Ring spinning production management and control platform based on industrial internet

Abstract

The invention belongs to the technical field of production management, and provides an industrial internet-based ring spinning production management and control platform, which comprises a data receiving and preprocessing unit, a data value evaluation unit, a data management unit and a model intensified training unit. According to the invention, by constructing a data value evaluation mechanism, operation data with high potential value can be automatically screened, the identification capability of a fault prediction model on a novel, rare and unobvious fault mode is significantly enhanced, and missing report and false report are reduced.

Description

Technical Field

[0001] This invention relates to the field of production management technology, and more specifically, to a ring spinning production control platform based on the Industrial Internet. Background Technology

[0002] Ring spinning, a core production process in the textile industry, directly impacts yarn quality and production costs due to the stability of its production process and the efficiency of its equipment operation. By deploying sensors, data acquisition modules, and edge computing devices, smart factories can acquire real-time equipment operating data (such as spindle speed, spindle vibration, temperature, and energy consumption), and use this data to monitor equipment status, provide early warnings of faults, and conduct energy efficiency analysis. Existing technologies already include systems that identify potential equipment faults by building fault prediction models (such as time-series data analysis models based on machine learning or deep learning), thereby supporting predictive maintenance. For example, fault classification models can be trained by analyzing historical data, or anomaly alarms can be triggered using real-time data streams.

[0003] However, existing industrial internet-based management and control platforms typically use all collected operational data directly for model training or analysis, lacking an assessment mechanism for the potential application value of the data in case of faults. The inclusion of large amounts of low-value or redundant data in the training set not only increases the computational burden but may also dilute the model's sensitivity to key features, affecting the accuracy of fault prediction. Furthermore, the performance of fault prediction models is highly dependent on the quality and representativeness of the training data. In actual production, equipment operating states change dynamically, and fault samples are relatively scarce. If new, high-value data (such as rare fault mode data) is not added in a timely manner, the model will struggle to adapt to new operating conditions, leading to prediction bias or missed detections.

[0004] Therefore, it is necessary to optimize the ring spinning production management and control platform based on the Industrial Internet. Specifically, it is necessary to optimize the mechanism for assessing the potential fault application value of the received operational data, thereby improving the accuracy and timeliness of fault prediction and thus enhancing the platform's management and control effectiveness. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides an industrial internet-based ring spinning production control platform, electronic equipment, and computer-readable storage medium.

[0006] This invention provides an industrial internet-based ring spinning production control platform, comprising a data receiving and preprocessing unit, a data value assessment unit, a data management unit, and a model reinforcement training unit. The data receiving and preprocessing unit receives multiple sets of operational data collected in real time from various ring spinning production equipment in a smart factory and preprocesses them. The data value assessment unit uses a preset set of anomaly rules to initially screen the preprocessed sets of operational data to obtain a first operational dataset. Furthermore, it performs correlation analysis on the first operational dataset based on a knowledge graph to determine the application value of each potential fault, and constructs a second operational dataset based on operational data whose potential fault application value exceeds a preset threshold. The knowledge graph is constructed based on a set of failed fault prediction cases. The data management unit formulates reinforcement training strategies based at least on the operational status information of the smart factory. The model reinforcement training unit reinforces and trains a fault prediction model based on the second operational dataset and the reinforcement training strategies.

[0007] The present invention also provides an electronic device applied to a ring spinning production control platform based on the Industrial Internet as described above; it includes a memory and a processor, wherein the memory stores a computer program.

[0008] The present invention also provides a computer-readable storage medium for use in a ring spinning production control platform based on the Industrial Internet as described above; the computer-readable storage medium stores computer instructions.

[0009] This invention constructs a data value assessment mechanism that automatically filters operational data with high potential value, significantly enhancing the fault prediction model's ability to identify novel, rare, and inconspicuous fault modes, and reducing false alarms and missed alarms. Simultaneously, by combining real-time factory operational status with training strategies, it achieves optimized allocation of computing resources and minimal production disruption to model updates. Ultimately, while ensuring equipment reliability and production efficiency, it achieves an effective balance between efficient data utilization, continuous model evolution, and stable production operation. Attached Figure Description

[0010] Figure 1 This is a schematic diagram of the structure of a ring spinning production control platform based on the Industrial Internet disclosed in an embodiment of the present invention.

