the cloud platform, ensuring the adjustment process is monitorable.

[0045] Finally, precise closed-loop control of the production process is achieved. After parameter adjustments, the cloud platform continuously collects actual operating parameters of the equipment, such as actual injection speed, actual mold temperature, and product quality data, such as actual defect rate. This data is compared with the optimization target in real time. For example, if the actual product defect rate is 5% higher than the predicted defect rate, it is determined that the parameter adjustment effect has not met expectations. The cloud platform immediately starts a new round of parameter optimization cycle, from parameter collection in S1 to optimization calculation in S4, generating a new optimal solution and issuing it for execution, ensuring that the production process is always under precise control.

[0046] In this embodiment, the data source in S1 further includes a cloud-based historical database, specifically including: The cloud platform obtains long-term production data for different mold types, material grades, and product batches from a historical database in the cloud. The cloud platform integrates long-term production data with real-time collected data to support offline training and periodic updates of the mold production dynamic prediction model.

[0047] Specifically, S1's data source further incorporates a cloud-based historical database, improving model performance through the fusion of historical and real-time data, as follows: First, the content stored in the cloud-based historical database must be clearly defined. This database stores long-term production data, with a storage period of the past three years, and is divided into three categories based on data type. The first category contains production data for different mold types, such as the parameter configurations and operation records for each batch of injection molds and stamping molds. The second category contains production data for different material grades, such as the processing parameters and corresponding quality and energy consumption data for commonly used mold materials like polypropylene (PP), acrylonitrile-butadiene-styrene copolymer (ABS), and polycarbonate (PC). The third category contains production data for different product batches, with each batch containing complete production cycle parameters, quality inspection reports, and equipment maintenance records to ensure data integrity and traceability.

[0048] Secondly, historical and real-time data are integrated. While collecting real-time data in the S1 phase, the cloud platform extracts historical data matching the current production scenario from the cloud-based historical database. For example, if PP material is currently used in injection mold production, production data of injection molds using PP material over the past year will be extracted. The integration process employs a data alignment and supplementation mechanism. In terms of time, the timestamp format of historical data is unified to millisecond-level timestamps consistent with real-time data, achieving time-series alignment. In terms of parameters, if there is a short-term missing parameter in the real-time data, such as a 5-minute data gap due to sensor malfunction, the average value of that parameter under the same production scenario in the historical data is used to supplement it, ensuring data continuity.

[0049] Finally, fused data is used to support offline training and periodic updates of the model. The cloud platform regularly trains the dynamic prediction model offline, for example, once a month. During the training process, fused data is used: historical data accounts for 70%, recent real-time data accounts for 30%, and new production scenario data, such as production data of new molds and new materials, is added to improve the model's generalization ability. At the same time, additional updates to the model are triggered according to changes in production scenarios, such as the introduction of new equipment or the change of material suppliers. For example, when a new injection molding machine is introduced, the initial production data of the equipment, such as the first 10 batches, is immediately extracted, fused with historical data, and the model is retrained to ensure that the model can adapt to the new production conditions.

[0050] In this embodiment, the prediction model further possesses self-learning and adaptive capabilities, specifically including: The cloud platform periodically or when it detects significant deviations in the production process triggers online retraining of the prediction model; Online retraining uses the latest real-time production and feedback data to iteratively update the parameters of the prediction model to adapt to dynamic changes in the production environment and equipment aging.

[0051] Specifically, the dynamic prediction model further possesses self-learning and adaptive capabilities, achieving dynamic adaptation to production changes through online retraining, as follows: First, the triggering conditions for online model retraining are defined. These conditions are categorized into two types: periodic triggering and event triggering. Periodic triggering executes at fixed intervals, utilizing idle production periods, such as initiating online retraining once every Sunday morning to avoid impacting normal production. Event triggering executes when a significant deviation in the production process is detected, such as a product defect rate exceeding a preset threshold of 1.5% for three consecutive production batches, unit product energy consumption exceeding the historical average by 10%, or equipment vibration exceeding the normal range. In these cases, the cloud platform automatically determines a significant change in production status and immediately triggers retraining.

