High-pressure glass fiber reinforced plastic pipe full-automatic layer winding simulation system based on digital twinning

Abstract

The application relates to the technical field of composite material intelligent manufacturing and digital twin simulation, in particular to a high-pressure glass steel pipe full-automatic layering and winding simulation system based on digital twinning, which comprises a physical data analysis module, a kinematic mapping fast ring module and a mechanical simulation slow ring module; spindle rotation, trolley translation and time sequence tension data are collected and cleaned, three-dimensional coordinates of fiber landing points, actual winding angles and residual tension values are obtained through mapping; when the deviation between the actual winding angle and the design winding angle exceeds a threshold value, a local perturbation correction unit is triggered, a global displacement field vector is updated based on low-rank matrix multiplication and matrix inversion lemma; when the threshold value is not exceeded, the state of the reference finite element model is maintained unchanged, and a normal displacement field calculation is performed, so that low-delay and high-fidelity mechanical synchronous simulation of the winding process is realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing and digital twin simulation technology of composite materials, specifically a fully automated layup and winding simulation system for high-pressure fiberglass pipes based on digital twins. Background Technology

[0002] Fully automated layup and winding simulation of high-pressure fiberglass pipes refers to the synchronous analysis and prediction of fiber layup trajectory and structural mechanical response during the winding and forming process of high-pressure fiberglass pipes, based on process information such as spindle rotation, carriage translation, and tension changes. Currently, there are three main methods for high-pressure fiberglass pipe winding analysis: offline process evaluation method based on equipment empirical parameters, layup monitoring method based on geometric trajectory reconstruction, and post-forming structural simulation method based on finite element model. However, when performing winding simulation based on existing technologies, on the one hand, it is difficult to accurately and in real time map the temporal motion data and tension data during the winding process to the actual fiber landing point, winding angle and local stress state. On the other hand, since existing simulation methods mostly adopt overall recalculation or post-forming analysis, there are also problems such as calculation delay exceeding the physical winding cycle and difficulty in balancing real-time performance and high-fidelity evaluation of local defects. All of these will reduce the accuracy of mechanical prediction of the winding process of high-pressure fiberglass pipe fittings. Summary of the Invention

[0003] The purpose of this invention is to provide a fully automated layup and winding simulation system for high-pressure fiberglass pipe fittings based on digital twins, and to solve the following technical problems: To avoid performing a global recalculation at every sampling point, which would cause the simulation delay to exceed the actual winding cycle, this method achieves low-delay, high-fidelity synchronous mechanical simulation during the winding process of high-pressure fiberglass pipes. It can also translate process deviations into structural performance information that can be used for quality assessment, thereby realizing a closed-loop connection from process monitoring to pressure risk prediction.

[0004] The objective of this invention can be achieved through the following technical solutions: A fully automated layup and winding simulation system for high-pressure fiberglass pipes based on digital twins. The system includes: a physical data analysis module, a kinematic mapping fast loop module, and a mechanical simulation slow loop module. The physical data parsing module connects to the encoder and tension sensor of the physical actuator, which includes a main shaft and a carriage. It collects the time-series status data of the main shaft rotation and carriage translation, as well as the time-series tension data, and performs data cleaning operations to remove outliers, generating a standard time-series dataset. The kinematic mapping fast loop module performs spatial geometric transformation operations on the standard time series dataset based on the established three-dimensional cylindrical coordinate system to calculate the three-dimensional spatial coordinates of the current fiber landing point and the actual winding angle; and calculates the residual tension value at the current fiber landing point based on the time series tension data and the tension transmission attenuation model including the friction coefficient and the contact wrap angle. The mechanical simulation slow loop module includes a reference finite element model and local perturbation correction elements. The reference finite element model is pre-initialized and generates a pre-stored global stiffness matrix inverse matrix, load vector, and global displacement field vector. The mechanical simulation slow loop module calculates the absolute value of the angular deviation between the actual winding angle and the design winding angle preset based on the pipe fitting design parameters. When the absolute value of the angle deviation is greater than the preset angle distortion judgment threshold, the local perturbation correction unit is triggered. Based on the current three-dimensional coordinates of the fiber landing point and the residual tension value, the local stiffness change matrix caused by the fiber deviation is calculated. The local stiffness change matrix is ​​decomposed into a low-rank matrix product, and the global displacement field vector of the reference finite element model is updated based on the local stiffness matrix direct correction algorithm containing the matrix inversion lemma. When the absolute value of the angle deviation is not greater than the angle distortion judgment threshold, the state of the reference finite element model remains unchanged, and the pre-stored global stiffness matrix inverse matrix is ​​multiplied with the load vector to perform conventional displacement field calculation to update the global displacement field vector.

