Self-sensing-self-analysis-self-decision-making textile intelligent manufacturing system construction method

Abstract

The invention relates to the technical field of intelligent textile manufacturing, in particular to a self-sensing-self-analysis-self-decision intelligent textile manufacturing system construction method, and aims to solve the problems that an existing system is insufficient in high-dimensional heterogeneous data processing, low in whole-process integration level, dependent on experience in operation and maintenance and difficult in intelligent upgrading of small and medium-sized enterprises. According to the method, upstream and downstream data of an industrial chain are transversely integrated, a 5G information platform is established, high-dimensional data are longitudinally processed, a digital twinborn visual monitoring system is constructed, a process rule base and a production process are optimized end to end, and data-driven operation and maintenance (health degree / importance index construction) are realized in combination with clustering, an Apriori algorithm and a D-S evidence theory. According to the method, the value of textile data is improved, full-process collaboration and'self-sensing-self-analysis-self-decision 'are realized, the operation and maintenance cost is reduced, the production flexibility and the quality stability are improved, and a generalizable scheme is provided for digital transformation of the textile industry.

Description

Technical Field

[0001] This invention relates to the field of intelligent textile manufacturing technology, specifically a method for constructing a self-sensing, self-analyzing, and self-decision-making intelligent textile manufacturing system. Background Technology

[0002] The textile industry is an important pillar industry of my country's national economy, and its transformation and upgrading are of great significance to promoting the high-quality development of the manufacturing industry.

[0003] Internationally, textile technology-advanced countries such as the US, Japan, and Germany have taken the lead in the deep application of intelligent manufacturing technologies. For example, the Trumpf smart factory in Chicago, USA, has introduced intelligent equipment and built an integrated manufacturing system platform with unified management and control. This platform allows for customized configuration of scenarios and software upgrades based on the needs of textile applications, revitalizing traditional factories. Carrington, a textile giant in the UK, has built a real-time tracking system, allowing employees to monitor production line status via iPads and using Beacon tracking devices to ensure accurate material scheduling and rapid error detection. Marui Textiles in Japan leverages IoT technology to integrate textile manufacturing with next-generation information technology, creating an on-demand production business model. Okada Textile Machinery Plant has integrated information and communication technologies to build a digital workshop, enabling rapid information collection, high-speed transmission, and centralized computing. These countries have developed advanced models such as "lights-out factories" and "super-digital spinning workshops," using data mining to achieve "single-spindle quality inspection" of spinning machines, resulting in a high level of full-process intelligence.

[0004] In my country, leading textile companies are also actively exploring intelligent manufacturing pathways: Shandong Zhengkai New Material Co., Ltd. has partnered with China Telecom to launch the "5G+Textile Industrial Internet" project, promoting the construction of the first multi-variety intelligent spinning system and industrial internet platform in China; Shandong Chaoyue Textile Co., Ltd. has joined forces with the Athena platform to create a "5G Smart Factory Big Data Platform," integrating eight centers to achieve order-driven full-process management; Shandong Weiqiao Group has built a 5G smart factory, whose automatic packaging production line is compatible with two packaging modes and features fully unmanned automatic transportation throughout the entire process. These practices have laid the foundation for intelligent manufacturing in my country's textile industry, but many problems still need to be addressed: First, the application of intelligent equipment has led to a geometric increase in textile process data, and existing data processing methods are unable to handle the efficient analysis of high-dimensional and heterogeneous data, resulting in insufficient data value mining; Second, most enterprises' intelligent manufacturing systems focus on a single link (such as production monitoring), lacking integrated management and control over the entire life cycle of product design, manufacturing, and product services, with low horizontal (upstream and downstream business), vertical (process-quality-materials), and end-to-end (full process) integration; Third, enterprise operation and maintenance decisions rely heavily on experience-based judgment, lacking accurate diagnosis of equipment failures and health status assessments, resulting in high operation and maintenance risks; Fourth, some small and medium-sized enterprises blindly follow the trend of building "smart factories," but due to unsystematic technical solutions and weak data collaboration capabilities, they are unable to achieve true intelligent upgrades.

[0005] Therefore, how to deeply integrate new-generation information technology with textile technology, based on the automation of production processes and equipment, and combined with technologies such as artificial intelligence and big data, to achieve flexible and intelligent upgrading of the entire textile process and build a human-machine collaborative intelligent manufacturing system with "self-sensing-self-analysis-self-decision-making" has become a key issue to support the digital transformation of my country's textile industry and realize the strategy of becoming a textile powerhouse. This invention is proposed accordingly. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of existing intelligent textile manufacturing systems, such as insufficient data processing capabilities, low integration throughout the entire lifecycle, and lack of intelligent operation and maintenance decision-making. It provides a method for constructing a self-sensing, self-analyzing, and self-decision-making intelligent textile manufacturing system, which realizes information flow-driven optimization of the entire textile process, enhances the system's "self-sensing" (data acquisition and status awareness), "self-analyzing" (data mining and feature extraction), and "self-decision-making" (operation and maintenance and production decision-making) capabilities, and forms an intelligent manufacturing model of human-machine integration.

[0007] The technical solution adopted by this invention to solve its technical problem is: a method for constructing a self-sensing-self-analysis-self-decision-making intelligent manufacturing system for textiles, comprising the following steps: (1) Horizontal integration of intelligent manufacturing in the textile industry: a) Relying on the enterprise's "digital spinning workshop" system, cluster analysis is used to collect and analyze business data of upstream fiber manufacturing and spinning, midstream fabric weaving and dyeing, and downstream apparel manufacturing and distribution. The correlation, autocorrelation and causal relationship between the data are used to characterize the complex high-dimensional textile data relationship. b) Explore information integration methods and data flow mining algorithms based on equipment, production, and business using the Monte Carlo method. After verification by examples, develop an information integration platform in conjunction with 5G technology to achieve horizontal integration of upstream and downstream businesses. (2) Vertical integration of intelligent manufacturing in the textile industry: a) Use random matrices to extract heterogeneous features from complex high-dimensional textile data, remove singular values ​​that do not conform to the prediction by using correlation matrices, perform singular value decomposition on the remaining feature count to analyze the correlation between features and classes, complete feature selection through redundancy analysis, and finally use semi-supervised regularization methods to construct objective functions and process vector features to achieve feature mapping dimensionality reduction. b) Based on virtualization technology, analyze the virtualization encapsulation characteristics and semantic verification mechanism of textile manufacturing resources, services and capabilities. Relying on the "digital spinning workshop" system, obtain process, quality and material dimension datasets and process missing and outlier values. Use deep convolutional autoencoders to predict datasets and achieve multi-dimensional integration through Stacking ensemble models. c) Utilize digital twin technology to construct a real-time visual monitoring system for textile workshops, study multi-source heterogeneous data modeling and transmission, event-driven virtual-real mapping, complex event processing logic modeling and information visualization push methods, and develop 3D visualization technology; (3) End-to-end integration of intelligent manufacturing in the textile industry: A rule base for collaborative operation of spinning, weaving, and AGV carts is constructed using digital control technology. The rule base is optimized through continuous textile process control and real-time feedback of production data. Combined with underlying sensing devices, AGV carts, control models, and production systems, end-to-end digital twins are studied to achieve dynamic optimization of the physical processes of spinning production and fabric forming. (4) Data-driven operation and maintenance decisions for textile enterprises: a) Utilize the K-means clustering algorithm to mine process, production, and management data throughout the entire lifecycle of textiles, extract single-dimensional state variable fault features, and analyze different business scenarios; b) The Apriori algorithm is used to mine data association rules in the scenario, a key performance matrix is ​​established, the spatiotemporal characteristics of equipment failure are analyzed using high-dimensional random matrix theory, and single-dimensional and multi-dimensional diagnostic results are synthesized using DS evidence theory. Based on the fault diagnosis criteria, equipment health index and importance index are established, and the risk of operation and maintenance decision-making is reduced by combining the system operating status.

