Production process regulation and control method and system based on big data

Abstract

The invention relates to the field of intelligent manufacturing, and discloses a production process regulation and control method based on big data, which comprises the following steps: collecting original data in a production process according to a sensor group, and carrying out data cleaning on the collected original data to obtain cleaned production process data; performing fusion processing according to the cleaned production process data to obtain a fused production process data set; establishing an initial prediction model according to the working condition classification data to obtain initial prediction model data corresponding to each working condition; the production process is regulated and controlled through big data analysis, the product quality can be monitored in real time, the process parameters are predicted, optimized and adjusted, so that the product quality is effectively improved, the quality deviation is reduced, key factors in the production process are accurately recognized and adjusted through the steps of data cleaning, fusion, classification, model optimization and the like, and the production efficiency is improved. Therefore, unnecessary production waste is reduced, and production efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing technology, and in particular to a production process control method and system based on big data. Background Art

[0002] Big data-based production process control refers to the use of big data technology to monitor, predict, and optimize various process parameters in real time. By collecting and analyzing large amounts of data generated during the production process, particularly sensor data, production logs, and quality inspection data, potential process issues and optimization opportunities can be identified, enabling refined management and control of the production process.

[0003] Existing equipment has the following shortcomings: traditional methods often rely on manual inspection or periodic testing and lack real-time monitoring, resulting in the inability to adjust product quality in a timely manner, and quality deviations are difficult to detect and correct in a timely manner. Traditional methods usually lack big data analysis and refined processing, making it difficult to identify key factors in the production process, which can easily lead to unnecessary production waste and inefficiency. Traditional methods are usually based on experience adjustment and cannot be effectively optimized through historical production data, resulting in the inability to achieve continuous and refined adjustment of process parameters and difficulty in achieving the optimal process state. Traditional methods rely on manual intervention for process adjustment and lack automation and intelligent support. The production process is easily affected by human factors, the adjustment response is slow, and the production process is unstable. Due to the lack of precise adjustment and optimization, traditional methods lead to a lot of waste of raw materials and production resources, a high proportion of defective products, and thus increase production costs. Traditional methods usually do not combine big data and artificial intelligence technologies. The management and optimization of the production process rely more on manual experience and manual operations, making it difficult to achieve intelligent manufacturing.

[0004] Therefore, this application now proposes a production process control method and system based on big data to solve the above problems. Summary of the Invention

[0005] The purpose of the present invention is to provide a production process control method and system based on big data to solve the problem that the above-mentioned traditional methods often rely on manual inspection or periodic inspection and lack real-time monitoring, resulting in the inability to adjust product quality in a timely manner and quality deviations are difficult to detect and correct in a timely manner. Traditional methods usually lack big data analysis and refined processing, making it difficult to identify key factors in the production process, and easily causing unnecessary production waste and inefficiency.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a production process control method based on big data, comprising: Collecting raw data during the production process according to the sensor group, performing data cleaning on the collected raw data to obtain cleaned production process data; and performing fusion processing on the cleaned production process data to obtain a fused production process data set; Identify and classify the fused production process data set to obtain working condition classification data; establish an initial prediction model based on the working condition classification data to obtain initial prediction model data corresponding to each working condition; evaluate and optimize the initial prediction model according to a cross-validation method to obtain optimized production process parameter prediction model data; Predicting product quality indicators based on the optimized production process parameter prediction model data to obtain product quality prediction data; comparing the product quality prediction data with a preset product quality target range to obtain product quality deviation data; and determining a production process parameter range that needs to be adjusted based on the product quality deviation data to obtain process parameter range data to be adjusted; Obtain historical production data similar to the current product; mine process parameter adjustment strategies related to product quality improvement from the historical production data based on the process parameter range data to obtain process parameter adjustment strategy data; make preliminary adjustments to the production process parameters based on the process parameter adjustment strategy data and the current production actual situation to obtain preliminary adjusted process parameter data; The process parameter data after preliminary adjustment is applied to the actual production process, and the multi-dimensional data of the production process is again collected in real time through the sensor group to obtain the adjusted production process data; the adjusted production process data is iteratively optimized until the product quality reaches the target range.

[0007] The iterative optimization process includes: cleaning and fusing the same data as the fused production process data set after adjustment to obtain the fused production process data set after adjustment; performing product quality prediction on the adjusted production process data according to the optimized production process parameter prediction model data to obtain adjusted product quality prediction data; and comparing the adjusted product quality prediction data with the product quality target range again, including: If the product quality reaches the target range, the process parameter adjustments, product quality changes, and related production data during the entire production process control process are recorded to obtain production process control record data; based on the production process control record data, the long-term stability and optimization direction of the production process are analyzed to obtain production process optimization trend data; based on the production process optimization trend data, a subsequent production process continuous improvement plan is formulated to obtain production process continuous improvement plan data; If the product quality still does not reach the target range, return to obtain historical production data similar to the current production product; based on the process parameter range data to be adjusted, mine the process parameter adjustment strategy related to product quality improvement from the historical production data to obtain process parameter adjustment strategy data: based on the process parameter adjustment strategy data, combined with the current production actual situation, make preliminary adjustments to the production process parameters to obtain preliminary adjusted process parameter data, re-mine the process parameter adjustment strategy based on the new product quality deviation data, and adjust the process parameters again.

[0008] The method includes collecting raw data in the production process according to the sensor group, performing data cleaning on the collected raw data to obtain cleaned production process data, and performing fusion processing on the cleaned production process data to obtain a fused production process data set, including: Performing spatiotemporal alignment calibration on the original time series data stream, and constructing a three-dimensional data mapping matrix according to the sensor installation position coordinates to obtain an initial spatiotemporal data set with spatial correlation; According to the improved dynamic time warping algorithm, the time series signals of each sensor channel in the initial spatiotemporal dataset are phase-synchronized and corrected to obtain a feature-enhanced spatiotemporal dataset. Construct a multi-sensor data fusion network based on the attention mechanism, input the feature-enhanced spatiotemporal dataset into the network for training to obtain a weighted fusion feature matrix; A spatiotemporal consistency check is performed based on the weighted fusion feature matrix to obtain a standardized fusion production process data set that has been integrity checked and anomaly filtered.