[0052] Secondly, the data sources and processing methods for online retraining are clearly defined. The data sources for retraining are the latest real-time production data and feedback data, including integrated feature vectors from the week prior to triggering retraining, actual product quality data (such as defect rate and dimensional deviation), actual energy consumption data (such as power consumption per unit product), and actual equipment status data (such as component temperature and vibration values). The data processing procedure is consistent with the preprocessing steps of S2, including data cleaning, normalization, temporal alignment, and feature dimensionality reduction, ensuring that the new data matches the model input format.

[0053] Next, online retraining and parameter updates of the model are performed. The retraining process uses incremental training, eliminating the need to retrain the entire model. Only some parameters are iteratively updated using new data; for example, for an LSTM prediction model, only the weights of the hidden layers and the biases of the output layer are updated. This method reduces computation and training time, typically keeping training time under 30 minutes. Mini-batch gradient descent is used during training, with a batch size of 32, a learning rate of 0.001, and 20 iterations. After each training iteration, the prediction error of the model on the new data is calculated. If the error drops below a preset threshold (e.g., mean squared error less than 0.01), training stops, and the updated model parameters are saved.

[0054] Finally, the performance of the retrained model was verified. The cloud platform used the retrained model to predict the latest 10 production batches of data, compared the prediction results with the actual data, and evaluated the model's prediction accuracy. If the prediction accuracy improved, the model was adopted as the new default prediction model; if the prediction accuracy did not improve or decreased, the model was reverted to the one before the update, the reasons for the data anomalies were analyzed, and retraining was performed again after the problem was resolved.

[0055] In this embodiment, the search strategy of the intelligent optimization algorithm specifically includes: Before generating parameter optimization schemes, the algorithm first uses sensitivity analysis to determine the key parameters that have the greatest impact on the optimization objective; During the search process, the algorithm uses historical optimization experience as prior knowledge to accelerate the convergence of the optimal solution; The algorithm weighs multiple optimization objectives to generate a set of Pareto optimal solutions for production decision-makers to choose from.

[0056] Specifically, the search strategy of intelligent optimization algorithms improves optimization efficiency and solution flexibility through sensitivity analysis, utilization of historical experience, and generation of Pareto optimal solution sets, as follows: First, sensitivity analysis was used to identify key parameters. Before generating the parameter optimization scheme, the cloud platform used the Sobol index analysis method to evaluate the impact of each production parameter on the optimization objective. The Sobol index is used to quantify the contribution of a single parameter or combination of parameters to the objective function; the larger the index, the more significant the parameter's impact on the objective. For example, the analysis showed that the Sobol index for injection speed was 0.35, for holding pressure it was 0.28, for ambient temperature it was 0.05, and for screw position it was 0.08. Clearly, the Sobol indices for injection speed and holding pressure were much higher than those for ambient temperature and screw position, thus identifying injection speed and holding pressure as key parameters. During the optimization process, the search precision for these two parameters was specifically set: 0.1 mm / s for injection speed and 0.5 MPa for holding pressure. This precision is higher than that of other parameters, such as ambient temperature (1℃), ensuring the optimization effect of the key parameters.

[0057] Secondly, historical optimization experience is leveraged to accelerate optimal solution convergence. The cloud platform establishes a historical optimization experience database, storing optimization schemes from the past year, including parameter combinations, optimization target values, and constraints, categorized by production scenario such as mold type and material grade. In a new optimization process, the algorithm first retrieves the historical optimal solution consistent with the current scenario from the database, using it as the initial global optimum position for the particle swarm optimization algorithm. Meanwhile, the initial particle swarm position is set within ±10% of the parameter values ​​of the historical best solution. For example, if the current production scenario is "PP material injection mold" and the injection speed of the historical best solution is 70 mm / s, then the initial particle swarm's injection speed search range is set to 63–77 mm / s. This avoids the algorithm searching in meaningless parameter regions and shortens the convergence time of the optimal solution by more than 30%.

