A system and method for monitoring a spinning production

Abstract

This invention discloses a monitoring system and method for spinning production, relating to the field of spinning monitoring technology. The invention sets a collection cycle for spinning production operation parameters, collects multi-dimensional operating parameters according to the set cycle, and organizes them into a monitoring dataset in chronological order. The dataset is then cleaned and standardized using data preprocessing methods. Simultaneously, a spinning production status identification model is constructed using a convolutional neural network algorithm combined with a long short-term memory network algorithm. The spinning production status is analyzed, and based on the analysis results, different spinning production anomaly levels are set, and a spinning production anomaly early warning mechanism is established. Anomaly early warning methods are generated based on this mechanism, and finally, anomalies are warned and handled, with the results recorded. The spinning production status identification model and anomaly early warning mechanism are continuously optimized, achieving intelligent and precise monitoring of the entire spinning production process.

Description

Technical Field

[0001] This invention relates to the field of spinning monitoring technology, specifically to a monitoring system and method for spinning production. Background Technology

[0002] Spinning production is the core link in the textile industry chain. The stability of production conditions directly affects the quality of finished yarn and the efficiency of production line operation. At present, traditional spinning production monitoring relies on extensive management methods such as manual inspection and single parameter threshold alarms, which have a low level of intelligence.

[0003] Existing technology, such as the invention patent application with publication number CN119571507A, discloses a monitoring system and method for spinning production. The method includes a cotton yarn detection module, a carding process fault diagnosis module, an analysis module, and a control monitoring module. In the carding process, photoelectric sensors are used to detect the uniformity of cotton yarn thickness. If the uniformity of cotton yarn thickness is poor, the carding machine and fibers are inspected. If the carding machine malfunctions, the cause of the malfunction is analyzed and repaired. If the fiber quality is substandard, the cause of the substandard fiber quality is analyzed and addressed. During carding machine repair or fiber treatment, the repair and treatment effects are monitored. If the effect is unsatisfactory, the repair or treatment plan is adjusted. This ensures the reliability of the cotton yarn detection results, accurately identifies the cause of poor cotton yarn thickness uniformity, and guarantees the effectiveness of the solution implementation and the quality of subsequent yarn production.

[0004] As can be seen from the above solutions, current spinning production monitoring focuses on post-production detection and troubleshooting of single processes and single quality parameters, lacking multi-dimensional collaborative monitoring of yarn operation, equipment conditions, and environmental parameters, and thus failing to comprehensively reflect the overall production status. Furthermore, existing solutions do not perform in-depth feature mining on massive amounts of time-series production data, cannot effectively capture parameter fluctuation trends and state evolution patterns, and have insufficient intelligent identification accuracy, failing to meet the high-precision and high-stability control requirements of modern spinning production. Therefore, a comprehensive intelligent monitoring solution for the entire process is urgently needed. Summary of the Invention

[0005] The purpose of this invention is to provide a monitoring system and method for spinning production, which solves the problems existing in the background art.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention provides a method for monitoring spinning production, characterized by comprising the following steps: S1, set the data collection cycle for spinning production operation parameters, collect multi-dimensional operating parameters of spinning production according to the set data collection cycle, and organize them into a monitoring dataset in chronological order; The multi-dimensional operating parameters for spinning production include: yarn operating parameters, equipment operating parameters, and environmental parameters. S2, the monitoring dataset is cleaned and standardized using data preprocessing methods to obtain the preprocessed monitoring dataset; S3, based on the preprocessed monitoring dataset, constructs a spinning production status identification model by combining a convolutional neural network algorithm with a long short-term memory network algorithm, and analyzes the spinning production status through the spinning production status identification model to obtain the spinning production status analysis results. S4. Based on the analysis results of spinning production status, set different levels of spinning production anomalies and establish a spinning production anomaly early warning mechanism. Generate anomaly early warning methods based on the anomaly early warning mechanism. S5, based on the generated anomaly warning method, provides early warning and handling for anomalies, records the anomaly warning and handling results, and continuously optimizes the spinning production status identification model and anomaly warning mechanism based on the recorded results.

[0007] Preferably, the step of setting the data collection cycle for spinning production operation parameters, collecting multi-dimensional operating parameters of spinning production according to the set data collection cycle, and organizing them into a monitoring dataset in chronological order includes the following steps: Set the collection cycle for multi-dimensional operating parameters of spinning production. Set the parameters of yarn operation and equipment condition as the first collection cycle and the environmental parameters as the second collection cycle, and record the collection time nodes of multi-dimensional operating parameters of spinning production. The first acquisition cycle of yarn operation and equipment condition parameters is set to 1 second for real-time acquisition of yarn operation and equipment condition parameters. The second acquisition cycle of environmental parameters is set to 1 minute for acquisition of environmental parameters that change relatively slowly, and the acquisition time nodes of multi-dimensional operating parameters of spinning production are recorded. Corresponding sensors are deployed at each workstation of the spinning production line. Tension sensors collect yarn tension values, yarn breakage detection sensors collect yarn breakage rates to form yarn operating parameters, speed sensors collect equipment speed, current sensors collect equipment operating current to form equipment condition parameters, temperature and humidity sensors collect production site temperature and humidity, and dust concentration sensors collect production site dust content to form environmental parameters. Each collected multi-dimensional operating parameter of the spinning production line is labeled with a corresponding timestamp, parameter type, workstation number, and specific collection location information. Based on the timestamp corresponding to each piece of collected data, the multi-dimensional operating parameters of spinning production are sorted in chronological order, and the multi-dimensional operating parameters of spinning production within the same period are integrated to form a monitoring dataset.