[0011] Figure 2 This is a schematic diagram of the system architecture on which the solution of the embodiment of the present invention is based.

[0012] Figure 3 This is a schematic diagram of the structure of the electronic device disclosed in the embodiments of the present invention.

[0013] Figure 4 This is a schematic diagram of the structure of a computer-readable storage medium disclosed in an embodiment of the present invention. Detailed Implementation

[0014] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0015] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] To facilitate understanding of the embodiments of the present invention, further explanations and descriptions will be provided below with reference to the accompanying drawings and specific embodiments. These embodiments do not constitute a limitation on the embodiments of the present invention.

[0017] like Figure 1 As shown in the figure, this embodiment of the invention discloses a ring spinning production control platform 100 based on the Industrial Internet, including a data receiving and preprocessing unit 11, a data value assessment unit 12, a data management unit 13, and a model reinforcement training unit 14.

[0018] The data receiving and preprocessing unit 11 receives multiple sets of operating data collected in real time from each ring spinning production equipment in the smart factory and preprocesses them.

[0019] like Figure 2 As shown, the ring spinning production control platform 100 is connected to relevant equipment in the smart factory via the Industrial Internet, enabling it to receive multiple sets of real-time operating data collected from various ring spinning production equipment (such as ring spinning frames, roving frames, and drawing frames). This operating data includes, but is not limited to, spindle speed, spindle vibration, drafting zone temperature, motor current, energy consumption, equipment start / stop status, and yarn tension.

[0020] Meanwhile, the received raw operational data is preprocessed, including data cleaning (removing outliers and filling in missing values), data normalization, time series alignment, and data format standardization, in order to eliminate noise and unify data scale.

[0021] The data value assessment unit 12 uses a preset set of anomaly rules to perform preliminary screening on multiple sets of preprocessed running data to obtain a first running dataset; and performs correlation analysis on the first running dataset based on a knowledge graph to obtain the application value of each potential fault, and constructs a second running dataset based on running data whose potential fault application value is higher than a preset threshold; wherein, the knowledge graph is constructed based on a set of fault prediction failure cases.

[0022] This unit performs rapid initial screening based on a preset set of anomaly rules (e.g., setting an alarm threshold for the effective value of vibration velocity), picking out obviously abnormal data segments to form the first operational dataset. The anomaly rule set includes threshold rules (e.g., vibration amplitude exceeding a safety threshold, sudden temperature rise) and pattern rules (e.g., abnormal vibration waveforms at specific frequencies) set based on the equipment's historical operating status and expert experience, used to quickly identify data segments that are obviously abnormal or have potential faults.

[0023] Subsequently, based on a pre-constructed knowledge graph, the system analyzes whether there is a correlation between current operational data and historically unpredictable faults, thereby inferring its potential application value for improving the model's predictive blind spots. The knowledge graph is constructed based on a set of historical failed fault prediction cases; its nodes include equipment components, fault types, environmental factors, and operating conditions, while edges represent causal, symbiotic, or temporal relationships. Operational data with potential fault application value exceeding a preset threshold are selected to construct a second operational dataset. This second dataset represents a set of high-value data samples that can effectively enhance model performance (for the identification of new, rare, and minor faults).

[0024] The data management unit 13 formulates reinforcement training strategies based at least on the operational status information of the smart factory.

[0025] This unit not only relies on the data itself but also comprehensively considers the real-time operational status of the entire smart factory. For example, it intelligently formulates model training plans based on factors such as the current production line's workload, whether it is in an alarm state, and the system's computing resource load. For instance, when selecting training times, it prioritizes scheduling training during periods of equipment idleness or low load to avoid disrupting normal production.

[0026] The developed reinforcement training strategy should include at least determining the timing of reinforcement training (such as idle periods or low-load periods), the priority of training data usage, and the allocation of training resources (computing resources and storage resources).

[0027] The model reinforcement training unit 14 performs reinforcement training on the fault prediction model based on the second running dataset and the reinforcement training strategy.