[0008] Specifically, the information integration platform described in step (1)b) integrates an IoT acquisition center, a real-time production data monitoring center, an intelligent production scheduling center, and an intelligent operation and maintenance center to achieve centralized management and real-time access to all business data.

[0009] Specifically, the objective function constructed by the semi-supervised regularization method in step (2) a) aims to minimize the dimension of vector features, and the optimal dimensionality reduction parameters are determined through iterative calculation.

[0010] Specifically, in step (2)b), missing value processing adopts the mean imputation method, and outlier processing adopts the 3σ criterion to remove data exceeding the mean ± 3 times the standard deviation.

[0011] Specifically, the monitoring system built by the digital twin technology described in step (2)c) supports real-time retrieval of equipment operating parameters, production progress and quality inspection data, and realizes automatic alarm for abnormal situations.

[0012] Specifically, the rule base optimization cycle in step (3) a) is every 24 hours, and the collaborative operation parameters are adjusted based on the deviation analysis of the production data of the previous cycle.

[0013] Specifically, the number of clusters in the K-means clustering algorithm described in step (4) a) is set to 5-8, based on the business stages of the entire life cycle of textiles.

[0014] Specifically, the key performance matrix mentioned in step (4)b) includes equipment operating efficiency, product qualification rate, material turnover cycle and energy consumption index.

[0015] Specifically, the equipment health index mentioned in step (4) b) is based on a percentage system, calculated by taking into account the equipment's running time, failure frequency, and maintenance records. A score of 80 or above indicates a healthy state.

[0016] Specifically, the method also includes conducting trial runs and debugging of the constructed intelligent manufacturing system, optimizing the parameters of each integrated module based on the trial run data, and ensuring the stability of the system's "self-sensing-self-analysis-self-decision-making" functions.

[0017] The beneficial effects of this invention are: Enhance the ability to process and extract value from high-dimensional heterogeneous textile data This method addresses the exponential growth in data brought about by the application of intelligent equipment in the textile industry. Through techniques such as cluster analysis, random matrix feature extraction, and semi-supervised regularization dimensionality reduction, it effectively processes high-dimensional, heterogeneous textile business data (covering dimensions such as process, quality, materials, and equipment). It eliminates redundant and noisy data, transforming high-dimensional data into low-dimensional analyzable data, thus overcoming the shortcomings of traditional data processing methods, such as low efficiency and insufficient value extraction. Simultaneously, by mining the correlations, autocorrelations, and causal relationships between data points, it clarifies the logical connections between data in various business processes, providing data support for production parameter optimization and quality control, and fully releasing the application value of textile data.

[0018] Achieving integrated and collaborative operation across the entire textile process Breaking away from the limitations of traditional intelligent textile manufacturing systems that focus on a single link (such as production monitoring only), this system constructs a full lifecycle management and control system through horizontal, vertical, and end-to-end triple integration: Horizontal integration covers the entire industrial chain data from upstream fiber manufacturing to downstream apparel distribution, enabling upstream and downstream business collaboration; Vertical integration merges multi-dimensional data on processes, quality, and materials, achieving deep integration of manufacturing resources, services, and capabilities; End-to-end integration connects the entire production process from raw material warehousing to finished product delivery, and uses digital twins to achieve dynamic optimization of physical processes. This triple integration significantly improves the overall system synergy, solving the problems of low integration and data fragmentation in traditional systems.

[0019] Implementing the core capabilities of "self-sensing, self-analysis, and self-decision-making" to promote human-machine collaborative intelligent manufacturing. At the self-sensing level: relying on the Internet of Things acquisition center and the digital twin 3D visualization monitoring system, the system collects equipment operating parameters, production progress, and quality inspection data in real time, enabling comprehensive and real-time perception of the status of the textile workshop. Abnormal situations can be automatically alarmed, solving the problems of lagging and limited coverage of traditional manual monitoring. Self-analysis level: Through technologies such as deep convolutional autoencoders, stacking ensemble models, K-means clustering, and Apriori association rule mining, production trends, equipment fault characteristics, and spatiotemporal features are accurately analyzed, and single-dimensional and multi-dimensional diagnostic results are synthesized to provide a scientific basis for decision-making and replace the traditional experience-based analysis mode. From the decision-making level: Based on a data-driven operation and maintenance decision-making model (equipment health and importance index) and dynamic optimization of the process collaborative operation rule base, the operation and maintenance strategy is automatically generated (emergency / planned maintenance work orders), and production parameters are dynamically adjusted, reducing reliance on manual decision-making, improving decision-making efficiency and accuracy, and forming a human-machine integrated intelligent manufacturing model.

[0020] Optimize equipment operation and maintenance management to reduce operation and maintenance risks and costs. By constructing an equipment health index (combining runtime, failure frequency, and maintenance records) and an importance index (based on the equipment's impact on production), and combining this with DS evidence theory, equipment failures are accurately diagnosed, enabling dynamic prioritization of maintenance and avoiding both over-maintenance and under-maintenance issues. Simultaneously, deep integration of maintenance decisions with the production system (MES system) allows for proactive adjustments to production tasks, reducing the impact of equipment failures on production, ensuring production continuity, lowering maintenance labor and spare parts costs, and improving equipment operational stability.

[0021] Enhance production flexibility and quality stability to adapt to changes in market demand. End-to-end digital twin technology supports simulation and optimization for different production scenarios (such as rush orders and equipment failures). It can quickly adjust physical process parameters such as spinning production and fabric forming, improve production flexibility, and better meet the needs of personalized and rush orders. Full-process data traceability and real-time control (such as process parameter deviation prediction and real-time quality data monitoring) can promptly detect and correct production anomalies, reduce the generation of defective products, improve product quality stability, and enhance the company's market competitiveness.

