Textile yarn intelligent production control system and method based on digital twinning

Abstract

The application discloses a textile yarn intelligent production control system and method based on digital twinning, which comprises the following steps: S1, obtaining the running data of a spinning frame; S2, establishing a virtual running body of the spinning frame to form a virtual-real working condition state; S3, inputting the virtual-real working condition state into an improved Reformer model, introducing a yarn state neighborhood resonance regularization mechanism in an LSH attention layer, executing neighborhood rearrangement and cross-bar neighborhood splicing, and generating continuous state quantities; S4, adopting a Hilbert-Huang transformation algorithm on the continuous state quantities, introducing a critical precursor stationary phase separation mechanism, and forming multi-scale working condition components; S5, forming a spinning working condition trajectory; S6, executing linkage matching to generate linkage control quantities; and S7, applying the linkage control quantities to the spinning frame to update the virtual-real working condition state. The application can realize continuous modeling and multi-parameter collaborative control of the spinning working condition, and improve the stability of the running state of the spinning frame and the production control precision.

Description

Technical Field

[0001] This invention relates to the fields of intelligent manufacturing and digital twin control technology in textiles, and particularly to an intelligent production control system and method for textile yarns based on digital twins. Background Technology

[0002] With the increasing demand for intelligent manufacturing and digital production in the textile industry, real-time sensing, state modeling, and dynamic control technologies for the operating status of spinning machines have received widespread attention. Existing textile yarn production control methods mainly rely on single sensor data or empirical rules to adjust the operating status of spinning machines, but these methods generally suffer from the following problems in practical applications:

[0003] Operational data such as tension in the drafting zone of the spinning frame, roller differential speed, spindle speed fluctuations, and workshop temperature and humidity suffer from inconsistent sampling frequencies, inconsistent time stamps, and data gaps during data collection. Existing time stamp alignment and operational condition normalization methods struggle to generate continuous and consistent spinning operational data, leading to biases in subsequent state analysis. Traditional state analysis methods based on empirical rules or simple statistical models are inadequate for depicting the continuous evolution of spinning operational conditions over time, lacking effective modeling tools for operational conditions under multivariate coupled changes, resulting in difficulties in accurately extracting key operational features. For long-term series operational data, existing attention-based models suffer from limitations in processing... The methods are susceptible to limitations of segmented calculations and local correlations, which can lead to breaks in continuous state information during the segmentation process, affecting the integrity of the state representation. For nonlinear and non-stationary spinning condition signals, traditional empirical mode decomposition methods suffer from mode aliasing and endpoint effects, making it difficult to distinguish between key oscillation components and interference components, thus reducing the accuracy of multi-scale condition component extraction. In the production control stage, existing methods mostly adopt independent adjustment of single parameters, lacking modeling of the linkage relationship between roller differential speed adjustment, spindle speed compensation, and environmental adjustment, making it difficult to form coordinated and consistent linkage control quantities, resulting in lag and insufficient stability in the adjustment response of the spinning machine's operating status.

[0004] Therefore, how to provide intelligent production control systems and methods for textile yarns based on digital twins is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose an intelligent production control system and method for textile yarn based on digital twins. This invention acquires the operating data of the spinning machine and forms spinning condition quantities. It establishes a virtual operating body of the spinning machine in the digital twin space and forms a virtual and real operating state. The virtual and real operating state is input into an improved Reformer model to generate continuous state quantities. Multi-scale operating condition components are obtained through the Hilbert-Huang transform algorithm, and a spinning condition trajectory is formed based on the multi-scale operating condition components. Furthermore, the roller differential speed adjustment, spindle speed compensation, and environmental adjustment are calculated to generate linkage control quantities, thereby realizing dynamic control of the spinning machine's operating state. This invention has the advantages of high accuracy in operating condition modeling, strong continuity of state evolution, and high stability in production control.

[0006] The intelligent production control method for textile yarn based on digital twins according to embodiments of the present invention includes the following steps:

[0007] S1. Obtain the operating data of the spinning machine, perform time scale alignment and working condition adjustment to form spinning working condition quantities;

[0008] S2. Based on the spinning condition parameters, a virtual operating body of the spinning machine is established in the digital twin space, and the spinning condition parameters are loaded into the virtual operating body of the spinning machine hourly to form a virtual and real operating condition state.

[0009] S3. Input the virtual and real working conditions into the improved Reformer model, introduce the yarn-state neighborhood resonance regularization mechanism in the LSH attention layer, construct the resonance indicator, and perform neighborhood rearrangement and cross-bucket adjacency splicing on the hash bucketing results generated by the LSH attention layer to generate continuous state variables.

[0010] S4. The Hilbert-Huang transform algorithm is used for continuous state variables. A critical precursor stationary phase separation mechanism is introduced. The main mode is obtained through empirical mode decomposition. The stationary phase criterion is constructed. The stationary phase region is identified from the main mode and weak oscillation components are separated to form multi-scale operating condition components.

[0011] S5. Based on the multi-scale working condition components, calculate the tension change trend in the drafting zone, the twisting stability change trend, and the winding rhythm change trend to form a spinning working condition trajectory.

[0012] S6. Based on the spinning condition trajectory, calculate the roller differential speed adjustment, spindle speed compensation and environmental adjustment respectively, perform linkage matching, and generate linkage control quantity;

[0013] S7. Apply the linkage control amount to the spinning machine, generate the controlled operating data, and update the virtual and real operating conditions.

[0014] Optionally, S1 specifically includes:

[0015] Collect data on tension in the drafting zone of the spinning machine, roller differential speed, spindle speed fluctuation, and workshop temperature and humidity, and add time stamps for the corresponding collection times.

[0016] According to the unified sampling interval, the tension, roller differential speed, spindle speed fluctuation and workshop temperature and humidity data of the spinning machine drafting zone after the additional time mark are resampled to obtain tension data, differential speed data, speed data and temperature and humidity data with consistent time position;

[0017] Tension data, differential speed data, rotational speed data, and temperature and humidity data with consistent time and location are paired and organized according to the same moment to obtain the working condition sampling group;

[0018] The operating condition sampling groups are sequentially arranged according to time sequence to form spinning operating condition quantities.

[0019] Optionally, the step of establishing a virtual operating entity of the spinning machine in the digital twin space based on spinning conditions specifically involves:

[0020] According to the correspondence of the physical structure of the spinning machine, the tension in the drafting zone of the spinning machine is correlated with the position of the drafting station parameters, the roller differential speed is correlated with the position of the roller transmission parameters, the spindle speed fluctuation is correlated with the position of the spindle operation parameters, and the workshop temperature and humidity data are correlated with the position of the environmental parameters.

[0021] A framework for the operating parameters of a spinning frame is constructed based on the positions of the drafting station parameters, roller drive parameters, spindle operation parameters, and environmental parameters.

[0022] The tension in the drafting zone of the spinning machine, roller differential speed, spindle speed fluctuation, and workshop temperature and humidity data are written into the corresponding parameter positions in the spinning machine operation parameter framework to form a virtual operating body of the spinning machine.