[0028] This unit is responsible for the self-learning and continuous optimization of the ring spinning production control platform 100. Specifically, it utilizes the high-quality second operational dataset selected by the data value assessment unit 12 and, in accordance with the reinforcement training strategy formulated by the management unit, performs targeted reinforcement training on the existing fault prediction model. This process focuses on enabling the model to learn the latest and more difficult-to-identify equipment fault state patterns, thereby continuously improving its prediction accuracy for new faults and less obvious faults, as well as its ability to adapt to production changes. It is understood that the fault prediction model in this invention can be any existing model, such as a model built and trained based on LSTM, Transformer, etc., without any specific limitation.

[0029] This invention constructs a data value assessment mechanism that automatically filters operational data with high potential value, significantly enhancing the fault prediction model's ability to identify novel, rare, and inconspicuous fault modes, and reducing false alarms and missed alarms. Simultaneously, by combining real-time factory operational status with training strategies, it achieves optimized allocation of computing resources and minimal production disruption to model updates. Ultimately, while ensuring equipment reliability and production efficiency, it achieves an effective balance between efficient data utilization, continuous model evolution, and stable production operation.

[0030] As an example, the data receiving and preprocessing unit 11 is also used to perform data augmentation processing on the preprocessed multiple sets of operating data, including at least: using time series interpolation algorithms to expand the sample size of each set of preprocessed operating data; in this method, in response to the problem of scarce key fault samples and imbalanced positive and negative samples that may exist in the operating data of the equipment, time series interpolation algorithms (such as linear interpolation, cubic spline interpolation, etc.) are used to reasonably insert new data points between the existing sparse or key data points, thereby effectively increasing the number of training samples, especially the amount of data under rare fault modes, without changing the overall trend of the data, and providing richer learning materials for the model.

[0031] Gaussian noise is added to each set of preprocessed operational data to generate adversarial examples. This method enhances the model's robustness and anti-interference capabilities in real industrial environments by adding a small amount of Gaussian noise (i.e., random noise following a normal distribution) to the preprocessed operational data. Generating adversarial examples in this way simulates unavoidable sensor measurement errors and transient electromagnetic interference in production environments. This forces the fault prediction model to not only fit the features of clean data during learning but also to learn to ignore these slight perturbations, thereby enhancing its generalization ability and reducing false alarms caused by minor noise.

[0032] The time-domain / frequency-domain transformation method is used to construct sample variants for each group of preprocessed running data.

[0033] In this approach, to help the model learn the characteristics of equipment operating status from multiple angles and dimensions, time-domain / frequency-domain transformation methods (such as Fast Fourier Transform (FFT) and wavelet transform) are used to process the original time-series data. For example, converting the vibration signal from the time domain to the frequency domain can highlight its energy distribution characteristics at different frequencies, thereby constructing new variations or views of the data. This is equivalent to providing multiple feature representations for the same segment of operating data, enriching the training set and helping the model to more comprehensively understand the mapping relationship between signal features and equipment status, which is particularly effective for identifying vibration-related faults.

[0034] In this embodiment, the data receiving and preprocessing unit 11 improves the quality and diversity of the training dataset by integrating the above-mentioned data augmentation techniques, thereby helping to update and derive a fault prediction model that is more generalizable (able to handle unseen data patterns well) and robust (insensitive to noise and interference).

[0035] As an example, the set of anomaly rules includes threshold rules set based on historical operating data of the equipment and pattern rules set based on expert experience; wherein, the threshold rules are used to identify data that exceeds the range of normal operating parameters of the equipment; and the pattern rules are used to identify data that matches known abnormal waveforms or fault sequences.

[0036] In this embodiment, the threshold rule is a quantified boundary set based on statistical analysis of long-term historical operating data of the equipment. The thresholds included, such as the maximum / minimum rotational speed, the safe upper limit of the effective value of vibration velocity, and the temperature alarm threshold, define the reasonable fluctuation range of various operating parameters of the equipment under normal operating conditions.

[0037] Threshold rules can efficiently identify and filter data points or data segments that clearly exceed the normal operating parameters of equipment. For example, spindle vibration data that momentarily exceeds historical highs is captured by this threshold rule. In other words, threshold rules are a numerically based, relatively simple and direct anomaly detection method that can quickly filter out significant abnormal data.

[0038] Pattern rules do not rely on simple numerical boundaries, but are based on waveform characteristics or event sequence patterns that represent specific faults or abnormal states, summarized and defined by domain experts. For example, specific frequency components of roller torsional vibration of a certain type of spinning machine, the vibration harmonic variation law caused by gradual spindle eccentricity, or a fault sequence consisting of a series of events (such as a sudden drop in current followed by a slow rise in temperature).