[0008] Preferably, the step of performing data cleaning and standardization preprocessing on the monitoring dataset using a data preprocessing method to obtain the preprocessed monitoring dataset includes the following steps: Missing and duplicate values ​​are detected in the monitoring dataset to remove time-series segments with missing and duplicate data, thus obtaining valid data. At the same time, based on the obtained valid data, the 3σ criterion anomaly detection method is used to identify and remove abnormal data, resulting in a cleaned monitoring dataset. Based on the cleaned monitoring dataset, a smoothing process is performed using the moving average filtering method to remove abnormal data caused by environmental interference and equipment vibration, resulting in a smoothed monitoring dataset. The moving average filtering method is shown below: ; in For the first The smoothed data obtained after moving average filtering The number of consecutive data points included in the average calculation. For the first The cleaned data; The smoothed monitoring dataset is processed by range normalization to obtain the preprocessed monitoring dataset.

[0009] Preferably, the process of constructing a spinning production status identification model based on the preprocessed monitoring dataset using a convolutional neural network algorithm combined with a long short-term memory network algorithm, and then analyzing the spinning production status using this model to obtain the spinning production status analysis results, includes the following steps: S31, the preprocessed monitoring dataset is divided into a training sample set and a test sample set, and each time series data in the monitoring dataset is labeled with the corresponding normal state, potential hidden danger and fault abnormality label according to the actual operation record of spinning production. S32, based on the training sample set, a feature extraction model is constructed using a convolutional neural network algorithm. The feature extraction model is used to extract features from the training sample set data to obtain the local fluctuation features and mutation features of the training sample set data. S33, a time series analysis model is constructed based on the long short-term memory network algorithm. The local fluctuation features and mutation features extracted by the convolutional neural network are input into the time series analysis model to obtain the analysis results. The model is tested through a test sample set. The feature extraction model and the time series analysis model are fused to construct a spinning production status identification model. S34. Collect real-time multi-dimensional operating parameters of spinning production according to the set data collection cycle, perform data preprocessing according to step S2 to obtain real-time preprocessed monitoring dataset, and input the real-time preprocessed monitoring dataset into the spinning production status identification model for identification and analysis to obtain spinning production status analysis results.

[0010] Preferably, the step of constructing a feature extraction model based on a training sample set using a convolutional neural network algorithm, and extracting features from the training sample set data using the feature extraction model to obtain the local fluctuation features and abrupt change features of the training sample set data includes the following steps: Convolutional neural networks are constructed based on training sample sets, and the kernel size and stride of the convolutional layers, the pooling window size of the pooling layers, and the number of neurons in the fully connected layers are set to construct a convolutional neural network feature extraction model. The convolutional neural network includes: convolutional layers, pooling layers, and fully connected layers; The training sample set is input into the constructed feature extraction model, and convolution operations are performed through convolutional layers to extract local features and perform weighted calculations on the training sample set data to obtain the original feature data. The formula for convolution is as follows: ; in For the convolutional layer Each output feature value The kernel length is 1. For the convolution kernel in the th Weight parameters for each position, For the first input training sample set One time series data point, For the convolutional layer The bias parameters corresponding to each output feature; The original feature data is reduced in dimensionality by using max pooling in a pooling layer. The feature response threshold is set according to the distribution characteristics of the feature data, and feature data whose feature response intensity meets the set threshold are filtered and retained to obtain the dimensionality-reduced feature data. The max pooling method is shown below: ; in For the pooling layer Each output feature value For the pooling window length, The pooling window covers the feature values ​​in the original feature data; The dimensionality-reduced feature data is input into the fully connected layer. The fully connected layer performs global integration and nonlinear mapping on the dimensionality-reduced feature data to form and output the local fluctuation and abrupt change characteristics of yarn operating parameters, equipment operating parameters, and environmental parameters.

[0011] Preferably, the step of constructing a time-series analysis model based on a long short-term memory network algorithm, inputting the local fluctuation features and abrupt change features extracted by a convolutional neural network into the time-series analysis model to obtain analysis results, and testing them through a test sample set, and fusing the feature extraction model and the time-series analysis model to construct a spinning production status identification model includes the following steps: A time series analysis model of a long short-term memory network is established based on the local fluctuation and mutation features extracted by a convolutional neural network, and hyperparameters such as the number of hidden layer neurons, the number of training iterations, and the learning rate are set. The long short-term memory network temporal analysis model includes: a forgetting gate, an input gate, and an output gate; The local fluctuation and mutation features output by the convolutional neural network are used as input data for the time series analysis model. The local fluctuation and mutation features under different acquisition periods are weighted and time-series encoded through forget gate, input gate and output gate. The feature data is updated in time sequence to form a time series feature expression. A fully connected layer is set at the output of the long short-term memory network temporal analysis model. The temporal feature expression is input into the fully connected layer, and the output of the fully connected layer is transformed into a probability distribution of spinning production status through an activation function. The category with the highest probability value is taken as the spinning production status identification result. The activation function is as follows: ; in For the first Predicted probability of spinning-like production conditions For the number of state categories, The first output of the fully connected layer The original output value corresponding to the spinning production state. It is a natural constant; By fusing the time series analysis model with the convolutional neural network feature extraction model, a spinning production status identification model in collaboration between convolutional neural networks and long short-term memory networks is obtained. The cross-entropy loss function is set as the model loss calculation function. The output spinning production status identification results are matched with the labeled normal status, potential hidden dangers and fault abnormality labels. The identification error is calculated by the cross-entropy loss function, and the spinning production status identification model is optimized based on the identification error. The cross-entropy loss function is shown below: ; in This is the loss value. One-hot encoding for the real label; The spinning production status recognition model is validated using a test sample set. The recognition accuracy and false alarm rate are statistically analyzed, and a validation threshold is set. When the validation result reaches the validation threshold, the final spinning production status recognition model is determined. If it does not meet the threshold, it is retrained and optimized until the preset accuracy is met.

