Virtual simulation method for glass fiber production line based on digital twinning

Abstract

The present application relates to the technical field of virtual simulation, in particular to a glass fiber production line virtual simulation method based on digital twinning, in the present application, multi-dimensional parameters such as equipment operation fluctuation and fiber morphology are extracted at the physical transfer boundary of key manufacturing links such as drawing, drying and winding to construct a global series data set, and in the digital twinning virtual space, combined with technical means such as nonlinear cross mapping rules and linear evolution matrix, the process state characteristics of future time nodes are deduced, the initial process disturbance experienced time delay, amplitude amplification and attenuation evolution strength during the transmission between different processes are accurately quantified and tracked. At the same time, according to the process difference data inversely restored in the prediction space, the product quality abnormal risk is judged in advance, and the control voltage command for the heating power of the bushing and the exhaust speed of the drying machine is generated directly according to the disturbance cross-process evolution characteristics.

Description

Technical Field

[0001] This invention relates to the field of virtual simulation technology, and in particular to a virtual simulation method for a glass fiber production line based on digital twins. Background Technology

[0002] The field of virtual simulation technology mainly involves using computer hardware and software to construct digital models that can reflect the operating laws of the real physical world, and allowing relevant personnel to interact and practice in the constructed virtual environment by inputting various parameters.

[0003] The traditional virtual simulation method for glass fiber production lines refers to establishing corresponding computer basic models for manufacturing processes such as glass fiber drawing, drying, and winding. It is mainly used to simulate the operating status of production equipment and the flow of materials before the actual production line is built or the process is adjusted, so as to discover physical interference or collision or process design defects in the production line layout or process parameters in advance.

[0004] Traditional virtual simulation methods for glass fiber production lines, while establishing corresponding computer-based models for the manufacturing process to simulate the operating status of production equipment and the flow of materials, aim to identify physical interference or collisions or process design defects in the production line layout or process parameters in advance. However, in actual continuous production, when process parameters fluctuate dynamically and local minor process disturbances spread across physical interfaces, there are problems such as difficulty in accurately tracking the nonlinear evolution trend of process deviations and the inability to implement proactive closed-loop intervention and control for product quality deterioration. Summary of the Invention

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a virtual simulation method for a glass fiber production line based on digital twins, comprising the following steps: S1: Collect the first process boundary state packet of the fiber drawing equipment during the production of glass fiber; S2: Collect the second process boundary state packet of the drying equipment and the winding tension change value during the process of glass fiber entering the winding equipment, combine the first process boundary state packet, and construct a global serial data set; S3: Predict the state of the global serial data set at the next time node to obtain the predicted global process state evolution feature set; S4: Based on the predicted global process state evolution feature set, analyze the cross-process evolution characteristics of future disturbances in the production line; S5: Based on the predicted global process state evolution feature set, determine the future glass fiber product quality of the production line, and generate control instructions by combining the cross-process evolution features of disturbances, thereby generating the virtual simulation control results of the glass fiber production line.

[0006] As a further aspect of the present invention, the first process boundary state package includes initial operating state parameters of the drawing process, single filament dimension parameters of the drawing stage junction boundary, and moisture content dimension parameters of the drawing stage junction boundary. The second process boundary state package includes temperature change state parameters of the drying process, single filament dimension parameters of the drying stage junction boundary, and moisture content dimension parameters of the drying stage junction boundary. The global serial data set includes a sequence of operating features of the drawing equipment, a sequence of operating features of the drying equipment, and a sequence of operating features of the winding equipment. The predicted global process state evolution feature set includes polynomial nonlinear mapping feature terms for future time nodes and state matrix evolution calculation feature terms for future time nodes. The predicted cross-process disturbance evolution features include cross-process disturbance propagation delay feature data, cross-process disturbance evolution amplification ratio data, and cross-process disturbance evolution attenuation ratio data. The virtual simulation control results of the glass fiber production line include spinneret heating power voltage control command parameters and dryer exhaust fan speed voltage control command parameters.

[0007] As a further aspect of the present invention, step S1 specifically comprises: S101: During the production of glass fiber in the drawing equipment, the values ​​of spinneret temperature disturbance, drawing speed change, and sizing agent coating amount change are collected, aligned and merged according to the time stamp order to generate an initial process state vector. S102: Monitor and characterize the position coordinates of the glass fiber bundle. When the position coordinates of the glass fiber bundle reach the first junction moment of the physical junction between the drawing and drying stages, collect the single filament diameter range value and the moisture content value at the first junction moment to generate the drawing junction boundary characteristic parameters. S103: Combine and encapsulate the wire drawing junction boundary feature parameters and the initial process state vector to generate the first process boundary state package.

[0008] As a further aspect of the present invention, step S2 specifically comprises: S201: When the position coordinates of the glass fiber bundle reach the second extended junction moment of the physical junction boundary between the drying stage and the winding stage, collect the single filament diameter range value and the moisture content value at the second extended junction moment to generate the drying junction boundary characteristic parameters. S202: Collect the drying temperature change value of the drying equipment, combine it with the drying handover boundary feature parameters, and integrate it into the second process boundary state package; S203: During the process of glass fiber entering the winding equipment, the value of the change in winding tension is collected, and the first process boundary state packet, the second process boundary state packet and the value of the change in winding tension are spliced ​​together to generate a global serial data set.

[0009] As a further aspect of the present invention, step S3 specifically comprises: S301: In the digital twin virtual space, the global concatenated dataset is used as an evolutionary inference benchmark to generate state observations; S302: Normalize the state observations to a unified numerical range, calculate the polynomial combination features of the normalized state observations using preset nonlinear rules, and generate a global process state evolution feature set. S303: Obtain the linear evolution matrix of the initial calibration stage, calculate the state evolution product between the global process state evolution feature set and the linear evolution matrix, and obtain the predicted global process state evolution feature set for the next time node.

[0010] As a further aspect of the present invention, step S4 specifically comprises: S401: Extract the state offset value caused by the leakage plate temperature disturbance value from the predicted global process state evolution feature set of multiple consecutive time nodes, as well as the feature timestamps representing the feature intervals corresponding to the first process boundary state package and the second process boundary state package, and calculate the time delay value experienced between the feature timestamp data when the state offset value propagates across processes over time. S402: Calculate the offset ratio between the state offset value after propagation and the state offset value before propagation; S403: When the offset ratio is greater than the preset first reference ratio, the corresponding offset ratio is determined to be the amplification value in the amplification state. When the offset ratio is less than the preset second reference ratio, the corresponding offset ratio is determined to be the attenuation value in the attenuation state. The time delay value, amplification value or attenuation value are integrated to generate the predicted disturbance cross-process evolution characteristics of future time nodes.