[0039] Based on pattern rules, processing techniques such as pattern matching and waveform recognition can identify data whose values ​​do not exceed thresholds but whose shape, trend, or sequence highly matches known abnormal waveforms or fault sequences. Pattern rules can uncover more subtle and complex potential fault symptoms, ensuring that the initial dataset contains not only obviously abnormal data but also high-value data suspected of being abnormal.

[0040] This embodiment combines statistical threshold rules and knowledge-based pattern rules to enable both high-speed, batch screening of explicit anomalies and intelligent capture of implicit, complex fault modes, ensuring the quality and comprehensiveness of the first running dataset.

[0041] As an example, the nodes of the knowledge graph include equipment components, fault types, environmental factors, and operating conditions, and the edges represent the causal, symbiotic, or temporal relationships between them; then, based on the knowledge graph, the correlation analysis of the first operating dataset is performed to obtain the application value of each potential fault, including: extracting the equipment operating features corresponding to the first operating dataset, and performing semantic matching of the equipment operating features with the nodes in the knowledge graph to determine the associated target node set.

[0042] In this step, key equipment operation features are extracted from the first running dataset. It is understood that these features are not raw readings, but processed feature information with clear physical meaning, such as "amplitude of spindle vibration in the 500Hz frequency band", "variance of spindle speed fluctuation", "temperature gradient in the drawing zone", etc.

[0043] The aforementioned equipment operation characteristics are semantically matched with nodes in the knowledge graph. For example, the extracted feature "500Hz vibration amplitude" is associated with nodes in the graph representing "spindle" and "bearing wear." Successfully matched nodes, which have a direct or indirect semantic connection to the current data features, are collectively formed into a target node set. This, in turn, maps the data into the fault semantic space constructed by the knowledge graph.

[0044] The correlation strength between the device's operating characteristics and each node in the target node set is calculated. This correlation strength is determined based on the edge type, weight, and path distance. In this step, the correlation strength quantifies the degree of correlation between the current device operating characteristics and each node in the target node set. It is understood that the calculation of this correlation strength is based on the attributes of the edges connecting these nodes in the target node set. Node attributes specifically include: edge type: causal edges (such as "bearing wear" leading to "increased vibration") typically have a higher weight than symbiotic edges (such as "high temperature" and "high humidity" often occurring simultaneously).

[0045] Edge weight: The higher the weight, the stronger or more common the association.

[0046] Path distance: The shorter the path between a data feature and a faulty node (e.g., directly connected), the stronger the association; the longer the path (requiring inference through multiple intermediate nodes), the weaker the association.

[0047] For example: Correlation strength = (W type ×W weight ) / (1+D path Among them, W type The type weight coefficient for edges is: 1.0 for causal edges, 0.6 for co-occurrence edges, and 0.8 for temporal edges (the specific value can be adjusted according to domain knowledge); W weight D represents the preset weight value of the corresponding edge in the knowledge graph, with a value range of [0,1]. path This represents the shortest path distance (number of hops) from the device's operating feature node to the target node.

[0048] Based on the correlation strength and the fault risk level corresponding to the target node, the potential fault application value of each running data in the first running dataset is calculated.

[0049] The potential fault application value score is the final quantitative output, which is derived from two aspects: the correlation strength and the fault risk level corresponding to the target node. Among them, the correlation strength represents the strength of the correlation between the operational data and a certain fault mode; the fault risk level corresponding to the target node is a predefined attribute in the knowledge graph, representing the severity (such as whether it will lead to downtime, the cost of maintenance) and probability of the fault occurring.

[0050] By employing methods such as linear weighting, combining high correlation strength with high risk level, a potential failure application value score for the operational data is calculated. Data with high scores indicates that it is not only highly correlated with a serious and likely potential failure, but is also likely a key sample that the current model has not fully learned.

[0051] This embodiment accurately identifies data samples with high value for improving the prediction blind spots of the model by deeply mining the semantic correlation between equipment operation data and failure modes.