[0012] Preferably, the method for setting different spinning production anomaly levels based on the spinning production status analysis results and establishing a spinning production anomaly early warning mechanism, and generating anomaly early warnings based on the anomaly early warning mechanism, includes the following steps: Formal spinning production process specifications, equipment rated operating parameters, equipment operation and maintenance standards, and yarn quality control requirements are collected and obtained from the spinning production line production management system, process document library, and equipment operation and maintenance manual. Combined with the spinning production status analysis results, abnormal judgment thresholds for yarn operating parameters, equipment operating parameters, and environmental parameters are set, and three abnormality levels are set: minor abnormality, general abnormality, and severe abnormality. Among them, a minor anomaly is when a single parameter is outside the standard range and the difference between the value and the standard range is less than the corresponding preset minor anomaly threshold; a general anomaly is when a single or multiple parameters are outside the standard range and the difference between the value and the standard range is greater than the minor anomaly threshold but less than the severe anomaly threshold; and a severe anomaly is when multiple parameters are outside the standard range at the same time and the difference between the value and the standard range is greater than or equal to the corresponding preset severe anomaly threshold. An anomaly handling mechanism is established based on the three-level anomaly classification, with differentiated early warning forms and handling priorities set according to the level, and anomaly information is recorded to form anomaly information reports. The real-time spinning production status analysis results are compared with the anomaly judgment threshold. When the corresponding anomaly level triggering conditions are met, the anomaly early warning mechanism is activated and a matching anomaly early warning method is generated.

[0013] Preferably, the anomaly handling mechanism based on the three-level anomaly classification, which sets differentiated early warning forms and handling priorities for each level, and records anomaly information to form an anomaly information report, includes the following steps: Based on the characteristics of minor, general, and severe anomalies, differentiated early warning methods are set according to the levels. Minor anomalies are set to send a text pop-up notification to the corresponding workstation operator to provide an early warning, without triggering a full-area early warning. General anomalies trigger a workstation-level audible and visual alarm, push information to the workstation operator and team manager, and mark the location of the anomaly and basic anomaly information. Severe anomalies trigger a full-area audible and visual alarm, and simultaneously push anomaly information via SMS and mobile APP to notify on-site maintenance personnel, team leaders, and production control personnel for early warning. The severity of anomalies is classified into hierarchical priority levels, and the handling requirements for each level of anomaly are set. Severe anomalies are listed as the highest priority, requiring immediate response and emergency handling to strictly control production risks. General anomalies are listed as medium priority, requiring rectification to be completed within a specified time limit. Minor anomalies are listed as routine priority, to be checked and handled by operators during daily inspections. The occurrence time, production station involved, abnormality level and related parameter type of various abnormalities are recorded throughout the process, abnormal data are recorded and integrated to form a standardized abnormal information report.

[0014] Preferably, the method for generating anomaly warnings and handling anomalies, and recording the results of anomaly warnings and handling, and continuously optimizing the spinning production status identification model and anomaly warning mechanism based on the recorded results, includes the following steps: According to the generated abnormality warning method and corresponding abnormality level, graded warning and handling operations are carried out. Based on the differentiated characteristics of minor abnormalities, general abnormalities and serious abnormalities, appropriate handling measures are adopted, and the abnormality handling process and spinning production operation status are monitored in real time. Record and store abnormal early warning information, handling procedures and handling results to form an abnormal handling report. Based on the abnormal handling report, verify and optimize the spinning production status identification model, and at the same time optimize and adjust the judgment threshold and handling logic of the abnormal early warning mechanism.

[0015] This embodiment also discloses a system for monitoring a spinning production method, characterized in that it includes: a multi-dimensional parameter acquisition module, a data preprocessing module, a production status identification module, an anomaly warning and handling module, and a model and mechanism optimization module; The multi-dimensional parameter acquisition module is used to set the acquisition cycle of spinning production operation parameters, collect spinning production operation parameters, and integrate them into a monitoring dataset by sorting them in time sequence. The data preprocessing module is used to detect missing and duplicate values ​​in the monitoring dataset, and to standardize it using the moving average filtering method and the range normalization method to obtain the preprocessed monitoring dataset. The production status identification module is used to construct a spinning production status identification model based on the preprocessed monitoring dataset, using a convolutional neural network algorithm combined with a long short-term memory network algorithm, and to complete the training and verification. At the same time, it combines the real-time preprocessed monitoring dataset to obtain the spinning production status analysis results. The anomaly warning and handling module is used to set anomaly judgment threshold and anomaly level based on the production status analysis results, set corresponding differentiated warning methods, and perform emergency handling and rectification operations according to the handling priority. The model and mechanism optimization module is used to verify the spinning production status identification model based on the abnormal handling report, supplement training data and optimize model parameters, and at the same time optimize the judgment threshold and handling logic of the abnormal early warning mechanism.

