Method and system for optimizing production process of discrete manufacturing industry by using data model

Abstract

The invention discloses a method and a system for optimizing a discrete manufacturing industry production process by using a data model. The method comprises the following steps: collecting all data in a manufacturing process; classifying the data and respectively preprocessing the data, wherein the classification types of the data comprise image data, video data and other data; key features influencing production efficiency and quality are extracted from the preprocessed data; respectively constructing a prediction model and an optimization model based on the key features influencing the production efficiency and quality; and optimizing the discrete manufacturing industry production process based on the prediction model and the optimization model. The method not only optimizes the traditional production process, but also promotes the intelligent upgrading of the whole industry chain.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing technology, and in particular to a method and system for optimizing a discrete manufacturing production process using a data model. Background Art

[0002] With the development of intelligent manufacturing, the discrete manufacturing industry is facing greater and greater data challenges. The discrete manufacturing industry is known for its product diversity and complex production process. The traditional experience-driven production management method often makes it difficult to accurately predict and control various variables in the production process, resulting in problems such as waste of resources, low production efficiency and unstable quality. Traditional product manufacturing methods are usually based on empirical formulas or trial and error methods, but these methods often cannot meet the personalized needs of production and research and development in the manufacturing industry, and cannot make full use of massive big data resources.

[0003] The rapid development of new generation information technologies such as big data and cloud computing, combined with modern manufacturing production technologies, has led to the innovation of many new production and manufacturing models, such as intelligent manufacturing and digital twins. Production data-driven manufacturing has become an effective way and important means to solve complex problems. With the effective application of CNC machine tools, sensors, PLCs and other data acquisition equipment, enterprises have accumulated a large amount of manufacturing data in the production process. These big data show typical big data characteristics. The implicit association rules and manufacturing knowledge behind the data need to be further analyzed. Data mining is needed to effectively guide product design optimization and production and manufacturing process improvement, thereby improving the adaptability of product R&D design to the manufacturing environment. The manufacturing process of the machine tool stamping industry usually involves multiple links. According to market demand and customer requirements, after determining the product design task, designers use professional design software to design and plan product R&D, design new products that meet the requirements, select appropriate production equipment according to process route planning, determine the layout of the workshop according to equipment selection and process route planning, trial-produce products and conduct product verification before putting them into formal production, modify and improve product design, determine the final design plan, and then produce and manufacture actual products.

[0004] Discrete manufacturing companies usually produce complex products, such as rockets, aircraft, weapons and equipment, ships, electronic equipment, machine tools, automobiles, etc. The production process of these products often includes multiple discontinuous processes, each of which requires different equipment, skills and materials. Therefore, the production and manufacturing process of discrete manufacturing requires a high degree of flexibility and adaptability to cope with the ever-changing market and customer demands.

[0005] Regarding the existing production and manufacturing processes of discrete manufacturing, the defects and problems of existing technologies are mainly reflected in the following aspects:

[0006] (1) The production process is not transparent

[0007] In traditional discrete manufacturing companies, the production process is often not transparent enough, resulting in problems such as high defective product rates and low production capacity being unable to be discovered and resolved in a timely manner.

[0008] The lack of effective real-time monitoring and control means makes it difficult for enterprises to grasp the production progress and product quality in real time.

[0009] (2) Production plan is uncontrollable

[0010] Discrete manufacturing companies are usually faced with complex production planning arrangements. If they rely solely on manual production planning and scheduling, problems such as inaccurate information and unreasonable arrangements are likely to occur.

[0011] The lack of an effective production planning management system results in frequent disruptions to production plans, making them unable to adapt to rapid market changes.

[0012] (3) Difficulty in material management

[0013] Discrete manufacturing companies have a wide variety of parts and components required for product processing, complex BOM structures, and great difficulty in material management.

[0014] Issues such as the timely supply of materials and the control of material inventory have an impact on the company's production process.

[0015] Lack of effective information sharing with suppliers and logistics companies leads to inefficiencies in the material supply chain.

[0016] (4) Highly complex technical processes and supply chains

[0017] There are many types of parts required for product processing, the design of process standards and process routes is complex, and the workload of producing drawings and process technical documents is large, which restricts production efficiency.

[0018] The product process often changes, which increases the complexity of production and management.

[0019] (5) Highly discrete manufacturing

[0020] Discrete manufacturing enterprises present an island distribution of process clusters, and the processing process flow of each component and part in the factory is diversified. Operators, equipment information, materials, production tasks, etc. are highly discrete, and the progress of task execution is difficult to monitor and the status of materials is difficult to track.

[0021] This seriously hinders planning management, resulting in the inability of work plans to keep up with changes and the uncontrolled execution of planned tasks.

[0022] (6) Long production cycle

[0023] Due to the complexity of production, the production cycle of some large equipment in discrete manufacturing enterprises can be between 6 and 18 months.

[0024] Due to the backward means of optimizing and scheduling manufacturing resources, coupled with the single-piece and small-batch production model, the production preparation time will be increased and a huge waste of equipment will be caused.

[0025] (7) Lack of real-time monitoring and control

[0026] In the production process, the lack of real-time monitoring and control means makes it impossible for enterprises to detect abnormal situations and make timely adjustments. This not only affects production efficiency, but may also have a serious impact on product quality and delivery time.

[0027] In summary, the discrete manufacturing industry faces problems in the existing production and manufacturing process, such as opaque production process, uncontrollable production plan, difficult material management, highly complex technical processes and supply chain, highly discrete production and manufacturing, long production and manufacturing cycle, and lack of real-time monitoring and control. These problems not only affect the production efficiency and product quality of enterprises, but also have a serious impact on the competitiveness of enterprises. Therefore, enterprises need to take effective measures to solve these problems, improve production efficiency and product quality, and enhance competitiveness. Summary of the invention

[0028] In view of the shortcomings existing in the above problems, the present invention provides a method and system for optimizing the production process of discrete manufacturing industry by using a data model.