[0052] As an example, the step of semantically matching the equipment operation characteristics with nodes in the knowledge graph to determine the associated target node set includes: evaluating the overall fault association complexity of each ring spinning production equipment in the smart factory based on the equipment fault association feature set, and dynamically determining the size threshold of the target node set according to the overall fault association complexity; the fault association feature set includes the coupling degree of equipment components, the historical frequency of multiple fault concurrency, and the number of fault propagation paths.

[0053] In this step, before starting semantic matching, the overall fault association complexity of a specific ring spinning production equipment in the current smart factory is quantitatively evaluated. This indicator is used to determine the reasonable range for subsequent knowledge graph analysis.

[0054] The overall fault correlation complexity is derived from an assessment based on a set of equipment fault correlation characteristics. These characteristics include: Equipment component coupling: This refers to the degree of interdependence between the mechanical and electrical components within the equipment. For example, the spindles, rings, and rollers of a spinning machine are tightly coupled; a failure in one component can easily trigger a chain reaction. The higher the equipment component coupling, the higher the overall fault correlation complexity.

[0055] Historical frequency of concurrent failures: The frequency with which multiple failures of this ring spinning production equipment occur simultaneously or sequentially throughout history. A higher frequency of concurrent failures indicates a more complex failure mode and a higher overall failure correlation complexity.

[0056] Number of fault propagation paths: In the knowledge graph, this refers to the number of paths that can reach other fault nodes from any component node of the ring spinning production equipment. The more paths there are, the more complex the causes and effects of the fault, and the higher the overall fault correlation complexity.

[0057] Based on the above three dimensions of indicators, the overall fault correlation complexity of the ring spinning production equipment is calculated and divided into different levels (such as high, medium, and low). It is understood that this calculation can involve scoring each component separately and then weighting and integrating them; details will not be elaborated further.

[0058] Next, based on the overall fault association complexity derived above, an acceptable target node set size threshold is automatically set for this semantic matching. The specific matching logic is as follows: the higher the overall fault association complexity, the larger the set size threshold. For example, for a device with a complex structure, frequent faults, and multiple propagation paths (high complexity), more related knowledge graph nodes are allowed to be matched for comprehensive and in-depth analysis, avoiding the omission of key associations. Conversely, for a device with a simple structure and a single fault mode (low complexity), a smaller size threshold is set to avoid unnecessary over-analysis, thereby improving efficiency.

[0059] The device operation characteristics are semantically matched with nodes in the knowledge graph to determine an associated target node set, the number of nodes in the target node set being adapted to the scale threshold.

[0060] In this step, under the constraint of the aforementioned scale threshold, a semantic matching algorithm between device operating features and knowledge graph nodes is executed. Specifically, the matching process continues until the number of associated nodes found reaches the preset scale threshold, or all possible nodes are traversed. The final target node set has a number of nodes that is adapted to (i.e., limited to or close to) this scale threshold. This setting ensures that the analysis depth matches the actual fault complexity of the device.

[0061] This embodiment achieves intelligent dynamic control over the size of the target node set by assessing the overall fault association complexity of the equipment. Specifically, for equipment with complex fault modes, the node set size is automatically expanded to ensure analysis depth and avoid missing key fault associations; for equipment with simple structures, the node set size is reduced to improve analysis efficiency and avoid wasting computational resources. This solution effectively addresses the problems of insufficient analysis on complex equipment and over-analysis on simple equipment in traditional fixed-size analysis methods, and is conducive to achieving the optimal balance between accuracy and computational efficiency in knowledge graph analysis.

[0062] As an example, reinforcement training strategies should be developed based on at least the operational status information of the smart factory, including: obtaining real-time operational status information of the smart factory, including equipment load rate, production plan urgency, fault warning level, and available computing resources.

[0063] This step involves real-time collection and integration of multi-dimensional status information from the production management system and computing platform, specifically including: Equipment load rate: reflecting the busyness of the production line. A high load rate usually indicates that the equipment is operating at full capacity to perform production tasks.

[0064] Production plan urgency level: reflects the priority of current production tasks and delivery pressure. High urgency level means that ensuring production stability must be prioritized.

[0065] Fault warning level: Reflects the overall health status of the system. A high-level warning indicates a potential fault, threatening system stability.

[0066] Availability of computing resources: This reflects whether the platform has sufficient computing (such as GPU / CPU), memory, and storage resources available for model training.