[0011] As a further aspect of the present invention, step S5 specifically comprises: S501: Calculate the inverse algebraic reconstruction features of the predicted global process state evolution feature set to generate a process prediction numerical set; S502: Based on the process prediction value set, obtain the difference value of uneven moisture content in the drying section, the abnormal deviation value of the winding density in the winding section, and the difference value of the end face forming deviation in the drawing section, and perform normalization processing on each to generate scalar values ​​of quality indicators. S503: Calculate the difference between the scalar value of the quality indicator and the preset percentage threshold of qualified products, obtain the absolute value of the difference, filter the absolute values ​​of the difference that are greater than zero, generate the predicted product quality anomaly judgment result for future time nodes, and generate control voltage value instructions to adjust the heating power of the spindle and the speed of the exhaust fan of the dryer based on the time delay value, the predicted product quality anomaly judgment result and the predicted cross-process evolution characteristics of the disturbance. Issue control voltage value instructions to adjust the drawing parameters and drying parameters, and generate the virtual simulation control result of the glass fiber production line.

[0012] Compared with the prior art, the advantages and positive effects of the present invention are as follows: This invention extracts multi-dimensional parameters such as equipment operation fluctuations and fiber morphology at the physical boundaries of key manufacturing processes such as drawing, drying, and winding to construct a globally serialized data set. In a digital twin virtual space, it combines nonlinear cross-mapping rules and linear evolution matrices to extrapolate the process state characteristics at future time points. This enables precise quantification and tracking of the time delay, amplitude amplification, and attenuation intensity experienced by initial process disturbances as they propagate between different processes. Simultaneously, based on the process difference data reconstructed from the predicted space, this invention proactively determines the risk of product quality anomalies and directly generates control voltage commands for the spindle heating power and dryer exhaust speed by linking the cross-process evolution characteristics of disturbances. This serves to perform nonlinear overdamping strong intervention compensation before physical quality defects actually form, completely establishing a data closed loop from virtual dynamic extrapolation to physical equipment parameter correction. This significantly improves the stability of the glass fiber production line's process anti-interference and the yield rate of the final product. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention. Detailed Implementation

[0015] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0016] Please see Figure 1 This invention provides a virtual simulation method for a glass fiber production line based on digital twins, comprising the following steps: S1: Collect the first process boundary state packet of the fiber drawing equipment during the production of glass fiber; S2: Collect the second process boundary state package of the drying equipment and the winding tension change value during the process of glass fiber entering the winding equipment, combine the first process boundary state package, and construct a global serial data set; S3: Predict the state of the global serial data set at the next time node to obtain the predicted global process state evolution feature set; S4: Based on the predicted global process state evolution feature set, analyze the cross-process evolution characteristics of future disturbances in the production line; S5: Based on the predicted global process state evolution feature set, determine the future glass fiber product quality of the production line, and generate control instructions by combining the cross-process evolution features of disturbances, thus generating the virtual simulation control results of the glass fiber production line.

[0017] The first process boundary state package includes the initial operating state parameters of the drawing process, the single filament dimension parameters of the drawing process junction boundary, and the moisture content dimension parameters of the drawing process junction boundary. The second process boundary state package includes the temperature change state parameters of the drying process, the single filament dimension parameters of the drying process junction boundary, and the moisture content dimension parameters of the drying process junction boundary. The global serial data set includes the operating feature slice sequence of the drawing equipment, the operating feature slice sequence of the drying equipment, and the operating feature slice sequence of the winding equipment. The predicted global process state evolution feature set includes the polynomial nonlinear mapping feature term of the future time node and the state matrix evolution calculation feature term of the future time node. The predicted cross-process disturbance evolution features include cross-process disturbance propagation delay feature data, cross-process disturbance evolution amplification ratio data, and cross-process disturbance evolution attenuation ratio data. The virtual simulation control results of the glass fiber production line include the spindle heating power voltage control command parameters and the dryer exhaust fan speed voltage control command parameters.

[0018] Please see Figure 2 Step S1 is as follows: S101: During the production of glass fiber in the drawing equipment, the values ​​of spinneret temperature disturbance, drawing speed change, and sizing agent coating amount change are collected, aligned and merged according to the time stamp order to generate an initial process state vector. In the process of producing glass fiber in a glass drawing machine, data acquisition and time series alignment and merging of multi-dimensional dynamic process parameters are performed. High-temperature resistant platinum-rhodium alloy temperature sensing probes are deployed at the physical location of the discharge port at the bottom of the glass furnace to continuously read the absolute temperature induced electrical signal and convert it into a real-time absolute temperature value reflecting thermal fluctuations. A sliding window data denoising, smoothing, and filtering process is performed on the received continuous real-time absolute temperature value sequence. A statistical sliding window containing 50 consecutive temperature sampling points is set. At each time step, all real-time absolute temperature values ​​covered within the statistical sliding window are extracted and sorted in descending order of value. The top three highest and bottom three lowest extreme temperature sampling points in the sequence are directly removed. The remaining 44 intermediate sampling temperature values ​​are arithmetically summed, and the sum is divided by the total number of valid sampling points (44) to calculate a smoothed reference temperature value representing the macroscopic trend at the current time point. The latest real-time absolute temperature value is extracted and a difference comparison calculation is performed. The deviation value derived by subtracting the smoothed reference temperature value from the real-time absolute temperature value is directly defined as the temperature disturbance value of the acquired stencil. Based on the actual working conditions, the real-time absolute temperature value of 1250.5 degrees Celsius is subtracted from the smoothed reference temperature value of 1250.0 degrees Celsius to calculate the stencil temperature disturbance value of 0.5 degrees Celsius. Simultaneously, the real-time rotational angular velocity parameter of the spindle is obtained through an incremental photoelectric encoder installed at the position of the wire drawing and winding spindle. The real-time rotational angular velocity parameter is multiplied and converted with the pre-calibrated physical outer circumference value of the winding spindle. The target standard wire drawing speed setting value given in the process standard formula data is retrieved. The target standard wire drawing speed setting value is subtracted from the real-time linear velocity value to derive the acquired wire drawing speed change value. The calculated real-time linear velocity value of 25.4 m / s is subtracted from the target standard wire drawing speed setting value of 25.0 m / s to obtain the wire drawing speed change value of 0.4 m / s. Simultaneously, the mass flow rate parameter of the sizing agent consumed, output by the Coriolis mass flow rate sensing node, is obtained at the sizing agent coating tank. This parameter is divided by the real-time mass flow rate parameter of the glass fiber pulled out using the stretching formula at the current time node to calculate the true sizing agent adhesion ratio per unit mass fiber. The true sizing agent adhesion ratio is then subtracted from the baseline coating ratio required by the process to calculate the change in the amount of sizing agent coated. Finally, the measured true sizing agent adhesion ratio of 7.5 g / kg is subtracted from the baseline coating ratio of 7.0 g / kg to obtain the change in the amount of sizing agent coated of 0.5 g / kg.For the obtained values ​​of 0.5 degrees Celsius stencil temperature disturbance, 0.4 meters per second wire drawing speed change, and 0.5 grams per kilogram wetting agent coating amount change, an alignment and merging operation is performed according to the time stamp order. A globally unified 5-millisecond reference time grid is set, and the nearest actual sampling points and historical data of each parameter data before and after the grid nodes are found. The time span ratio weight of the distance from the previous sampling point is calculated, and the difference between the preceding and following values ​​is multiplied by the ratio weight and added to the preceding value to generate synchronous interpolation alignment data. The time-aligned parameter indicators are spliced ​​to generate the initial process state vector.