[0058] Finally, a set of Pareto optimal solutions is generated for decision-making. Since objectives in multi-objective optimization may conflict—for example, increasing the pass rate might lead to increased energy consumption—the algorithm is not limited to a single optimal solution but generates a set of Pareto optimal solutions. A Pareto optimal solution is defined as one where no other solution can improve the performance of another objective without reducing the performance of at least one objective. For example, five sets of Pareto optimal solutions are generated: Solution 1 focuses on the pass rate (99% pass rate, 0.8 kW·h / piece energy consumption); Solution 2 focuses on energy consumption (97% pass rate, 0.6 kW·h / piece energy consumption); Solution 3 focuses on equipment protection (98% pass rate, 1.2% equipment wear rate); Solution 4 focuses on production efficiency (96% pass rate, 50-second production cycle); and Solution 5 is a comprehensive balanced solution (98% pass rate, 0.7 kW·h / piece energy consumption, 1.5% wear rate, 55-second cycle). The cloud platform presents these solutions to production decision-makers in a visual interface, such as tables or line graphs, allowing decision-makers to select the final solution to implement based on actual production needs.

[0059] In this embodiment, adaptive and dynamic adjustment further includes: After the mold production parameters are adjusted, the cloud platform continuously monitors the actual operation effect and product quality of the mold production process; The cloud platform uses actual operational results and product quality data as feedback signals to verify the effectiveness of parameter optimization schemes in real time. When the cloud platform detects a deviation between the production results and the optimization goals, it immediately starts a new round of parameter optimization cycle to ensure continuous optimization of the mold production process.

[0060] Specifically, adaptive dynamic adjustment ensures continuous optimization of the production process through continuous monitoring, feedback verification, and iterative optimization, as follows: First, continuous monitoring of the production process is achieved. After parameter adjustments, the cloud platform collects two types of data in real time through sensors and detection equipment distributed throughout the production site. The first type is the actual operating parameters of the equipment, including the actual injection speed, actual holding pressure, and actual mold temperature of the injection molding machine; the actual heating power of the mold temperature controller; and the actual coolant flow rate of the cooling machine. The data is collected at 100 millisecond intervals to ensure that dynamic changes in parameters can be captured. The second type is product quality data. The machine vision inspection system captures surface images of each product in real time, identifies defects such as scratches and bubbles, and calculates the defect rate. At the same time, the deviation values ​​of key dimensions of the product are detected by a dimensional measuring instrument. The quality data collection interval is synchronized with the production cycle. The specific interval can be referenced to the production cycle, for example, data is collected every 60 seconds, which corresponds exactly to one production cycle.

[0061] Next, the effectiveness of the feedback signals is verified. The cloud platform compares the collected actual data with the optimization targets to establish a verification indicator system. Regarding product quality, the actual defect rate is compared with the predicted defect rate; if the actual defect rate is lower than or equal to the predicted defect rate, the quality target is considered met. Regarding energy consumption, the actual energy consumption per unit product is compared with the predicted energy consumption; if the actual energy consumption is lower than or equal to the predicted energy consumption, the energy consumption target is considered met. Regarding equipment status, the actual equipment wear rate is compared with the predicted wear rate; if the actual wear rate is lower than or equal to the predicted wear rate, the equipment protection target is considered met. Regarding production efficiency, the actual production cycle is compared with the predicted cycle; if the actual cycle is shorter than or equal to the predicted cycle, the efficiency target is considered met. If all indicators meet the targets, the current parameter scheme is maintained; if any indicator fails to meet the targets, such as actual energy consumption exceeding predicted energy consumption by 8%, the parameter optimization scheme is deemed insufficiently effective, triggering a feedback signal.