[0067] Based on the operational status information, determine the timing of reinforcement training, the priority of training data usage, and the training resource allocation scheme; in this step, based on the above information, make intelligent decisions to generate the following three key strategy elements: (1) Determine the timing of reinforcement training: training should be prioritized to be triggered during periods of low equipment load (such as when shifts end or equipment is idle) and low production urgency; conversely, when production is busy or tasks are urgent, postpone or suspend training tasks to avoid interference with production by reinforcement training of the model.

[0068] (2) Determine the priority of training data: When the fault warning level is high, prioritize the use of the data most relevant to the current warning fault mode (from the second running dataset) for training, aiming to quickly strengthen the model's ability to cope with imminent risks, thereby improving the training effect.

[0069] (3) Determine the training resource allocation scheme: dynamically allocate the upper limit of resources that training tasks can occupy based on the availability of computing resources. When resources are sufficient, allocate more resources to accelerate training, and when resources are scarce, restrict resource usage to ensure that the performance of core services such as online monitoring and control is not affected.

[0070] The reinforcement training strategy is generated based on the triggering timing, usage priority, and training resource allocation scheme.

[0071] In this step, the three elements of triggering timing, usage priority, and training resource allocation scheme determined above are integrated and encapsulated to form a complete reinforcement training strategy that can be directly executed by the model reinforcement training unit 14.

[0072] As an example, the model reinforcement training unit employs a combined reinforcement training algorithm that integrates online incremental learning with periodic batch learning.

[0073] As an example, the online incremental learning fine-tunes the parameters of the fault prediction model based on real-time high-value data samples in the second running dataset; the periodic batch learning optimizes the structure and resets the parameters of the fault prediction model based on historical high-value data samples collected in a preset period.

[0074] In this embodiment, the present invention sets up a model reinforcement training unit that employs a combined strategy to reinforce the fault prediction model. This combined strategy integrates two learning modes with different rhythms and objectives: online incremental learning and periodic batch learning. It can simultaneously take into account the model's immediate adaptability and long-term stability, which is beneficial for addressing the complex and ever-changing needs in industrial scenarios.

[0075] The specific responsibilities and collaboration methods of the two learning modes are as follows: Online incremental learning is responsible for agile response and rapid adaptation. It is initiated immediately once the data value assessment unit 12 produces new real-time high-value data samples (i.e., the second running dataset). It fine-tunes the parameters of the existing fault prediction model. Online incremental learning does not change the basic structure of the model, but rather makes small, rapid adjustments based on the new samples and the current model parameters. This allows the model to quickly absorb the latest and critical equipment status information, adapt instantly to new operating conditions or newly emerging minor fault symptoms, thereby maintaining the model's timeliness and sensitivity, and enabling rapid response to sudden anomalies.

[0076] Periodic batch learning is responsible for comprehensive optimization and consolidation of the foundation. It does not use fragmented real-time data, but rather a larger, more comprehensive dataset composed of all historical high-value data samples collected within a period (e.g., every 24 hours, week, or month). This allows for deeper structural optimization and parameter resetting of the fault prediction model, including adjusting the model's network structure, reinitializing some parameters, and performing full training. This enables the model to learn fundamental patterns from a broader data distribution and corrects potential biases (such as catastrophic forgetting) that may arise from online incremental learning, ensuring the model's long-term stability and generalization ability.

[0077] like Figure 3 As shown, this embodiment of the invention also discloses an electronic device 200, which is applied to a ring spinning production control platform based on the Industrial Internet as described above; it includes a memory and a processor, wherein the memory stores a computer program.

[0078] like Figure 4 As shown, this embodiment of the invention also discloses a computer-readable storage medium 300, which is applied to a ring spinning production control platform based on the Industrial Internet as described above; the computer-readable storage medium stores computer instructions.

[0079] Although the invention has been specifically shown and described with reference to preferred embodiments, those skilled in the art will understand that various modifications in form and detail may be made without departing from the spirit and scope of the invention. Accordingly, the disclosed invention should be considered merely illustrative and limited only by the scope specified in the appended claims.