[0009] Among them, identification and classification are performed based on the fused production process data set to obtain working condition classification data; an initial prediction model is established based on the working condition classification data to obtain initial prediction model data corresponding to each working condition; and the initial prediction model is evaluated and optimized based on a cross-validation method to obtain optimized production process parameter prediction model data, including: Extract multi-dimensional time series features from the fused production process dataset and construct a three-level feature pyramid consisting of time-domain statistical features, frequency-domain energy features, and time-frequency joint features. Dynamically segment production conditions based on the hierarchical similarity of the feature pyramid to obtain condition classification data clusters with time-series evolution characteristics. For each operating condition classification data cluster, a dual-channel deep learning model is constructed. The pre-trained industrial process model parameters are initialized to the dual-channel network through a transfer learning mechanism to obtain an initial prediction model set with cross-operating condition generalization capabilities. Build a model evaluation framework based on adversarial verification, and train a generative adversarial network to simulate the distribution of real-world operating data to obtain an optimized prediction model with enhanced anti-interference capabilities; Deploying the optimized prediction model to the edge computing node, building a real-time working condition model routing mechanism, and building a closed-loop feedback loop to continuously inject the deviation value between the model prediction result and the actual production parameter into the model update queue.

[0010] The product quality prediction data is compared with a preset product quality target range to obtain product quality deviation data, including: The optimized production process parameter prediction model data is input into a production site edge computing terminal to build a real-time data flow pipeline; and product quality key indicators under the current working condition are predicted in advance according to dynamic simulation results of a virtual production environment to obtain multi-dimensional product quality prediction data containing a space-time evolution track; An adaptive quality threshold system is built, and a reinforcement learning algorithm is used to dynamically adjust the product quality target range according to real-time production environment parameters; the product quality prediction data is compared with the dynamically adjusted target range in multiple dimensions to obtain structured product quality deviation data containing deviation direction, deviation amplitude, and deviation duration; A deviation traceability module based on a causal reasoning network is developed; through network topology structure analysis, key process parameter links causing quality deviation are automatically identified to obtain a deviation traceability analysis report with causal relationship annotation; A closed-loop feedback loop of deviation data process parameters is built, and the key parameter links in the deviation traceability analysis report are mapped to a production control system; the process parameter correction amount is automatically calculated according to the deviation amplitude, and the corrected parameters are fed back to the optimized production process parameter prediction model to obtain a continuously improved quality optimization cycle.

[0011] The product quality deviation data is used to determine the production process parameter range that needs to be adjusted to obtain process parameter range data to be adjusted, including: A causal reasoning deviation traceability map is built, and key process parameter links causing quality deviation are located through node activation propagation analysis to obtain a deviation traceability report containing parameter influence weight and propagation path; A process parameter dynamic constraint map is built, and the key parameter links in the deviation traceability report are mapped to the constraint map to obtain a candidate parameter adjustment range that meets multiple constraint conditions; The candidate parameter adjustment range is input into an optimization model based on the NSGA-III algorithm to obtain a Pareto optimal parameter adjustment scheme set; A digital twin verification module of the parameter adjustment scheme is built; the parameter adjustment scheme is comprehensively evaluated based on simulation results to obtain process parameter range data to be adjusted that meets actual production requirements.

[0012] Among them, historical production data similar to the current production product is obtained; based on the range data of the process parameters to be adjusted, process parameter adjustment strategies related to product quality improvement are mined from the historical production data to obtain process parameter adjustment strategy data, including: Using natural language processing technology to parse the technical documentation and process requirements of current production products to obtain a structured product feature tag library. Based on this feature tag library, a hybrid recommendation algorithm is used to obtain a cross-dimensional historical production dataset that includes process parameter trajectories, quality fluctuation records, and equipment status logs. Perform deep cleaning based on the cross-dimensional historical production dataset; construct a multimodal feature encoder to map parameters to a unified feature space to obtain a structured process adjustment knowledge graph; The process parameter range data to be adjusted is used as query input to locate similar parameter adjustment scenarios in the process adjustment knowledge graph to obtain a multi-dimensional candidate strategy set that includes parameter coupling relationships, adjustment steps, and expected effects. The multi-dimensional candidate strategy set is injected into the virtual production line for parallel simulation and deduction; a multi-attribute decision matrix is ​​constructed to screen the optimal strategy to obtain a virtually verified process parameter adjustment strategy data package.

[0013] Among them, according to the process parameter adjustment strategy data, combined with the current actual production situation, the production process parameters are preliminarily adjusted to obtain the preliminarily adjusted process parameter data, including: Converting the process parameter adjustment strategy data into a structured operation instruction set; obtaining a parameter adjustment decision tree integrating the strategy status through a virtual reality data mapping engine; Input the parameter adjustment decision tree into a deep Q-network model for training; construct a reward function that considers multi-objective optimization to obtain a preliminary adjustment action sequence that takes into account both quality improvement and process stability; Convert the preliminary adjustment action sequence into instruction code that can be executed by the device; through the two-way closed-loop control of the digital twin and the physical system, realize the synchronization of virtual verification and actual adjustment; Construct an effect evaluation model based on the spatiotemporal attention mechanism to conduct real-time evaluation of the process parameters after preliminary adjustment, so as to obtain the process parameter data after preliminary adjustment that has undergone adaptive optimization.

[0014] The process parameter data after preliminary adjustment is applied to the actual production process, and the multi-dimensional data of the production process is collected again in real time through the sensor group to obtain the adjusted production process data, including: Convert the initially adjusted process parameter data into a digital instruction set executable by the equipment; develop a parameter deployment verification platform based on digital twins to obtain a virtual verification report that includes equipment thermodynamic behavior, material flow characteristics, and quality fluctuation trends; Build an industrial IoT gateway with real-time response capabilities to synchronously push digital instruction sets to each execution terminal on the production line through a time-sensitive network; build a real-time monitoring interface for parameter deployment, and overlay the display of the equipment operating status and expected results after parameter adjustment in the physical space; Build a multimodal data fusion center to collect adjusted production process data in real time through edge computing nodes, build a data fusion algorithm based on the spatiotemporal attention mechanism, and map multi-source heterogeneous data into a unified feature space to obtain a structured adjusted production process dataset that includes spatiotemporal evolution laws; The structured and adjusted production process data set is input into the long short-term memory network model; by comparing the deviation between the actual data and the virtual verification report, parameter fine-tuning instructions are automatically generated and fed back to the production equipment through the industrial Internet of Things gateway, forming a closed-loop optimization circuit of "parameter deployment-real-time monitoring-data fusion-dynamic calibration".