[0029] To achieve the above object, the present invention provides a method for optimizing the production process of discrete manufacturing industry using a data model, comprising:

[0030] Collect all data in the manufacturing process;

[0031] Classifying the data and preprocessing them respectively, wherein the types of the data classification include image data, video data and other data;

[0032] Extracting key features that affect production efficiency and quality from the preprocessed data;

[0033] Based on the key features affecting production efficiency and quality, a prediction model and an optimization model are respectively constructed;

[0034] The production process of discrete manufacturing industry is optimized based on the prediction model and the optimization model.

[0035] Preferably, the data includes system data and equipment data, the system data includes ERP system data, MES system data, PLM system data, BPM system data and VMS system data, the ERP system data includes raw material purchase orders, inventory information and supplier information; the MES system data includes production plans, order status, equipment usage, material consumption and production progress; the PLM system data includes product design, engineering changes and product configuration information; the BPM system data includes business process monitoring and management data; the VMS system data includes monitoring and maintenance production equipment information; the equipment data includes processing data, model data, process data and quality inspection data.

[0036] Preferably, classifying and preprocessing the data respectively comprises cleaning, deduplicating and normalizing the data in sequence;

[0037] The data cleaning step includes:

[0038] Missing value processing: the missing parts of the image data and the video data are processed by repair technology or cropping, and the missing values ​​of the other data are filled according to the business logic;

[0039] Outlier processing: abnormal frames or areas of the image data and the video data are screened or replaced, and abnormal values ​​of other data are identified by setting thresholds or statistical methods, and corrected or deleted;

[0040] The step of deduplicating the data includes:

[0041] Unique identifier comparison: compare the unique identifiers in the other data to identify and delete duplicate records;

[0042] Hash value comparison, calculating the hash values ​​of the image data and the video data, and identifying and deleting duplicate files or segments by comparing the hash values;

[0043] Matching method: matching multiple fields in the other data to identify and delete duplicate data;

[0044] The step of normalizing the data includes:

[0045] Rotate the corresponding normalization method for different fields in the other data according to business requirements and data analysis objectives;

[0046] Standardizing the image data by adjusting the pixel value range and the color space;

[0047] The frame rate and resolution parameters of the video data are adjusted and standardized.

[0048] Preferably, extracting key features affecting production efficiency and quality from the preprocessed data comprises selecting PCA, LDA, TF-IDF algorithm and deep learning for feature extraction;

[0049] Using PCA algorithm to extract features from the processed data;

[0050] Using LDA algorithm to extract features from the process data;

[0051] Use the TF-IDF algorithm to extract features from data in text form;

[0052] Use deep learning and select the corresponding deep neural network structure to extract features from the remaining data.

[0053] Preferably, the prediction model building step includes:

[0054] Collect historical data during the stamping production process of machine tools, including production volume, equipment status, and product quality;

[0055] Cleaning, converting and normalizing the historical data;

[0056] Extracting features related to the prediction target from the processed historical data;

[0057] Training the prediction model using features related to the prediction target;

[0058] The trained prediction model is evaluated.

[0059] Preferably, based on the different data, the prediction model includes a time series prediction model, a regression prediction model and a machine learning prediction model.

[0060] Preferably, the optimization model building step includes:

[0061] Determine key indicators that need to be optimized, including production efficiency, product quality and cost;

[0062] According to the optimization goal, a corresponding mathematical model is established, wherein the mathematical model includes decision variables, objective functions and constraint conditions;

[0063] The mathematical model is solved using an optimization algorithm to obtain an optimal solution.

[0064] Preferably, the optimization model includes a production scheduling optimization model, a quality control optimization model, and a supply chain optimization model.

[0065] The present invention also provides a system for optimizing a discrete manufacturing production process using a data model, comprising:

[0066] The acquisition module is used to collect all data in the manufacturing process;

[0067] A preprocessing module, used to classify the data and preprocess them respectively, the data classification types include image data, video data and other data;

[0068] An extraction module, used to extract key features that affect production efficiency and quality from the preprocessed data;

[0069] A construction module, used to construct a prediction model and an optimization model based on the key features affecting production efficiency and quality;

[0070] The optimization module is used to optimize the discrete manufacturing production process based on the prediction model and the optimization model.

[0071] Compared with the prior art, the present invention has the following beneficial effects:

[0072] The present invention not only optimizes the traditional production process, but also promotes the intelligent upgrading of the entire industrial chain. Its in-depth beneficial effects can be summarized as follows:

[0073] Refined energy consumption management and green manufacturing: Through real-time monitoring and intelligent analysis of energy consumption in the production process, we can identify energy consumption anomalies and potential energy-saving space to achieve precise energy management. Combined with machine learning algorithms to predict energy consumption trends, we can optimize production scheduling to balance peak electricity consumption, reduce carbon footprint, and promote green and sustainable development;

[0074] Predictive maintenance and asset health management: Utilize the large amount of operating data collected by sensors, combined with advanced analytical techniques (such as fault prediction models), to identify potential equipment failures in advance and achieve a shift from passive maintenance to active maintenance. This can not only effectively extend the service life of equipment, but also significantly reduce production losses caused by unexpected downtime;

[0075] Supply chain collaborative optimization: By integrating data from suppliers, manufacturers, logistics providers, etc., a global supply chain digital twin model is constructed to achieve transparency and collaborative optimization of all links in the supply chain. Advanced algorithms are used to simulate supply chain responses in different scenarios, improve the ability to respond to market fluctuations, reduce inventory backlogs, and ensure accurate matching of material supply with market demand;

[0076] Personalized customization and flexible production: Based on big data analysis of consumer behavior and market trends, we can quickly respond to personalized needs, flexibly adjust production line configurations, and achieve large-scale customized production. Through the intelligent scheduling system, we can dynamically optimize production plans, balance production capacity and orders, and improve market competitiveness;

[0077] Intelligent prevention and control of safety risks: Use AI technology to conduct comprehensive analysis of multi-source heterogeneous data such as video surveillance and equipment status at the production site to promptly identify safety hazards and prevent safety accidents. Establish an emergency response mechanism to quickly handle emergencies and ensure production safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 It is a flow chart of a method for optimizing production processes in discrete manufacturing industry using a data model according to the present invention. DETAILED DESCRIPTION

[0079] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are 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 creative work are within the scope of protection of the present invention.