[0023] Optionally, the step of loading the spinning condition parameters into the virtual operating entity of the spinning machine hourly to form a virtual-real condition state specifically involves:

[0024] According to the time sequence in the spinning condition data, extract the tension in the drafting zone of the spinning machine, roller differential speed, spindle speed fluctuation and workshop temperature and humidity data corresponding to each time position;

[0025] Write the tension in the drafting zone of the spinning machine corresponding to each time position into the drafting station parameter position in the virtual operating body of the spinning machine; write the roller differential speed corresponding to each time position into the roller transmission parameter position in the virtual operating body of the spinning machine; write the spindle speed fluctuation corresponding to each time position into the spindle operation parameter position in the virtual operating body of the spinning machine; and write the workshop temperature and humidity data corresponding to each time position into the environmental parameter position in the virtual operating body of the spinning machine.

[0026] After each write operation, the virtual operating status at the corresponding time position is obtained; the virtual operating status is continuously arranged in chronological order and synchronized with the corresponding time position data in the spinning condition data to form a virtual and real operating status.

[0027] Optionally, the improved Reformer model specifically includes a sequence embedding layer, an LSH attention layer, an invertible residual layer, and a block feedforward layer;

[0028] The sequence embedding layer arranges the tension in the drafting zone of the spinning machine, roller differential speed, spindle speed fluctuation, and workshop temperature and humidity data in the virtual and real working conditions in chronological order, and combines the tension in the drafting zone of the spinning machine, roller differential speed, spindle speed fluctuation, and workshop temperature and humidity data at the corresponding time position into a state vector.

[0029] The LSH attention layer converts the state vector into a query vector and a key vector, respectively, and divides the state vector into different hash buckets based on the random hash values ​​of the query vector and the key vector, forming a hash bucketing result;

[0030] After generating the hash binning results, a yarn-state neighborhood resonance regularization mechanism is introduced. Resonance indicators are constructed based on the differences in drafting zone tension, roller speed, spindle speed, and temperature and humidity at adjacent time positions. The state vectors within the same hash bin are reordered according to the direction of change of the resonance indicators, and state vectors with continuous resonance indicators at the boundaries of adjacent hash bins are connected as adjacent segments. The attention results within the bin are calculated based on the reordered state vectors and adjacent segments, and the attention results within the bin are restored and arranged according to time position to form the neighborhood-regularized state vectors.

[0031] The reversible residual layer splits the neighborhood-normalized state vector into a first sub-vector and a second sub-vector, and inputs the first sub-vector and the second sub-vector into the reversible residual unit respectively. An intermediate state vector is formed by adding the vectors of the front and back layers.

[0032] The segmented feedforward layer divides the intermediate state vector into segments according to a preset length, inputs each segment of the intermediate state vector into the feedforward network in sequence, and splices each segment of the feedforward output according to the original time position to form a continuous state quantity.

[0033] Optionally, S4 specifically includes:

[0034] For each segment of the state curve in the continuous state variable, extract the local maximum and local minimum points respectively. Construct an upper envelope based on the local maximum points and a lower envelope based on the local minimum points. Calculate the local mean curve based on the upper and lower envelopes.

[0035] Subtract the corresponding local mean curve from each segment of the state curve in the continuous state variables to obtain the screening result. Repeat the extraction of local maxima, local minima, upper envelope, lower envelope, and local mean curves from the screening result until the main mode that satisfies the empirical mode decomposition conditions is obtained. The empirical mode decomposition conditions include that, within the same time interval, the difference between the sum of the number of local maxima and the number of local minima and the number of zero crossings does not exceed 1, and the local mean curves calculated from the upper and lower envelopes are close to 0 within the time interval, and the amplitude range enclosed by the upper and lower envelopes is symmetrically distributed.

[0036] Perform Hilbert transform on the continuous state variables to obtain the phase and envelope values ​​at the corresponding time positions, and construct the stationary phase criterion based on the phase difference and envelope difference between adjacent time positions;

[0037] The stationary phase region is formed by determining the time segment in the main mode where the phase change slows down and the envelope change is concentrated according to the stationary phase criterion. The local oscillation component is separated along the fluctuation direction of the stationary phase region in the main mode to obtain the weak oscillation component. The remaining components in the main mode and the weak oscillation component are retained respectively, and multi-scale operating condition components are formed according to the time position correspondence.

[0038] Optionally, S5 specifically includes:

[0039] Based on the time-location correspondence, the tension-related components, twisting-related components, and winding-related components in the multi-scale working condition components are extracted;

[0040] The tension change in the stretching zone is constructed based on the numerical difference of the tension-related components at adjacent time positions, and the tension change in the stretching zone is arranged in chronological order to form the tension change trend in the stretching zone.

[0041] The twisting stability variation is constructed based on the difference in fluctuation amplitude and fluctuation interval of the twisting-related components at adjacent time positions, and the twisting stability variation is arranged in chronological order to form the twisting stability variation trend.

[0042] The winding beat variation is constructed based on the beat interval difference between adjacent time positions of the winding-related components, and the winding beat variation is arranged in chronological order to form the winding beat variation trend.

[0043] The trends of tension change in the drafting zone, twisting stability change, and winding rhythm change are arranged according to the same time position to form a spinning condition trajectory.

[0044] Optionally, S6 specifically includes:

[0045] Extract the trends of tension variation in the drafting zone, twisting stability variation, and winding rhythm variation in the spinning process trajectory based on the time-location correspondence.

[0046] The roller differential speed adjustment is constructed based on the difference in the change trend of tension in the stretching zone at adjacent time positions, and the roller differential speed adjustment is arranged in chronological order.

[0047] Spindle speed compensation is constructed based on the difference in the change trend of twisting stability at adjacent time positions, and the spindle speed compensation is arranged in chronological order.

[0048] Environmental regulation quantities are constructed based on the difference in the change trend of winding cycle at adjacent time positions and the difference in the change of workshop temperature and humidity data at adjacent time positions, and the environmental regulation quantities are arranged in chronological order.

[0049] The roller differential speed adjustment, spindle speed compensation, and environmental adjustment are organized according to the same time position; and the linkage matching is performed according to the combination relationship of the roller differential speed adjustment, spindle speed compensation, and environmental adjustment at the same time position to form a linkage control quantity.

[0050] Optionally, S7 specifically includes:

[0051] The roller differential speed adjustment corresponding to each time position is applied to the roller transmission parameter position of the spinning machine, the spindle speed compensation corresponding to each time position is applied to the spindle operation parameter position of the spinning machine, and the environmental adjustment corresponding to each time position is applied to the environmental parameter position.

[0052] After loading is completed at each time point, data on tension in the drafting zone of the spinning machine, roller differential speed, spindle speed fluctuation, and workshop temperature and humidity are collected at the corresponding time point to form the adjusted operating data.

[0053] According to the time sequence in the adjusted operating data, the tension in the drafting zone of the spinning machine is written into the drafting station parameter position, the roller differential speed is written into the roller transmission parameter position, the spindle speed fluctuation is written into the spindle operation parameter position, and the workshop temperature and humidity data is written into the environmental parameter position to obtain the updated virtual operating status.

[0054] The updated virtual operating states are arranged sequentially according to time, and synchronized with the corresponding time and position data in the adjusted operating data to update the virtual and real operating conditions.