[0022] Provide scalable system solutions for the digital transformation of the textile industry This method uses a "digitalized spinning workshop" as its platform, with a clear technical path (forming a closed loop from data collection to decision implementation). It also caters to the application needs of textile enterprises of different sizes, particularly addressing the challenges faced by SMEs in intelligent upgrading due to unsystematic technical solutions and weak data collaboration capabilities. Its implementation experience can be extended to textile enterprises in different sub-sectors such as cotton spinning, weaving, and dyeing and finishing, providing key technical support for my country's textile industry to transform from "large to strong" and supporting the national strategy of becoming a textile powerhouse. Attached Figure Description

[0023] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0024] Figure 1 This is a schematic diagram representing the complex high-dimensional textile data relationships in a self-sensing-self-analysis-self-decision-making intelligent textile manufacturing system construction method provided by the present invention. Figure 2 A schematic diagram of a three-dimensional visualization method for a textile workshop based on digital twins in a self-sensing-self-analysis-self-decision-making intelligent manufacturing system construction method provided by the present invention. Figure 3 A schematic diagram of the physical process dynamic optimization method based on digital twin in the self-sensing-self-analysis-self-decision-making intelligent textile manufacturing system construction method provided by the present invention; Figure 4 This is a schematic diagram of the intelligent operation and maintenance decision-making method in the self-sensing-self-analysis-self-decision-making intelligent textile manufacturing system construction method provided by the present invention. Detailed Implementation

[0025] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0026] like Figures 1-4 As shown, the present invention discloses a method for constructing a self-sensing, self-analyzing, and self-decision-making intelligent textile manufacturing system. Using an intelligent spinning workshop as a carrier and end-to-end data flow as a foundation, it constructs a "self-sensing, self-analyzing, and self-decision-making" intelligent textile manufacturing system through four core steps: horizontal integration, vertical integration, end-to-end integration, and data-driven operation and maintenance decision-making. The specific details are as follows: (1) Horizontal integration of intelligent manufacturing in the textile industry This step aims to achieve data collaboration and information integration across the upstream and downstream businesses of the textile industry chain, supporting "self-sensing" data collection and full business coverage: Characterizing complex, high-dimensional textile data relationships: Relying on the company's existing "digital spinning workshop" system, we collect comprehensive business data from upstream fiber manufacturing (e.g., fiber composition and strength data), spinning (e.g., spinning speed and twist data), midstream fabric weaving (e.g., weaving density and breakage frequency data), dyeing and finishing (e.g., dye concentration and temperature data), and downstream apparel manufacturing and distribution (e.g., cutting accuracy and logistics timeliness data). We employ cluster analysis methods (e.g., an improved algorithm of K-means clustering) to group and analyze the collected high-dimensional data, focusing on mining the correlations (e.g., the relationship between fiber strength and yarn breakage rate), autocorrelation (e.g., the relationship between the speed of the same spinning machine at different times), and causal relationships (e.g., the causal relationship between dye concentration and fabric color fastness). This clarifies the logical connections between data in each business segment, laying the foundation for subsequent data integration.

[0027] Information Integration Platform: First, the Monte Carlo method is used to simulate different information integration schemes (such as the fusion of equipment data and production data, and the transmission protocol of business data). This explores information integration methods and data flow mining algorithms based on equipment, production, and business (such as data flow mining based on association rules). Actual production data from a spinning workshop is selected for case verification to ensure the feasibility of the integration methods. Subsequently, combined with 5G technology (supporting high-speed, low-latency data transmission), an information integration platform is developed. This platform integrates an IoT acquisition center (for real-time acquisition of equipment sensor data), a real-time production data monitoring center (for dynamic display of production progress), an intelligent scheduling center (for adjusting production plans based on orders), and an intelligent operation and maintenance center (for monitoring equipment status). This enables centralized management and real-time sharing of horizontal business data from upstream fiber manufacturing to downstream apparel distribution, supporting the "self-sensing" of the entire industry chain.

[0028] (2) Vertical integration of intelligent manufacturing in the textile industry This step focuses on the deep integration and data dimensionality reduction analysis of various dimensions of textile manufacturing (process, quality, materials) to enhance "self-analysis" capabilities: Dimensionality Reduction through Heterogeneous Feature Mapping for High-Dimensional Data: Addressing the high-dimensional and heterogeneous nature of textile data, this approach first extracts heterogeneous features from complex high-dimensional data (such as equipment operating parameters, process parameters, and quality inspection data) using a random matrix, obtaining an initial feature set. Then, the correlation between each feature and the prediction target (e.g., product qualification rate) is calculated using a correlation matrix, eliminating redundant singular values ​​with correlations below a threshold (e.g., 0.1) to reduce noise interference. Singular value decomposition is performed on the remaining features to obtain a correlation matrix between features and data categories, analyzing feature redundancy (e.g., retaining only one of two highly correlated process parameters) to complete feature selection. Finally, a semi-supervised regularization method is used to construct an objective function that minimizes the dimensionality of vector features. The optimal dimensionality reduction parameters are determined through iterative calculations (e.g., gradient descent), processing vector features to achieve feature mapping dimensionality reduction, transforming high-dimensional data into low-dimensional analyzable data, and improving data processing efficiency.

[0029] Multi-dimensional integration approach: Based on virtualization technology (such as cloud computing virtualization), analyze the virtualization encapsulation characteristics of textile manufacturing resources (such as spinning machines and weaving machines), services (such as production scheduling services), and capabilities (such as quality inspection capabilities) (e.g., standardized resource interfaces, callable service protocols), clarify their virtualization structure and semantic verification mechanism (ensuring semantic consistency between virtual and physical resources); rely on the "digital spinning workshop" system to obtain datasets in three dimensions: process (such as spinning twist, weaving speed), quality (such as fabric defect rate, color fastness), and materials (such as raw material inventory, material delivery time), and adopt a uniform... The value imputation method handles missing values ​​in the data (e.g., missing quality inspection data for a certain period is filled with the average value of the same period on that day), and the 3σ criterion is used to remove outliers (e.g., spinning speed data that exceeds the mean ± 3 times the standard deviation). A deep convolutional autoencoder is used to predict the three-dimensional dataset (e.g., predicting material demand and process parameter deviations for the next 2 hours). Then, a stacking ensemble model (using the prediction results of multiple basic models as input to build a meta-model for final prediction) is used to integrate process, quality, and material data to achieve deep integration of multi-dimensional data and support "self-analysis" data mining and prediction.