[0015] A production process control system based on big data, which is applicable to the above-mentioned production process control method based on big data, includes: a fusion processing unit, the fusion processing unit being configured to collect raw data during the production process according to the sensor group, perform data cleaning on the collected raw data to obtain cleaned production process data; and perform fusion processing on the cleaned production process data to obtain a fused production process data set; An evaluation and optimization unit, the evaluation and optimization unit being configured to identify and classify the fused production process data set to obtain operating condition classification data; establish an initial prediction model based on the operating condition classification data to obtain initial prediction model data corresponding to each operating condition; and evaluate and optimize the initial prediction model based on a cross-validation method to obtain optimized production process parameter prediction model data; a data prediction unit, the data prediction unit being configured to predict product quality indicators based on the optimized production process parameter prediction model data to obtain product quality prediction data; compare the product quality prediction data with a preset product quality target range to obtain product quality deviation data; and determine a production process parameter range that needs to be adjusted based on the product quality deviation data to obtain process parameter range data to be adjusted; A preliminary adjustment unit is configured to obtain historical production data similar to the currently produced product; mine process parameter adjustment strategies related to product quality improvement from the historical production data based on the process parameter range data to obtain process parameter adjustment strategy data; and perform preliminary adjustments to the production process parameters based on the process parameter adjustment strategy data and in combination with the current production actual situation to obtain preliminary adjusted process parameter data; An iterative optimization unit is used to apply the preliminarily adjusted process parameter data to the actual production process, and again collect multi-dimensional data of the production process in real time through the sensor group to obtain adjusted production process data; the adjusted production process data is iteratively optimized until the product quality reaches the target range.

[0016] Compared with the prior art, the present invention has the following beneficial effects: (1) The present invention regulates the production process through big data analysis, can monitor product quality in real time, and adjust process parameters through prediction and optimization, thereby effectively improving product quality and reducing quality deviations.

[0017] (2) The present invention enables the key factors in the production process to be accurately identified and adjusted through steps such as data cleaning, fusion, classification and model optimization, thereby reducing unnecessary production waste and improving production efficiency.

[0018] (3) The present invention mines historical production data and combines it with actual production conditions to continuously optimize and adjust process parameters, making the production process more refined and achieving the optimal process state.

[0019] (4) The present invention uses sensors to collect data in real time and continuously adjust the process. The system can automatically adjust the production process parameters based on real-time feedback, ensuring that the production process is always in the best state and reducing human intervention.

[0020] (5) The present invention can reduce the waste of raw materials and defective products in the production process by optimizing process parameters and improving production efficiency, thereby reducing production costs.

[0021] (6) The present invention provides a practical solution for intelligent manufacturing by combining big data with artificial intelligence technology, which can realize intelligent management and optimization of the production process and meet the development needs of modern manufacturing industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Fig. 1 A schematic diagram of the steps of the overall method in one embodiment of the present invention; Fig. 2 FIG. 1 is a schematic diagram of the system architecture structure of the overall system in one embodiment of the present invention.

[0023] In the figure: 1. Fusion processing unit; 2. Evaluation and optimization unit; 3. Data prediction unit; 4. Preliminary adjustment unit; 5. Iterative optimization unit. DETAILED DESCRIPTION

[0024] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0025] See also Figs. 1-2 The present invention provides a technical solution: a production process control method based on big data, comprising: Raw data from the production process is collected according to the sensor group, and the collected raw data is cleaned to obtain cleaned production process data; and fusion processing is performed on the cleaned production process data to obtain a fused production process data set; Identify and classify the fused production process data set to obtain working condition classification data; establish an initial prediction model based on the working condition classification data to obtain initial prediction model data corresponding to each working condition; evaluate and optimize the initial prediction model using the cross-validation method to obtain optimized production process parameter prediction model data; Predict product quality indicators based on the optimized production process parameter prediction model data to obtain product quality prediction data; compare the product quality prediction data with the preset product quality target range to obtain product quality deviation data; determine the production process parameter range that needs to be adjusted based on the product quality deviation data to obtain process parameter range data to be adjusted; Obtain historical production data similar to the current product; mine process parameter adjustment strategies related to product quality improvement from the historical production data based on the process parameter range data to obtain process parameter adjustment strategy data; make preliminary adjustments to the production process parameters based on the process parameter adjustment strategy data and the current production actual situation to obtain preliminary adjusted process parameter data; The process parameter data after preliminary adjustment is applied to the actual production process, and the multi-dimensional data of the production process is again collected in real time through the sensor group to obtain the adjusted production process data; the adjusted production process data is iteratively optimized until the product quality reaches the target range.

[0026] It should be noted that, during operation, the sensor group refers to multiple sensors used to monitor various parameters in the production process, such as temperature, humidity, pressure, and flow. These sensors collect data in real time, providing raw information for subsequent data analysis. Data cleaning is the process of processing raw data, primarily to remove erroneous, inconsistent, incomplete, or redundant data to ensure the accuracy and reliability of subsequent analysis. Data fusion integrates data from different sensors or sources, combining multi-dimensional information to produce a more comprehensive and accurate dataset. Condition classification data is generated by classifying operating conditions during the production process according to different operating states (such as normal, fault, and overload), generating a classified dataset for subsequent analysis. The initial prediction model is a preliminary model constructed based on the condition classification data and is used to predict key parameters or product quality indicators in the production process under different operating conditions. Cross-validation is a technique for evaluating model performance. A dataset is divided into multiple subsets and then used alternately for training and validation. The goal is to evaluate the model's performance on different datasets and select the optimal model. Optimized production process parameter prediction model data is refined through cross-validation and other optimization techniques to provide more accurate process parameter predictions. Product quality indicators are parameters used to measure product quality, such as product dimensions, strength, and stability. Predicting these indicators can help promptly identify potential production issues. Product quality deviation data is obtained by comparing actual product quality predictions with pre-set product quality target ranges. This data can help identify areas for improvement in the production process. Historical production data refers to data from past production processes, including production time, process parameters, and quality results. By analyzing historical data, strategies for improving production processes and product quality can be identified. Process parameter adjustment strategies are developed during the production process based on historical data analysis to achieve higher product quality and production efficiency. Iterative optimization is a process of repeated adjustment and improvement, typically performed over multiple rounds of production data, with the results of each round approaching or reaching the predetermined target values. In this way, process parameters are continuously optimized until quality targets are met. Smart manufacturing combines big data, artificial intelligence, the Internet of Things and other technologies to make data-driven intelligent decisions and optimize processes during the production process, thereby achieving automated and intelligent production management.