[0055] Optional, the intelligent production control system for textile yarn based on digital twins includes the following modules:

[0056] The data acquisition module is used to acquire the operating data of the spinning machine, perform time scale alignment and working condition correction, and form spinning working condition quantities;

[0057] The twin modeling module is used to establish a virtual operating body of the spinning machine in the digital twin space based on the spinning condition parameters, and to load the spinning condition parameters into the virtual operating body of the spinning machine hourly to form a virtual and real operating condition state.

[0058] The state generation module is used to input the virtual and real working conditions into the improved Reformer model, introduce a yarn-state neighborhood resonance regularization mechanism in the LSH attention layer, construct a resonance indicator, and perform neighborhood rearrangement and cross-bucket adjacency splicing on the hash bucketing results generated by the LSH attention layer to generate continuous state variables.

[0059] The component generation module is used to apply the Hilbert-Huang transform algorithm to continuous state variables, introduce a critical precursor stationary phase separation mechanism, obtain the main mode through empirical mode decomposition, construct the stationary phase criterion quantity, identify the stationary phase region from the main mode and separate the weak oscillation component to form multi-scale operating condition components.

[0060] The trajectory construction module is used to calculate the tension change trend of the drafting zone, the twisting stability change trend, and the winding rhythm change trend based on the multi-scale working condition components, and form a spinning working condition trajectory.

[0061] The control generation module is used to calculate the roller differential speed adjustment, spindle speed compensation and environmental adjustment based on the spinning working condition trajectory, perform linkage matching and generate linkage control quantities;

[0062] The feedback update module is used to apply the linkage control amount to the spinning machine, generate the adjusted operating data, and update the virtual and real operating conditions.

[0063] The beneficial effects of this invention are:

[0064] (1) By aligning the time stamp and adjusting the working conditions of the spinning machine's operating data, spinning working condition quantities are formed, and a virtual operating body of the spinning machine is established in the digital twin space, so that the virtual and real working conditions are consistent at the same time and position, effectively solving the problems of inconsistent time of multi-source operating data and discrete expression of working conditions, and improving the continuity and consistency of spinning working condition quantities.

[0065] (2) In the improved Reformer model, a neighborhood resonance regularization mechanism is introduced. By constructing resonance indicators and performing neighborhood rearrangement and cross-bucket adjacency splicing on the hash bucket results, the continuous state variables maintain the continuity of working condition changes on the time axis, reduce the state breakage caused by hash bucketing, and improve the expression accuracy of continuous state variables.

[0066] (3) In the Hilbert-Huang transform algorithm, a critical precursor stationary phase separation mechanism is introduced. By constructing a stationary phase criterion, the stationary phase region is identified and weak oscillation components are separated, so that the multi-scale operating condition components can distinguish between effective operating condition changes and interference oscillations, reduce mode aliasing, and improve the accuracy of multi-scale operating condition component extraction. Attached Figure Description

[0067] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0068] Figure 1 This is a flowchart of the intelligent production control method for textile yarn based on digital twin proposed in this invention;

[0069] Figure 2 This is a schematic diagram of the improved Reformer model proposed in this invention;

[0070] Figure 3 This is a schematic diagram of the intelligent production control system for textile yarn based on digital twin proposed in this invention. Detailed Implementation

[0071] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0072] refer to Figures 1-3 A digital twin-based intelligent production control method for textile yarn includes the following steps:

[0073] S1. Obtain the operating data of the spinning machine, perform time scale alignment and working condition adjustment to form spinning working condition quantities;

[0074] S2. Based on the spinning condition parameters, a virtual operating body of the spinning machine is established in the digital twin space. The spinning condition parameters are loaded into the virtual operating body of the spinning machine hourly to form a virtual and real operating condition state.

[0075] S3. Improve the Reformer model by inputting virtual and real working conditions, introduce a yarn-state neighborhood resonance regularization mechanism in the LSH attention layer, construct a resonance indicator, and perform neighborhood rearrangement and cross-bucket adjacency splicing on the hash bucketing results generated by the LSH attention layer to generate continuous state variables.

[0076] S4. The Hilbert-Huang transform algorithm is used for continuous state variables. A critical precursor stationary phase separation mechanism is introduced. The main mode is obtained through empirical mode decomposition. The stationary phase criterion is constructed. The stationary phase region is identified from the main mode and weak oscillation components are separated to form multi-scale operating condition components.

[0077] S5. Based on the multi-scale working condition components, calculate the tension change trend in the drafting zone, the twisting stability change trend, and the winding rhythm change trend to form a spinning working condition trajectory.

[0078] S6. Calculate the roller differential speed adjustment, spindle speed compensation and environmental adjustment based on the spinning working condition trajectory, perform linkage matching and generate linkage control quantity;

[0079] S7. Apply the linkage control amount to the spinning machine, generate the controlled operating data, and update the virtual and real operating conditions.

[0080] In this embodiment, S1 specifically refers to:

[0081] Data on tension in the drafting zone, roller differential speed, spindle speed fluctuation, and temperature and humidity in the workshop were acquired at the drafting zone of the spinning machine, the roller drive position, the spindle running position, and the workshop environment position, and a time stamp corresponding to the acquisition time was added to each data point.

[0082] The tension, roller differential, spindle speed fluctuation, and workshop temperature and humidity data of the drawing zone after the addition of time markers are time-aligned according to the uniform sampling interval. The data of the corresponding time position is obtained between adjacent time positions according to the relationship between the values ​​and time, so as to obtain tension data, differential speed data, speed data and temperature and humidity data with consistent time positions.

[0083] The tension data, differential speed data, rotational speed data, and temperature and humidity data are combined according to the same time position, so that one time position corresponds to one set of tension data, differential speed data, rotational speed data, and temperature and humidity data, thus obtaining the working condition sampling group. The working condition sampling groups are arranged continuously in chronological order, so that the working condition sampling groups of adjacent time positions are connected in sequence to form the spinning working condition quantity.

[0084] In this embodiment, a virtual operating entity of the spinning machine is established in the digital twin space based on the spinning condition parameters, specifically as follows:

[0085] Based on the physical structure of the spinning machine, the spinning machine is divided into drafting station, roller drive position, spindle operation position and environmental parameter position, and parameter positions corresponding to the drafting station, roller drive position, spindle operation position and environmental parameter position are set in the digital twin space respectively.

[0086] The tension in the drafting zone of the spinning machine in the spinning condition parameters is mapped to the drafting station parameter position, the roller differential speed is mapped to the roller drive parameter position, the spindle speed fluctuation is mapped to the spindle operation parameter position, and the workshop temperature and humidity data are mapped to the environmental parameter position, so that one type of operating data corresponds to one parameter position.

[0087] The positions of the drafting station parameters, roller drive parameters, spindle operation parameters, and environmental parameters are arranged according to the connection relationships in the physical structure of the spinning frame. This ensures that the positions of the drafting station parameters and roller drive parameters maintain a transmission correspondence, the positions of the roller drive parameters and spindle operation parameters maintain an operation correspondence, and the positions of the environmental parameters maintain an environmental interaction relationship with the positions of the drafting station parameters, roller drive parameters, and spindle operation parameters, thus forming the framework of the spinning frame's operating parameters.