[0019] S102: Monitor and characterize the position coordinates of the glass fiber bundle. When the position coordinates of the glass fiber bundle reach the first junction moment of the physical junction between the drawing and drying stages, collect the single filament diameter range value and the moisture content value at the first junction moment to generate the drawing junction boundary characteristic parameters. The real-time linear velocity value obtained from the previous permutation calculation is continuously accumulated through position integration to monitor and characterize the position coordinates of the glass fiber bundle. Within each 5-millisecond reference time grid, the current real-time linear velocity value is multiplied by a time span of 0.005 seconds to calculate the single minute displacement value. All single minute displacement values ​​generated from the starting point of drawing to the current tracking node are continuously summed to calculate the absolute position coordinates of the current glass fiber bundle in its three-dimensional physical path. The position coordinate reference value of the physical boundary between the drawing and drying stages, distance from the starting point of drawing, is read. During continuous position coordinate monitoring, the real-time accumulated absolute position coordinate value is continuously compared with the 8.5-meter position coordinate reference value. When the absolute position coordinate value is first greater than or equal to the 8.5-meter position coordinate reference value, the currently recorded global absolute timestamp is extracted and directly defined as the first intersection moment when the glass fiber bundle's position coordinates reach the physical boundary between the drawing and drying stages. When the trigger condition is met at the first handover moment, a detection command is issued to the dual-axis cross-laser scanning detection node arranged at the physical handover boundary. Within a 100-millisecond detection time window, the cross-sectional outer diameter values ​​of a large number of glass fiber monofilaments within the light curtain are continuously captured. The 1000 captured outer diameter values ​​are quickly sorted to rearrange them in ascending order. The value at the 50th percentile of the sorted sequence is extracted as the lower limit value of the diameter of this batch of filament bundles, and the value at the 950th percentile is extracted as the upper limit value of the diameter. The extracted lower limit value of 12.8 micrometers and the upper limit value of 13.5 micrometers are bound to each other as endpoint data. The combined acquisition of the monofilament diameter range at the first handover moment is 12.8 micrometers to 13.5 micrometers. The synchronously controlled microwave resonance moisture sensing node emits a constant-frequency high-frequency microwave signal to the passing glass fiber bundle. It receives and collects the microwave signal attenuation amplitude data after penetrating the fiber. It retrieves the linear proportional mapping coefficient between the microwave signal attenuation amplitude data and the moisture content, calibrated based on drying experiments. The real-time collected microwave signal attenuation amplitude data is multiplied by this mapping coefficient to calculate the percentage of liquid moisture mass inside and on the surface of the fiber bundle. The resulting percentage of 11.2% is collected as the moisture content value. Following the structural arrangement of the preceding monofilament diameter range and the subsequent moisture content value, the monofilament diameter range values ​​at the first junction (12.8 μm to 13.5 μm) are merged and spliced ​​with the 11.2% moisture content value to generate the fiber drawing junction boundary characteristic parameters.

[0020] S103: Combine the feature parameters of the encapsulation wire drawing junction boundary and the initial process state vector to generate the first process boundary state package; A joint packaging process is performed on the previously independently generated cross-domain feature parameter dataset, combining and encapsulating the drawing junction boundary feature parameters and the initial process state vector. An initial process state vector, generated by strictly aligning and merging timestamps, is extracted. This vector contains values ​​for 0.5°C stencil temperature disturbance, 0.4 m / s drawing speed change, and 0.5 g / kg sizing agent coating change. Simultaneously, the drawing junction boundary feature parameters, representing the physical interface between the drawing and drying stages, are extracted. These parameters specifically cover the single filament diameter range of 12.8 μm to 13.5 μm and the moisture content of 11.2%. Following the forward progression of the time series, a data splicing format rule is set. The initial process state vector, recording the thermal and dynamic fluctuations at the front of the drawing section, is placed at the front of the main structure to retain the source state evolution information. Subsequently, the drawing junction boundary feature parameters, recording the dimensions of the physical interface endpoints and moisture indices, are tightly connected as a continuous data stream and appended to the end of the initial process state vector. Cyclic redundancy check is performed at the splicing point between the initial front-end state data and the boundary state data at the back-end to ensure the reliability of the continuous connection of the floating-point sequence. This merges and binds the front-end data reflecting dynamic process disturbances with the back-end features reflecting the geometric and physical state of the product at the process handover point, generating the first process boundary state package. This step achieves complete coverage of multi-dimensional parameters throughout the entire process from the wire drawing source to the first boundary handover, establishing the first stable boundary input condition for the subsequent continuous spatiotemporal evolution correlation network calculation of the entire process.