[0027] In one embodiment, the adjusted production process data and the fused production process data set are cleaned and fused to obtain an adjusted fused production process data set; product quality prediction is performed on the adjusted production process data based on the optimized production process parameter prediction model data to obtain adjusted product quality prediction data; the adjusted product quality prediction data is compared again with the product quality target range, including, if the product quality reaches the target range, recording the process parameter adjustment, product quality change and related production data during the entire production process control process to obtain production process control record data; based on the production process control record data, the long-term stability and optimization direction of the production process are analyzed. To obtain production process optimization trend data; based on the production process optimization trend data, formulate subsequent production process continuous improvement plans to obtain production process continuous improvement plan data; if the product quality still does not reach the target range, return to obtain historical production data similar to the current production products; based on the process parameter range data to be adjusted, mine the process parameter adjustment strategy related to product quality improvement from the historical production data to obtain process parameter adjustment strategy data: based on the process parameter adjustment strategy data, combined with the current production actual situation, make preliminary adjustments to the production process parameters to obtain preliminary adjusted process parameter data, re-mine the process parameter adjustment strategy based on the new product quality deviation data, and adjust the process parameters again.

[0028] This design, through data cleaning and fusion processing, yields an adjusted production process dataset. Next, the optimized predictive model is used to predict data quality and compare it against the target range to determine whether quality standards are met. If quality meets the target, the process adjustments and quality changes are recorded, and the stability and optimization trends of the production process are analyzed to provide a basis for continuous improvement. If quality does not meet the target, historical production data is reused to explore optimization strategies and adjust process parameters, forming a new adjustment cycle until quality reaches the target. The entire process emphasizes data feedback and optimization to ensure continuous improvement and production stability.

[0029] In one embodiment, raw data collected during a production process is collected by a sensor group, and data cleaning is performed on the collected raw data to obtain cleaned production process data; fusion processing is performed on the cleaned production process data to obtain a fused production process data set, including: performing spatiotemporal alignment calibration on the original time series data stream, and constructing a three-dimensional data mapping matrix based on the sensor installation position coordinates to obtain an initial spatiotemporal data set with spatial correlation; performing phase synchronization correction on the time series signals of each sensor channel in the initial spatiotemporal data set based on an improved dynamic time warping algorithm to obtain a feature-enhanced spatiotemporal data set; constructing a multi-sensor data fusion network based on an attention mechanism, and inputting the feature-enhanced spatiotemporal data set into the network for training to obtain a weighted fusion feature matrix; and performing spatiotemporal consistency verification based on the weighted fusion feature matrix to obtain a standardized fused production process data set that has undergone integrity verification and anomaly filtering.

[0030] Data cleaning formula: in, represents the cleaned dataset, represents a data point in the original data, represents the original dataset, Represents the data quality assessment function, which calculates the quality of the data. Indicates the set quality threshold, data below which is considered invalid.

[0031] This design forms a preliminary spatiotemporal dataset by performing spatiotemporal alignment and three-dimensional data mapping on the time series data collected by the sensors. Next, a dynamic time warping algorithm is used to perform phase synchronization correction, improving the data's feature expression capabilities. A multi-sensor data fusion network based on an attention mechanism is then trained and weighted to fuse data from different sensors, generating a fused feature matrix. Finally, a spatiotemporal consistency check is performed to verify integrity and filter out anomalies, resulting in a standardized fused production dataset. This process effectively processes complex, multi-dimensional sensor data, ensuring data accuracy and reliability, and providing a high-quality data foundation for subsequent production process optimization.

[0032] In one embodiment, identification and classification are performed based on the fused production process data set to obtain working condition classification data; an initial prediction model is established based on the working condition classification data to obtain initial prediction model data corresponding to each working condition; the initial prediction model is evaluated and optimized based on a cross-validation method to obtain optimized production process parameter prediction model data, including: extracting multi-dimensional time series features from the fused production process data set, constructing a three-level feature pyramid including time domain statistical features, frequency domain energy features and time-frequency joint features, and dynamically segmenting the production working conditions according to the hierarchical similarity of the feature pyramid to obtain a model with time series evolution characteristics. Working condition classification data cluster; for each working condition classification data cluster, a dual-channel deep learning model is constructed, and the pre-trained industrial process model parameters are initialized to the dual-channel network through the transfer learning mechanism to obtain an initial prediction model set with cross-working condition generalization capability; a model evaluation framework based on adversarial verification is constructed, and a generative adversarial network is trained to simulate the distribution of real working condition data to obtain an optimized prediction model with enhanced anti-interference capability; the optimized prediction model is deployed to the edge computing node, a real-time working condition model routing mechanism is constructed, and a closed-loop feedback loop is constructed to continuously inject the deviation value between the model prediction result and the actual production parameter into the model update queue.

[0033] Feature pyramid construction formula: in, represents the feature pyramid, Indicates the The characteristics of the layer include time domain statistical characteristics, frequency domain energy characteristics, time-frequency joint characteristics, etc. Indicates the number of layers of the feature pyramid.

[0034] The three-level feature pyramid achieves macro- to micro-level operating condition characterization through hierarchical feature extraction, supporting dynamic operating condition segmentation and cross-condition generalized modeling: The bottom layer (time-domain statistical features) captures the global statistical patterns of the signal and reflects the steady-state characteristics of the operating condition. The middle layer (frequency-domain energy features) reveals the signal's periodic components and frequency band energy distribution, characterizing the dynamic patterns of the operating condition. The top layer (joint time-frequency features) characterizes the signal's time-varying characteristics and transient events, capturing the evolution of the operating condition.

[0035] Time domain statistical characteristic formula: in, Represents the sample mean (or arithmetic mean), which represents the central tendency of a set of data. Indicates the number of samples (the total number of data points), that is, the length of the time series involved in the calculation, Indicates the time series The sample value (observation value) at a moment.