[0016] The beneficial effects of this invention are as follows: (1) This invention sets a collection cycle for spinning production operation parameters, collects multi-dimensional spinning production operation parameters according to the set collection cycle, and organizes them into a monitoring dataset according to the time sequence. Then, the monitoring dataset is cleaned and standardized by a data preprocessing method to obtain a preprocessed monitoring dataset. At the same time, based on the preprocessed monitoring dataset, a spinning production status identification model is constructed by combining a convolutional neural network algorithm with a long short-term memory network algorithm. The spinning production status is analyzed by the spinning production status identification model to obtain the spinning production status analysis results. Then, based on the spinning production status analysis results, different spinning production abnormality levels are set, and a spinning production abnormality early warning mechanism is established. An abnormality early warning method is generated according to the abnormality early warning mechanism. Finally, an abnormality is warned and dealt with based on the generated abnormality early warning method, and the abnormality early warning and handling results are recorded. The spinning production status identification model and abnormality early warning mechanism are continuously optimized based on the recorded results, thus realizing intelligent and precise monitoring of the entire spinning production process. (2) This invention constructs a spinning production status identification model by integrating convolutional neural networks and long short-term memory networks. It can deeply explore the local mutation characteristics and long-term evolution laws of multi-dimensional time series parameters. By implementing differentiated collection cycle design for yarn operation, equipment conditions and environmental parameters, it provides accurate data support and decision-making basis for the stable operation of spinning production.

[0017] (3) By establishing a three-level anomaly classification early warning system, the present invention can match the corresponding early warning method and handling priority according to the anomaly level. At the same time, relying on the historical data of anomaly early warning and handling, the status identification model parameters and anomaly judgment threshold are continuously optimized, so that the monitoring system can adapt to changes in different production conditions, process requirements and equipment operating status, thereby improving the production line operating efficiency. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the monitoring system and method for spinning production according to the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1 Please see Figure 1 This embodiment discloses a method for monitoring spinning production, including the following steps: S1, set the data collection cycle for spinning production operation parameters, collect multi-dimensional operating parameters of spinning production according to the set data collection cycle, and organize them into a monitoring dataset in chronological order; The multi-dimensional operating parameters for spinning production include: yarn operating parameters, equipment operating parameters, and environmental parameters. S2, the monitoring dataset is cleaned and standardized using data preprocessing methods to obtain the preprocessed monitoring dataset; S3, based on the preprocessed monitoring dataset, constructs a spinning production status identification model by combining a convolutional neural network algorithm with a long short-term memory network algorithm, and analyzes the spinning production status through the spinning production status identification model to obtain the spinning production status analysis results. S4. Based on the analysis results of spinning production status, set different levels of spinning production anomalies and establish a spinning production anomaly early warning mechanism. Generate anomaly early warning methods based on the anomaly early warning mechanism. S5, based on the generated anomaly warning method, provides early warning and handling for anomalies, records the anomaly warning and handling results, and continuously optimizes the spinning production status identification model and anomaly warning mechanism based on the recorded results.

[0021] Further, please refer to Figure 1 The process involves setting a data collection cycle for spinning production operation parameters, collecting multi-dimensional operating parameters of spinning production according to the set cycle, and organizing them into a monitoring dataset in chronological order. This includes the following steps: Set the collection cycle for multi-dimensional operating parameters of spinning production. Set the parameters of yarn operation and equipment condition as the first collection cycle and the environmental parameters as the second collection cycle, and record the collection time nodes of multi-dimensional operating parameters of spinning production. The first acquisition cycle of yarn operation and equipment condition parameters is set to 1 second for real-time acquisition of yarn operation and equipment condition parameters. The second acquisition cycle of environmental parameters is set to 1 minute for acquisition of environmental parameters that change relatively slowly, and the acquisition time nodes of multi-dimensional operating parameters of spinning production are recorded. Corresponding sensors are deployed at each workstation of the spinning production line. Tension sensors collect yarn tension values, yarn breakage detection sensors collect yarn breakage rates to form yarn operating parameters, speed sensors collect equipment speed, current sensors collect equipment operating current to form equipment condition parameters, temperature and humidity sensors collect production site temperature and humidity, and dust concentration sensors collect production site dust content to form environmental parameters. Each collected multi-dimensional operating parameter of the spinning production line is labeled with a corresponding timestamp, parameter type, workstation number, and specific collection location information. Based on the timestamp corresponding to each piece of collected data, the multi-dimensional operating parameters of spinning production are sorted in chronological order, and the multi-dimensional operating parameters of spinning production within the same period are integrated to form a monitoring dataset.

[0022] Further, please refer to Figure 1 The monitoring dataset is cleaned and standardized using data preprocessing methods to obtain the preprocessed monitoring dataset, which includes the following steps: Missing and duplicate values ​​are detected in the monitoring dataset to remove time-series segments with missing and duplicate data, thus obtaining valid data. At the same time, based on the obtained valid data, the 3σ criterion anomaly detection method is used to identify and remove abnormal data, resulting in a cleaned monitoring dataset. Based on the cleaned monitoring dataset, a smoothing process is performed using the moving average filtering method to remove abnormal data caused by environmental interference and equipment vibration, resulting in a smoothed monitoring dataset. The moving average filtering method is shown below: ; in For the first The smoothed data obtained after moving average filtering The number of consecutive data points included in the average calculation. For the first The cleaned data; The smoothed monitoring dataset is processed by the range normalization method to obtain the preprocessed monitoring dataset; The formula for the range normalization method is shown below: ; in The data in the normalized monitoring dataset, To smooth the data in the monitoring dataset, The minimum value in the smoothed monitoring dataset. This represents the maximum value in the smoothed monitoring dataset.