[0080] Reference Figure 1 The present invention provides a method for optimizing the production process of discrete manufacturing industry by using a data model, comprising:

[0081] Collect all data in the manufacturing process;

[0082] Specifically, a comprehensive data collection system is built to cover the entire production chain from raw material procurement, production planning, processing and assembly to finished product inspection, collecting multi-dimensional data including but not limited to equipment operating status, material flow, operation time, quality inspection, etc.

[0083] As a typical representative of discrete manufacturing, the machine tool industry needs to deal with various types of sensors, such as shaft temperature sensors, vibration detection modules, clutch pressure sensors, etc. In addition, sensors such as clutch cooling flow and balancer air pressure also need to be taken into consideration. Through sensors, industrial Internet of Things and other equipment deployed in discrete manufacturing workshops, various data in the production process are collected in real time, including equipment operating status, production efficiency, product quality, etc. It also includes material flow records, operator allocation information, environmental monitoring data (such as temperature, humidity), etc.

[0084] During the manufacturing process, mechanical sensors and speed sensors generate time series data, while visual sensors are responsible for capturing images and generating image data. At the same time, the product data management system (PDM) also generates relevant data during operation. The main characteristics of manufacturing big data are its diversity, complexity and uncertainty. Therefore, how to effectively express this data has become a key difficulty in solving the data fusion problem.

[0085] The fusion of time series data refers to the process of matching and integrating high-speed (40-80fps) time series data collected by machine tool and workpiece status monitoring with other types of data. In order to achieve this goal, it is necessary to downsample the time series data and use methods such as mean and variance to reflect the processing status at a certain stage. It can be random downsampling, randomly selecting a part of the data points from the original time series data, or fixed interval downsampling, selecting a data point at a certain time interval. It can also be based on statistical downsampling, selecting the mean, median or maximum value in each time window as a representative. Based on statistical fusion methods, the average method: calculate the average value of the downsampled data to reflect the overall level of the processing status at a certain stage. Variance method: calculate the variance of the data to measure the discreteness and fluctuation of the data, and help understand the stability of the processing status.

[0086] Image data fusion refers to processing image data in the manufacturing system to extract information and express it in a structured manner. Traditionally, image information extraction needs to be done manually, which is inefficient. However, with the application of convolutional neural network models in the field of computer vision, the image perception ability of machines has been greatly improved. These models can be used to extract information from image data and realize the conversion of unstructured image data into structured data information. This not only improves the efficiency and accuracy of the manufacturing system, but also makes better use of image data to optimize the manufacturing process and improve product quality.

[0087] In a product data management system, data can be collected in various ways, such as sensors, barcode scanners, RFID tags, etc., and automatically entered into a database. The system also provides query tools and reporting functions to facilitate users to access and analyze data.

[0088] The data sources collected mainly include industrial control management software, production scheduling system data, and various industrial equipment data such as PLC, instruments, power equipment, CNC, sensors, etc. The data content includes system data such as ERP, MES, PLM, BPM, VMS, equipment data, machine tool data, model data, process data and other equipment inspection and monitoring data.

[0089] Furthermore, ERP system data includes raw material purchase orders, inventory information, supplier information, etc., which are used to track the source and status of raw materials. MES system data involves production plans, order status, equipment use, material consumption, production progress, etc., and is the core data source for production site management. PLM system data contains information such as product design, engineering changes, product configuration, etc., to ensure that product data in the production process is consistent with design data. BPM system data provides monitoring and management data for business processes to ensure efficient execution of production processes. VMS system data is used to monitor and maintain production equipment, providing information such as equipment status, maintenance records, and fault warnings. Industrial equipment data collects real-time data such as equipment operating status, temperature, pressure, speed, position, etc. through industrial equipment such as PLC, instrumentation, power equipment, CNC machine tools, sensors, etc. Equipment data: including equipment operating time, fault records, maintenance cycles, energy consumption, etc., used to evaluate equipment performance and formulate maintenance plans. Machine tool data: detailed records of CNC machine tool processing parameters, tool usage, processing quality, etc., used to optimize processing technology and improve product quality. Model data: Contains 3D models, CAD drawings, etc. in the product design and manufacturing process, used to guide production and quality inspection. Process data: Records process parameters, processing sequence, operation requirements, etc. in the product processing process to ensure the consistency and stability of product processing. Quality inspection data: Includes data such as size, appearance, and performance of finished product inspection, used to evaluate product quality and conduct quality traceability. Data collection technology includes API interface: Establish API interface with ERP, MES, PLM and other systems to achieve real-time synchronization and transmission of data. OPC UA: Use OPC UA (OLE for Process Control Unified Architecture) protocol to achieve communication and data collection with PLC, instrumentation and other equipment. MQTT: For IoT devices that require low power consumption and low bandwidth, use MQTT (Message Queuing Telemetry Transport) protocol for data collection. RFID: Use RFID (Radio Frequency Identification) technology in the material flow link to achieve automatic identification and tracking of materials. Sensor technology: Use various sensors to collect real-time operating data of equipment, such as temperature, pressure, speed, etc. In addition, the image data of discrete manufacturing industry includes equipment operation status images, material flow images, process images, etc., as shown in Table 1:

[0090] Table 1

[0091]

[0092]

[0093] The video data of discrete manufacturing industry includes production site monitoring videos, equipment troubleshooting videos, quality inspection videos, etc., as shown in Table 2:

[0094] Table 2

[0095]

[0096] Classify the data and pre-process them respectively. The types of data classification include image data, video data and other data;

[0097] Specifically, the collected data is cleaned, deduplicated, normalized and other preprocessing operations are performed to improve the accuracy and efficiency of subsequent analysis. Specifically, Python is used to write scripts to read data from the database, and text mining, feature extraction and other technologies are used to preprocess the data, including removing duplicate values, missing values ​​and outliers; performing operations such as word segmentation, stop word removal and stem extraction on text data; and normalizing, standardizing or discretizing numerical data. The preprocessing process for different types of data is as follows:

[0098] Data cleaning:

[0099] Data identification and classification: First, identify and classify data according to data type (such as system data, images, videos) to clarify the processing requirements of different data.

[0100] Missing value processing: For missing values ​​in system data, fill them in (such as using mean, median, etc.) or delete them directly according to business logic. For image and video data, the missing parts may need to be processed through repair technology or cropping.

[0101] Outlier processing: For outliers in system data, identify them by setting thresholds or statistical methods, and correct or delete them. For abnormal frames or areas in image and video data, filter or replace them.

[0102] Unified data format: Ensure that all system data follows a unified format and standard to facilitate subsequent processing and analysis.

[0103] Data deduplication:

[0104] Unique identifier comparison: For system data, use unique identifiers such as order numbers and product IDs to compare and identify and delete duplicate records.

[0105] Image and video hash value comparison: For image and video data, calculate their hash values, identify and delete duplicate files or clips by comparing the hash values.

[0106] Data matching method: Combine multiple fields in system data to match, identify and delete duplicate data across systems.

[0107] Data normalization:

[0108] Make data from different sources and formats have a unified scale to facilitate subsequent data analysis and mining.

[0109] System data normalization: For different fields in the system data, select appropriate normalization methods (such as maximum and minimum normalization, Z-score normalization, etc.) based on business needs and data analysis objectives.

[0110] Image and video data standardization: For image data, standardization can be performed by adjusting the pixel value range, color space, etc. For video data, standardization can be performed by adjusting parameters such as frame rate and resolution.

[0111] Extract key features that affect production efficiency and quality from the preprocessed data;

[0112] Specifically, the big data analysis technology is used to deeply analyze and mine the processed data to find potential problems and bottlenecks in the production process. Discrete manufacturing big data is the result of digital expression of discrete manufacturing systems and discrete manufacturing processes. With the continuous accumulation of data in discrete manufacturing systems, the complete description of the entire discrete manufacturing system has become more accurate and detailed, which is very helpful for data mining. However, this also brings a problem-dimensionality disaster. Dimensionality disaster refers to one of the most important problems encountered when processing high-dimensional data. It will directly affect the time and space complexity of data analysis algorithms, and even cause the algorithm to fail to converge. Therefore, when processing discrete manufacturing big data, effective dimensionality reduction methods need to be used to deal with the dimensionality disaster problem. Commonly used dimensionality reduction methods include principal component analysis (PCA), linear discriminant analysis (LDA), etc. Through these methods, the dimension of the data can be effectively reduced and the main information in the data can be retained, thereby improving the efficiency and accuracy of data analysis.

[0113] Based on domain knowledge, key features are extracted from the preprocessed data to identify key factors affecting production efficiency and quality, such as the die replacement frequency in the stamping process, the qualified rate of welding points, and the material loss ratio in the spraying process. The various data collected in the discrete manufacturing process usually have a certain correlation, such as the voltage and current in the processing process, the speed and cutting speed of the machine tool spindle, etc. This correlation will cause data redundancy and increase unnecessary computing burden. Therefore, when processing these data, data dimensionality reduction is particularly important. Data dimensionality reduction refers to retaining appropriate feature data from a high-dimensional data space and eliminating redundant data, thereby reducing the dimension of the data. The data after dimensionality reduction can not only retain the original amount of information, but also avoid the dimensionality disaster problem caused by data redundancy.

[0114] According to the preprocessed data set, select a suitable feature extraction algorithm to extract meaningful feature vectors. Specifically, PCA, LDA, TF-IDF and other algorithms can be used for feature extraction. In addition, deep learning and other technologies can also be used for feature extraction. The following is the specific extraction process of these technologies:

[0115] PCA (Principal Component Analysis) steps:

[0116] During the data preprocessing process, PCA was used to normalize the data to unify the scale of each feature. Then, based on the preprocessed data set, the covariance matrix between samples was calculated, which describes the correlation between features. The covariance matrix was decomposed by eigenvalue to obtain eigenvalues ​​and eigenvectors. The first k principal components were selected according to the size of the eigenvalues, and the eigenvectors corresponding to these principal components formed a new feature space. Finally, the original data was projected into the new feature space to obtain the reduced-dimensional data.

[0117] Using the PCA (Principal Component Analysis) algorithm to extract key features from machine tool manufacturing data is an effective method for data dimensionality reduction and feature extraction. This algorithm is mainly used to extract key features from machine tool manufacturing data to help optimize the processing process and improve processing quality and efficiency. PCA can identify the main variation directions in the data and project the original data into these directions, thereby retaining the most important information while removing noise and redundancy. Model building process:

[0118] (1) Data collection and preprocessing

[0119] First, various data are collected during the stamping machine processing, such as stamping speed, stamping force, die temperature, material thickness, dimensional deviation, surface quality, etc. These data constitute the original data set X, where each row represents a sample (such as a processing process) and each column represents a feature.