[0005] In one possible implementation, the physical data parsing module also includes a time-series noise reduction unit, which performs Kalman filtering on the time-series tension data to extract the true tension sequence, eliminate mechanical resonance noise interference, and fuse the true tension sequence into the standard time-series dataset.

[0006] In one possible implementation, when the local perturbation correction unit executes the local stiffness matrix direct correction algorithm, it calls a pre-set local stiffness matrix direct correction formula based on the matrix inversion lemma. Based on the pre-stored product of the global stiffness matrix inverse and the low-rank matrix, it calculates the local defect dimension reduction inverse term and combines the local defect dimension reduction inverse term with the pre-stored global stiffness matrix inverse to solve for the updated global displacement field vector.

[0007] In one possible implementation, the low-rank matrix product is formed by multiplying the transpose of a first low-dimensional matrix and a second low-dimensional matrix, where the first low-dimensional matrix represents the influence distribution of the local mesh degrees of freedom association, and the second low-dimensional matrix represents the corresponding state feedback. The local perturbation correction element maps the current three-dimensional coordinates of the fiber landing point to the local mesh elements divided in the reference finite element model, and determines the number of columns of the first low-dimensional matrix and the second low-dimensional matrix based on the number of degrees of freedom of the local mesh elements.

[0008] In one possible implementation, when performing spatial geometric transformation operations, the kinematic mapping fast loop module uses preset core model radius parameters and kinematic mapping equations that map time, displacement, and rotation angles to spatial coordinates to map the standard time-series dataset collected based on the time dimension to a three-dimensional cylindrical coordinate system, generating a point cloud coordinate set containing spatial topological relationships.

[0009] In one possible implementation, the mechanical simulation slow loop module also includes a performance defect evaluation unit, which calculates the local stress and strain distribution based on the updated global displacement field vector and extracts the characteristic values ​​of the local fiber micro-buckling evolution state. When the characteristic value of the local fiber microbuckling evolution state exceeds the preset material failure envelope boundary based on multiaxial stress state, the performance defect assessment unit outputs a defect warning signal containing the predicted value of burst pressure reduction. When the characteristic value of the local fiber microbuckling evolution state does not exceed the boundary of the material failure envelope, the performance defect evaluation unit outputs a normal state signal.

[0010] In one possible implementation, the baseline finite element model is constructed based on the constitutive equations of a three-dimensional orthogonal anisotropic material; The constitutive equation of three-dimensional orthogonal anisotropic materials establishes the mathematical mapping relationship between macroscopic mechanical performance parameters and microscopic fiber volume fraction and design winding angle, and pre-calculates and generates the inverse global stiffness matrix.

[0011] In one possible implementation, the system is deployed in a layered network architecture that includes edge computing nodes and cloud servers; The physical data parsing module and the kinematic mapping fast loop module reside on the edge computing node, performing low-latency data acquisition and spatial coordinate mapping. The slow-loop mechanical simulation module resides on the cloud server and performs asynchronous mechanical reconstruction based on the current three-dimensional coordinates of the fiber landing point, the actual winding angle, and the residual tension value output by the fast-loop kinematic mapping module.

[0012] In one possible implementation, the timing status data includes spindle rotation angle data and trolley displacement data; The physical data parsing module is connected to an absolute encoder. The physical data parsing module obtains the spindle rotation angle data and the trolley displacement data through the absolute encoder, and performs timestamp alignment processing on the spindle rotation angle data and the trolley displacement data to eliminate acquisition delay errors.

[0013] In one possible implementation, the operating frequency of the kinematic mapping fast loop module is synchronized with the data sampling frequency of the physical actuator; When the mechanical simulation slow loop module does not receive a trigger command from the local perturbation correction unit, it writes the actual winding angle and residual tension value output by the kinematic mapping fast loop module into the timing feature cache pool to generate feature data for offline process review.