[0036] Frequency domain energy characteristic formula: in, Represents the power spectral density function, which indicates the signal at frequency The average power per unit frequency bandwidth at represents the theoretical limit, which represents the analysis of infinite length signal, Indicates the length of the observation time window (total number of samples), that is, the duration of the signal involved in the calculation, Represents a discrete-time signal at The sampling value at the moment (real or complex), Represents frequency variables, indicating the frequency points of analysis, represents the complex exponential function, which constitutes the basis function of the discrete-time Fourier transform, Indicates that the discrete-time Fourier transform calculates the signal at frequency The complex spectrum at .

[0037] Time-frequency joint feature formula: in, Indicates signal The short-time Fourier transform of provides a time-frequency representation, represents the continuous-time signal to be analyzed, Indicates that it is used to intercept signals In time Nearby local fragments, represents the complex exponential basis function, which is used to transform the local signal fragment from the time domain to the frequency domain. represents the central time position of the window function, indicating the time localization of the analysis, represents the frequency variable of the analysis, , integral operation, represents the Fourier transform of the local signal segment covered by the window function.

[0038] With this design, by identifying and classifying the fused data, working condition classification data is obtained, and an initial prediction model is built based on this. Next, cross-validation is used to evaluate and optimize the model, extract multi-dimensional time series features, construct a three-level feature pyramid, and dynamically segment the working conditions to obtain working condition classification data clusters with time series evolution characteristics. Then, for each classified data cluster, a dual-channel deep learning model is used, and transfer learning is used to improve the model's generalization ability across working conditions. The anti-interference ability of the model is enhanced by constructing a generative adversarial network. The final optimized prediction model is deployed to the edge computing node, and the deviation between the model and the actual production data is monitored and fed back in real time. Through this closed-loop feedback mechanism, the model is continuously updated and optimized to improve the prediction accuracy and stability of the production process.

[0039] In one embodiment, product quality indicators are predicted based on optimized production process parameter prediction model data to obtain product quality prediction data; the product quality prediction data is compared with a preset product quality target range to obtain product quality deviation data, including: inputting the optimized production process parameter prediction model data into the edge computing terminal at the production site to build a real-time data flow pipeline; based on the dynamic simulation results of the virtual production environment, a forward-looking prediction is made on the key product quality indicators under the current working conditions to obtain multi-dimensional product quality prediction data including a spatiotemporal evolution trajectory; an adaptive quality threshold system is constructed to dynamically adjust the product quality target range according to the real-time production environment parameters using a reinforcement learning algorithm; The product quality prediction data is compared with the dynamically adjusted target range in multiple dimensions to obtain structured product quality deviation data including deviation direction, deviation amplitude and deviation duration; a deviation traceability module based on a causal reasoning network is developed; through network topology analysis, the key process parameter links that cause quality deviations are automatically identified to obtain a deviation traceability analysis report with causal relationship annotations; a closed-loop feedback loop of the deviation data process parameters is constructed, and the key parameter links in the deviation traceability analysis report are mapped to the production control system; the process parameter correction amount is automatically calculated according to the deviation amplitude, and the corrected parameter combination is fed back to the optimized production process parameter prediction model to obtain a continuously improved quality optimization cycle.

[0040] With this design, the optimized production process parameter prediction model is used to predict product quality and compare it with the preset quality target range. Based on the dynamic simulation results, key product quality indicators are predicted, and the quality target range is dynamically adjusted using a reinforcement learning algorithm. By comparing the predicted data with the target range, product quality deviation data is obtained, including the deviation direction, magnitude, and duration. Next, a causal reasoning network is used to perform deviation traceability analysis, identify key process parameters that affect quality, and generate a traceability report. By establishing a closed-loop feedback mechanism, the deviation correction amount is automatically fed back to the production control system to further optimize the production process parameters and ultimately achieve continuous improvement in product quality. This process ensures that quality issues in the production process can be identified and corrected in a timely manner, thereby improving the stability and optimization effect of the production process.

[0041] In an embodiment, according to product quality deviation data, a production process parameter range needing to be adjusted is determined to obtain process parameter range data to be adjusted, including: constructing a deviation traceability graph of causal reasoning, locating a key process parameter link causing quality deviation through node activation propagation analysis to obtain a deviation traceability report containing parameter influence weight and propagation path; constructing a process parameter dynamic constraint graph, mapping the key parameter link in the deviation traceability report to the constraint graph to obtain a candidate parameter adjustment range meeting multiple constraint conditions; inputting the candidate parameter adjustment range into an optimization model based on the NSGA-III algorithm to obtain a Pareto optimal parameter adjustment scheme set (temperature 150℃≤T≤300℃, pressure 3MPa≤P≤7MPa); constructing a digital twin verification module of the parameter adjustment scheme; and based on simulation results, comprehensively evaluating the parameter adjustment scheme to obtain process parameter range data to be adjusted meeting actual production requirements.

[0042] In this way, through the causal reasoning network, the quality deviation is traced and analyzed, the key process parameter link is identified, and a traceability report containing the parameter influence weight is generated. Then, the traceability report is mapped to the constraint graph using the process parameter dynamic constraint graph to determine the candidate parameter adjustment range meeting the multiple constraint conditions. Next, multi-objective optimization is performed based on the NSGA-III algorithm to obtain the optimal parameter adjustment scheme set. Finally, the digital twin verification module is used to simulate and evaluate the adjustment scheme to ensure its effectiveness in actual production. Through this process, the production process parameter range needing to be adjusted can be accurately determined, the production process is optimized, and the product quality and production efficiency are improved.

[0043] In an embodiment, historical production data similar to the current production product is obtained; and from the historical production data, a process parameter adjustment strategy related to product quality improvement is mined according to the process parameter range data to be adjusted to obtain process parameter adjustment strategy data, including: analyzing technical documents and process requirements of the current production product through natural language processing technology to obtain a structured product feature tag library; performing a hybrid recommendation algorithm according to the feature tag library to obtain a cross-dimensional historical production data set containing process parameter trajectories, quality fluctuation records, and equipment state logs; performing deep cleaning according to the cross-dimensional historical production data set; constructing a multi-modal feature encoder to map the parameters to a unified feature space to obtain a structured process adjustment knowledge graph; inputting the process parameter range data to be adjusted as a query to locate a similar parameter adjustment scene in the process adjustment knowledge graph to obtain a multi-dimensional candidate strategy set containing parameter coupling relationships, adjustment steps, and expected effects; injecting the multi-dimensional candidate strategy set into a virtual production line (for pre-lesson learning) to perform parallel simulation and reasoning; constructing a multi-attribute decision matrix to filter the optimal strategy to obtain a process parameter adjustment strategy data package verified virtually.