[0120] Data preprocessing steps may include missing value processing, outlier detection and replacement, data standardization (such as Z-score standardization), etc. to ensure the effectiveness of PCA analysis.

[0121] (2) Application of PCA algorithm

[0122] a. Calculate the covariance matrix

[0123] Let X be an n×p data matrix (n samples, p features), first calculate its covariance matrix S or correlation coefficient matrix R (if the data has been standardized). In PCA, the correlation coefficient matrix R is usually used because it is not affected by the feature dimension.

[0124]

[0125] (Note: Here X is the standardized data matrix, and the above formula is more accurately expressed as the calculation method of the correlation coefficient matrix under standardized data, rather than the direct covariance matrix.)

[0126] b. Eigenvalue decomposition

[0127] Perform eigenvalue decomposition on the covariance matrix S or the correlation coefficient matrix R to obtain the eigenvalues ​​λ1,λ2,…,λp and the corresponding eigenvectors ν1,ν2,…,ν p .

[0128] c. Select the principal component

[0129] According to the size of the eigenvalue, the eigenvectors corresponding to the first k largest eigenvalues ​​are selected, and these eigenvectors constitute a new feature space (principal component). The criterion for selecting k is that the cumulative contribution rate of the eigenvalue reaches a certain threshold (such as 90%).

[0130] d. Projection data

[0131] Project the original data X onto the selected principal component to obtain the reduced-dimensional data Y.

[0132] Y=XV k

[0133] Where V k is a matrix consisting of the first k eigenvectors.

[0134] Example scenario for stamping machine processing data. During the stamping machine processing, the quality and efficiency of the processing process are monitored by collecting the following types of data:

[0135] Punching speed (times / minute): indicates the number of punches completed by the punching machine per minute.

[0136] Punching force (tons): Indicates the maximum force applied to the workpiece during the punching process.

[0137] Mold temperature (℃): The temperature of the mold has an important impact on the quality of stamping parts and the life of the mold.

[0138] Material thickness (mm): The thickness of the material being stamped. Materials of different thicknesses require different stamping parameters.

[0139] Dimensional deviation (μm): The deviation between the actual size of the stamping part and the designed size.

[0140] Surface quality (rating or quantitative indicators): such as surface finish, number of scratches, etc., is used to evaluate the appearance quality of stamping parts.

[0141] By applying the PCA algorithm, the data are reduced in dimension and feature extracted to identify the key factors affecting the quality and efficiency of stamping processing.

[0142] After PCA analysis, it is found that the cumulative contribution rate of the first two principal components (PC1 and PC2) has reached more than 90%. These two principal components have already contained most of the important information in the data. The following results are obtained:

[0143] The first principal component (PC1): represents the combined effect of punching speed and punching force, as well as their indirect effects on die temperature, material thickness, and final dimensional deviation. This principal component reveals the best parameter combination to improve production efficiency while maintaining stable quality.

[0144] The second principal component (PC2): mainly reflects the direct impact of material thickness and mold temperature on surface quality. This principal component helps identify which material thickness and mold temperature combinations can produce the best surface quality, thus avoiding defects such as scratches and pits.

[0145] After PCA dimensionality reduction, we simplified the original six-dimensional data (stamping speed, stamping force, die temperature, material thickness, dimensional deviation, surface quality) into two dimensions (PC1 and PC2), which makes it more convenient in visualization, analysis and optimization. These key features (i.e. PC1 and PC2) are further used to establish a prediction model to predict the quality of stamping parts under different processing conditions.

[0146] LDA (Linear Discriminant Analysis) steps:

[0147] First, the mean vector of each class of samples, as well as the intra-class scatter matrix and inter-class scatter matrix of each class are calculated. Then, by maximizing the ratio of the inter-class scatter matrix to the intra-class scatter matrix, the projection matrix W is obtained. Then, the original data is projected into the low-dimensional space determined by W to obtain the reduced-dimensional data.

[0148] For process parameter characteristics, LDA may identify process parameter characteristics that are closely related to the quality of the machining process, such as cutting speed, feed rate, cutting depth, etc. Changes in these characteristics may directly affect the quality indicators such as dimensional accuracy and surface roughness of the machined parts.

[0149] The application of LDA in extracting key features guides the dimensionality reduction process by assuming the existence of one or more class labels. In the case of stamping machine process parameters, the importance of distinguishing different process results (such as product quality, production efficiency, etc.) is identified based on the classification labels of process parameters (e.g., classification based on product quality).

[0150] The process parameter data of the stamping machine, including stamping speed, stamping force, die temperature, material thickness, etc., are divided into two categories according to the quality of the final product: "high quality" and "low quality". LDA is used to identify which process parameters are most critical to distinguish between the two categories.

[0151] Data preparation collects N samples, each with d process parameters (features), and divides these samples into C categories. The number of samples in the i-th category is N i The total number of samples is

[0152]

[0153] The calculation formula for the intra-class scatter matrix is: Among them, X i is the sample set of the i-th category, μ i is the sample mean of the ith category:

[0154] Calculate the inter-class scatter matrix S B , the calculation formula is: Among them, μ is the global mean of all samples:

[0155] LDA is used to find the linear combination coefficient vector w, so that the projected data is as separated as possible between classes and as compact as possible within a class. This is achieved by maximizing the Fisher criterion: The maximum ω is achieved by solving the generalized eigenvalue: S B ω=λS w ω, where λ is the eigenvalue and w is the corresponding eigenvector.