[0023] Further, please refer to Figure 1 Based on the preprocessed monitoring dataset, a spinning production status identification model is constructed using a convolutional neural network algorithm combined with a long short-term memory network algorithm. The spinning production status is then analyzed using this model to obtain the spinning production status analysis results, including the following steps: S31, the preprocessed monitoring dataset is divided into a training sample set and a test sample set, and each time series data in the monitoring dataset is labeled with the corresponding normal state, potential hidden danger and fault abnormality label according to the actual operation record of spinning production. S32, based on the training sample set, a feature extraction model is constructed using a convolutional neural network algorithm. The feature extraction model is used to extract features from the training sample set data to obtain the local fluctuation features and mutation features of the training sample set data. S33, a time series analysis model is constructed based on the long short-term memory network algorithm. The local fluctuation features and mutation features extracted by the convolutional neural network are input into the time series analysis model to obtain the analysis results. The model is tested through a test sample set. The feature extraction model and the time series analysis model are fused to construct a spinning production status identification model. S34. Collect real-time multi-dimensional operating parameters of spinning production according to the set data collection cycle, perform data preprocessing according to step S2 to obtain real-time preprocessed monitoring dataset, and input the real-time preprocessed monitoring dataset into the spinning production status identification model for identification and analysis to obtain spinning production status analysis results.

[0024] Further, please refer to Figure 1 Based on the training sample set, a feature extraction model is constructed using a convolutional neural network algorithm. This model then extracts features from the training sample set data to obtain local fluctuation and abrupt change features. The steps include: Convolutional neural networks are constructed based on training sample sets, and the kernel size and stride of the convolutional layers, the pooling window size of the pooling layers, and the number of neurons in the fully connected layers are set to construct a convolutional neural network feature extraction model. The convolutional neural network includes: convolutional layers, pooling layers, and fully connected layers; The training sample set is input into the constructed feature extraction model, and convolution operations are performed through convolutional layers to extract local features and perform weighted calculations on the training sample set data to obtain the original feature data. The formula for convolution is as follows: ; in For the convolutional layer Each output feature value The kernel length is 1. For the convolution kernel in the th Weight parameters for each position, For the first input training sample set One time series data point, For the convolutional layer The bias parameters corresponding to each output feature; The original feature data is reduced in dimensionality by using max pooling in a pooling layer. The feature response threshold is set according to the distribution characteristics of the feature data, and feature data whose feature response intensity meets the set threshold are filtered and retained to obtain the dimensionality-reduced feature data. The max pooling method is shown below: ; in For the pooling layer Each output feature value For the pooling window length, The pooling window covers the feature values ​​in the original feature data; The dimensionality-reduced feature data is input into the fully connected layer. The fully connected layer performs global integration and nonlinear mapping on the dimensionality-reduced feature data to form and output the local fluctuation and abrupt change characteristics of yarn operating parameters, equipment operating parameters, and environmental parameters.

[0025] Further, please refer to Figure 1 A time-series analysis model is constructed based on the Long Short-Term Memory (LSTM) network algorithm. Local fluctuation and abrupt change features extracted by a convolutional neural network are input into the time-series analysis model to obtain analysis results. These results are then tested using a test sample set. The integration of the feature extraction model and the time-series analysis model to construct a spinning production status identification model includes the following steps: A time series analysis model of a long short-term memory network is established based on the local fluctuation and mutation features extracted by a convolutional neural network, and hyperparameters such as the number of hidden layer neurons, the number of training iterations, and the learning rate are set. The long short-term memory network temporal analysis model includes: a forgetting gate, an input gate, and an output gate; The local fluctuation and mutation features output by the convolutional neural network are used as input data for the time series analysis model. The local fluctuation and mutation features under different acquisition periods are weighted and time-series encoded through forget gate, input gate and output gate. The feature data is updated in time sequence to form a time series feature expression. A fully connected layer is set at the output of the long short-term memory network temporal analysis model. The temporal feature expression is input into the fully connected layer, and the output of the fully connected layer is transformed into a probability distribution of spinning production status through an activation function. The category with the highest probability value is taken as the spinning production status identification result. The activation function is as follows: ; in For the first Predicted probability of spinning-like production conditions For the number of state categories, The first output of the fully connected layer The original output value corresponding to the spinning production state. It is a natural constant; By fusing the time series analysis model with the convolutional neural network feature extraction model, a spinning production status identification model in collaboration between convolutional neural networks and long short-term memory networks is obtained. The cross-entropy loss function is set as the model loss calculation function. The output spinning production status identification results are matched with the labeled normal status, potential hidden dangers and fault abnormality labels. The identification error is calculated by the cross-entropy loss function, and the spinning production status identification model is optimized based on the identification error. The cross-entropy loss function is shown below: ; in This is the loss value. One-hot encoding for the real label; The spinning production status recognition model is validated using a test sample set. The recognition accuracy and false alarm rate are statistically analyzed, and a validation threshold is set. When the validation result reaches the validation threshold, the final spinning production status recognition model is determined. If it does not meet the threshold, it is retrained and optimized until the preset accuracy is met.

[0026] Further, please refer to Figure 1 Based on the analysis results of spinning production status, different levels of spinning production anomalies are set, and a spinning production anomaly early warning mechanism is established. The method for generating anomaly early warnings based on the anomaly early warning mechanism includes the following steps: Formal spinning production process specifications, equipment rated operating parameters, equipment operation and maintenance standards, and yarn quality control requirements are collected and obtained from the spinning production line production management system, process document library, and equipment operation and maintenance manual. Combined with the spinning production status analysis results, abnormal judgment thresholds for yarn operating parameters, equipment operating parameters, and environmental parameters are set, and three abnormality levels are set: minor abnormality, general abnormality, and severe abnormality. Among them, a minor anomaly is when a single parameter is outside the standard range and the difference between the value and the standard range is less than the corresponding preset minor anomaly threshold; a general anomaly is when a single or multiple parameters are outside the standard range and the difference between the value and the standard range is greater than the minor anomaly threshold but less than the severe anomaly threshold; and a severe anomaly is when multiple parameters are outside the standard range at the same time and the difference between the value and the standard range is greater than or equal to the corresponding preset severe anomaly threshold. An anomaly handling mechanism is established based on the three-level anomaly classification, with differentiated early warning forms and handling priorities set according to the level, and anomaly information is recorded to form anomaly information reports. The real-time spinning production status analysis results are compared with the anomaly judgment threshold. When the corresponding anomaly level triggering conditions are met, the anomaly early warning mechanism is activated and a matching anomaly early warning method is generated.