[0014] The beneficial effects of this invention are: 1. This invention adopts a dual-loop architecture of kinematic mapping fast loop and mechanical simulation slow loop, which transforms temporal motion and tension data into three-dimensional landing point, actual winding angle and residual tension in real time; by pre-storing the inverse global stiffness matrix, local perturbation correction based on low-rank matrix decomposition is triggered only when the angle deviation exceeds the threshold, avoiding the computational delay caused by global recalculation and greatly improving the accuracy and synchronization of mechanical prediction. 2. This invention uses an absolute encoder to acquire the spindle rotation angle and trolley displacement and aligns them with timestamps, eliminating false landing points and trajectory misjudgments caused by acquisition delay; at the same time, it performs Kalman filtering on the time-series tension data to extract the true sequence, effectively filtering out mechanical resonance noise during the winding process, avoiding false defect identification caused by pseudo-fluctuations and unnecessary computational resource consumption; 3. This invention can accurately map the actual landing point to local mesh elements and construct a low-dimensional matrix based on the affected degrees of freedom. It can directly call the local stiffness matrix to directly correct the formula and calculate the dimension reduction inverse term. This method limits the abnormal mechanical effects to a finite-dimensional calculation range. It can solve the updated global displacement field without inverting the whole machine and achieve low-latency mechanical reconstruction within the physical beat. 4. This invention utilizes the core mold radius parameter and mapping equation to map the time-series acquisition dataset to a three-dimensional cylindrical coordinate system, generating a point cloud coordinate set with spatial topological relationships. This method successfully transforms the abstract one-dimensional process sampling sequence into an intuitive spatial trajectory containing actual geometric continuity, crowding or sparsity, laying a reliable data foundation for subsequent accurate angle evaluation and force reconstruction. 5. This invention constructs a mechanical base that truly reflects the properties of composite materials using the constitutive equation of three-dimensional orthogonal anisotropic materials, and extracts local fiber micro-buckling characteristic values ​​after updating the displacement field; by comparing with the boundary of the multiaxial failure envelope surface, it can not only output defect warning, but also directly predict the reduction value of burst pressure, successfully translating process deviations into pressure risk assessment indicators, and realizing a closed loop of quality throughout the entire process. 6. This invention adopts a layered architecture that coordinates edge and cloud. Edge nodes are responsible for low-latency high-frequency acquisition mapping, while cloud servers are responsible for asynchronous mechanical reconstruction, which takes into account both the stability of on-site acquisition and the scalability of simulation. At the same time, with the help of a time-series feature cache pool, key features during normal periods are retained offline, which enables the visualization and traceability of historical processes and the review of sample quality without crowding out online computing power. Attached Figure Description

[0015] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a schematic diagram of the modules of the fully automated layup and winding simulation system for high-pressure fiberglass pipes based on digital twins provided in the embodiments of this application; Figure 2 This is a schematic diagram of the workflow of the fully automated layup and winding simulation system for high-pressure fiberglass pipes based on digital twins provided in this application embodiment. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figure 1 A fully automated layup and winding simulation system for high-pressure fiberglass pipe fittings based on digital twins. The system includes: a physical data analysis module, a kinematic mapping fast loop module, and a mechanical simulation slow loop module. The physical data parsing module connects to the encoder and tension sensor of the physical actuator, which includes a main shaft and a carriage. It collects the time-series status data of the main shaft rotation and carriage translation, as well as the time-series tension data, and performs data cleaning operations to remove outliers, generating a standard time-series dataset. The kinematic mapping fast loop module performs spatial geometric transformation operations on the standard time series dataset based on the established three-dimensional cylindrical coordinate system to calculate the three-dimensional spatial coordinates of the current fiber landing point and the actual winding angle; and calculates the residual tension value at the current fiber landing point based on the time series tension data and the tension transmission attenuation model including the friction coefficient and the contact wrap angle. The mechanical simulation slow loop module includes a reference finite element model and local perturbation correction elements. The reference finite element model is pre-initialized and generates a pre-stored global stiffness matrix inverse matrix, load vector, and global displacement field vector. The mechanical simulation slow loop module calculates the absolute value of the angular deviation between the actual winding angle and the design winding angle preset based on the pipe fitting design parameters. When the absolute value of the angle deviation is greater than the preset angle distortion judgment threshold, the local perturbation correction unit is triggered. Based on the current three-dimensional coordinates of the fiber landing point and the residual tension value, the local stiffness change matrix caused by the fiber deviation is calculated. The local stiffness change matrix is ​​decomposed into a low-rank matrix product, and the global displacement field vector of the reference finite element model is updated based on the local stiffness matrix direct correction algorithm containing the matrix inversion lemma. When the absolute value of the angle deviation is not greater than the angle distortion judgment threshold, the state of the reference finite element model remains unchanged, and the pre-stored global stiffness matrix inverse matrix is ​​multiplied with the load vector to perform conventional displacement field calculation to update the global displacement field vector.