[0021] Please see Figure 3 Step S2 is as follows: S201: When the position coordinates of the glass fiber bundle reach the second extended junction moment of the physical junction boundary between the drying stage and the winding stage, collect the single filament diameter range value and the moisture content value at the second extended junction moment to generate the drying junction boundary characteristic parameters. The preceding position tracking and accumulation logic continues to operate independently, continuously superimposing the single minute displacement value calculated based on the product of real-time linear velocity and time span onto the previous absolute position coordinate value, maintaining continuous real-time monitoring of the glass fiber bundle's position coordinates after detaching from the drawing stage. The extended position coordinate reference value of the physical junction boundary between the drying and winding stages, from the starting point of the drawing stage, is extracted. During the continuous accumulation tracking and value comparison process, the real-time absolute position coordinate value is compared with the 22.4-meter extended position coordinate reference value. When the accumulated result of the absolute position coordinate value is first greater than or equal to 22.4 meters, the timestamp of the operating system in the current trigger state is captured, and this timestamp is confirmed as the second extended junction moment when the glass fiber bundle's position coordinates reach the physical junction boundary between the drying and winding stages. When the second extended junction moment is triggered, the second set of dual-axis cross-laser scanning detection nodes, arranged in the corresponding physical junction boundary area, is activated to collect the outer diameter of the monofilament cross-section after heat treatment at the dried outlet end within a fixed time window. The large number of intercepted outer diameter values ​​were sorted using the same incremental sorting operation as at the drawing end. Extreme endpoint values ​​at the 50th and 950th percentiles were extracted. Due to the long-distance heat shrinkage and drying water loss effects, the extracted and combined single filament diameter range at the second extension junction was reduced to 12.5 to 13.2 micrometers. Simultaneously, the second set of microwave resonant moisture sensing nodes deployed in the boundary region were driven to measure the microwave attenuation amplitude data penetrating this section of the dried filament bundle. This microwave attenuation amplitude data was multiplied by a pre-determined mapping coefficient to calculate the moisture percentage. After forced heat exchange and evaporation of moisture in the drying equipment, the value obtained through multiplication dropped significantly to 0.6%, which was then extracted as the moisture content value at the second extension junction. The obtained single filament diameter range values ​​of 12.5 to 13.2 micrometers at the second extension junction were sequentially spliced ​​and bound with the 0.6% moisture content value to generate the drying junction boundary feature parameters.

[0022] S202: Collect the drying temperature change values ​​of the drying equipment, combine them with the characteristic parameters of the drying handover boundary, and integrate them into the second process boundary state package; Platinum resistance temperature probes deployed in the core area of ​​the exhaust duct inside the drying equipment continuously and in real-time acquire the absolute temperature sensing analog signal of the hot airflow passing through this exhaust area. After high-precision analog-to-digital conversion, a real-time exhaust temperature value characterizing the transient heat exchange status of the drying equipment is generated. Simultaneously, the standard drying reference temperature value set in the current glass fiber production batch's standard process control document is retrieved. This real-time exhaust temperature value is subtracted from the standard drying reference temperature value to derive the deviation result. This deviation result, representing the fluctuation of environmental thermal energy, is extracted and collected as the drying temperature change value of the drying equipment. In specific operational calculations, when the standard drying reference temperature value set in the process guidance document is 150.0 degrees Celsius, and the exhaust temperature value obtained from the real-time conversion by the probe is 153.5 degrees Celsius, a drying temperature change value of 3.5 degrees Celsius is calculated by subtracting 150.0 degrees Celsius from 153.5 degrees Celsius. After calculating the value reflecting the deviation characteristics of the thermal environment, the drying handover boundary feature parameter generated by the previous operation is extracted. The drying temperature change value of 3.5 degrees Celsius is placed at the beginning of this data sequence. Combined with the data set of drying handover boundary feature parameters, which includes values ​​of 12.5 micrometers to 13.2 micrometers of single filament diameter and values ​​of 0.6% moisture content, these two types of parameters are combined by performing a data appending operation according to a specific dimension order of first environmental characteristics and then product morphology characteristics, and integrated into the second process boundary state package.

[0023] S203: During the process of glass fiber entering the winding equipment, collect the winding tension change value, splice the first process boundary state packet, the second process boundary state packet and the winding tension change value, and generate a global serial data set. During the process of the glass fiber tow completely leaving the high-temperature drying environment and entering the winding equipment for winding, strain gauge tension sensing nodes deployed at the tension buffer swing arm shaft of the winding machine are activated. These nodes continuously acquire the absolute tension values ​​of the swing arm and yarn at a high-frequency sampling rate of 200 Hz, and retrieve the optimal winding tension reference value matching the characteristics of the current tow number. The initial tension deviation value, reflecting equipment vibration and shrinkage fluctuations, is derived by subtracting the optimal winding tension reference value from the absolute tension value. A first-order low-frequency smoothing filter is applied to the initial tension deviation value to eliminate high-frequency mechanical noise interference. The initial tension deviation value at the current time point is extracted and multiplied by a preset new value retention coefficient. Similarly, the smoothed tension change value at the previous time point is extracted and multiplied by a preset historical retention coefficient. These two multiplication results are then summed to derive the winding tension change value collected after filtering out high-frequency vibrations. In this step, the new value retention coefficient is set to 0.2 and the historical retention coefficient is set to 0.8. If the current initial tension deviation value is 2.0 centinewtons and the previous node value is 1.0 centinewtons, the coefficients are substituted and multiplied to obtain 0.4 and 0.8 respectively. Adding 0.4 and 0.8, the currently determined winding tension change value is calculated to be 1.2 centinewtons. After collecting all cross-process evolution feature parameters, the first process boundary state package generated in the first segment, the second process boundary state package generated in the middle segment, and the winding tension change value derived from the above calculations are extracted. The first process boundary state package, which includes the temperature and speed variations and transition characteristics at the source of the drawing process, is arranged at the beginning of the global sequence. The second process boundary state package, which includes the influence of the heat treatment environment and the drying transition, is arranged in the middle. The 1.2 cmN winding tension change value, which reflects the dynamic mechanics of the final winding, is arranged at the end. The first process boundary state package, the second process boundary state package, and the winding tension change value are spliced ​​together in strict accordance with the physical evolution timeline to generate a global serial data set covering the fluctuations of all key process indicators from the source of drawing to the end of winding.