[0030] Digital Twin 3D Visualization Method: Utilizing digital twin technology, a virtual mapping model of the textile workshop is constructed, which is synchronized with the physical workshop in real time. Research is conducted on modeling and transmission methods for multi-source heterogeneous textile data (e.g., encapsulating data in JSON format and transmitting via the MQTT protocol), event-driven virtual-physical mapping methods (e.g., physical equipment failure triggering alarm displays for corresponding components in the virtual model), complex event-handling textile workshop logic modeling methods (e.g., transforming the chain event of "yarn breakage - machine shutdown for maintenance - material replenishment" into logical rules for the virtual model), and information visualization and push methods (e.g., pushing equipment operation reports via web or mobile devices). Based on the above research, 3D visualization technology for textile workshops is developed to achieve real-time visual monitoring of workshop equipment status, production progress, and quality data, improving the intuitiveness of "self-sensing" and the accuracy of "self-analysis."

[0031] (3) End-to-end integration of intelligent manufacturing in the textile industry This step enables collaborative optimization of the entire textile production process, promoting "self-decision-making" production control: End-to-end digital twin approach: Employing digital control technologies (such as PLC programming control), this approach streamlines the processes of spinning (such as roving and spinning), weaving (such as rapier loom operation), and AGV (automated guided vehicle) collaboration (such as raw material delivery and finished product transportation), constructing a collaborative operation rule base (such as the rule that AGVs prioritize yarn delivery when the spinning machine breaks down). Through continuous textile process control (such as real-time monitoring of production cycle time for each process) and real-time feedback of production data (such as process completion rate and material consumption rate), the rule base is optimized every 24 hours (e.g., based on excessively long AGV waiting times in the previous cycle). The problem is to adjust the delivery route rules. By combining the underlying sensing textile equipment (such as sensors and RFID tags), AGVs, control models (such as PID control models), and production systems (such as MES systems), an end-to-end digital twin model of the textile workshop is constructed. This model covers the entire process from raw material warehousing to finished product warehousing. Based on the digital twin model, different production scenarios (such as expedited orders and equipment failures) are simulated to study the dynamic optimization of physical processes such as spinning production and fabric forming (such as adjusting the spinning speed to match the demand for expedited orders and optimizing the weaving process to reduce the defect rate), so as to achieve "self-decision" control of the entire process.

[0032] (4) Data-driven operation and maintenance decision-making for textile enterprises This step enables intelligent decision-making in equipment operation and maintenance, improving the "self-decision-making" capability: Textile lifecycle business scenario analysis: Using the K-means clustering algorithm, cluster analysis is performed on the process (such as spinning process parameters for different products), production (such as production time for different batches), and management (such as equipment maintenance records) data of the entire textile lifecycle. The number of clusters is set to 5-8 (based on business stages such as raw material preparation, spinning, weaving, dyeing and finishing, finished products, logistics, and after-sales service). Based on the clustering results, single-dimensional state quantity fault features are extracted (such as spinning machine bearing temperature and weaving machine yarn breakage count), and typical fault types in each business scenario are analyzed (such as excessive fiber impurities in the raw material preparation stage and roller wear in the spinning stage), clarifying the operation and maintenance requirements of different scenarios.

[0033] Intelligent Operation and Maintenance Decision Model Construction: The Apriori algorithm is used to mine data association rules under different business scenarios (e.g., "bearing temperature > 80℃ and running time > 1000h → failure probability increases by 50%). Based on these association rules, a key performance matrix is ​​established, including equipment operating efficiency (e.g., equipment uptime), product qualification rate, material turnover cycle, and energy consumption indicators. High-dimensional random matrix theory is used to analyze the spatiotemporal characteristics of equipment failures (e.g., concentrated periods of failure in spinning machines in a certain area, and failure propagation paths). DS evidence theory is used to combine single-dimensional diagnostic results (e.g., temperature-based fault diagnosis) with multi-dimensional diagnostic results (e.g., combining temperature...). Information is synthesized for fault diagnosis of temperature, vibration, and noise to address the uncertainty of diagnostic results. Based on the synthesized fault diagnosis criteria, an equipment health index (out of 100, calculated by combining equipment runtime, fault frequency, and maintenance records, with a score of 80 or above considered healthy) and an importance index (scored according to the degree of equipment's impact on production, such as spinning machines being more important than auxiliary equipment) are established. Taking into account the system's operating status (such as current production order volume and equipment load), operation and maintenance strategies are formulated (such as prioritizing shutdown and maintenance of equipment with a health score below 60 and high importance), reducing the risk of equipment operation and maintenance decisions and achieving "self-decision-making" operation and maintenance management.

[0034] To make the technical solution of this invention clearer and easier to understand, the following specific implementation method is provided in conjunction with the intelligent spinning workshop transformation project of a medium-sized textile enterprise in Shandong (hereinafter referred to as "the target enterprise"), to explain in detail the implementation process and effects of this invention. The target enterprise mainly engages in cotton yarn spinning and grey fabric weaving. Its original production lines suffered from problems such as scattered data, reliance on experience for operation and maintenance, and slow production response. After adopting the method of this invention to build an intelligent manufacturing system, significant upgrades were achieved.

[0035] Example 1: Horizontal Integration Implementation of Intelligent Manufacturing in the Textile Industry Implementation Preparation: The target enterprise first improves its "digital spinning workshop" system. This involves installing raw material detection sensors (collecting fiber composition, length, and strength data) in the upstream fiber procurement stage; installing speed, twist, and temperature sensors (collecting spinning speed, twist deviation, and equipment temperature data) on the roving frames, spinning frames, and winding machines in the spinning workshop; installing density sensors and thread breakage detectors on the rapier looms in the midstream weaving workshop (collecting weaving density and thread breakage frequency data); deploying dye concentration detectors and temperature controllers in the dyeing and finishing workshop (collecting dye concentration and dyeing and finishing temperature data); and connecting downstream garment processing partners and logistics links to order completion rate and logistics delivery timeliness data via API interfaces. Simultaneously, a technical team is formed to be responsible for data acquisition and algorithm debugging, using Python as the data processing language and the TensorFlow framework to support algorithm execution.

[0036] Implementation of Complex High-Dimensional Data Relationship Characterization: The technical team collected approximately 500,000 production data points from each of the above stages over a continuous month (including 28 process parameters, 15 quality parameters, and 12 material parameters). An improved K-means clustering algorithm (introducing a silhouette coefficient to optimize the number of clusters, ultimately determining 6 clusters) was used to analyze the data. For example, in the clustering of data from the spinning stage, a cluster group with "fiber strength > 5 cN / dtex and spinning speed < 1200 m / min" was found, with a breakage rate of only 0.5%, far lower than other groups (average breakage rate 2.3%), clearly demonstrating the negative correlation between fiber strength and spinning speed. In the dyeing and finishing stage, causal relationship analysis determined that "dye concentration of 3%-5% and temperature of 80-85℃" are key causal conditions for achieving the required fabric color fastness, providing a basis for subsequent production parameter optimization.