[0156] The optimal projection direction w is obtained, and the original data is projected to this direction to obtain the reduced-dimensional data. The projection formula is: i =ω T x i , where x i is the i-th sample in the original data, y i It is the data after projection. By observing the projected data, it is identified that the original process parameters dominate the projection (that is, the parameters have a larger weight). These parameters are considered to be the key features to distinguish different categories, so as to check whether the extracted key features are consistent with the actual situation and improve the process control of the stamping machine.

[0157] TF-IDF (Term Frequency-Inverse Document Frequency) is mainly for text data:

[0158] When the machining and manufacturing data of machine tools exist in text form, TF-IDF is used to extract key features and applied to text data related to stamping machine tools (such as technical documents, operation manuals, user reviews, etc.) to identify key terms or phrases related to the functions, performance, or faults of stamping machine tools.

[0159] First, calculate the term frequency (TF) to count the number of times each word appears in the document. Then calculate the inverse document frequency (IDF). According to the ratio of the number of documents containing the word to the total number of documents, calculate the IDF value. IDF reflects the rarity of the word, and the rarer the word, the higher the IDF value. Finally, calculate the TF-IDF value by multiplying TF and IDF to obtain the TF-IDF value, which takes into account both the frequency of the word in the document and its rarity in the entire document set.

[0160] Use the TF-IDF algorithm to identify the most important words or phrases in these texts, which may represent the key features of the stamping machine tool (such as model, material compatibility, maximum punching force, fault mode, etc.).

[0161] (1) First, perform text preprocessing:

[0162] Clean the text data to remove punctuation marks, stop words (such as common but meaningless words like "of", "is", etc.), numbers, etc.

[0163] Convert the text to lowercase to eliminate the influence of case differences.

[0164] Optionally, perform stemming or lemmatization to convert the vocabulary to its basic form (such as converting "running" and "ran" both to "run").

[0165] (2) Calculate the term frequency (TF): For each document, calculate the number of times each vocabulary (or phrase, if dealing with phrases) appears.

[0166] The formula for term frequency (TF) is usually: n i,j is the number of times the vocabulary i appears in the document j, and the denominator is the sum of the number of times all vocabularies appear in the document j.

[0167] (3) Calculate the inverse document frequency (IDF). For each vocabulary, calculate its distribution in the entire document set. The formula for the inverse document frequency (IDF) is: where, is the total number of documents in the document set, and is the number of documents containing the vocabulary (plus 1 to avoid a zero denominator).

[0168] (4) Calculate the TF-IDF value by multiplying the term frequency (TF) of each vocabulary by its inverse document frequency (IDF) to obtain the TF-IDF value of the vocabulary.

[0169] The TF-IDF value reflects how important a term is in a specific document and how common it is in the entire document set.

[0170] (5) Extract key features, sort the words according to the TF-IDF value, and select the words or phrases with the highest value as key features. These key features represent the terms or phrases most relevant to stamping machines, which may reflect their performance, specifications, problems, or user concerns.

[0171] Through the above calculation steps, the document set contains the following two short descriptions about stamping machines:

[0172] Text 1: "This punching machine has a maximum punching force of 500 tons and is suitable for sheet metal processing."

[0173] Text 2: “New stamping machines have improved productivity, but some users have reported overheating problems.”

[0174] After preprocessing and TF-IDF calculation, it was found that the words “punching pressure”, “sheet metal processing”, “production efficiency” and “overheating problem” had high TF-IDF values ​​and were therefore identified as key features.

[0175] Deep learning feature extraction steps:

[0176] First, a deep neural network model is established. According to the task requirements and data characteristics, a suitable deep neural network structure is selected, such as convolutional neural network (CNN) for image data, recurrent neural network (RNN) or long short-term memory network (LSTM) for sequence data, etc. Through forward propagation calculation, the data is sent to the neural network for forward propagation calculation, and the abstract features of the data are extracted layer by layer through nonlinear transformation of multiple hidden layers. Through back propagation and parameter update, the gradient is calculated using the back propagation algorithm, and the network parameters are updated using optimization algorithms such as gradient descent to minimize the loss function. During or after training, features can be extracted from a hidden layer or output layer of the neural network, which are usually more discriminative and expressive than the original data.

[0177] The advantage of deep learning for feature extraction is that it can automatically learn high-level abstract representations of data and adapt to different tasks and data types.

[0178] Build prediction models and optimization models based on key features that affect production efficiency and quality;

[0179] Specifically, for model selection and training, machine learning algorithms (such as decision trees, support vector machines, deep neural networks, etc.) are used.

[0180] Combine historical data to build prediction models and optimization models to predict production bottlenecks, identify causes of defective products, and optimize production resource allocation. Use machine learning algorithms provided by the scikit-learn library (such as cluster analysis, association rule mining, regression analysis, etc.) to analyze preprocessed data, mine the rules and trends, and build a preliminary model based on feature vectors using machine learning algorithms. The performance of the model is evaluated through methods such as cross-validation.

[0181] For the optimization of the machine tool stamping production and manufacturing process, we can consider using prediction models and optimization models to improve production efficiency, reduce production bottlenecks, identify the causes of defective products, and optimize production resource allocation. The following are specific models, establishment steps, and application instructions, as shown in Table 3:

[0182]

[0183] By predicting key indicators such as equipment failure rate and production time, potential production bottlenecks can be discovered in advance. When it is predicted that the equipment failure rate will increase or the production time will be extended, equipment maintenance and adjustments can be carried out in advance to avoid production interruptions.

[0184] By analyzing the causes of defective products, such as raw material problems, equipment failures, operational errors, etc., corresponding improvement measures are formulated. Using the quality control optimization model, production parameters and processes are optimized to reduce the rate of defective products. Using the production scheduling optimization model, production tasks and production sequences are reasonably arranged to reduce line change time and adjustment time and improve production efficiency. Using the supply chain optimization model, material supply and inventory management are optimized to ensure that the material supply in the production process is sufficient and timely.