[0024] Please see Figure 4 Step S3 is as follows: S301: In the digital twin virtual space, global concatenated datasets are used as the basis for evolutionary inference to generate state observations; Decoupled from the underlying hardware acquisition environment, a multi-layer feature transmission network model architecture is constructed within a digital twin virtual space to simulate the dynamic evolution and cross-stage mapping of complex nonlinear processes. This multi-layer feature transmission network model consists of a data receiving input layer, multiple hidden layers responsible for deep extraction of cross-features, and a feature reconstruction output layer. A globally concatenated dataset containing all stage representation parameters is used as the evolutionary inference benchmark and imported into the data receiving input layer of this network model, guiding the multi-dimensional process data signal to be forward transmitted to the internal feature extraction hidden layers. Within the neural processing nodes of any feature extraction hidden layer, matrix multiplication and algebraic calculation logic is performed on the feature values ​​input from the previous layer network nodes. Specifically, each input feature value is multiplied by the pre-calibrated weight coefficient matrix values ​​on the corresponding node connection path. All product calculation results received by the same node are summed, and a preset bias constant value is superimposed on the sum by adding the sum. Subsequently, the accumulated sum carrying the bias constant is input into the modified linear threshold activation function to perform nonlinear activation mapping judgment processing. Its built-in mapping judgment logic is that if the accumulated sum is detected to be greater than zero, the value is output to the next layer intact; if the accumulated sum is detected to be less than or equal to zero, a truncation protection mechanism is triggered to directly output zero. After the feature weights are multiplied, nodes are summed, and nonlinear activation mapping is processed step by step in the three concatenated feature extraction hidden layers, the original one-dimensional time-series process data is derived to the state feature output layer. In the corresponding node set of the output layer, the information carrier composed of multiple floating-point feature values ​​is summarized and output, thereby generating the state observation.

[0025] S302: Normalize the state observations to a unified numerical range, calculate the polynomial combination characteristics of the normalized state observations through preset nonlinear rules, and generate a global process state evolution feature set; the nonlinear rules include multiplying the spindle temperature disturbance value and the drawing speed change value, performing square power calculation on the moisture content value and the drying temperature change value at the first handover moment, and performing high-order multiplication on the winding tension change value. The generated state observations undergo a scaling transformation to a unified data range. For each independent dimensional feature value in the state observations, the minimum and maximum historical extreme values ​​of the corresponding dimension, recorded long-term in the system database, are retrieved. The absolute excess is calculated by subtracting the corresponding minimum historical extreme value from the current dimension's state observation value, and the global range is calculated by subtracting the corresponding maximum historical extreme value from the minimum historical extreme value. The obtained absolute excess is then divided by the global range to calculate a mapping ratio between zero and one, thus normalizing the state observations to a unified numerical range. After normalization, specific dimensional parameters in the normalized state observations are retrieved using preset nonlinear rules for correlation and combination operations to calculate the polynomial combination feature of the normalized state observations. This nonlinear rule encompasses three specific data cross-calculation protocols. The first nonlinear rule involves multiplying the perturbation value of the stencil temperature and the change value of the drawing speed. These two normalized values ​​are then directly multiplied using floating-point numbers to derive the thermodynamic coupling characteristics. If the perturbation value of the stencil temperature is 0.6 and the change value of the drawing speed is 0.5, the resulting coupling characteristic is 0.30. The second nonlinear rule involves performing a power-square calculation on the moisture content value and the change value of the drying temperature at the first transition moment. The operation sequence is as follows: first, the normalized moisture content value and the normalized drying temperature change value are multiplied to obtain a preliminary intermediate result. Then, this preliminary intermediate result is multiplied again with itself to obtain a square value for deriving the moisture evaporation potential. If the input values ​​are 0.4 and 0.8, the preliminary multiplication yields 0.32, and multiplying 0.32 by itself and squaring it yields 0.1024. The third nonlinear rule involves performing a high-order multiplication of the yarn tension change values. This involves sequentially extracting the normalized yarn tension change values ​​from the current tracking time node and the two preceding consecutive historical time nodes, multiplying them to derive the cumulative tension fluctuation characteristics. If the values ​​for the three consecutive time nodes are 0.5, 0.6, and 0.7, the high-order multiplication yields a result of 0.21. The results derived through multiplication, power-law calculations, and high-order multiplication are then combined with other normalized indices to generate a global set of process state evolution characteristics that reveals underlying potential relationships.

[0026] S303: Obtain the linear evolution matrix of the initial calibration stage, calculate the state evolution product between the global process state evolution feature set and the linear evolution matrix, and obtain the predicted global process state evolution feature set for the next time node; The initial calibration phase involves obtaining a linear evolution matrix established through iterative training using long-term historical data. The calibration and optimization process for the weight elements within this linear evolution matrix strictly employs a mean squared error (MSE) approximation feedback mechanism. In the historical validation set, the difference between the predicted feature values ​​of the network model and the actual recorded process state feature values ​​is calculated. The squares of all differences are accumulated and divided by the total number of samples to derive the MSE, representing the degree of prediction deviation. The partial derivative of this MSE with respect to each weight element in the matrix is ​​calculated, and the weight values ​​are updated by gradually decreasing the step size along the gradient descent direction until the error reduction tends to stagnate. This method solidifies the weights of each network parameter, generating the linear evolution matrix. The global process state evolution feature set generated by concatenating the current tracking time nodes is converted into a row vector structure with horizontal dimension arrangement. The calibrated linear evolution matrix is ​​used as a multi-layered, multi-column transformation operator, and the state evolution product between the global process state evolution feature set and the linear evolution matrix is ​​calculated strictly following the linear algebra matrix multiplication rules. The state evolution product operation involves sequentially multiplying each independent feature element in the row vector of the global process state evolution feature set with each weight coefficient element in the corresponding column of the linear evolution matrix. Then, all the product results are summed, and the sum is used as the single element at the corresponding position in the newly generated high-dimensional vector. By systematically traversing and summing the product operations between the feature set row vector and all columns of the linear evolution matrix, a new set of numerical feature vectors with predictive direction is obtained. This set of vectors is then used to directly define and derive the predicted global process state evolution feature set for the next time node.