[0037] Information integration platform development and implementation: First, the Monte Carlo method was used to simulate three information integration schemes (Scheme 1: equipment data and production data are stored in separate databases and synchronized periodically; Scheme 2: real-time fusion storage with a caching mechanism; Scheme 3: integration after preprocessing based on edge computing). By simulating the transmission latency and data loss rate of each scheme, it was found that the average transmission latency of Scheme 2 was 15ms (far lower than 50ms of Scheme 1 and 30ms of Scheme 3), and the data loss rate was only 0.1%, thus Scheme 2 was determined to be the optimal integration method. Subsequently, production data (approximately 80,000 records) from a week in the spinning workshop of the target company was selected for case verification. The verification results showed that Scheme 2 can achieve real-time matching of equipment data and production data, with a data mining accuracy rate of 88%, meeting the actual needs. Based on this solution and combined with China Telecom's 5G network (100Mbps bandwidth, latency within 10ms), an information integration platform was developed. The platform includes an IoT data acquisition center (collecting sensor data in real time via the MQTT protocol), a real-time production data monitoring center (displaying production progress using ECharts visualization components), an intelligent scheduling center (developing production plans based on genetic algorithms), and an intelligent operation and maintenance center (monitoring equipment temperature and vibration data in real time). After the platform went live, it achieved horizontal data integration from upstream fiber procurement to downstream logistics, reducing data query response time from 30 seconds to 2 seconds, and achieving 100% data sharing across the entire industry chain.

[0038] Implementation Results: After the horizontal integration was implemented, the target enterprise achieved "self-aware" collection and centralized management of all business data. The data coverage dimensions increased from the original 18 to 55, and the data processing efficiency improved by 45%, laying a data foundation for subsequent vertical integration and end-to-end integration. At the same time, through data relationship analysis, the matching parameters of spinning speed and fiber strength were optimized, the spinning breakage rate was reduced by 60%, and the color fastness qualification rate in the dyeing and finishing process was improved to 98%.

[0039] Example 2: Vertical Integration Implementation of Intelligent Manufacturing in the Textile Industry Implementation Preparation: Based on horizontal integration, the target enterprise sorts out high-dimensional data in production (a total of 60 dimensions, including 18 equipment operating parameters, 22 process parameters, 12 quality parameters, and 8 material parameters), and uses MATLAB software for data dimensionality reduction and modeling; at the same time, a virtualization server (using VMware ESXi system) is deployed to build a cloud computing platform for virtualization processing of multi-dimensional data; digital twin development tools (such as Unity3D) are purchased, and a 3D modeling team is formed to be responsible for building the virtual model of the workshop.

[0040] Dimensionality reduction implementation for heterogeneous feature mapping of high-dimensional data: First, heterogeneous features of 60-dimensional data are extracted using a random matrix, resulting in an initial feature set containing 45 features. The correlation between each feature and "product qualification rate" is calculated using a correlation matrix. Twelve features with a correlation below 0.1 (such as workshop humidity and raw material warehousing time, which have minimal impact on the qualification rate) are removed, leaving 33 features. Singular value decomposition is performed on these 33 features to obtain the correlation matrix between features and data categories. It is found that the correlation between "spinning twist" and "yarn strength" reaches 0.92, indicating serious redundancy. "Yarn strength" is further removed (since "spinning twist" can be directly controlled), ultimately retaining 32 features. A semi-supervised regularization method is used to construct an objective function, expressed as: min||X-WY|| 2 +λ||L(Y)|| 2 Where X is the original feature matrix, W is the mapping matrix, Y is the feature matrix after dimensionality reduction, λ is the regularization parameter, and L(Y) is the semi-supervised regularization term. Through iterative calculation using the gradient descent method, it was determined that the objective function is minimized when λ=0.01. At this point, the feature dimension after dimensionality reduction is 10, the data processing speed is improved by 70%, and the prediction accuracy of the data for product qualification rate after dimensionality reduction is still 92% (only 3% different from the prediction accuracy of the original data), which meets the analysis requirements.

[0041] Implementation of a multi-dimensional integration approach: Based on virtualization servers, textile manufacturing resources (10 roving frames, 20 spinning frames, and 15 looms) are virtualized and encapsulated, with standardized interfaces (such as the OPC UA protocol) defined to ensure that virtual resources can be invoked in real time. The service virtualization structure (e.g., production scheduling services are divided into three sub-services: order reception, plan generation, and task allocation) and semantic verification mechanism are clearly defined (using OWL language to describe semantics, ensuring consistency between virtual and physical services). Data sets (totaling 100,000 records) across three dimensions—process, quality, and materials—are acquired through an information integration platform. 500 missing data records are processed using the mean imputation method (e.g., missing spinning frame speed data for a certain period is filled with the average of 1150 m / min for that period). 280 abnormal data records are removed using the 3σ criterion (e.g., a spinning frame speed data of 1800 m / min exceeds the average range of 1150 ± 3 × 80 = 1150 ± 240, and is therefore considered abnormal and removed). A deep convolutional autoencoder was used to predict the data from the dataset. This autoencoder consists of an input layer (10 neurons, corresponding to the dimensionality-reduced data), a convolutional layer (32 convolutional kernels, 3×3 size), a pooling layer (max pooling, 2×2 size), a deconvolutional layer (32 deconvolutional kernels, 3×3 size), and an output layer (10 neurons). Through training (1000 iterations, with the loss function converging to 0.02), it achieved predictions of process parameter deviations (89% accuracy) and material requirements (91% accuracy) for the next two hours. Subsequently, a Stacking ensemble model was constructed, using Random Forest, XGBoost, and Support Vector Machine as the base models and Logistic Regression as the meta-model. The prediction results from the three-dimensional dataset were input into the meta-model, ultimately achieving multi-dimensional data integration. The integrated model achieved a 95% accuracy in predicting product qualification rate, a 10%-15% improvement over the single model.