[0185] According to the optimization goal, the corresponding mathematical model is established, including decision variables, objective functions and constraints. There are many relevant decision variables for the process of stamping production line, including:

[0186]

[0187] The objective functions involved include:

[0188]

[0189]

[0190] The corresponding constraints include:

[0191]

[0192] By determining the decision variables, objective functions and constraints, a mathematical model applicable to different links in the stamping production line field is established to optimize the production process of the discrete manufacturing industry. Then, the optimization algorithm is used to solve the mathematical model, and the parameters of the model are adjusted according to the evaluation results and actual conditions to improve the performance and generalization ability of the model. Repeat the above steps until a relatively stable adaptive model is obtained. In the present invention, by comprehensively using the prediction model and the optimization model, the comprehensive optimization of the machine tool stamping production process can be achieved, the production efficiency can be improved, the production cost can be reduced, the product quality can be improved, and the generation of defective products can be reduced.

[0193] Intelligent diagnosis and optimization suggestions: Anomaly detection uses statistical analysis and machine learning methods to automatically identify abnormal situations in the production process. After the fault prediction model predicts different fault scenarios or different equipment components, the generated warning will be associated with specific equipment and components, and solutions will be provided for the troubleshooting and diagnosis of different models. Automation for models includes automated modeling and incremental automated optimization after the model is launched.

[0194] Aiming at the needs of fine equipment control and efficient operation and maintenance, the real-time collection of equipment operation data and process parameters by digital sensors, relying on the equipment management system, integrating industrial mechanisms and data models, realizes visual monitoring of equipment operation status, comprehensive analysis of operation efficiency and performance, as well as fault diagnosis and failure warning. By collecting data such as equipment operating conditions and process parameters in real time, the equipment operation status and key performance indicators are dynamically displayed through big data analysis and data visualization technology. Based on industrial mechanisms and data models, an equipment health prediction model is built to analyze equipment operation data in real time, and automatic alarms are issued when there are abnormal data analysis or parameter values ​​exceed the wide range. For example, based on machine transmission belt failure monitoring, lubrication unit device health monitoring and abnormal warning, etc. Based on algorithms such as cluster regression, deep learning, decision trees, knowledge graphs, etc., equipment fault analysis models and maintenance knowledge bases are built to extract fault features, analyze fault causes, decide on repair strategies, and generate maintenance work orders in a linked manner, such as rapid matching of CNC machine fault diagnosis and maintenance solutions. Online operation monitoring and fault diagnosis realize the optimization of data-driven equipment scheduling and operation and maintenance guarantee, improve the overall efficiency of equipment, and reduce the risk of non-fault downtime; at the same time, fault diagnosis and maintenance planning are carried out based on data analysis to improve the efficiency of fault repair and reduce downtime loss. Industrial cameras are used to collect the appearance or surface image data of the object under test, and the defect analysis model constructed by combining industrial mechanism models, big data analysis and deep learning algorithms can adaptively identify and locate quality defects, such as stamping plate surface defect detection.

[0195] Optimize discrete manufacturing production processes based on prediction models and optimization models.

[0196] Using detection and diagnosis technology, deep learning and other algorithms are used to extract key features and judge quality compliance. By setting thresholds and pattern recognition, abnormal vibration patterns of the stamping machine can be discovered in time, and maintenance warnings can be issued one week in advance to avoid sudden failures. The process of expressing, modeling and animating data through image processing, computer graphics and user interface technologies. It can present valuable information in large data sets in the form of scatter plots, matrix plots, surfaces, etc., deepen users' understanding of data and speed up the acquisition of information. According to the data type, data visualization can be divided into high-dimensional data visualization, time series data visualization, hierarchical data visualization and network data visualization.

[0197] By visually analyzing the data collected from the equipment, the root cause of the equipment failure can be quickly discovered, and repairs can be carried out in time to avoid production interruptions. At the same time, by visually analyzing the production management data, problems in the production process can be discovered and adjusted in time to improve production efficiency and product quality. In terms of optimization suggestions, the model analysis shows that the adjustment of the operation sequence in the welding process can significantly reduce the waiting time of the workpiece between each workstation, propose an optimized operation sequence, and simulate the expected reduction of the overall production cycle by 5%.

[0198] Implementation and feedback loop: Apply the optimization suggestions to actual production, continuously monitor changes in production data, evaluate the effects of improvement measures, and iterate the optimization model based on feedback information to form a closed-loop continuous improvement mechanism. For example, when determining product design parameters, the knowledge obtained from data mining can be used to assist decision-making and optimize product design. When determining product architecture, data mining can be used to discover the correlation between different design schemes and select the optimal scheme. When determining the process flow, the knowledge obtained from data mining can be used to identify the key factors affecting the process flow and make optimization adjustments. Adjust the operation sequence according to the optimization suggestions proposed by the model and implement a preventive maintenance plan. At the same time, continue to collect new data after implementation and compare and analyze it with the data before optimization. In the feedback loop, it was found that although the paint loss rate in the spraying process has decreased, there is still room for improvement. The team decided to introduce more sophisticated environmental parameter monitoring and retrain the model to further refine the optimization strategy.

[0199] Result analysis: Use adaptive models to analyze data, draw conclusions and make recommendations. Specifically, the results can be displayed in the form of charts through visualization tools (such as Tableau); or the results can be stored in a database for subsequent query and analysis.

[0200] Implementation effect:

[0201] The changes in product qualification rates were compared before and after the improvements were implemented. The qualification rate was 85% before the improvements and increased to 90% after the improvements.

[0202] In terms of production efficiency, after the improvement measures were implemented, the average production time of the production line was reduced from 15 minutes to 13 minutes.