[0044] This design uses natural language processing to parse the product's technical documents and process requirements, and builds a structured product feature tag library. Then, using a hybrid recommendation algorithm, process parameter adjustment strategies related to product quality improvement are mined from historical production data and deep cleansed. Next, a multimodal feature encoder is used to map different parameters to a unified feature space to form a process adjustment knowledge graph. By querying the range of process parameters to be adjusted, similar historical adjustment scenarios are located, and multi-dimensional candidate strategies are generated. Finally, a virtual production line is used for parallel simulation and multi-attribute decision matrix screening to obtain a verified optimal process parameter adjustment strategy. This process can accurately find the process adjustment strategy that suits current production needs, improving production efficiency and quality stability.

[0045] In one embodiment, according to the process parameter adjustment strategy data and in combination with the current actual production situation, the production process parameters are preliminarily adjusted to obtain preliminarily adjusted process parameter data, including: converting the process parameter adjustment strategy data into a structured operation instruction set; obtaining a parameter adjustment decision tree that integrates the strategy status through a virtual reality data mapping engine; inputting the parameter adjustment decision tree into a deep Q network model for training; constructing a reward function that considers multi-objective optimization to obtain a preliminarily adjusted action sequence that takes into account both quality improvement and process stability; converting the preliminarily adjusted action sequence into an instruction code executable by the equipment; realizing the simultaneous virtual verification and actual adjustment through bidirectional closed-loop control of digital twins and physical systems; constructing an effect evaluation model based on the spatiotemporal attention mechanism to perform real-time evaluation of the preliminarily adjusted process parameters; so as to obtain preliminarily adjusted process parameter data that has undergone adaptive optimization.

[0046] This design converts process parameter adjustment strategy data into an operational instruction set, and a parameter adjustment decision tree is generated using a virtual reality data mapping engine. The decision tree is then trained using a deep Q-network model to construct a reward function, optimize quality improvement and process stability, and generate a preliminary adjustment action sequence. This adjustment action sequence is then converted into executable instruction code for the device, and virtual verification and actual adjustments are synchronized through the digital twin and physical system. Finally, an effect evaluation model based on a spatiotemporal attention mechanism is used to evaluate the initially adjusted process parameters in real time. The parameters are then updated through adaptive optimization to ensure that the adjustments meet production requirements. This process enables precise adjustment of process parameters, improving production quality and stability.

[0047] In one embodiment, the process parameter data after preliminary adjustment is applied to the actual production process, and the multi-dimensional data of the production process is again collected in real time through the sensor group to obtain the adjusted production process data, including: converting the process parameter data after preliminary adjustment into a digital instruction set executable by the equipment; developing a parameter deployment verification platform based on digital twins to obtain a virtual verification report containing the thermodynamic behavior of the equipment, material flow characteristics and quality fluctuation trends; building an industrial Internet of Things gateway with real-time response capabilities, and synchronously pushing the digital instruction set to each execution terminal of the production line through a time-sensitive network; building a real-time monitoring interface for parameter deployment, and superimposing the display in the physical space. The system displays the equipment operating status and expected effects after parameter adjustment; builds a multimodal data fusion center, collects adjusted production process data in real time through edge computing nodes, constructs a data fusion algorithm based on the spatiotemporal attention mechanism, maps multi-source heterogeneous data to a unified feature space, and obtains a structured adjusted production process data set that contains spatiotemporal evolution laws; inputs the structured adjusted production process data set into the long-short-term memory network model; automatically generates parameter fine-tuning instructions by comparing the deviation between actual data and virtual verification reports, and feeds them back to production equipment through the industrial Internet of Things gateway, forming a closed-loop optimization loop of "parameter deployment-real-time monitoring-data fusion-dynamic calibration".

[0048] Closed-loop optimization feedback formula: in, Represents the final adjusted process parameters, which are obtained based on real-time feedback and dynamic calibration. Represents the process parameters after preliminary adjustment, which are obtained from virtual verification and preliminary parameter optimization. Indicates the feedback coefficient, which controls the influence of the fine-tuning instruction on the final process parameter adjustment. Indicates the process parameter fine-tuning amount calculated based on the deviation between actual data and virtual verification report.

[0049] This design converts the initially adjusted process parameters into a digital instruction set executable by the equipment. The equipment's thermodynamic behavior and quality fluctuation trends are then evaluated through a virtual verification platform. The digital instructions are then synchronously pushed to the execution terminal via the Industrial Internet of Things gateway, and a real-time monitoring interface is constructed to display the equipment's operating status and expected results. Simultaneously, a multimodal data fusion center collects adjusted production process data through edge computing nodes, fusing multi-source heterogeneous data using a spatiotemporal attention mechanism. This fused data is then fed into a long-short-term memory network model, comparing the actual data with the virtual verification report. Fine-tuning instructions are then automatically generated and fed back to the production equipment. This ultimately forms a closed-loop optimization circuit, enabling dynamic calibration of process parameters and improving the accuracy and stability of the production process.

[0050] A production process control system based on big data, which is applicable to the above-mentioned production process control method based on big data, includes: The fusion processing unit 1 is used to collect raw data in the production process according to the sensor group, perform data cleaning on the collected raw data to obtain cleaned production process data; and perform fusion processing on the cleaned production process data to obtain a fused production process data set; Evaluation and optimization unit 2 is used to identify and classify the fused production process data set to obtain working condition classification data; establish an initial prediction model based on the working condition classification data to obtain initial prediction model data corresponding to each working condition; evaluate and optimize the initial prediction model according to the cross-validation method to obtain optimized production process parameter prediction model data; The data prediction unit 3 is used to predict the product quality index based on the optimized production process parameter prediction model data to obtain product quality prediction data; compare the product quality prediction data with the preset product quality target range to obtain product quality deviation data; and determine the production process parameter range that needs to be adjusted based on the product quality deviation data to obtain process parameter range data to be adjusted; The preliminary adjustment unit 4 is used to obtain historical production data similar to the current production product; mine the process parameter adjustment strategy related to product quality improvement from the historical production data based on the process parameter range data to obtain process parameter adjustment strategy data; based on the process parameter adjustment strategy data and the current production actual situation, the production process parameters are preliminarily adjusted to obtain the preliminarily adjusted process parameter data; Iterative optimization unit 5, iterative optimization unit 5 is used to apply the process parameter data after preliminary adjustment to the actual production process, and again collect multi-dimensional data of the production process in real time through the sensor group to obtain adjusted production process data; iterative optimization processing is performed on the adjusted production process data until the product quality reaches the target range.