[0042] Implementation of Digital Twin 3D Visualization Technology: The 3D modeling team used Unity3D to construct a virtual model of the textile workshop, restoring the workshop layout (including equipment locations, material warehouses, and AGV paths) at a 1:1 scale. This model was then connected to physical equipment sensors via an SDK interface to achieve real-time synchronization between the virtual model and the physical equipment (synchronization frequency of 1 time / second). The team researched modeling and transmission methods for multi-source heterogeneous data, encapsulating equipment operating parameters (such as temperature and speed) and production data (such as output and pass rate) in JSON format and transmitting them to the virtual model via the MQTT protocol. Event-driven virtual-physical mapping rules were designed; for example, when the physical spinning machine temperature exceeds 90℃, the corresponding spinning machine in the virtual model displays a red alarm and triggers a pop-up notification. A complex event processing logic model was constructed, transforming the chain event of "spinning machine yarn breakage → stop signal transmission → AGV yarn delivery → restart" into a logical flow within the virtual model, achieving visualized simulation of the events. Develop an information visualization push function to push equipment operation reports (such as daily fault statistics) and production progress charts (such as order completion rate) through the web (supports computer access) and mobile (develop WeChat mini program), which can be viewed by managers in real time.

[0043] Implementation Results: After vertical integration, the target enterprise's high-dimensional data processing efficiency improved by 70%, and the prediction accuracy after multi-dimensional data integration reached 95%. It can predict process deviations and material shortages 2 hours in advance, avoiding batch non-conforming products caused by abnormal parameters (non-conforming product rate reduced by 35%). Digital twin 3D visualization technology enables real-time monitoring of workshop status, shortening the equipment failure detection time from 30 minutes to 1 minute, improving failure handling efficiency by 80%, and significantly enhancing the system's "self-analysis" capability.

[0044] Example 3: End-to-end integration and operation and maintenance decision-making implementation of intelligent manufacturing in the textile industry Implementation Preparation: Based on horizontal and vertical integration, the target enterprise streamlines its entire production process (raw material warehousing → fiber inspection → spinning → weaving → dyeing and finishing → finished product inspection → warehousing → logistics) and clarifies the collaborative needs of each process; Siemens PLC controllers are selected to program and control spinning machines, weaving machines, and AGVs, and an MES (Manufacturing Execution System) is deployed for production management; equipment operation and maintenance data (fault records, maintenance time, and spare parts replacement status for the past two years) are collected for training the operation and maintenance decision model.

[0045] End-to-end digital twin technology implementation: PLC programming control is used to streamline the process flow of spinning (roving, spinning, winding), weaving (warping, sizing, weaving), and AGV cart collaboration (raw material delivery, semi-finished product transfer, finished product transportation), and to formulate a collaborative operation rule library. For example, "When the yarn breaks on the spinning machine, the AGV cart's priority is raised to the highest level, and it must deliver yarn within 5 minutes." "When the finished products in the weaving workshop accumulate to 50 pieces, the AGV cart automatically transfers them to the dyeing and finishing workshop." The production cycle of each process is monitored in real time through the MES system (such as the hourly output of the spinning process and the time per meter of the weaving process), and production data (such as AGV cart waiting time and process completion rate) is collected every 24 hours to optimize the rule library. For example, it was found that the waiting time of the AGV cart was too long during the peak period of yarn breakage from 8-10 am (average 12 minutes). The rule was adjusted to "add 2 AGV carts as backup during peak periods, giving priority to serving the spinning workshop." After optimization, the waiting time was shortened to 4 minutes. By integrating underlying sensing devices (such as RFID tags for material tracking and vibration sensors for equipment status monitoring), 10 AGVs (supporting path planning), a PID control model (for spinning speed adjustment), and an MES system, an end-to-end digital twin model is constructed, covering the entire process from raw material warehousing to logistics. Based on the digital twin model, different scenarios are simulated: for example, for an urgent order requiring delivery 3 days ahead of schedule, the model simulates adjusting the spinning speed (increasing it from 1150m / min to 1200m / min) and adding 2 looms to production, predicting that the order can be completed 3.2 days ahead of schedule while maintaining a 98% product qualification rate; in a simulated spinning machine failure scenario, the model automatically schedules a backup spinning machine and adjusts the AGV delivery route to ensure that production interruption time does not exceed 10 minutes. The simulated and optimized parameters are then applied to physical production to achieve dynamic optimization of the entire process.

[0046] Data-driven operation and maintenance decision implementation: K-means clustering algorithm was used to cluster the entire lifecycle data of textiles (nearly one year of process, production, and management data, totaling 150,000 records), with 7 clusters (corresponding to 7 business scenarios: raw material preparation, spinning, weaving, dyeing and finishing, finished products, logistics, and after-sales service). In the clustering results of the spinning scenario, single-dimensional state variables and fault features were extracted: such as the bearing temperature of the spinning machine (normal range 60-80℃), the degree of roller wear (normal range 0-0.1mm), and the number of yarn breaks (normal range <5 times / hour). The typical faults in this scenario were analyzed as "bearing overheating" (accounting for 40% of the total faults) and "roller wear" (accounting for 30%). Association rules were mined using the Apriori algorithm, with a minimum support of 10% and a minimum confidence of 80%. Key rules such as "bearing temperature > 85℃ and running time > 1200h → failure probability > 80%" and "roller wear > 0.15mm and spinning speed > 1180m / min → yarn breakage rate > 10%" were identified. A key performance matrix was established based on these rules, with matrix indicators including equipment operating efficiency (running machine uptime above 95% is considered excellent), product qualification rate (above 98% is considered excellent), material turnover cycle (< 2 hours is considered excellent), and energy consumption (< 15kW·h / ton of yarn is considered excellent). The spatiotemporal characteristics of equipment failures were analyzed using high-dimensional random matrix theory. It was found that the three spinning machines on the east side of the spinning workshop experienced a higher failure frequency every Monday morning from 9-11 am (possibly related to restarting after weekend shutdowns). The failure propagation path was "bearing overheating → roller wear → yarn breakage". The diagnostic results are synthesized using the DS evidence theory: for example, if the bearing temperature of a spinning machine is 88℃ (70% probability of failure in single-dimensional diagnosis), combined with a vibration value of 0.3mm / s (90% probability of failure in multi-dimensional diagnosis), the synthesized failure probability reaches 95%, thus resolving the uncertainty of single-dimensional diagnosis. An equipment health index is established based on diagnostic criteria. Health status = 100 - (Fault frequency × 5 + Maintenance interval overdue days × 2 + (Running time - Rated maintenance cycle) / 10 + Spare parts replacement delay days × 3), where the definitions and values ​​of each parameter are based on the following: Failure frequency: Count the number of times the equipment has failed in the past 30 days (e.g., if the spinning machine fails twice in the past 30 days, this item will be deducted 2×5=10 points). Maintenance interval overdue days: The difference between the actual maintenance time and the planned maintenance time (e.g., if the planned maintenance cycle is 30 days and the actual overdue period is 5 days, this item will be deducted 5 × 2 = 10 points). Rated maintenance cycle: set according to the type of textile equipment. For example, the rated maintenance cycle of a spinning machine is 1000h. If the current running time of the equipment is 1200h, the excess part is 200h, and the deduction for this item is 200 / 10=20 points. Spare parts replacement delay days: The delay time from spare parts application to actual replacement (e.g., if the replacement of bearing spare parts is delayed by 3 days, this item will be deducted 3×3=9 points).