[0027] Please see Figure 5 Step S4 is as follows: S401: Extract the state offset values ​​caused by the perturbation values ​​of the slot plate temperature from the predicted global process state evolution feature set of multiple consecutive time nodes, as well as the feature timestamps representing the feature intervals corresponding to the first process boundary state package and the second process boundary state package. Calculate the time delay values ​​experienced between the feature timestamp data when the state offset values ​​propagate across processes over time. Specifically, the feature interval corresponding to the first process boundary state package is the data dimension range of the values ​​corresponding to the single filament diameter interval at the first handover time and the values ​​corresponding to the moisture content at the first handover time in the predicted global process state evolution feature set. Specifically, the feature interval corresponding to the second process boundary state package is the data dimension range of the values ​​corresponding to the drying temperature change, the single filament diameter interval at the second extended handover time, and the values ​​corresponding to the moisture content at the second extended handover time in the predicted global process state evolution feature set. Following the timeline of model deduction, the system retrieves the predicted global process state evolution feature sets for multiple time nodes within the future prediction period generated during the preceding operations. Multi-dimensional process state offset measurement and response delay tracking are then implemented. The system uses the actual predicted feature results, which include the measured perforated plate temperature disturbance values, forward propagated. This is subtracted from the baseline predicted feature results generated during the simulation by forcibly zeroing the perforated plate temperature disturbance values ​​and re-propagating them forward. The difference between the two is calculated, and the absolute value of the deviation is obtained. This difference stripping algorithm extracts the state offset values ​​caused by the perforated plate temperature disturbance values ​​from the predicted global process state evolution feature sets for multiple consecutive time nodes. Within the data structure definition of the evolution feature sets, boundary markers representing the data ranges of different process flow transition points are located and extracted. Feature timestamps representing the feature intervals corresponding to the first process boundary state package and the second process boundary state package are also extracted. The feature interval corresponding to the first process boundary state package is specifically defined as the continuous arrangement range of data dimensions in the predictive global process state evolution feature set, corresponding to the single filament diameter range value at the first handover time and the moisture content value at the first handover time. The feature interval corresponding to the second process boundary state package is specifically defined as the continuous arrangement range of data dimensions in the predictive global process state evolution feature set, corresponding to the drying temperature change value, the single filament diameter range value at the second extended handover time, and the moisture content value at the second extended handover time. Peak scanning is performed along the fluctuation sequence formed by the state offset values ​​on the prediction time axis. The physical absolute time record points corresponding to the local maximum peak change of the state offset values ​​in the feature intervals of the two independent processes are extracted as feature timestamps. The second feature timestamp recorded in the subsequent stage is subtracted from the first feature timestamp that occurred in the initial stage. The time delay value experienced between feature timestamp data when the state offset values ​​propagate across processes over time is calculated. In the sequence tracking, it is found that the position of the second feature timestamp is 8500 milliseconds and the position of the first feature timestamp is 1500 milliseconds. Subtracting 1500 milliseconds from 8500 milliseconds gives the time delay value experienced across processes as 7000 milliseconds.

[0028] S402: Calculate the offset ratio between the state offset value after propagation and the state offset value before propagation; Based on the spatiotemporal location nodes determined by the preceding peak scanning and tracking, the absolute magnitude of the maximum peak value occurring within the data intervals corresponding to two different physical processes is further extracted to perform a relative comparison calculation of the process disturbance evolution intensity. The absolute magnitude of the maximum local peak value scanned and intercepted within the characteristic interval corresponding to the boundary state packet of the second process flow position is identified as the propagated state offset value reflecting the residual intensity of the disturbance intensification. Simultaneously, the absolute magnitude of the original initial maximum local peak value scanned and intercepted within the characteristic interval corresponding to the boundary state packet of the first process at the process start point is identified as the propagated state offset value reflecting the absolute intensity of the source outbreak before propagation. The propagated state offset value representing the terminal intensification intensity is used as the numerator parameter of the division operation, and the propagated state offset value representing the initial source intensity is used as the denominator parameter of the division operation. The offset ratio between the propagated state offset value and the propagated state offset value is calculated using a direct division comparison rule. Substituting the measured prediction results, when the magnitude of the propagated state offset extracted from the network prediction evolution sequence is 4.5 and the magnitude of the extracted pre-propagation state offset is 3.0, by performing an absolute division operation by dividing 4.5 by 3.0, the offset ratio that objectively characterizes the overall fluctuation trend of the original temperature disturbance after being propagated through the complex heat treatment environment of drying is calculated to be 1.5.

[0029] S403: When the offset ratio is greater than the preset first reference ratio, the corresponding offset ratio is determined to be the amplification value in the amplification state. When the offset ratio is less than the preset second reference ratio, the corresponding offset ratio is determined to be the attenuation value in the attenuation state. The time delay value, amplification value or attenuation value are integrated to generate the predicted cross-process evolution characteristics of the disturbance at future time nodes. By pre-determining the first and second benchmark ratios for classifying and judging the degree of deviation through a historical quality control data grid traversal and comparison method, the system continuously adjusts the ratio combinations during backtesting of product quality across 500 historical batches to find the threshold values ​​that maximize the system's accuracy. The set of values ​​with the highest accuracy is selected as the first benchmark ratio of 1.30 and the second benchmark ratio of 0.80. The calculated deviation ratios are then substituted into the numerical comparison control condition node for judgment. When the condition node determines that the measured deviation ratio is greater than the preset first benchmark ratio, the system's high-risk escalation warning logic is triggered, and the corresponding deviation ratio exceeding the benchmark limit is judged as an amplified value. For example, if the measured deviation ratio of 1.5 is determined to be greater than the first benchmark ratio limit of 1.30, it is directly judged and assigned the corresponding deviation ratio of 1.5 as a clear amplification value. Conversely, when the offset ratio calculated by the condition node is less than the preset second benchmark ratio, the low-risk natural dissipation judgment logic is triggered, and the corresponding smaller ratio is judged as the attenuation degree value in the attenuation state; if the offset ratio calculated in a single iteration is 0.6 and it is judged to be less than the second benchmark ratio of 0.80, then 0.6 is assigned and judged as the attenuation degree value. After clarifying the macro trend of evolution fluctuations, the 7000-millisecond time delay value determined by the aforementioned subtraction tracking is extracted, and together with the amplification degree value of 1.5 obtained by comparison and judgment in this stage or the attenuation degree value under the corresponding branch, multi-dimensional key features are merged and spliced. The time delay value, amplification degree value or attenuation degree value are integrated, and the dual state parameters encompassing time domain lag and frequency domain amplitude variation rate are packaged to generate the predicted perturbation cross-process evolution characteristics of future time nodes.

[0030] Please see Figure 6 Step S5 is as follows: S501: Calculate the inverse algebraic reconstruction features of the global process state evolution feature set to generate a process prediction numerical set. The algebraic mapping reconstruction component used to connect the nonlinear digital feature evolution network model with the physical environment of the physical process control is invoked. The row and column positions of the forward projection transformation matrix generated by the system network training are interchanged to generate the transpose matrix. The transpose matrix is ​​multiplied by the original projection transformation matrix to obtain the corresponding intermediate matrix. The inverse matrix is ​​obtained by calculating the adjoint matrix of the intermediate matrix and dividing by the determinant. This inverse matrix is ​​then multiplied again by the transpose matrix generated in the first step to derive the pseudo-inverse mapping matrix used for physical inverse solution. The multidimensional vector matrix of the predicted global process state evolution feature set located at the end of the time-series evolution prediction chain is extracted. This vector matrix is ​​then multiplied by the newly generated pseudo-inverse mapping matrix using element-wise algebraic dimensionality reduction to calculate the inverse algebraic reconstruction feature of the predicted global process state evolution feature set. Through this mapping multiplication solution step, the high-dimensional dimensionless network evolution characteristic parameters derived in the early stage are completely transformed into a sequence of conventional physical parameters carrying various actual equipment measurement units in accordance with the algebraic space transformation rules. The series of equipment execution parameters calculated are summarized and merged in the output register layer, thereby generating a set of process prediction values ​​that can directly reflect the future physical production trend.