[0051] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A production process control method based on big data, characterized in that: include: collecting raw data during the production process using the sensor group, and performing data cleaning on the collected raw data to obtain cleaned production process data; Performing fusion processing on the cleaned production process data to obtain a fused production process data set; Identify and classify the fused production process data set to obtain working condition classification data; Establishing an initial prediction model based on the working condition classification data to obtain initial prediction model data corresponding to each working condition; evaluating and optimizing the initial prediction model based on a cross-validation method to obtain optimized production process parameter prediction model data; Predicting product quality indicators based on the optimized production process parameter prediction model data to obtain product quality prediction data; comparing the product quality prediction data with a preset product quality target range to obtain product quality deviation data; and determining a production process parameter range that needs to be adjusted based on the product quality deviation data to obtain process parameter range data to be adjusted; Obtain historical production data for products similar to those currently being produced; Mining process parameter adjustment strategies related to product quality improvement from historical production data based on the process parameter range data to obtain process parameter adjustment strategy data: Based on the process parameter adjustment strategy data and combined with the current production actual situation, preliminary adjustments are made to the production process parameters to obtain preliminary adjusted process parameter data; The process parameter data after preliminary adjustment is applied to the actual production process, and the multi-dimensional data of the production process is again collected in real time through the sensor group to obtain the adjusted production process data; the adjusted production process data is iteratively optimized until the product quality reaches the target range.

2. The method for controlling a production process based on big data according to claim 1, characterized in that: The iterative optimization process includes: cleaning and fusing the same data as the adjusted production process data and the fused production process data set to obtain an adjusted fused production process data set; performing product quality prediction on the adjusted production process data based on the optimized production process parameter prediction model data to obtain adjusted product quality prediction data; and comparing the adjusted product quality prediction data with the product quality target range again, including: If the product quality reaches the target range, the process parameter adjustments, product quality changes, and related production data during the entire production process control process are recorded to obtain production process control record data; based on the production process control record data, the long-term stability and optimization direction of the production process are analyzed to obtain production process optimization trend data; based on the production process optimization trend data, a subsequent production process continuous improvement plan is formulated to obtain production process continuous improvement plan data; If the product quality still does not reach the target range, return to obtain historical production data similar to the current production product; based on the process parameter range data to be adjusted, mine the process parameter adjustment strategy related to product quality improvement from the historical production data to obtain process parameter adjustment strategy data: based on the process parameter adjustment strategy data, combined with the current production actual situation, make preliminary adjustments to the production process parameters to obtain preliminary adjusted process parameter data, re-mine the process parameter adjustment strategy based on the new product quality deviation data, and adjust the process parameters again.

3. The method for controlling a production process based on big data according to claim 2, characterized in that: collecting raw data during the production process using the sensor group, and performing data cleaning on the collected raw data to obtain cleaned production process data; The cleaned production process data is fused to obtain a fused production process data set, including: Performing spatiotemporal alignment calibration on the original time series data stream, and constructing a three-dimensional data mapping matrix according to the sensor installation position coordinates to obtain an initial spatiotemporal data set with spatial correlation; According to the improved dynamic time warping algorithm, the time series signals of each sensor channel in the initial spatiotemporal dataset are phase-synchronized and corrected to obtain a feature-enhanced spatiotemporal dataset. Construct a multi-sensor data fusion network based on the attention mechanism, input the feature-enhanced spatiotemporal dataset into the network for training to obtain a weighted fusion feature matrix; A spatiotemporal consistency check is performed based on the weighted fusion feature matrix to obtain a standardized fusion production process data set that has been integrity checked and anomaly filtered.

4. The method for controlling a production process based on big data according to claim 3, characterized in that: Identify and classify the fused production process data set to obtain working condition classification data; establish an initial prediction model based on the working condition classification data to obtain initial prediction model data corresponding to each working condition; The initial prediction model is evaluated and optimized using the cross-validation method to obtain optimized production process parameter prediction model data, including: Extract multi-dimensional time series features from the fused production process dataset and construct a three-level feature pyramid consisting of time-domain statistical features, frequency-domain energy features, and time-frequency joint features. Dynamically segment production conditions based on the hierarchical similarity of the feature pyramid to obtain condition classification data clusters with time-series evolution characteristics. For each operating condition classification data cluster, a dual-channel deep learning model is constructed. The pre-trained industrial process model parameters are initialized to the dual-channel network through a transfer learning mechanism to obtain an initial prediction model set with cross-operating condition generalization capabilities. Build a model evaluation framework based on adversarial verification, and train a generative adversarial network to simulate the distribution of real-world operating data to obtain an optimized prediction model with enhanced anti-interference capabilities; Deploy the optimized prediction model to the edge computing node, build a real-time working condition model routing mechanism, build a closed-loop feedback loop, and continuously inject the deviation value between the model prediction results and the actual production parameters into the model update queue.

5. The method for controlling a production process based on big data according to claim 4, characterized in that: Predicting product quality indicators based on the optimized production process parameter prediction model data to obtain product quality prediction data; Comparing the product quality prediction data with the preset product quality target range to obtain product quality deviation data, including: The optimized production process parameter prediction model data is input into the edge computing terminal at the production site to build a real-time data flow pipeline. Based on the dynamic simulation results of the virtual production environment, the key product quality indicators under the current working conditions are predicted forward-lookingly to obtain multi-dimensional product quality prediction data including spatiotemporal evolution trajectories. Build an adaptive quality threshold system that uses reinforcement learning algorithms to dynamically adjust product quality target ranges based on real-time production environment parameters. Perform a multi-dimensional comparison of product quality prediction data with the dynamically adjusted target ranges to obtain structured product quality deviation data that includes deviation direction, deviation magnitude, and deviation duration. Develop a deviation traceability module based on a causal reasoning network; automatically identify key process parameter links that lead to quality deviations through network topology analysis to generate a deviation traceability analysis report with causal relationship annotations; Build a closed-loop feedback loop for deviation data process parameters, map the key parameter links in the deviation traceability analysis report to the production control system; automatically calculate the process parameter correction amount based on the deviation amplitude, and feed the corrected parameter combination back to the optimized production process parameter prediction model to obtain a continuous improvement quality optimization cycle.