[0203] After one quarter of practice, the overall efficiency of the production line increased by 10%, equipment downtime decreased by 20%, and defective rate decreased by 3%, with significant improvement in direct economic benefits. In addition, through continuous feedback loops and model iterations, the intelligence level and response speed of the production process have been continuously optimized.

[0204] In summary, the intelligent manufacturing process improvement method based on discrete manufacturing data model analysis, through the deep integration and innovative application of interdisciplinary technologies, has not only greatly improved the level of intelligence in the discrete manufacturing industry, but also brought significant economic benefits and social value to enterprises, demonstrating a high degree of professionalism and innovation. The core of the invention to be protected lies in its innovative data model construction, the application of deep reinforcement learning in production scheduling, the realization of intelligent decision support systems, and the contribution of big data analysis to production process improvement, which together constitute an efficient and intelligent discrete manufacturing process improvement method.

[0205] The present invention also provides a system for optimizing a discrete manufacturing production process using a data model, comprising:

[0206] The acquisition module is used to collect all data in the manufacturing process;

[0207] A preprocessing module is used to classify and preprocess the data respectively, and the types of data classification include image data, video data and other data;

[0208] The extraction module is used to extract the key features that affect production efficiency and quality from the preprocessed data;

[0209] Building modules for building prediction models and optimization models based on key features that affect production efficiency and quality, respectively;

[0210] The optimization module is used to optimize the discrete manufacturing production process based on prediction models and optimization models.

[0211] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for optimizing the production process of discrete manufacturing industry using a data model, characterized in that: include: Collect all data in the manufacturing process; Classifying the data and preprocessing them respectively, wherein the types of the data classification include image data, video data and other data; Extracting key features that affect production efficiency and quality from the preprocessed data; Based on the key features affecting production efficiency and quality, a prediction model and an optimization model are respectively constructed; The production process of discrete manufacturing industry is optimized based on the prediction model and the optimization model.

2. The method for optimizing the production process of discrete manufacturing industry using a data model according to claim 1, characterized in that: The data includes system data and equipment data. The system data includes ERP system data, MES system data, PLM system data, BPM system data and VMS system data. The ERP system data includes raw material purchase orders, inventory information and supplier information; the MES system data includes production plans, order status, equipment usage, material consumption and production progress; the PLM system data includes product design, engineering changes and product configuration information; the BPM system data includes business process monitoring and management data; the VMS system data includes monitoring and maintenance production equipment information; the equipment data includes processing data, model data, process data and quality inspection data.

3. The method for optimizing the production process of discrete manufacturing industry using a data model according to claim 2, characterized in that: Classifying the data and preprocessing the data respectively include cleaning, deduplication and normalization of the data in sequence; The data cleaning step includes: Missing value processing: the missing parts of the image data and the video data are processed by repair technology or cropping, and the missing values ​​of the other data are filled according to the business logic; Outlier processing: abnormal frames or areas of the image data and the video data are screened or replaced, and abnormal values ​​of other data are identified by setting thresholds or statistical methods, and corrected or deleted; The step of deduplicating the data includes: Unique identifier comparison: compare the unique identifiers in the other data to identify and delete duplicate records; Hash value comparison, calculating the hash values ​​of the image data and the video data, and identifying and deleting duplicate files or segments by comparing the hash values; Matching method: matching multiple fields in the other data to identify and delete duplicate data; The step of normalizing the data includes: Rotate the corresponding normalization method for different fields in the other data according to business requirements and data analysis objectives; Standardizing the image data by adjusting the pixel value range and the color space; The frame rate and resolution parameters of the video data are adjusted and standardized.

4. The method for optimizing the production process of discrete manufacturing industry using a data model according to claim 3 is characterized in that: Extracting key features that affect production efficiency and quality from the preprocessed data respectively includes selecting PCA, LDA, TF-IDF algorithms and deep learning for feature extraction; Using PCA algorithm to extract features from the processed data; Using LDA algorithm to extract features from the process data; Use the TF-IDF algorithm to extract features from data in text form; Use deep learning and select the corresponding deep neural network structure to extract features from the remaining data.

5. The method for optimizing the production process of discrete manufacturing industry using a data model according to claim 4, characterized in that: The prediction model building step comprises: Collect historical data during the stamping production process of machine tools, including production volume, equipment status, and product quality; Cleaning, converting and normalizing the historical data; Extracting features related to the prediction target from the processed historical data; Training the prediction model using features related to the prediction target; The trained prediction model is evaluated.

6. The method for optimizing the production process of discrete manufacturing industry using a data model according to claim 5, characterized in that: Based on the different data, the prediction models include time series prediction models, regression prediction models and machine learning prediction models.

7. The method for optimizing the production process of discrete manufacturing industry using a data model according to claim 6, characterized in that: The optimization model building step comprises: Determine key indicators that need to be optimized, including production efficiency, product quality and cost; According to the optimization goal, a corresponding mathematical model is established, wherein the mathematical model includes decision variables, objective functions and constraint conditions; The mathematical model is solved using an optimization algorithm to obtain an optimal solution.

8. The method for optimizing the production process of discrete manufacturing industry using a data model according to claim 7, characterized in that: The optimization models include a production scheduling optimization model, a quality control optimization model, and a supply chain optimization model.

9. A system for optimizing the production process of discrete manufacturing industry using a data model according to any one of claims 1 to 8, characterized in that: include: The acquisition module is used to collect all data in the manufacturing process; A preprocessing module, used to classify the data and preprocess them respectively, the data classification types include image data, video data and other data; An extraction module, used to extract key features that affect production efficiency and quality from the preprocessed data; A construction module, used to construct a prediction model and an optimization model based on the key features affecting production efficiency and quality; The optimization module is used to optimize the discrete manufacturing production process based on the prediction model and the optimization model.

Images