[0088] Write the tension in the drafting zone of the spinning machine into the drafting station parameter position, write the roller differential speed into the roller transmission parameter position, write the spindle speed fluctuation into the spindle operation parameter position, and write the workshop temperature and humidity data into the environmental parameter position, so that each parameter position in the spinning machine operation parameter framework has corresponding operation data.

[0089] The positions of the drafting station parameters, roller drive parameters, spindle operation parameters, and environmental parameters after the running data is written are correlated as a whole to obtain a virtual running body of the spinning machine corresponding to the running status of the spinning machine.

[0090] In this embodiment, the spinning condition parameters are loaded hourly into the virtual operating entity of the spinning frame to form a virtual and real operating condition state, specifically as follows:

[0091] The spinning condition sampling groups are arranged in chronological order according to time, and the tension in the drafting zone of the spinning machine, roller differential speed, spindle speed fluctuation and workshop temperature and humidity data corresponding to one spinning condition sampling group are read.

[0092] Write the tension in the drafting zone of the spinning machine in a working condition sampling group into the drafting station parameter position in the virtual running body of the spinning machine; write the roller differential speed in a working condition sampling group into the roller transmission parameter position; write the spindle speed fluctuation in a working condition sampling group into the spindle operation parameter position; and write the workshop temperature and humidity data in a working condition sampling group into the environmental parameter position.

[0093] After writing the data of the drawing station parameter position, roller drive parameter position, spindle operation parameter position and environmental parameter position at the same time and location, the drawing station parameter position, roller drive parameter position, spindle operation parameter position and environmental parameter position are correlated accordingly to obtain a virtual running state at a time position;

[0094] The next working condition sampling group is read repeatedly in chronological order, and the corresponding data writing and association are repeated to obtain the virtual operating status of continuous time positions;

[0095] The virtual running state at the previous time position is connected sequentially with the virtual running state at the next time position to form a group of virtual running states arranged in chronological order.

[0096] Each time position in the virtual operating status group is mapped to the same time position in the spinning condition quantity, so that the same time position simultaneously has the actual operating data of the spinning machine and the virtual operating status of the spinning machine, forming a virtual and real operating condition state.

[0097] In this embodiment, the improved Reformer model specifically includes a sequence embedding layer, an LSH attention layer, an invertible residual layer, and a block feedforward layer;

[0098] The sequence embedding layer arranges the data of tension in the drafting zone of the spinning machine, roller differential speed, spindle speed fluctuation, and workshop temperature and humidity in the virtual and real working conditions in chronological order. At the same time position, the data of tension in the drafting zone of the spinning machine, roller differential speed, spindle speed fluctuation, and workshop temperature and humidity are combined in sequence to form a state vector. The first dimension of the state vector is the tension value of the drafting zone of the spinning machine, the second dimension is the roller differential speed value, the third dimension is the spindle speed fluctuation value, the fourth dimension is the workshop temperature value, and the fifth dimension is the workshop humidity value.

[0099] The LSH attention layer multiplies the state vector by the query transformation matrix and the key transformation matrix respectively to obtain the query vector and the key vector. Any component in the query vector is equal to the sum of the coefficients in the corresponding column of the query transformation matrix multiplied by all components in the state vector respectively. Any component in the key vector is equal to the sum of the coefficients in the corresponding column of the key transformation matrix multiplied by all components in the state vector respectively.

[0100] The query vector and key vector are respectively multiplied by the random projection vector group to obtain the query hash value and key hash value. The query hash value is equal to the sum of the product of each component of the query vector and the corresponding component of the random projection vector group. The key hash value is equal to the sum of the product of each component of the key vector and the corresponding component of the random projection vector group. Hash buckets are divided according to the numerical range of the query hash value and the key hash value. State vectors whose query hash value and key hash value fall into the same range are grouped into the same hash bucket to form the hash bucketing result.

[0101] After generating the hash binning results, a yarn-state neighborhood resonance regularization mechanism is introduced. The tension in the drafting zone of the spinning machine, roller differential speed, spindle speed fluctuation, workshop temperature, and workshop humidity data at adjacent time positions are read. The difference in drafting zone tension, roller differential speed, spindle speed, workshop temperature, and workshop humidity are obtained by subtracting the value of the previous time position from the value of the later time position. The difference in workshop temperature and workshop humidity is then added together and divided by 2 to obtain the temperature and humidity difference.

[0102] The tension difference in the drawing zone, the differential speed difference of the rollers, the difference in spindle speed, and the difference in temperature and humidity are combined according to the same time position to form a resonance indicator. The resonance indicator is equal to the sum of the absolute values ​​of the tension difference in the drawing zone, the differential speed difference of the rollers, the difference in spindle speed, and the difference in temperature and humidity.

[0103] According to the direction of change of the resonance indicator on the time axis, the state vectors in the same hash bucket are rearranged, with the state vectors with smaller resonance indicators placed first and the state vectors with larger resonance indicators placed last.

[0104] Compare the state vectors at the boundaries of adjacent hash buckets. When the difference between the resonance indicator corresponding to the state vector at the boundary of the next hash bucket and the resonance indicator corresponding to the state vector at the boundary of the previous hash bucket is not greater than the average difference of the resonance indicators of adjacent state vectors within the same hash bucket, connect the state vector at the boundary of the previous hash bucket and the state vector at the boundary of the next hash bucket in sequence to form an adjacent segment.

[0105] Within each hash bucket, the query vector and key vector are read sequentially. The correlation value is obtained by multiplying the corresponding components of the query vector and key vector respectively and then summing them. The normalized correlation value is obtained by dividing the correlation value by the square root of the number of dimensions of the query vector. Then, the maximum normalized correlation value in the same hash bucket is subtracted from all normalized correlation values ​​in the same hash bucket, and the exponent is taken. The exponent is then divided by the sum of all exponent values ​​to obtain the attention result in the bucket. The attention results in the bucket are restored and arranged according to their original time positions to form a neighborhood-normalized state vector.

[0106] The reversible residual layer splits the neighborhood-normalized state vector into a first sub-vector and a second sub-vector along the vector dimension. The first sub-vector and the second sub-vector have the same number of dimensions. The first sub-vector is input into the first residual mapping relation, and the second sub-vector is input into the second residual mapping relation. The output of the first residual mapping relation is added element-wise to the second sub-vector to obtain the first intermediate vector. The output of the second residual mapping relation is added element-wise to the first intermediate vector to obtain the second intermediate vector. The first intermediate vector and the second intermediate vector are then combined to form an intermediate state vector.

[0107] The segmented feedforward layer divides the intermediate state vector into segments according to a preset length. The preset length is determined by the number of time positions contained in one control cycle of the spinning machine. The number of time positions contained in one control cycle of the spinning machine is equal to the control cycle duration divided by a uniform sampling interval. Each segment of the intermediate state vector is sequentially input into the feedforward network. Any component of the output vector of each layer in the feedforward network is equal to the sum of all components of the previous layer's output vector multiplied by the corresponding coefficient of the current layer, and then added to the current layer's bias value. Each segment of the feedforward output is sequentially connected according to the original time position order to form a continuous state quantity.