[0062] Finally, a new round of parameter optimization is initiated. Upon receiving the feedback signal, the cloud platform immediately starts a new round of parameter optimization, following the same process as the initial optimization. First, multi-source parameter acquisition (S1) is executed, focusing on real-time data within the 10 minutes prior to the feedback signal trigger. Then, data preprocessing and global perception data model updates (S2) are performed. Next, the current production status is reassessed using the dynamic prediction model (S3), such as predicting energy consumption trends after parameter adjustments. Then, a new optimal parameter scheme is generated using the multi-objective intelligent optimization algorithm (S4). Finally, the new scheme is distributed to the equipment and adjustments are implemented (S5). The interval between optimization cycles is determined based on the degree of deviation. For example, if the deviation is small, the next round of optimization is initiated after 30 minutes; if the deviation is large, the next round of optimization is initiated immediately to ensure the production process can quickly return to its optimal state.

[0063] In summary, this invention collects data from multiple heterogeneous data sources in real time through a cloud platform, including equipment operation, material performance, environmental status parameters, and historical data from the cloud. After standardized preprocessing and multi-dimensional feature fusion, a globally perceptive data model is constructed, overcoming the limitations of traditional methods in terms of local information and lack of global collaborative analysis. A dynamic prediction model is then built using deep learning to accurately assess molding quality, predict energy consumption trends, and assess equipment risks, addressing the highly coupled nature of multi-variable mold production. A multi-objective intelligent optimization algorithm generates optimal parameter schemes, balancing product qualification rate, energy consumption, equipment wear, and production cycle. Finally, instructions are issued via the OPCUA protocol with closed-loop feedback adjustments to respond to production fluctuations in real time, solving the problems of unstable molding quality, high energy consumption, and significant equipment wear caused by insufficient adaptability in traditional optimization methods, thus achieving continuous optimization of mold production.

[0064] 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 process, method, article, or apparatus.

[0065] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A cloud platform-based mold production parameter intelligent optimization method, characterized in that, Comprise: S1, the cloud platform collects the equipment operation parameters, material performance parameters and environmental state parameters involved in the mold production process from multiple heterogeneous data sources in real time; S2, the cloud platform standardizes and pre-processes the collected parameters and multi-dimensional feature fusion, and constructs a global perception data model covering the whole process of mold production; S3, the cloud platform constructs a multivariate and strongly coupled mold production dynamic prediction model based on the global perception data model through deep learning technology, which can evaluate mold forming quality, predict energy consumption trend and identify potential equipment risks in real time; S4, the cloud platform uses multi-objective intelligent optimization algorithm to calculate and generate the optimal mold production parameter optimization scheme in real time according to the evaluation results output by the dynamic prediction model and the preset production optimization target; S5, the cloud platform sends the optimal mold production parameter optimization scheme to the corresponding mold production execution equipment to drive the mold production execution equipment to realize the adaptive and dynamic adjustment of production parameters. 2.The cloud platform-based mold production parameter intelligent optimization method of claim 1, wherein, In S1, specifically comprising: The cloud platform acquires mold cavity temperature, injection pressure, holding pressure, cooling time, injection speed, screw position, clamping force, raw material batch number, material melt index, environmental temperature, environmental humidity and product defect rate from injection molding machine, mold temperature sensor, mold pressure sensor, cooling liquid flowmeter, feeding system, environmental temperature and humidity sensor and machine vision detection system through industrial communication interface. 3.The cloud platform-based mold production parameter intelligent optimization method of claim 2, wherein, In S2, the cloud platform standardizes and pre-processes the collected parameters and multi-dimensional feature fusion, specifically comprising: The cloud platform performs data cleaning on the collected parameters, including removing duplicates, handling missing data and correcting abnormal data; The cloud platform normalizes and standardizes the cleaned parameters to eliminate the influence of different dimension data; The cloud platform aligns the parameters of different sources and time stamps in time sequence, and uses principal component analysis or independent component analysis method for feature dimension reduction to extract integrated feature vectors representing the core state of mold production. 4.The cloud platform-based mold production parameter intelligent optimization method of claim 3, wherein, In S3, the cloud platform constructs a multivariate and strongly coupled mold production dynamic prediction model based on the global perception data model through deep learning technology, specifically comprising: The cloud platform uses historical production batch data, known equipment parameter configuration and corresponding product quality detection report to construct a labeled data set; The cloud platform uses convolutional neural network or recurrent neural network structure to train the integrated feature vectors and learn the nonlinear mapping relationship between mold production parameters and forming quality, energy consumption and equipment state; The cloud platform optimizes the performance of the prediction model through cross-validation and hyperparameter tuning to ensure its accurate perception of dynamic changes in mold production.