[0027] Further, please refer to Figure 1 An anomaly handling mechanism is established based on the three-level anomaly classification, with differentiated early warning forms and handling priorities set for each level. Anomaly information is recorded and an anomaly information report is generated, including the following steps: Based on the characteristics of minor, general, and severe anomalies, differentiated early warning methods are set according to the levels. Minor anomalies are set to send a text pop-up notification to the corresponding workstation operator to provide an early warning, without triggering a full-area early warning. General anomalies trigger a workstation-level audible and visual alarm, push information to the workstation operator and team manager, and mark the location of the anomaly and basic anomaly information. Severe anomalies trigger a full-area audible and visual alarm, and simultaneously push anomaly information via SMS and mobile APP to notify on-site maintenance personnel, team leaders, and production control personnel for early warning. The severity of anomalies is classified into hierarchical priority levels, and the handling requirements for each level of anomaly are set. Severe anomalies are listed as the highest priority, requiring immediate response and emergency handling to strictly control production risks. General anomalies are listed as medium priority, requiring rectification to be completed within a specified time limit. Minor anomalies are listed as routine priority, to be checked and handled by operators during daily inspections. The occurrence time, production station involved, abnormality level and related parameter type of various abnormalities are recorded throughout the process, abnormal data are recorded and integrated to form a standardized abnormal information report.

[0028] Further, please refer to Figure 1 The method for generating anomaly warnings involves issuing warnings and handling anomalies, recording the results of the warnings and handling, and continuously optimizing the spinning production status identification model and anomaly warning mechanism based on the recorded results. This includes the following steps: According to the generated abnormality warning method and corresponding abnormality level, graded warning and handling operations are carried out. Based on the differentiated characteristics of minor abnormalities, general abnormalities and serious abnormalities, appropriate handling measures are adopted, and the abnormality handling process and spinning production operation status are monitored in real time. Record and store abnormal early warning information, handling procedures and handling results to form an abnormal handling report. Based on the abnormal handling report, verify and optimize the spinning production status identification model, and at the same time optimize and adjust the judgment threshold and handling logic of the abnormal early warning mechanism.

[0029] Example 2 This embodiment also discloses a system for monitoring a spinning production method, characterized in that it includes: a multi-dimensional parameter acquisition module, a data preprocessing module, a production status identification module, an anomaly warning and handling module, and a model and mechanism optimization module; The multi-dimensional parameter acquisition module is used to set the acquisition cycle of spinning production operation parameters, collect spinning production operation parameters, and integrate them into a monitoring dataset by sorting them in time sequence. The data preprocessing module is used to detect missing and duplicate values ​​in the monitoring dataset, and to standardize it using the moving average filtering method and the range normalization method to obtain the preprocessed monitoring dataset. The production status identification module is used to construct a spinning production status identification model based on the preprocessed monitoring dataset, using a convolutional neural network algorithm combined with a long short-term memory network algorithm, and to complete the training and verification. At the same time, it combines the real-time preprocessed monitoring dataset to obtain the spinning production status analysis results. The anomaly warning and handling module is used to set anomaly judgment threshold and anomaly level based on the production status analysis results, set corresponding differentiated warning methods, and perform emergency handling and rectification operations according to the handling priority. The model and mechanism optimization module is used to verify the spinning production status identification model based on the abnormal handling report, supplement training data and optimize model parameters, and at the same time optimize the judgment threshold and handling logic of the abnormal early warning mechanism.

[0030] It should be noted that The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A method for monitoring spinning production, characterized in that, Includes the following steps: S1, set the data collection cycle for spinning production operation parameters, collect multi-dimensional operating parameters of spinning production according to the set data collection cycle, and organize them into a monitoring dataset in chronological order; The multi-dimensional operating parameters for spinning production include: yarn operating parameters, equipment operating parameters, and environmental parameters. S2, the monitoring dataset is cleaned and standardized using data preprocessing methods to obtain the preprocessed monitoring dataset; S3, based on the preprocessed monitoring dataset, constructs a spinning production status identification model by combining a convolutional neural network algorithm with a long short-term memory network algorithm, and analyzes the spinning production status through the spinning production status identification model to obtain the spinning production status analysis results. S4. Based on the analysis results of spinning production status, set different levels of spinning production anomalies and establish a spinning production anomaly early warning mechanism. Generate anomaly early warning methods based on the anomaly early warning mechanism. S5, based on the generated anomaly warning method, provides early warning and handling for anomalies, records the anomaly warning and handling results, and continuously optimizes the spinning production status identification model and anomaly warning mechanism based on the recorded results.