[0108] In this embodiment, both the improved Reformer model and the Reformer model include a sequence embedding layer, an attention computation structure based on locality-sensitive hashing, a reversible residual structure, and a block feedforward structure. Both employ a hash bucketing method using query vectors and key vectors to reduce the computational complexity of long sequences. The sequence embedding layer maintains the vectorized representation of time-ordered data, while combining data on tension in the drafting zone of the spinning frame, roller differential speed, spindle speed fluctuations, and workshop temperature and humidity into a state vector according to a unified time position, allowing the state vector to directly correspond to the operating state in the spinning process parameters. In LSH... In the attention layer, a yarn-state neighborhood resonance regularization mechanism is introduced on top of the original random hash binning. This mechanism calculates and combines the differences in drafting zone tension, roller speed, spindle speed, and temperature / humidity at adjacent time points to form a resonance indicator reflecting the intensity of changes in spinning conditions. This ensures that the hash binning results are no longer solely dependent on random projection relationships but are also constrained by the continuity of condition changes. Within the same hash bucket, the state vectors are rearranged according to the direction of change of the resonance indicator, ensuring that state vectors with consistent trends remain adjacent on the time axis. At the boundaries of adjacent hash buckets, state vectors with continuously changing resonance indicators are connected, re-establishing the correlation between continuous condition segments originally divided by hash binning, thereby reducing the impact of bucket boundaries on temporal continuity. During the attention calculation stage, the rearranged state vectors and adjacent segments are used as unified inputs, allowing the attention calculation to consider both buckets simultaneously. By improving internal correlation and cross-bucket continuity, a neighborhood regularization result that better conforms to the evolution law of spinning conditions is obtained. In the reversible residual structure, the vector splitting and element-wise superposition method are maintained to ensure the reversible transmission of information between the front and back layers, while ensuring the stable propagation of long sequence state information in the multi-layer structure. In the block feedforward structure, the intermediate state vector is segmented according to the number of time positions corresponding to the spinning machine control cycle, so that each segment corresponds to a continuous working condition interval. Then, the segmented data is processed by layer-by-layer linear combination and reconnected according to time position to obtain continuous state variables. Through the above improvements, the state vector takes into account both random distribution characteristics and continuous working condition changes in the hash bucketing process, so that local abnormal changes and continuous change segments in long sequence data can be captured simultaneously. This improves the expression accuracy of continuous state variables for the evolution process of spinning conditions, reduces the impact of information breakage caused by hash bucketing, and improves the stability and accuracy of subsequent multi-scale working condition component decomposition.

[0109] In this embodiment, S4 specifically refers to:

[0110] Read a segment of the state curve in the continuous state variables in chronological order, and compare the values ​​of adjacent time positions point by point in the segment of the state curve. Points with values ​​greater than the previous time position and greater than the next time position are recorded as local maxima, and points with values ​​less than the previous time position and less than the next time position are recorded as local minima.

[0111] Connect adjacent local maxima in chronological order to form an upper envelope, and connect adjacent local minima in chronological order to form a lower envelope. Add the values ​​of the upper and lower envelopes at the same time position and divide by 2 to obtain the local mean curve.

[0112] The screening result is obtained by subtracting the value of the local mean curve at the same time position from the value of the state curve at the same time position. The local maximum point, local minimum point, upper envelope, lower envelope and local mean curve are repeatedly extracted from the screening result until the difference between the sum of the number of local maximum points and the number of local minimum points and the number of zero crossing points does not exceed 1, the value of the local mean curve in the time interval is close to 0, and the amplitude range enclosed by the upper envelope and the lower envelope is symmetrically distributed.

[0113] For a segment of a state curve in a continuous state variable, perform a Hilbert transform, add the squares of the original value and the transformed value at the same time position, take the square root to obtain the envelope value, divide the transformed value by the original value at the same time position, and take the arctangent value to obtain the phase value.

[0114] Read the phase and envelope values ​​of adjacent time positions, subtract the phase value of the previous time position from the phase value of the next time position to obtain the phase difference, subtract the envelope value of the previous time position from the envelope value of the next time position to obtain the envelope difference, and add the absolute value of the phase difference to the absolute value of the envelope difference to obtain the stationary phase criterion quantity.

[0115] By comparing the stationary phase criteria in chronological order, when the phase difference between adjacent time positions continues to decrease and the envelope difference continues to concentrate in the same segment, the corresponding time segment is determined as the stationary phase region.

[0116] Local fluctuations are read along the peak and trough directions of the main mode in the stationary phase region. Local fluctuations that are small in amplitude and persistent relative to the average fluctuation of the main mode in the stationary phase region are separated from the main mode to obtain weak oscillation components.

[0117] The remaining components after subtracting the weak oscillation component from the main mode are retained. The weak oscillation component is retained separately, and the remaining components are arranged in correspondence with the weak oscillation component according to the time position relationship to form multi-scale working condition components.

[0118] In this embodiment, S5 specifically refers to:

[0119] According to the time-location correspondence, the multi-scale working condition components are read, and according to the source location of the multi-scale working condition components in the formation process, the components from the tension change in the stretching zone are recorded as tension-related components, the components from the spindle speed fluctuation and the local oscillation change of the main mode are recorded as twisting-related components, and the components from the periodic fluctuation change during the winding period are recorded as winding-related components.

[0120] Read the values ​​of tension-related components at adjacent time positions in chronological order, subtract the value of the previous time position from the value of the later time position to obtain the tension change in the stretching zone, and arrange the tension change in the stretching zone obtained from consecutive time positions in sequence to form the tension change trend in the stretching zone.

[0121] Read the peak and trough values ​​of the twisting-related components at adjacent time positions in chronological order. Subtract the trough value from the peak value at the same time position to obtain the fluctuation amplitude. Subtract the fluctuation amplitude of the previous time position from the fluctuation amplitude of the later time position to obtain the fluctuation amplitude difference.

[0122] Read the peak or trough of the twisted component in chronological order at adjacent time positions. Subtract the peak of the previous time position from the peak of the later time position to obtain the fluctuation interval. Subtract the fluctuation interval of the previous time position from the fluctuation interval of the later time position to obtain the fluctuation interval difference.

[0123] The difference in fluctuation amplitude at the same time position is added to the difference in fluctuation interval to obtain the change in twisting stability. The changes in twisting stability obtained at consecutive time positions are arranged in sequence to form the trend of twisting stability change.

[0124] Read the periodic fluctuation positions in the winding-related components in chronological order. Subtract the time value of the previous periodic fluctuation position from the time value of the next periodic fluctuation position to obtain the beat interval. Subtract the beat interval of the previous time position from the beat interval of the next time position to obtain the winding beat change. Arrange the winding beat changes obtained from consecutive time positions in sequence to form the winding beat change trend.

[0125] The trends of tension change in the drafting zone, twisting stability change, and winding rhythm change are correlated according to the same time position. The changes in tension in the drafting zone, twisting stability change, and winding rhythm change at the same time position are combined sequentially to form the spinning condition trajectory.

[0126] In this embodiment, S6 specifically refers to:

[0127] According to the time-location correspondence, read the tension change trend of the drafting zone, the twisting stability change trend, and the winding rhythm change trend in the spinning condition trajectory, and extract the tension change, twisting stability change, and winding rhythm change at the same time location respectively.

[0128] Subtract the change in tension in the stretching zone from the change in tension in the previous time position to obtain the difference in the trend of tension change in the stretching zone between adjacent time positions.