5. The method of claim 4, wherein the method further comprises: In S4, specifically comprising: The cloud platform takes the mold forming quality prediction value, energy consumption trend prediction value and equipment risk index output by the prediction model as the input of the optimization algorithm; The cloud platform sets a multi-objective optimization function including maximizing product pass rate, minimizing unit product energy consumption, minimizing equipment wear rate and shortening production cycle; The cloud platform adopts a reinforcement learning algorithm based on high-level search strategies such as particle swarm optimization, genetic algorithm or Bayesian optimization, and dynamically searches for the optimal injection speed curve, holding pressure curve, mold heating / cooling temperature curve and cooling time combination under the condition of meeting the physical constraints of the equipment and the constraints of the material process.

6. The mold production parameter intelligent optimization method based on a cloud platform according to claim 5, characterized in that, The S5 specifically includes: The cloud platform converts the mold production parameter optimization scheme into executable control instructions through an industrial communication interface based on the OPCUA protocol; The cloud platform transmits the control instructions to the controllers of the mold production execution equipment such as injection molding machines, mold temperature machines and cooling machines in real time; The controller of the mold production execution equipment receives the control instructions and automatically adjusts the parameters of each link in the injection molding process according to the instructions to realize precise closed-loop control of the production process. 7.The cloud platform-based mold production parameter intelligent optimization method of claim 6, wherein, The data source in S1 further includes a cloud historical database, specifically including: The cloud platform obtains long-term production data of different mold types, different material brands and different product batches from the cloud historical database; The cloud platform fuses the long-term production data with the real-time collected data to support offline training and periodic updating of the mold production dynamic prediction model. 8.The cloud platform-based mold production parameter intelligent optimization method of claim 4, wherein, The prediction model further has self-learning and self-adaptive capabilities, specifically including: The cloud platform triggers online retraining of the prediction model periodically or when it detects significant deviations in the production process; Online retraining iteratively updates the parameters of the prediction model using the latest real-time production data and feedback data to adapt to the dynamic changes in the production environment and equipment aging. 9.The cloud platform-based mold production parameter intelligent optimization method of claim 5, wherein, The search strategy of the intelligent optimization algorithm specifically includes: Before generating the parameter optimization scheme, the algorithm first determines the key parameters that have the greatest impact on the optimization target through sensitivity analysis; During the search process, the algorithm uses historical optimization experience as prior knowledge to accelerate the convergence of the optimal solution; The algorithm weighs between multiple optimization targets to generate a set of Pareto optimal solutions for production decision-makers to choose from. 10.The cloud platform-based mold production parameter intelligent optimization method of claim 6, wherein, The adaptive and dynamic adjustment further includes: After adjusting the mold production parameters, the cloud platform continuously monitors the actual operation effect and product quality of the mold production process; The cloud platform uses the actual operation effect and product quality data as feedback signals to real-time check the effectiveness of the parameter optimization scheme; When the cloud platform detects deviations between the production effect and the optimization target, it immediately starts a new round of parameter optimization cycle to ensure continuous optimization of the mold production process.

Images