Claims

1. A ring spinning production control platform based on the Industrial Internet, characterized in that: It includes a data receiving and preprocessing unit, a data value assessment unit, a data management unit, and a model reinforcement training unit; the data receiving and preprocessing unit receives multiple sets of operational data collected in real time from each ring spinning production equipment in the smart factory and preprocesses them; the data value assessment unit uses a preset set of anomaly rules to perform preliminary screening on the preprocessed multiple sets of operational data to obtain a first operational dataset. Furthermore, the first operational dataset is analyzed for correlation based on a knowledge graph to determine the application value of each potential fault, and a second operational dataset is constructed based on operational data whose potential fault application value exceeds a preset threshold; wherein, the knowledge graph is constructed based on a set of failed fault prediction cases; and the data management unit formulates reinforcement training strategies based at least on the operational status information of the smart factory. The model reinforcement training unit performs reinforcement training on the fault prediction model based on the second running dataset and the reinforcement training strategy.

2. The ring spinning production control platform based on the Industrial Internet according to claim 1, characterized in that: The data receiving and preprocessing unit is further configured to perform data augmentation processing on the preprocessed multiple sets of running data, including at least one of the following: using a time series interpolation algorithm to expand the sample size of each set of preprocessed running data; adding Gaussian noise to each set of preprocessed running data to generate adversarial examples; The time-domain / frequency-domain transformation method is used to construct sample variants for each group of preprocessed running data.

3. The ring spinning production control platform based on the Industrial Internet according to claim 1, characterized in that: The set of abnormal rules includes threshold rules set based on historical operating data of the equipment and pattern rules set based on expert experience; wherein, the threshold rules are used to identify data that exceeds the range of normal operating parameters of the equipment; and the pattern rules are used to identify data that matches known abnormal waveforms or fault sequences.

4. The ring spinning production control platform based on the Industrial Internet according to claim 1, characterized in that: The nodes of the knowledge graph include equipment components, fault types, environmental factors, and operating conditions, while edges represent causal, symbiotic, or temporal relationships between them. Based on the knowledge graph, correlation analysis is performed on the first operating dataset to derive the application value of each potential fault. This includes: extracting equipment operating features corresponding to the first operating dataset; semantically matching the equipment operating features with nodes in the knowledge graph to determine a set of associated target nodes; calculating the correlation strength between the equipment operating features and each node in the target node set, where the correlation strength is determined based on edge type, weight, and path distance; and calculating the potential fault application value of each operating data point in the first operating dataset based on the correlation strength and the fault risk level corresponding to the target node.

5. The ring spinning production control platform based on the Industrial Internet according to claim 4, characterized in that: The process of semantically matching the equipment operation characteristics with nodes in the knowledge graph to determine the associated target node set includes: assessing the overall fault association complexity of each ring spinning production equipment in the smart factory based on a set of equipment fault association characteristics; dynamically determining a size threshold for the target node set based on the overall fault association complexity; the fault association characteristic set includes equipment component coupling degree, historical frequency of multiple fault concurrency, and number of fault propagation paths; and semantically matching the equipment operation characteristics with nodes in the knowledge graph to determine the associated target node set, wherein the number of nodes in the target node set is adapted to the size threshold.

6. The ring spinning production control platform based on the Industrial Internet according to claim 1, characterized in that: At least based on the operational status information of the smart factory, a reinforcement training strategy is formulated, including: acquiring real-time operational status information of the smart factory, including equipment load rate, production plan urgency, fault warning level, and available computing resources; determining the triggering time for reinforcement training, the priority of training data usage, and the training resource allocation scheme based on the operational status information; and generating the reinforcement training strategy based on the triggering time, usage priority, and training resource allocation scheme.

7. The ring spinning production control platform based on the Industrial Internet according to claim 1, characterized in that: The model reinforcement training unit adopts a combined reinforcement training algorithm, which integrates online incremental learning and periodic batch learning.

8. A ring spinning production control platform based on the Industrial Internet as described in claim 7, characterized in that: The online incremental learning is based on real-time high-value data samples in the second running dataset to fine-tune the parameters of the fault prediction model; the periodic batch learning is based on historical high-value data samples collected in a preset period to optimize the structure and reset the parameters of the fault prediction model.

9. An electronic device, characterized in that: The device is applied to an industrial internet-based ring spinning production control platform as described in any one of claims 1-8; it includes a memory and a processor, wherein the memory stores a computer program.

10. A computer-readable storage medium, characterized in that: The device is applied to the industrial internet-based ring spinning production control platform as described in any one of claims 1-8; the computer-readable storage medium stores computer instructions.

Images