[0129] The difference in the trend of tension change in the stretching zone at adjacent time positions is mapped to the direction of roller differential speed adjustment. When the difference is positive, the roller differential speed adjustment amount is recorded as the decrease amount; when the difference is negative, the roller differential speed adjustment amount is recorded as the increase amount; when the difference is 0, the roller differential speed adjustment amount is recorded as the hold amount. The roller differential speed adjustment amounts are arranged in chronological order.

[0130] Subtract the change in twisting stability at the previous time position from the change in twisting stability at the next time position to obtain the difference in the trend of twisting stability at adjacent time positions.

[0131] The difference in the trend of twisting stability at adjacent time positions is mapped to the direction of spindle speed compensation. When the difference is positive, the spindle speed compensation amount is recorded as a decrease. When the difference is negative, the spindle speed compensation amount is recorded as an increase. When the difference is 0, the spindle speed compensation amount is recorded as a maintenance amount. The spindle speed compensation amounts are arranged in chronological order.

[0132] Subtract the change in winding rhythm at the previous time position from the change in winding rhythm at the next time position to obtain the difference in the trend of winding rhythm change at adjacent time positions.

[0133] Subtract the workshop temperature value from the previous time position to obtain the temperature change difference. Subtract the workshop humidity value from the previous time position to obtain the humidity change difference. Add the temperature change difference and the humidity change difference and divide by 2 to obtain the change difference of workshop temperature and humidity data between adjacent time positions.

[0134] The difference in the trend of winding cycle at adjacent time positions is added to the difference in the temperature and humidity data at adjacent time positions to obtain the environmental regulation baseline. The environmental regulation baseline is then mapped to the environmental regulation direction. When the environmental regulation baseline is positive, the environmental regulation amount is recorded as a decrease. When the environmental regulation baseline is negative, the environmental regulation amount is recorded as an increase. When the environmental regulation baseline is 0, the environmental regulation amount is recorded as a maintenance amount. The environmental regulation amounts are then arranged in chronological order.

[0135] The roller differential speed adjustment, spindle speed compensation, and environmental adjustment are matched one-to-one according to the same time position, so that there is a set of roller differential speed adjustment, spindle speed compensation, and environmental adjustment at the same time position.

[0136] When the directions of change of roller differential speed adjustment, spindle speed compensation, and environmental adjustment are consistent, these three adjustments are directly combined sequentially. When the directions of change of roller differential speed adjustment, spindle speed compensation, and environmental adjustment are inconsistent, the direction of change of roller differential speed adjustment corresponding to the tension change trend in the drafting zone is maintained, and the direction of change of spindle speed compensation corresponding to the twisting stability change trend is maintained. The direction of environmental adjustment is adjusted according to the positive or negative relationship between the change trend of winding cycle and the difference in workshop temperature and humidity data. Then, the roller differential speed adjustment, spindle speed compensation, and environmental adjustment are combined sequentially to form a linkage control quantity.

[0137] In this embodiment, S7 specifically refers to:

[0138] The roller differential speed adjustment corresponding to each time position is applied to the roller transmission parameter position of the spinning machine, the spindle speed compensation corresponding to each time position is applied to the spindle operation parameter position of the spinning machine, and the environmental adjustment corresponding to each time position is applied to the environmental parameter position.

[0139] After loading is completed at each time point, data on tension in the drafting zone of the spinning machine, roller differential speed, spindle speed fluctuation, and workshop temperature and humidity are collected at the corresponding time point to form the adjusted operating data.

[0140] According to the time sequence in the adjusted operating data, the tension in the drafting zone of the spinning machine is written into the drafting station parameter position, the roller differential speed is written into the roller transmission parameter position, the spindle speed fluctuation is written into the spindle operation parameter position, and the workshop temperature and humidity data is written into the environmental parameter position to obtain the updated virtual operating status.

[0141] The updated virtual operating states are arranged sequentially according to time, and synchronized with the corresponding time and position data in the adjusted operating data to update the virtual and real operating conditions.

[0142] In this embodiment, the intelligent production control system for textile yarn based on digital twins includes the following modules:

[0143] The data acquisition module is used to acquire the operating data of the spinning machine, perform time scale alignment and working condition correction, and form spinning working condition quantities;

[0144] The twin modeling module is used to establish a virtual operating body of the spinning machine in the digital twin space based on the spinning condition parameters, and to load the spinning condition parameters into the virtual operating body of the spinning machine hourly to form a virtual and real operating condition state.

[0145] The state generation module is used to improve the Reformer model by inputting virtual and real working conditions. It introduces a yarn-state neighborhood resonance regularization mechanism in the LSH attention layer to construct resonance indicators. It performs neighborhood rearrangement and cross-bucket adjacency splicing on the hash bucketing results generated by the LSH attention layer to generate continuous state variables.

[0146] The component generation module is used to apply the Hilbert-Huang transform algorithm to continuous state variables, introduce a critical precursor stationary phase separation mechanism, obtain the main mode through empirical mode decomposition, construct the stationary phase criterion quantity, identify the stationary phase region from the main mode and separate the weak oscillation component to form multi-scale operating condition components.

[0147] The trajectory construction module is used to calculate the trends of tension change in the drafting zone, twisting stability change, and winding rhythm change based on multi-scale working condition components, thereby forming a spinning working condition trajectory.

[0148] The control generation module is used to calculate the roller differential speed adjustment, spindle speed compensation and environmental adjustment based on the spinning working condition trajectory, perform linkage matching and generate linkage control quantities.

[0149] The feedback update module is used to apply the linkage control amount to the spinning machine, generate the adjusted operating data, and update the virtual and real operating conditions.

[0150] Example 1: To verify the feasibility of this invention in practice, it was applied to 12 spinning machines in the spinning workshop of a cotton spinning enterprise. Six machines served as the test unit using the method of this invention, while six machines served as the control unit using the traditional method. Production conditions were kept consistent, and the yarn specification was 18.5 tex pure cotton yarn.

[0151] During the experiment, traditional methods mainly relied on operators to manually judge based on tension in the drafting zone, yarn breakage, and winding status, and to adjust roller differential speed and spindle speed. Workshop temperature and humidity were adjusted at fixed time intervals. The method of this invention, however, acquires the operating data of the spinning machine to form spinning condition quantities. A virtual operating entity of the spinning machine is established in a digital twin space to form a virtual-real operating state. This virtual-real operating state is input into an improved Reformer model to generate continuous state quantities. Multi-scale operating condition components are obtained through the Hilbert-Huang transform algorithm, further forming the spinning condition trajectory. Roller differential speed adjustment, spindle speed compensation, and environmental adjustment are calculated. Linkage matching generates linkage control quantities to achieve dynamic control. The overall operating results are shown in Table 1.