2. The method for monitoring spinning production according to claim 1, characterized in that, The process of setting a data collection cycle for spinning production operation parameters, collecting multi-dimensional operating parameters of spinning production according to the set data collection cycle, and organizing them into a monitoring dataset in chronological order includes the following steps: Set the collection cycle for multi-dimensional operating parameters of spinning production. Set the parameters of yarn operation and equipment condition as the first collection cycle and the environmental parameters as the second collection cycle, and record the collection time nodes of multi-dimensional operating parameters of spinning production. Corresponding sensors are deployed at each workstation of the spinning production line. Tension sensors collect yarn tension values, yarn breakage detection sensors collect yarn breakage rates to form yarn operating parameters, speed sensors collect equipment speed, current sensors collect equipment operating current to form equipment condition parameters, temperature and humidity sensors collect production site temperature and humidity, and dust concentration sensors collect production site dust content to form environmental parameters. Each collected multi-dimensional operating parameter of the spinning production line is labeled with a corresponding timestamp, parameter type, workstation number, and specific collection location information. Based on the timestamp corresponding to each piece of collected data, the multi-dimensional operating parameters of spinning production are sorted in chronological order, and the multi-dimensional operating parameters of spinning production within the same period are integrated to form a monitoring dataset.

3. The method for monitoring spinning production according to claim 1, characterized in that, The process of cleaning and standardizing the monitoring dataset using data preprocessing methods to obtain the preprocessed monitoring dataset includes the following steps: Missing and duplicate values ​​are detected in the monitoring dataset to remove time-series segments with missing and duplicate data, thus obtaining valid data. At the same time, based on the obtained valid data, the 3σ criterion anomaly detection method is used to identify and remove abnormal data, resulting in a cleaned monitoring dataset. Based on the cleaned monitoring dataset, a smoothing process is performed using the moving average filtering method to remove abnormal data caused by environmental interference and equipment vibration, resulting in a smoothed monitoring dataset. The smoothed monitoring dataset is processed by range normalization to obtain the preprocessed monitoring dataset.

4. The method for monitoring spinning production according to claim 1, characterized in that, Based on the preprocessed monitoring dataset, a spinning production status identification model is constructed using a convolutional neural network algorithm combined with a long short-term memory network algorithm. The spinning production status is then analyzed using this model to obtain the spinning production status analysis results, including the following steps: S31, the preprocessed monitoring dataset is divided into a training sample set and a test sample set, and each time series data in the monitoring dataset is labeled with the corresponding normal state, potential hidden danger and fault abnormality label according to the actual operation record of spinning production. S32, based on the training sample set, a feature extraction model is constructed using a convolutional neural network algorithm. The feature extraction model is used to extract features from the training sample set data to obtain the local fluctuation features and mutation features of the training sample set data. S33, a time series analysis model is constructed based on the long short-term memory network algorithm. The local fluctuation features and mutation features extracted by the convolutional neural network are input into the time series analysis model to obtain the analysis results. The model is tested through a test sample set. The feature extraction model and the time series analysis model are fused to construct a spinning production status identification model. S34. Collect real-time multi-dimensional operating parameters of spinning production according to the set data collection cycle, perform data preprocessing according to step S2 to obtain real-time preprocessed monitoring dataset, and input the real-time preprocessed monitoring dataset into the spinning production status identification model for identification and analysis to obtain spinning production status analysis results.

5. The method for monitoring spinning production according to claim 4, characterized in that, The process of constructing a feature extraction model based on a training sample set using a convolutional neural network algorithm, and then extracting features from the training sample set data using this model to obtain the local fluctuation and abrupt change features of the training sample set data includes the following steps: Convolutional neural networks are constructed based on training sample sets, and the kernel size and stride of the convolutional layers, the pooling window size of the pooling layers, and the number of neurons in the fully connected layers are set to construct a convolutional neural network feature extraction model. The convolutional neural network includes: convolutional layers, pooling layers, and fully connected layers; The training sample set is input into the constructed feature extraction model, and convolution operations are performed through convolutional layers to extract local features and perform weighted calculations on the training sample set data to obtain the original feature data. The formula for convolution is as follows: ； in For the convolutional layer Each output feature value The kernel length is 1. For the convolution kernel in the th Weight parameters for each position, For the first input training sample set One time series data point, For the convolutional layer The bias parameters corresponding to each output feature; The original feature data is reduced in dimensionality by using max pooling in a pooling layer. The feature response threshold is set according to the distribution characteristics of the feature data, and feature data whose feature response intensity meets the set threshold are filtered and retained to obtain the dimensionality-reduced feature data. The max pooling method is shown below: ； in For the pooling layer Each output feature value For the pooling window length, The pooling window covers the feature values ​​in the original feature data; The dimensionality-reduced feature data is input into the fully connected layer. The fully connected layer performs global integration and nonlinear mapping on the dimensionality-reduced feature data to form and output the local fluctuation and abrupt change characteristics of yarn operating parameters, equipment operating parameters, and environmental parameters.