[0031] S502: Based on the set of process prediction values, obtain the difference values ​​of uneven moisture content in the drying section, the abnormal deviation values ​​of the winding density in the winding section, and the difference values ​​of the end face forming deviation in the drawing section, and perform normalization processing on each value to generate scalar values ​​of quality indicators. Based on the predetermined data field identification names in the generated process prediction numerical set, a targeted retrieval operation is performed to accurately obtain three types of key process differences characterizing the micro and macro defect risks of finished products from the complex set. Specifically, the difference in moisture content in the drying section, which characterizes the degree of residual dispersion caused by uneven axial moisture evaporation of glass fibers after drying and shaping treatment, is obtained; the abnormal deviation in the winding density of the winding section, which characterizes the degree of misalignment of the winding structure induced by tension accumulation fluctuations in the final product, is obtained; and the difference in the dimensional deviation of the end face forming of the drawing section, which characterizes the degree of decrease in the roundness of the cross-section caused by thermal variations, is obtained. After extracting these three key process differences, the maximum upper limit range specification for the allowable fluctuation of each parameter is retrieved from the national and industry process safety and quality tolerance standard library. The extracted difference values ​​are then divided by the corresponding maximum upper limit range and normalized accordingly. In actual quantitative operations, the extracted 1.2 percentage point difference in moisture content unevenness in the drying section is divided by the allowable range of 2.0 percentage points to obtain a normalized result of 0.60. The abnormal deviation in package density of the winding section of 2.5 kg / m³ is divided by the range of 5.0 kg / m³ to obtain a normalized result of 0.50. The dimensional difference in end-face forming deviation of the drawing section of 0.8 mm is divided by the range of 1.0 mm to obtain a normalized result of 0.80. Within the system, these three transformed unified scale indicators are each assigned a weighted conversion factor of 33.3, each accounting for one-third of the total. The three normalized results are multiplied by 33.3 respectively, and then continuously added together to obtain the global sum. That is, 0.60 multiplied by 33.3, 0.50 multiplied by 33.3, and 0.80 multiplied by 33.3 are summed to obtain a value of 63.3. This weighted addition operation generates a scalar value of quality indicator for comprehensively assessing the overall defect risk trend.

[0032] S503: Calculate the difference between the scalar value of the quality indicator and the preset percentage threshold of qualified products, obtain the absolute value of the difference, filter the absolute values ​​of the difference that are greater than zero, generate the predicted product quality anomaly judgment result for future time nodes, and generate control voltage value commands to adjust the spindle heating power and the exhaust fan speed of the dryer based on the time delay value, the predicted product quality anomaly judgment result, and the predicted cross-process evolution characteristics of the disturbance. Issue control voltage value commands to adjust the drawing parameters and drying parameters, and generate the virtual simulation control result of the glass fiber production line. The percentage threshold of qualified products is set based on the allowable fluctuation tolerance of the moisture content of the drying section, the upper limit of the standard qualified deviation of the winding density of the winding section, and the limit allowable dimensional tolerance of the end face of the drawing section in the historical qualified quality inspection data of glass fiber. Large-sample evaluation data from past production records were collected from the company's historical product inspection and release database. The percentage of products meeting all dimensional and physical requirements was statistically analyzed. The percentage threshold for qualified products was set based on historical qualified quality inspection data for glass fiber, including allowable fluctuations in moisture content during drying, upper limits of acceptable deviations in winding density, and allowable dimensional tolerances at the end face of the drawing section. A percentage point constant was obtained by dividing batches of products meeting all these constraints by the total output. This baseline constant was set as the preset qualified product percentage threshold of 98.0%. A predicted qualified rate variable of 36.7 was calculated by subtracting 63.3 from the constant base of 100. The difference between the converted quality indicator variable and the preset qualified product percentage threshold was calculated. Subtracting 98.0% from the predicted qualified rate variable of 36.7 yielded a negative 61.3. The absolute value of this negative difference was then calculated to obtain an absolute value of 61.3. The system performs a size judgment condition comparison operation to filter for absolute drop values ​​greater than zero. Since 61.3 is greater than the zero baseline, it meets the alarm triggering limit, generating a predicted product quality anomaly judgment result for future time nodes. The absolute drop value of 61.3 is extracted and multiplied by the proportional adjustment gain coefficient fixed in the control module to generate an instantaneous proportional adjustment component. The summation result of historical absolute drop values ​​within the integral time window is retrieved and multiplied by the integral adjustment gain coefficient to generate an integral adjustment component to eliminate steady-state error. The proportional adjustment component and the integral adjustment component are added together and forcibly multiplied by an amplification value of 1.5 to enhance the control strength against severe disturbances. Based on the time delay value, the predicted product quality anomaly judgment result, and the predicted cross-process evolution characteristics of disturbances, after completing the proportional-integral product conversion, control voltage commands are generated to adjust the spindle heating power to -1.2 volts and the dryer exhaust fan speed to +2.5 volts. These control voltage commands are sent to the heating drive power supply and frequency converter via the industrial communication bus to adjust the drawing and drying parameters. The key evolution indicators are confirmed to return to a smooth threshold, generating the virtual simulation control result for the glass fiber production line.

[0033] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A virtual simulation method for a glass fiber production line based on digital twins, characterized in that, Includes the following steps: S1: Collect the first process boundary state packet of the fiber drawing equipment during the production of glass fiber; S2: Collect the second process boundary state packet of the drying equipment and the winding tension change value during the process of glass fiber entering the winding equipment, combine the first process boundary state packet, and construct a global serial data set; S3: Predict the state of the global serial data set at the next time node to obtain the predicted global process state evolution feature set; S4: Based on the predicted global process state evolution feature set, analyze the cross-process evolution characteristics of future disturbances in the production line; S5: Based on the predicted global process state evolution feature set, determine the future glass fiber product quality of the production line, and generate control instructions by combining the cross-process evolution features of disturbances, thereby generating the virtual simulation control results of the glass fiber production line.