[0152] Table 1. Comparison of Production Performance between the Method of the Invention and Traditional Methods

[0153] Average breakage rate (times / thousand spindles·hour) 18.7 11.9 36.36% CV value of stem (%) 13.4 11.1 17.16% Tension fluctuation amplitude in the stretching zone (cN) 22.6 14.8 34.51% Spindle rotation speed fluctuation range (r / min) 312 201 35.58% Winding cycle time deviation (ms) 86 49 43.02% Rolla differential average settling time (s) 15.4 3.8 75.32% Electricity consumption per 100 kilograms (kWh) 52.8 49.6 6.06%

[0154] As shown in Table 1, the method of this invention outperforms traditional methods in several key production indicators. The significantly reduced yarn breakage rate indicates that the spinning process trajectory can identify abnormal changes in advance and output corresponding control parameters; the reduced yarn evenness CV value indicates a more stable synergistic relationship between the tension in the drafting zone and the roller differential speed; the simultaneous decrease in winding cycle deviation and spindle speed fluctuation demonstrates that multi-scale process components can effectively characterize the coupled changes in the twisting and winding processes; the significantly shortened roller differential speed adjustment time reflects the rapid response capability of the digital twin space and the improved Reformer model to changes in process conditions; the increased first-grade yield and reduced energy consumption further illustrate the improvement in both production quality and economic efficiency.

[0155] Table 2 Comparison of Stability under Average Operating Conditions

[0156] Average tension (cN) 128.5 127.9 Tension fluctuation amplitude (cN) 22.6 14.8 Average spindle speed (r / min) 15450 15449 Spindle speed fluctuation range (r / min) 312 201 Average relative humidity (%) 60.4 63.8 Winding cycle time deviation (ms) 86 49 Breakage rate (times / thousand spindles·hour) 18.7 11.9 First-class product rate (%) 92.1 96.4

[0157] As shown in Table 2, the method of the present invention significantly reduces the amplitude of tension fluctuation and spindle speed fluctuation while maintaining the average tension and average spindle speed at a basically consistent level. This indicates that continuous state variables and multi-scale working condition components can effectively constrain working condition fluctuations. The average relative humidity increases and remains stable, corresponding to a significant decrease in winding cycle deviation, indicating an effective linkage between environmental adjustment and the trend of winding cycle change. The breakage rate continues to decrease and the first-grade product rate significantly increases, indicating that by coordinating the control of roller differential speed adjustment, spindle speed compensation, and environmental adjustment through linkage control, the operating status of the spinning frame can be maintained stably under multi-parameter coupling conditions. Overall, the results verify that the method of the present invention has higher stability and better production quality control capabilities under complex working conditions.

[0158] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent production control of textile yarn based on digital twins, characterized in that, Includes the following steps: S1. Obtain the operating data of the spinning machine, perform time scale alignment and working condition adjustment to form spinning working condition quantities; S2. Based on the spinning condition parameters, a virtual operating body of the spinning machine is established in the digital twin space, and the spinning condition parameters are loaded into the virtual operating body of the spinning machine hourly to form a virtual and real operating condition state. S3. Input the virtual and real working conditions into the improved Reformer model, introduce the yarn-state neighborhood resonance regularization mechanism in the LSH attention layer, construct the resonance indicator, and perform neighborhood rearrangement and cross-bucket adjacency splicing on the hash bucketing results generated by the LSH attention layer to generate continuous state variables. S4. The Hilbert-Huang transform algorithm is used for continuous state variables. A critical precursor stationary phase separation mechanism is introduced. The main mode is obtained through empirical mode decomposition. The stationary phase criterion is constructed. The stationary phase region is identified from the main mode and weak oscillation components are separated to form multi-scale operating condition components. S5. Based on the multi-scale working condition components, calculate the tension change trend in the drafting zone, the twisting stability change trend, and the winding rhythm change trend to form a spinning working condition trajectory. S6. Based on the spinning condition trajectory, calculate the roller differential speed adjustment, spindle speed compensation and environmental adjustment respectively, perform linkage matching, and generate linkage control quantity; S7. Apply the linkage control amount to the spinning machine, generate the controlled operating data, and update the virtual and real operating conditions.

2. The intelligent production control method for textile yarn based on digital twins according to claim 1, characterized in that, Specifically, S1 is: Collect data on tension in the drafting zone of the spinning machine, roller differential speed, spindle speed fluctuation, and workshop temperature and humidity, and add time stamps for the corresponding collection times. According to the unified sampling interval, the tension, roller differential speed, spindle speed fluctuation and workshop temperature and humidity data of the spinning machine drafting zone after the additional time mark are resampled to obtain tension data, differential speed data, speed data and temperature and humidity data with consistent time position; Tension data, differential speed data, rotational speed data, and temperature and humidity data with consistent time and location are paired and organized according to the same moment to obtain the working condition sampling group; The operating condition sampling groups are sequentially arranged according to time sequence to form spinning operating condition quantities.

3. The intelligent production control method for textile yarn based on digital twins according to claim 1, characterized in that, The establishment of a virtual operating entity for the spinning machine in the digital twin space based on spinning conditions specifically involves: According to the correspondence of the physical structure of the spinning machine, the tension in the drafting zone of the spinning machine is correlated with the position of the drafting station parameters, the roller differential speed is correlated with the position of the roller transmission parameters, the spindle speed fluctuation is correlated with the position of the spindle operation parameters, and the workshop temperature and humidity data are correlated with the position of the environmental parameters. A framework for the operating parameters of a spinning frame is constructed based on the positions of the drafting station parameters, roller drive parameters, spindle operation parameters, and environmental parameters. The tension in the drafting zone of the spinning machine, roller differential speed, spindle speed fluctuation, and workshop temperature and humidity data are written into the corresponding parameter positions in the spinning machine operation parameter framework to form a virtual operating body of the spinning machine.

4. The intelligent production control method for textile yarn based on digital twins according to claim 1, characterized in that, The step of loading spinning condition parameters hourly into the virtual operating entity of the spinning machine to form a virtual-real condition state specifically involves: According to the time sequence in the spinning condition data, extract the tension in the drafting zone of the spinning machine, roller differential speed, spindle speed fluctuation and workshop temperature and humidity data corresponding to each time position; Write the tension in the drafting zone of the spinning machine corresponding to each time position into the drafting station parameter position in the virtual operating body of the spinning machine; write the roller differential speed corresponding to each time position into the roller transmission parameter position in the virtual operating body of the spinning machine; write the spindle speed fluctuation corresponding to each time position into the spindle operation parameter position in the virtual operating body of the spinning machine; and write the workshop temperature and humidity data corresponding to each time position into the environmental parameter position in the virtual operating body of the spinning machine. After each write operation, the virtual operating status at the corresponding time position is obtained; the virtual operating status is continuously arranged in chronological order and synchronized with the corresponding time position data in the spinning condition data to form a virtual and real operating status.