6. The method for monitoring spinning production according to claim 4, characterized in that, The time series analysis model constructed based on the Long Short-Term Memory (LSTM) network algorithm, inputting the local fluctuation features and abrupt change features extracted by the convolutional neural network into the time series analysis model to obtain the analysis results, and testing them through a test sample set, and fusing the feature extraction model and the time series analysis model to construct a spinning production status identification model includes the following steps: A time series analysis model of a long short-term memory network is established based on the local fluctuation and mutation features extracted by a convolutional neural network, and hyperparameters such as the number of hidden layer neurons, the number of training iterations, and the learning rate are set. The long short-term memory network temporal analysis model includes: a forgetting gate, an input gate, and an output gate; The local fluctuation and mutation features output by the convolutional neural network are used as input data for the time series analysis model. The local fluctuation and mutation features under different acquisition periods are weighted and time-series encoded through forget gate, input gate and output gate. The feature data is updated in time sequence to form a time series feature expression. A fully connected layer is set at the output of the long short-term memory network temporal analysis model. The temporal feature expression is input into the fully connected layer, and the output of the fully connected layer is transformed into a probability distribution of spinning production status through an activation function. The category with the highest probability value is taken as the spinning production status identification result. The activation function is as follows: ； in For the first Predicted probability of spinning-like production conditions For the number of state categories, The first output of the fully connected layer The original output value corresponding to the spinning production state. It is a natural constant; By fusing the time series analysis model with the convolutional neural network feature extraction model, a spinning production status identification model in collaboration between convolutional neural networks and long short-term memory networks is obtained. The cross-entropy loss function is set as the model loss calculation function. The output spinning production status identification results are matched with the labeled normal status, potential hidden dangers and fault abnormality labels. The identification error is calculated by the cross-entropy loss function, and the spinning production status identification model is optimized based on the identification error. The cross-entropy loss function is shown below: ； in This is the loss value. One-hot encoding for the real label; The spinning production status recognition model is validated using a test sample set. The recognition accuracy and false alarm rate are statistically analyzed, and a validation threshold is set. When the validation result reaches the validation threshold, the final spinning production status recognition model is determined. If it does not meet the threshold, it is retrained and optimized until the preset accuracy is met.

7. The method for monitoring spinning production according to claim 1, characterized in that, The method for setting different levels of spinning production anomalies based on the analysis results of spinning production status, and establishing a spinning production anomaly early warning mechanism, and generating anomaly early warnings based on the anomaly early warning mechanism, includes the following steps: Formal spinning production process specifications, equipment rated operating parameters, equipment operation and maintenance standards, and yarn quality control requirements are collected and obtained from the spinning production line production management system, process document library, and equipment operation and maintenance manual. Combined with the spinning production status analysis results, abnormal judgment thresholds for yarn operating parameters, equipment operating parameters, and environmental parameters are set, and three abnormality levels are set: minor abnormality, general abnormality, and severe abnormality. An anomaly handling mechanism is established based on the three-level anomaly classification, with differentiated early warning forms and handling priorities set according to the level, and anomaly information is recorded to form anomaly information reports. The real-time spinning production status analysis results are compared with the anomaly judgment threshold. When the corresponding anomaly level triggering conditions are met, the anomaly early warning mechanism is activated and a matching anomaly early warning method is generated.

8. The method for monitoring spinning production according to claim 7, characterized in that, The anomaly handling mechanism established based on the three-level anomaly classification, which sets differentiated early warning forms and handling priorities for each level, and records anomaly information to form an anomaly information report, includes the following steps: Based on the characteristics of minor, general, and severe anomalies, differentiated early warning methods are set according to the levels. Minor anomalies are set to send a text pop-up notification to the corresponding workstation operator to provide an early warning, without triggering a full-area early warning. General anomalies trigger a workstation-level audible and visual alarm, push information to the workstation operator and team manager, and mark the location of the anomaly and basic anomaly information. Severe anomalies trigger a full-area audible and visual alarm, and simultaneously push anomaly information via SMS and mobile APP to notify on-site maintenance personnel, team leaders, and production control personnel for early warning. The severity of anomalies is classified into hierarchical priority levels, and the handling requirements for each level of anomaly are set. Severe anomalies are listed as the highest priority, requiring immediate response and emergency handling to strictly control production risks. General anomalies are listed as medium priority, requiring rectification to be completed within a specified time limit. Minor anomalies are listed as routine priority, to be checked and handled by operators during daily inspections. The occurrence time, production station involved, abnormality level and associated parameter type of all kinds of abnormalities are recorded in the whole process and integrated into a standardized abnormal information report.

9. The method for monitoring spinning production according to claim 1, characterized in that, The anomaly early warning method based on generation provides early warnings and handles anomalies, records the results of the anomaly early warning and handling, and continuously optimizes the spinning production status identification model and anomaly early warning mechanism based on the recorded results, including the following steps: According to the generated abnormality warning method and corresponding abnormality level, graded warning and handling operations are carried out. Based on the differentiated characteristics of minor abnormalities, general abnormalities and serious abnormalities, appropriate handling measures are adopted, and the abnormality handling process and spinning production operation status are monitored in real time. Record and store abnormal early warning information, handling procedures and handling results to form an abnormal handling report. Based on the abnormal handling report, verify and optimize the spinning production status identification model, and at the same time optimize and adjust the judgment threshold and handling logic of the abnormal early warning mechanism.

10. A system for implementing the monitoring method for spinning production as described in claim 1, characterized in that, include: Multi-dimensional parameter acquisition module, data preprocessing module, production status identification module, anomaly early warning and handling module, and model and mechanism optimization module; The multi-dimensional parameter acquisition module is used to set the acquisition cycle of spinning production operation parameters, collect spinning production operation parameters, and integrate them into a monitoring dataset by sorting them in time sequence. The data preprocessing module is used to detect missing and duplicate values ​​in the monitoring dataset, and to standardize it using the moving average filtering method and the range normalization method to obtain the preprocessed monitoring dataset. The production status identification module is used to construct a spinning production status identification model based on the preprocessed monitoring dataset, using a convolutional neural network algorithm combined with a long short-term memory network algorithm, and to complete the training and verification. At the same time, it combines the real-time preprocessed monitoring dataset to obtain the spinning production status analysis results. The anomaly warning and handling module is used to set anomaly judgment threshold and anomaly level based on the production status analysis results, set corresponding differentiated warning methods, and perform emergency handling and rectification operations according to the handling priority. The model and mechanism optimization module is used to verify the spinning production status identification model based on the abnormal handling report, supplement training data and optimize model parameters, and at the same time optimize the judgment threshold and handling logic of the abnormal early warning mechanism.

Images