2. The virtual simulation method for a glass fiber production line based on digital twins according to claim 1, characterized in that, The first process boundary state package includes initial operating state parameters of the drawing process, single filament dimension parameters of the drawing stage junction boundary, and moisture content dimension parameters of the drawing stage junction boundary. The second process boundary state package includes temperature change state parameters of the drying process, single filament dimension parameters of the drying stage junction boundary, and moisture content dimension parameters of the drying stage junction boundary. The global serial data set includes a sequence of operating characteristics of the drawing equipment, a sequence of operating characteristics of the drying equipment, and a sequence of operating characteristics of the winding equipment. The predicted global process state evolution feature set includes polynomial nonlinear mapping feature terms for future time nodes and state matrix evolution calculation feature terms for future time nodes. The predicted cross-process disturbance evolution features include cross-process disturbance propagation delay feature data, cross-process disturbance evolution amplification ratio data, and cross-process disturbance evolution attenuation ratio data. The virtual simulation control results of the glass fiber production line include spinneret heating power voltage control command parameters and dryer exhaust fan speed voltage control command parameters.

3. The virtual simulation method for a glass fiber production line based on digital twins according to claim 1, characterized in that, Step S1 is as follows: S101: During the production of glass fiber in the drawing equipment, the values ​​of spinneret temperature disturbance, drawing speed change, and sizing agent coating amount change are collected, aligned and merged according to the time stamp order to generate an initial process state vector. S102: Monitor and characterize the position coordinates of the glass fiber bundle. When the position coordinates of the glass fiber bundle reach the first junction moment of the physical junction between the drawing and drying stages, collect the single filament diameter range value and the moisture content value at the first junction moment to generate the drawing junction boundary characteristic parameters. S103: Combine and encapsulate the wire drawing junction boundary feature parameters and the initial process state vector to generate the first process boundary state package.

4. The virtual simulation method for a glass fiber production line based on digital twins according to claim 3, characterized in that, Step S2 is as follows: S201: When the position coordinates of the glass fiber bundle reach the second extended junction moment of the physical junction boundary between the drying stage and the winding stage, collect the single filament diameter range value and the moisture content value at the second extended junction moment to generate the drying junction boundary characteristic parameters. S202: Collect the drying temperature change value of the drying equipment, combine it with the drying handover boundary feature parameters, and integrate it into the second process boundary state package; S203: During the process of glass fiber entering the winding equipment, the value of the change in winding tension is collected, and the first process boundary state packet, the second process boundary state packet and the value of the change in winding tension are spliced ​​together to generate a global serial data set.

5. The virtual simulation method for a glass fiber production line based on digital twins according to claim 4, characterized in that, Step S3 is as follows: S301: In the digital twin virtual space, the global concatenated dataset is used as an evolutionary inference benchmark to generate state observations; S302: Normalize the state observations to a unified numerical range, calculate the polynomial combination features of the normalized state observations using preset nonlinear rules, and generate a global process state evolution feature set. S303: Obtain the linear evolution matrix of the initial calibration stage, calculate the state evolution product between the global process state evolution feature set and the linear evolution matrix, and obtain the predicted global process state evolution feature set for the next time node.

6. The virtual simulation method for a glass fiber production line based on digital twins according to claim 5, characterized in that, Step S4 is as follows: S401: Extract the state offset value caused by the leakage plate temperature disturbance value from the predicted global process state evolution feature set of multiple consecutive time nodes, as well as the feature timestamps representing the feature intervals corresponding to the first process boundary state package and the second process boundary state package, and calculate the time delay value experienced between the feature timestamp data when the state offset value propagates across processes over time. S402: Calculate the offset ratio between the state offset value after propagation and the state offset value before propagation; S403: When the offset ratio is greater than the preset first reference ratio, the corresponding offset ratio is determined to be the amplification value in the amplification state. When the offset ratio is less than the preset second reference ratio, the corresponding offset ratio is determined to be the attenuation value in the attenuation state. The time delay value, amplification value or attenuation value are integrated to generate the predicted disturbance cross-process evolution characteristics of future time nodes.

7. The virtual simulation method for a glass fiber production line based on digital twins according to claim 6, characterized in that, Step S5 is as follows: S501: Calculate the inverse algebraic reconstruction features of the predicted global process state evolution feature set to generate a process prediction numerical set; S502: Based on the process prediction value set, obtain the difference value of uneven moisture content in the drying section, the abnormal deviation value of the winding density in the winding section, and the difference value of the end face forming deviation in the drawing section, and perform normalization processing on each to generate scalar values ​​of quality indicators. S503: Calculate the difference between the scalar value of the quality indicator and the preset percentage threshold of qualified products, obtain the absolute value of the difference, filter the absolute values ​​of the difference that are greater than zero, generate the predicted product quality anomaly judgment result for future time nodes, and generate control voltage value instructions to adjust the heating power of the spindle and the speed of the exhaust fan of the dryer based on the time delay value, the predicted product quality anomaly judgment result and the predicted cross-process evolution characteristics of the disturbance. Issue control voltage value instructions to adjust the drawing parameters and drying parameters, and generate the virtual simulation control result of the glass fiber production line.

8. The virtual simulation method for a glass fiber production line based on digital twins according to claim 5, characterized in that, The nonlinear rules include multiplying the values ​​of the spinneret temperature disturbance and the drawing speed change, performing a power-square calculation on the values ​​of the moisture content at the first handover moment and the drying temperature change, and performing a high-order multiplication on the values ​​of the winding tension change.

9. The virtual simulation method for a glass fiber production line based on digital twins according to claim 6, characterized in that, The feature interval corresponding to the first process boundary state package is specifically the data dimension range of the values ​​of the single filament diameter range at the first handover time and the values ​​of the moisture content at the first handover time in the predicted global process state evolution feature set. The feature interval corresponding to the second process boundary state package is specifically the data dimension range of the values ​​of the drying temperature change, the single filament diameter range at the second extension handover time, and the values ​​of the moisture content at the second extension handover time in the predicted global process state evolution feature set.

10. The virtual simulation method for a glass fiber production line based on digital twins according to claim 7, characterized in that, The percentage threshold for qualified products is set based on the allowable fluctuation tolerance of moisture content in the drying section, the upper limit of the standard qualified deviation of the winding density in the winding section, and the limit allowable dimensional tolerance of the end face in the drawing section, all from the historical qualified quality inspection data of glass fiber.

Images