[0047] Taking the core equipment of the target company—the spinning machine of model FA506—as an example, its parameters in a certain period are: failure frequency 1 time, maintenance interval overdue by 2 days, running time 1100h (rated maintenance cycle 1000h), spare parts replacement delay of 1 day. Substituting into the formula, we get: health level = 100 - (1×5 + 2×2 + (1100-1000) / 10 + 1×3) = 100 - (5 + 4 + 10 + 3) = 78 points, which is judged as "sub-healthy state" and needs to be included in the planned maintenance.

[0048] The equipment importance index is based on the "impact of equipment downtime on production schedule," using a 10-point scoring system. 10 points: If core production equipment (such as spinning machines and rapier looms) is shut down for 1 hour, the output loss will exceed 5 tons, or the entire production process will be interrupted. 8-9 points: Key auxiliary equipment (such as winding machines and sizing machines) will result in a production loss of 3-5 tons or cause local process to stop if the machine is shut down for 1 hour; 5-7 points: For general production equipment (such as warping machines and fabric inspection machines), a 1-hour downtime results in a loss of 1-3 tons of output, which can be temporarily replaced by backup equipment; 3-4 points: Auxiliary service equipment (such as workshop air conditioning, raw material conveyor belts) has a production loss of less than 1 ton per hour of downtime, and has a small impact on production; 1-2 points: Logistics support equipment (such as workshop lighting, office computers), shutdown does not directly affect production.

[0049] The target company's 15 rapier looms were all rated 10 points (importance index 10), 5 warping machines were rated 6 points (importance index 6), and 3 raw material conveyor belts were rated 4 points (importance index 4), providing a basis for prioritizing operation and maintenance.

[0050] Automated execution and system linkage of operation and maintenance decisions Based on the aforementioned indices, the target enterprise builds an "automated operation and maintenance decision-making platform." This platform is deeply integrated with a horizontally integrated information integration platform, a vertically integrated digital twin system, and a MES (Manufacturing Execution System) system, achieving a complete closed loop of "data acquisition - index calculation - decision generation - work order execution - production adaptation." Data synchronization: Every day at midnight, the platform automatically retrieves equipment operation data (fault records, runtime), obtains status deviation data of virtual equipment mapping (such as vibration values, temperature deviations) from the digital twin system, and extracts production plans and order progress data from the MES system to ensure the real-time nature of the calculation basis; Index Calculation: The platform has a built-in health and importance index calculation module. It automatically outputs the index results of all devices at 6:00 every day and generates an "Equipment Maintenance Priority List". The list calculates the comprehensive priority score according to "(100 - health score) × importance index". The higher the score, the higher the priority (e.g., a spinning machine with a health score of 58 and an importance score of 10 has a comprehensive score of (100 - 58) × 10 = 420 points; a conveyor belt with a health score of 65 and an importance score of 4 has a comprehensive score of 35 × 4 = 140 points, so the spinning machine has a higher priority). Decision generation: The platform triggers different operation and maintenance strategies based on the comprehensive priority score. Emergency Maintenance: When the overall score is ≥300 points (corresponding to a health score ≤70 points and an importance score ≥8 points), the platform will automatically generate an "Emergency Maintenance Work Order", which includes the equipment number, fault risk point (such as "overheating risk of spinning machine bearing"), required spare parts model (such as "6205 bearing"), and suggested maintenance duration (such as "2 hours"), and push it to the operation and maintenance team via SMS and WeChat. Planned maintenance: With a comprehensive score of 100-299 (corresponding to a health score of 70-85 and an importance score of 5-7), the work order is automatically included in the "Weekly Maintenance Plan" and synchronized to the MES system. The MES system adjusts the production task of the equipment in advance (such as transferring the weaving task of the equipment to the standby equipment). Routine Inspection: If the overall score is less than 100 points (corresponding to a health score of ≥85 points or an importance score of <5 points), the platform will only generate a "routine inspection record" to remind the inspection personnel to focus on checking key parts of the equipment (such as abnormal noise from the conveyor belt motor). Execution feedback: After the maintenance team completes the maintenance, they enter the maintenance results into the platform (such as "bearing has been replaced and operating temperature has returned to 72℃"). The platform automatically updates the equipment health index (such as the health index of the above-mentioned spinning machine increasing to 88 points after maintenance) and synchronizes the maintenance record to the digital twin system to update the status of the virtual equipment, forming a closed-loop management.

[0051] The Implementation Effect of End-to-End Integration and Operation and Maintenance Decisions After three months of trial operation and parameter optimization, the target company achieved significant results in end-to-end integration and operation and maintenance decision-making modules. Specific data is as follows: Improved production continuity: The dynamic optimization function of the end-to-end digital twin system reduces the response time for adjustments to physical processes such as spinning production and fabric forming from 4 hours to 30 minutes. For example, if a batch of cotton yarn needs to have its spinning speed adjusted due to deviations in raw material fiber strength, the digital twin model completes the simulation optimization in just 25 minutes, determining to reduce the spinning speed from 1200m / min to 1120m / min, thus avoiding the generation of 1.2 tons of substandard cotton yarn due to untimely parameter adjustments. The automated execution of equipment operation and maintenance decisions reduces the emergency fault handling time from an average of 4 hours to 1.5 hours, reduces the total production interruption time by 60 hours per month, and improves production continuity to 99.2%.

[0052] Reduced maintenance costs: Data-driven maintenance decisions avoided the problems of "over-maintenance" and "missed maintenance." During the trial operation, the number of spare parts replaced decreased by 18% compared to before (for example, the replacement cycle of the spinning machine bearing was optimized from 800 hours to 1000 hours, which meets the rated maintenance cycle), and the number of temporary calls for external maintenance personnel decreased from 12 times per month to 3 times. The total maintenance cost (including spare parts and labor) decreased by RMB 126,000 per month, a reduction of 22%.

[0053] Product quality and order delivery optimization: End-to-end integration enables traceability and control of quality data throughout the entire process, reducing the fabric defect rate from 2.3% to 0.8% (e.g., in the dyeing and finishing process, digital twin simulation is used to predict dye concentration deviations in advance, and the color fastness pass rate remains above 98.5% after adjustment); Operation and maintenance decisions ensure stable equipment operation, and the on-time order delivery rate has increased from 85% to 98%, with the on-time delivery rate of expedited orders increasing from 70% to 95%, meeting customers' personalized needs.