6. The method for controlling a production process based on big data according to claim 5, characterized in that: Based on the product quality deviation data, determine the range of production process parameters that need to be adjusted to obtain the range data of process parameters to be adjusted, including: Construct a deviation traceability map for causal reasoning, locate the key process parameter links that lead to quality deviations through node activation and propagation analysis, and obtain a deviation traceability report that includes parameter impact weights and propagation paths; Construct a dynamic constraint map for process parameters and map the key parameter links in the deviation traceability report to the constraint map to obtain the candidate parameter adjustment range that meets multiple constraints. Inputting the candidate parameter adjustment range into an optimization model based on the NSGA-III algorithm to obtain a Pareto optimal set of parameter adjustment solutions; Build a digital twin verification module for the parameter adjustment plan; conduct a comprehensive evaluation of the parameter adjustment plan based on the simulation results to obtain the range of process parameters to be adjusted that meets actual production needs.

7. The method for controlling a production process based on big data according to claim 6, characterized in that: Obtain historical production data similar to the current product; mine process parameter adjustment strategies related to product quality improvement from historical production data based on the process parameter range data to obtain process parameter adjustment strategy data, including: Using natural language processing technology to parse the technical documentation and process requirements of current production products to obtain a structured product feature tag library. Based on this feature tag library, a hybrid recommendation algorithm is used to obtain a cross-dimensional historical production dataset that includes process parameter trajectories, quality fluctuation records, and equipment status logs. Perform deep cleaning based on the cross-dimensional historical production dataset; construct a multimodal feature encoder to map parameters to a unified feature space to obtain a structured process adjustment knowledge graph; The process parameter range data to be adjusted is used as query input to locate similar parameter adjustment scenarios in the process adjustment knowledge graph to obtain a multi-dimensional candidate strategy set that includes parameter coupling relationships, adjustment steps, and expected effects. The multi-dimensional candidate strategy set is injected into the virtual production line for parallel simulation and deduction; a multi-attribute decision matrix is ​​constructed to screen the optimal strategy to obtain a virtually verified process parameter adjustment strategy data package.

8. The method for controlling production process based on big data according to claim 7, characterized in that: According to the process parameter adjustment strategy data and combined with the current production actual situation, the production process parameters are preliminarily adjusted to obtain the preliminarily adjusted process parameter data, including: Converting the process parameter adjustment strategy data into a structured operation instruction set; obtaining a parameter adjustment decision tree integrating the strategy status through a virtual reality data mapping engine; Input the parameter adjustment decision tree into a deep Q-network model for training; construct a reward function that considers multi-objective optimization to obtain a preliminary adjustment action sequence that takes into account both quality improvement and process stability; Convert the preliminary adjustment action sequence into instruction code that can be executed by the device; through the two-way closed-loop control of the digital twin and the physical system, realize the synchronization of virtual verification and actual adjustment; Construct an effect evaluation model based on the spatiotemporal attention mechanism to conduct real-time evaluation of the process parameters after preliminary adjustment, so as to obtain the preliminary adjusted process parameter data after adaptive optimization.

9. The method for controlling a production process based on big data according to claim 8, characterized in that: Apply the initially adjusted process parameter data to the actual production process, and again use the sensor group to collect multi-dimensional data in real time during the production process to obtain adjusted production process data, including: Convert the initially adjusted process parameter data into a digital instruction set executable by the equipment; develop a parameter deployment verification platform based on digital twins to obtain a virtual verification report that includes equipment thermodynamic behavior, material flow characteristics, and quality fluctuation trends; Build an industrial IoT gateway with real-time response capabilities to synchronously push digital instruction sets to each execution terminal on the production line through a time-sensitive network; build a real-time monitoring interface for parameter deployment, and overlay the display of the equipment operating status and expected results after parameter adjustment in the physical space; Build a multimodal data fusion center to collect adjusted production process data in real time through edge computing nodes, build a data fusion algorithm based on the spatiotemporal attention mechanism, and map multi-source heterogeneous data into a unified feature space to obtain a structured adjusted production process dataset that includes spatiotemporal evolution laws; The structured and adjusted production process data set is input into the long short-term memory network model. By comparing the deviation between the actual data and the virtual verification report, parameter fine-tuning instructions are automatically generated and fed back to the production equipment via the Industrial Internet of Things gateway, forming a closed-loop optimization circuit of "parameter deployment-real-time monitoring-data fusion-dynamic calibration".

10. A production process control system based on big data, which is applicable to the production process control method based on big data according to any one of claims 1 to 9, characterized in that: include, A fusion processing unit (1), the fusion processing unit (1) is used to collect raw data in the production process according to the sensor group, and perform data cleaning on the collected raw data to obtain cleaned production process data; Performing fusion processing on the cleaned production process data to obtain a fused production process data set; An evaluation and optimization unit (2), the evaluation and optimization unit (2) is used to identify and classify the fused production process data set to obtain working condition classification data; establish an initial prediction model based on the working condition classification data to obtain initial prediction model data corresponding to each working condition; evaluate and optimize the initial prediction model according to a cross-validation method to obtain optimized production process parameter prediction model data; A data prediction unit (3), the data prediction unit (3) is used to predict the product quality index based on the optimized production process parameter prediction model data to obtain product quality prediction data; compare the product quality prediction data with a preset product quality target range to obtain product quality deviation data; and determine the production process parameter range that needs to be adjusted based on the product quality deviation data to obtain process parameter range data to be adjusted; A preliminary adjustment unit (4) is used to obtain historical production data similar to the current production product; mine a process parameter adjustment strategy related to product quality improvement from the historical production data based on the process parameter range data to obtain process parameter adjustment strategy data; and perform preliminary adjustments on the production process parameters based on the process parameter adjustment strategy data and in combination with the current production actual situation to obtain preliminary adjusted process parameter data; An iterative optimization unit (5) is used to apply the preliminarily adjusted process parameter data to the actual production process, and again collect multi-dimensional data in the production process in real time through the sensor group to obtain adjusted production process data; and iteratively optimize the adjusted production process data until the product quality reaches the target range.

Images