5. The intelligent production control method for textile yarn based on digital twins according to claim 1, characterized in that, The improved Reformer model specifically includes a sequence embedding layer, an LSH attention layer, an invertible residual layer, and a block feedforward layer; The sequence embedding layer arranges the tension in the drafting zone of the spinning machine, roller differential speed, spindle speed fluctuation, and workshop temperature and humidity data in the virtual and real working conditions in chronological order, and combines the tension in the drafting zone of the spinning machine, roller differential speed, spindle speed fluctuation, and workshop temperature and humidity data at the corresponding time position into a state vector. The LSH attention layer converts the state vector into a query vector and a key vector, respectively, and divides the state vector into different hash buckets based on the random hash values ​​of the query vector and the key vector, forming a hash bucketing result; After generating the hash binning results, a yarn-state neighborhood resonance regularization mechanism is introduced. Resonance indicators are constructed based on the differences in drafting zone tension, roller speed, spindle speed, and temperature and humidity at adjacent time positions. The state vectors within the same hash bin are reordered according to the direction of change of the resonance indicators, and state vectors with continuous resonance indicators at the boundaries of adjacent hash bins are connected as adjacent segments. The attention results within the bin are calculated based on the reordered state vectors and adjacent segments, and the attention results within the bin are restored and arranged according to time position to form the neighborhood-regularized state vectors. The reversible residual layer splits the neighborhood-normalized state vector into a first sub-vector and a second sub-vector, and inputs the first sub-vector and the second sub-vector into the reversible residual unit respectively. An intermediate state vector is formed by adding the vectors of the front and back layers. The segmented feedforward layer divides the intermediate state vector into segments according to a preset length, inputs each segment of the intermediate state vector into the feedforward network in sequence, and splices each segment of the feedforward output according to the original time position to form a continuous state quantity.

6. The intelligent production control method for textile yarn based on digital twins according to claim 1, characterized in that, Specifically, S4 is: For each segment of the state curve in the continuous state variable, extract the local maximum and local minimum points respectively. Construct an upper envelope based on the local maximum points and a lower envelope based on the local minimum points. Calculate the local mean curve based on the upper and lower envelopes. Subtract the corresponding local mean curve from each segment of the state curve in the continuous state variables to obtain the screening result. Repeat the extraction of local maxima, local minima, upper envelope, lower envelope, and local mean curves from the screening result until the main mode that satisfies the empirical mode decomposition conditions is obtained. The empirical mode decomposition conditions include that, within the same time interval, the difference between the sum of the number of local maxima and the number of local minima and the number of zero crossings does not exceed 1, and the local mean curves calculated from the upper and lower envelopes are close to 0 within the time interval, and the amplitude range enclosed by the upper and lower envelopes is symmetrically distributed. Perform Hilbert transform on the continuous state variables to obtain the phase and envelope values ​​at the corresponding time positions, and construct the stationary phase criterion based on the phase difference and envelope difference between adjacent time positions; The stationary phase region is formed by determining the time interval in the main mode where the phase change slows down and the envelope change is concentrated according to the stationary phase criterion; the local oscillation component is separated along the fluctuation direction of the stationary phase region in the main mode to obtain the weak oscillation component. The remaining components and weak oscillatory components in the main mode are retained respectively, and multi-scale working condition components are formed according to the time position correspondence.

7. The intelligent production control method for textile yarn based on digital twins according to claim 1, characterized in that, Specifically, S5 is: Based on the time-location correspondence, the tension-related components, twisting-related components, and winding-related components in the multi-scale working condition components are extracted; The tension change in the stretching zone is constructed based on the numerical difference of the tension-related components at adjacent time positions, and the tension change in the stretching zone is arranged in chronological order to form the tension change trend in the stretching zone. The twisting stability variation is constructed based on the difference in fluctuation amplitude and fluctuation interval of the twisting-related components at adjacent time positions, and the twisting stability variation is arranged in chronological order to form the twisting stability variation trend. The winding beat variation is constructed based on the beat interval difference between adjacent time positions of the winding-related components, and the winding beat variation is arranged in chronological order to form the winding beat variation trend. The trends of tension change in the drafting zone, twisting stability change, and winding rhythm change are arranged according to the same time position to form a spinning condition trajectory.

8. The intelligent production control method for textile yarn based on digital twins according to claim 1, characterized in that, Specifically, S6 is: Extract the trends of tension variation in the drafting zone, twisting stability variation, and winding rhythm variation in the spinning process trajectory based on the time-location correspondence. The roller differential speed adjustment is constructed based on the difference in the change trend of tension in the stretching zone at adjacent time positions, and the roller differential speed adjustment is arranged in chronological order. Spindle speed compensation is constructed based on the difference in the change trend of twisting stability at adjacent time positions, and the spindle speed compensation is arranged in chronological order. Environmental regulation quantities are constructed based on the difference in the change trend of winding cycle at adjacent time positions and the difference in the change of workshop temperature and humidity data at adjacent time positions, and the environmental regulation quantities are arranged in chronological order. The roller differential speed adjustment, spindle speed compensation, and environmental adjustment are organized according to the same time position; and the linkage matching is performed according to the combination relationship of the roller differential speed adjustment, spindle speed compensation, and environmental adjustment at the same time position to form a linkage control quantity.

9. The intelligent production control method for textile yarn based on digital twins according to claim 1, characterized in that, Specifically, S7 is: The roller differential speed adjustment corresponding to each time position is applied to the roller transmission parameter position of the spinning machine, the spindle speed compensation corresponding to each time position is applied to the spindle operation parameter position of the spinning machine, and the environmental adjustment corresponding to each time position is applied to the environmental parameter position. After loading is completed at each time point, data on tension in the drafting zone of the spinning machine, roller differential speed, spindle speed fluctuation, and workshop temperature and humidity are collected at the corresponding time point to form the adjusted operating data. According to the time sequence in the adjusted operating data, the tension in the drafting zone of the spinning machine is written into the drafting station parameter position, the roller differential speed is written into the roller transmission parameter position, the spindle speed fluctuation is written into the spindle operation parameter position, and the workshop temperature and humidity data is written into the environmental parameter position to obtain the updated virtual operating status. The updated virtual operating states are arranged sequentially according to time, and synchronized with the corresponding time and position data in the adjusted operating data to update the virtual and real operating conditions.

10. A digital twin-based intelligent production control system for textile yarn, executing the digital twin-based intelligent production control method for textile yarn as described in any one of claims 1 to 9, characterized in that, Includes the following modules: The data acquisition module is used to acquire the operating data of the spinning machine, perform time scale alignment and working condition correction, and form spinning working condition quantities; The twin modeling module is used to establish a virtual operating body of the spinning machine in the digital twin space based on the spinning condition parameters, and to load the spinning condition parameters into the virtual operating body of the spinning machine hourly to form a virtual and real operating condition state. The state generation module is used to input the virtual and real working conditions into the improved Reformer model, introduce a yarn-state neighborhood resonance regularization mechanism in the LSH attention layer, construct a resonance indicator, and perform neighborhood rearrangement and cross-bucket adjacency splicing on the hash bucketing results generated by the LSH attention layer to generate continuous state variables. The component generation module is used to apply the Hilbert-Huang transform algorithm to continuous state variables, introduce a critical precursor stationary phase separation mechanism, obtain the main mode through empirical mode decomposition, construct the stationary phase criterion quantity, identify the stationary phase region from the main mode and separate the weak oscillation component to form multi-scale operating condition components. The trajectory construction module is used to calculate the tension change trend of the drafting zone, the twisting stability change trend, and the winding rhythm change trend based on the multi-scale working condition components, and form a spinning working condition trajectory. The control generation module is used to calculate the roller differential speed adjustment, spindle speed compensation and environmental adjustment based on the spinning working condition trajectory, perform linkage matching and generate linkage control quantities; The feedback update module is used to apply the linkage control amount to the spinning machine, generate the adjusted operating data, and update the virtual and real operating conditions.

Images