[0054] The "self-sensing-self-analysis-self-decision-making" capability has been implemented: the system achieves automatic data collection throughout the entire process (self-sensing), dimensionality reduction and correlation analysis of high-dimensional data (self-analysis), and automatic adjustment of operation and maintenance strategies and production parameters (self-decision-making). For example, at 3:00 AM on a certain workday, the system sensed through sensors that the bearing temperature of a spinning machine had risen to 88℃ (self-sensing). After high-dimensional data correlation analysis (combined with a running time of 1100 hours and a vibration value of 0.3 mm / s), it determined that the machine was at "high failure risk" (self-analysis), automatically generated an emergency maintenance work order, and simultaneously adjusted the production plan in the MES system (self-decision-making). The operation and maintenance team completed the maintenance within 1 hour without causing production interruption, fully demonstrating the intelligent manufacturing model of human-machine collaboration.

[0055] Summary of Implementation Examples Through Implementation Examples 1 (horizontal integration), Implementation Example 2 (vertical integration), and Implementation Example 3 (end-to-end integration and operation and maintenance decision-making), the target enterprise successfully constructed a "self-sensing-self-analysis-self-decision-making" intelligent manufacturing system for textiles. Horizontally, it achieves data sharing across the upstream and downstream of the industrial chain; vertically, it completes deep integration of process, quality, and materials; end-to-end, it enables dynamic optimization of the entire production process; and operation and maintenance decisions achieve data-driven automated management. During the system's trial operation, production efficiency increased by 40%, product qualification rate increased to 99.2%, and operation and maintenance costs decreased by 22%, verifying the feasibility and advancement of the technical solution of this invention. It can be promoted to textile enterprises of different sizes, providing a feasible technical path for the digital transformation of the textile industry.

[0056] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of protection claimed by the present invention. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A self-awareness-self-analysis-self-decision textile intelligent manufacturing system construction method, characterized in that, Comprising the following steps: (1) Horizontal integration of intelligent manufacturing in the textile industry: a) Based on the "digital spinning workshop" system, use clustering analysis method to collect and analyze the business data of upstream fiber manufacturing, spinning, midstream fabric weaving, dyeing and finishing, downstream apparel manufacturing and circulation. The correlation, autocorrelation and causality between the data represent the complex high-dimensional textile data relationship; b) Use Monte Carlo method to explore information integration method and data flow mining algorithm based on equipment, production and business. After instance verification, combine 5G technology to develop information integration platform to realize horizontal integration of the upstream to downstream business; (2) Vertical integration of intelligent manufacturing in the textile industry: a) Use random matrix to extract heterogeneous features of complex high-dimensional textile data. Remove singular values that do not meet the prediction through correlation matrix. Perform singular value decomposition on the remaining feature quantity to analyze the correlation between features and classes. Then complete feature selection through redundancy analysis. Finally, use semi-supervised regularization method to build objective function and process vector features to achieve feature mapping and dimension reduction; b) Based on virtualization technology, analyze the virtualization packaging characteristics and semantic verification mechanism of textile manufacturing resources, services and capabilities. Obtain process, quality and material dimension data sets from the "digital spinning workshop" system and process missing values and outliers. Use deep convolutional autoencoder to predict data sets. Realize multi-dimensional integration through Stacking integrated model; c) Use digital twin technology to build a visual real-time monitoring system for textile workshops. Research multi-source heterogeneous data modeling and transmission, event-driven virtual-real mapping, complex event processing logic modeling and information visualization pushing methods. Develop three-dimensional visualization technology; (3) End-to-end integration of intelligent manufacturing in the textile industry: Use digital control technology to build process coordination operation rule library for spinning operation, weaving operation and AGV car coordination. Optimize the rule library through continuous textile process control and production data real-time feedback. Combine bottom layer sensing equipment, AGV car, control model and production system to research end-to-end digital twin and realize dynamic optimization of spinning production and fabric forming physical process; (4) Data-driven operation and maintenance decision of textile enterprises: a) Use K-means clustering algorithm to mine process, production and management data of textile products throughout their life cycle. Extract single-dimensional state quantity fault features and analyze different business scenarios; b) Use Apriori algorithm to mine data association rules in the scenarios. Establish key performance matrix. Use high-dimensional random matrix theory to analyze spatial and temporal characteristics of equipment failure. Then synthesize single-dimensional and multi-dimensional diagnostic results through D-S evidence theory. Based on fault diagnosis criteria, establish equipment health index and importance index. Combine system operation state to reduce operation and maintenance decision risk.

2. The self-aware, self-analytic, self-decisional textile smart manufacturing system construction method of claim 1, wherein: The information integration platform in step (1) b) integrates Internet of Things collection center, production data real-time monitoring center, intelligent scheduling center and intelligent operation and maintenance center to realize centralized management and real-time calling of all business data.

3. The self-aware, self-analytic, self-decisional textile smart manufacturing system construction method of claim 1, wherein: The objective function built by the semi-supervised regularization method in step (2) a) aims to minimize the dimension of vector features. Optimal dimension reduction parameters are determined through iterative calculation.

4. The self-aware, self-analytic, self-decisional textile smart manufacturing system construction method of claim 1, wherein: The missing value processing in step (2) b) adopts mean filling method, and the outlier processing adopts 3σ criterion to eliminate data exceeding the mean value ± 3 times the standard deviation.

5. The self-aware, self-analytic, self-decisional textile smart manufacturing system construction method of claim 1, wherein: The monitoring system constructed by the digital twin technology in step (2) c) supports real-time retrieval of equipment operation parameters, production progress and quality detection data, and realizes automatic alarm of abnormal conditions.

6. The self-aware, self-analytic, self-decisional textile smart manufacturing system construction method of claim 1, wherein: The rule base optimization period in step (3) a) is every 24 hours, and the collaborative operation parameters are adjusted based on the deviation analysis of the production data of the previous period.

7. The self-aware, self-analytic, self-decisional textile smart manufacturing system construction method of claim 1, wherein: The cluster number of the K-means clustering algorithm in step (4) a) is set to 5-8 categories, and the scene is divided according to the business stage of the whole life cycle of the textile product.

8. The self-aware, self-analytic, self-decisional textile smart manufacturing system construction method of claim 1, wherein: The key performance matrix in step (4) b) includes equipment operation efficiency, product qualification rate, material turnover cycle and energy consumption index.

9. The self-aware, self-analytic, self-decisional textile smart manufacturing system construction method of claim 1, wherein: The equipment health index in step (4) b) adopts a percentage system, which is calculated by combining equipment operation time, fault frequency and maintenance records, and a health state is above 80 points.

10. The self-aware, self-analytic, self-decisional textile smart manufacturing system construction method of claim 1, wherein: The method further comprises trial operation debugging of the constructed intelligent manufacturing system, and optimizing the parameters of each integrated module based on the trial operation data.

Images