[0018] This embodiment provides a fully automated layup and winding simulation mechanism for high-pressure fiberglass pipe fittings based on digital twins. Specifically, the main application scenario is set as follows: in a fully automated winding production line for preparing high-pressure hydrogen storage fiberglass pipe fittings, the parts to be processed include a cylindrical section and a head transition section, and the winding equipment lays up the circumferential layer and the spiral layer in sequence according to a predetermined process. The system does not output control commands to the machine tool, but instead builds and updates a virtual mechanical model simultaneously during the winding process to predict the displacement response, local instability risk, and burst pressure change trend of the batch of pipe fittings in subsequent pressure-bearing use. The physical data analysis module collects two types of basic data from the actual equipment: one is the time-series state data formed by the spindle rotation and the carriage translation, and the other is the tension time-series data on the guide wire path; The above-mentioned data collection objects correspond to the actual process conditions: the rotation of the main shaft determines the circumferential advancement of the fiber wrapping, the translation of the carriage determines the axial speed of the fiber, and the tension determines the degree of adhesion of the fiber bundle to the core mold surface and the additional load on the under-layer incompletely cured material. The aforementioned data cleaning operation is not simply deleting abnormal values, but rather removing abnormal records that clearly do not conform to mechanical continuity, such as suddenly changing angle values, displacement back-off values ​​caused by short-term code breaks, and distorted tension peaks caused by instantaneous impacts; after cleaning, the system obtains a standard time-series dataset that can reflect the continuous winding process. Based on this, the kinematic mapping fast loop module converts the process information in the time dimension into fiber landing point information in the spatial dimension; for the cylindrical mandrel, the spindle angular displacement corresponds to the circumferential position, and the carriage displacement corresponds to the axial position. The two together determine the real-time landing point of the fiber on the mandrel surface. For the head transition section, it can be projected onto the surface of the three-dimensional model according to the corresponding surface parameters; the three-dimensional coordinates of the current fiber landing point output by this module are used to characterize the actual pressing position of the current fiber bundle on the pipe surface; the actual winding angle output is used to quantify the angular deviation of the current fiber bundle direction from the design path; Meanwhile, tension will not be transmitted to the landing point without physical attenuation, because the fiber will pass through several guide surfaces and wrapping paths from the guide nozzle to the mandrel, and friction and wrap angle will consume some tension; therefore, the system combines the friction coefficient and the contact wrap angle to establish a tension transmission attenuation model in order to obtain a residual tension value that is more consistent with the actual stress state. The tension transmission attenuation model is constructed based on Euler's winch formula, and its calculation expression is as follows:

[0019] in, This represents the residual tension value at the current fiber landing point. This refers to the initial tension data output by the guide nozzle. The coefficient of sliding friction between the guide surface and the fiber. This refers to the contact angle radius between the guide surface and the fiber. The aforementioned residual tension is key boundary information in mechanical simulation, corresponding to the actual compression and pulling effects applied by the fiber to the landing area; Furthermore, the mechanical simulation slow loop module starts with a pre-established benchmark finite element model; this model has been initialized according to the designed laminated structure, geometry and material parameters before the production line is started, and the global stiffness matrix inverse matrix, load vector and initial global displacement field vector are pre-stored. The reason for adopting the dual-loop structure of fast loop + slow loop is that the physical state of the winding equipment changes frequently, and the computational cost of large-scale mechanical reconstruction is large. If a global recalculation is performed at each sampling point, the simulation delay will be greater than the actual winding cycle, thus losing the meaning of synchronization. The system extracts the allowable in-plane shear strain parameters of the currently used composite material. and the upper limit of the allowable angular deviation of the winding equipment's mechanical tolerance. Establish an angle distortion judgment threshold. The mapping relationship is:

[0020] This generates a precise local perturbation triggering reference; Therefore, the system introduces asynchronous triggering judgment logic: only when the deviation between the actual winding angle and the designed winding angle exceeds the angle distortion judgment threshold, it is considered that the fiber laying at that point is no longer a normal process fluctuation, but may change the local lamination direction, load transfer path and local stability. When the above deviation exceeds the threshold, the local perturbation correction unit is triggered; the unit determines the location of the anomaly based on the three-dimensional coordinates of the current landing point, judges the level of the additional load at that location based on the residual tension, and estimates the local stiffness change at that location accordingly. In engineering terms, the aforementioned local stiffness change corresponds to the shift in the main load-bearing direction of the area after the actual fiber laying angle deviates. If the residual tension is low at the same time, the fiber compaction will be insufficient, the risk of porosity will increase, and the interlayer constraint will be weakened; if the residual tension is high, the lower layer will be easily subjected to additional compression, inducing the initiation of micro buckling. The specific logic for quantifying stiffness changes is as follows: The system establishes a local material coordinate system for the fiber and a global geometric coordinate system for the pipe fitting. First, based on the absolute value of the deviation between the actual winding angle and the designed winding angle, the rotation deviation of the stiffness matrix caused by the deflection of the fiber principal direction is calculated through the coordinate transformation tensor formed by the direction cosine matrix from the local material coordinate system to the global geometric coordinate system. Then, based on the ratio of the decrease or increase of residual tension relative to the standard process value, the reduction or strengthening coefficient of transverse and shear stiffness caused by insufficient interlayer compaction or excessive compression is quantitatively evaluated. By combining and superimposing the two and applying them to the initial stiffness matrix corresponding to the local mesh, the local stiffness variation matrix relative to the reference finite element model at that point can be calculated. The system uses a singular value decomposition algorithm on the local stiffness variation matrix to extract its non-zero or greater than the truncation set value and the corresponding eigenvectors, and constructs the first low-dimensional matrix and the second low-dimensional matrix accordingly. Since anomalies are usually confined to a specific local mesh region, such as a few mesh cells near the reversal of the end cap, the system represents the local stiffness change as a low-rank matrix product and uses the direct correction algorithm corresponding to the matrix inversion lemma to locally update the existing global response, without having to reassemble the entire pipe fitting and perform global inversion. The updated results are reflected in the new global displacement field vector, providing a basis for subsequent stress, strain and defect assessments; When the angle deviation does not exceed the threshold, the system considers that the current laying is still within the design allowable fluctuation range, does not start local reconstruction, but keeps the state of the reference finite element model unchanged, and directly uses the cached global stiffness matrix inverse matrix and load vector to refresh the conventional displacement field. The essence of this processing logic is to treat the normal layer as a continuation of the baseline state, and the abnormal layer as a local deviation that needs to be specially explained, thereby balancing real-time performance and mechanical fidelity. As a fault-tolerant mechanism for abnormal situations, if encoder data is lost within a certain sampling period, the physical data parsing module can mark that period as an untrusted sample and suspend the spatial landing point mapping at the corresponding time. If tension data is continuously missing or fluctuates beyond the sensor's range, the system can retain the pose mapping results for that interval, but adopt a conservative estimate of the most recent stable interval for the residual tension, and add a low confidence mark to the simulation results for that segment to avoid false high-frequency fluctuations triggering unnecessary mechanical reconstruction. If the local anomaly range expands to multiple consecutive adjacent regions, making the low-rank approximation no longer applicable, the slow loop can be switched to the partitioned reorganization mode to reassemble the larger local sub-models without having to fall back to the whole machine for global recalculation. During the winding process of a high-pressure fiberglass pipe with a nominal working pressure of 70MPa, the production line first completes the conventional spiral winding of the cylindrical section. At this time, the spindle rotation and the trolley feed remain stable. The fast loop continuously outputs the regularly distributed landing point coordinates and the winding angle close to the design value, while the slow loop only performs the conventional displacement field refresh. When the equipment enters the head transition section, due to the change in curvature and the switching of the guide nozzle posture, the system detects that the actual winding angle of a certain local area deviates significantly from the design value. At the same time, the residual tension decreases compared to the previous area, indicating that the fiber has both directional deviation and insufficient adhesion at this point. The slow loop triggers a local perturbation correction, which updates the mesh stiffness only within a limited range at that point, obtains the new global displacement field in a timely manner, and marks the anomalous area as a key evaluation area for subsequent evaluation. The purpose of this step is to stably map the time-series process data continuously generated on the machine tool side into spatial position, angle and force information that can be directly used by the finite element model, and to achieve low-delay, high-fidelity synchronous mechanical simulation of the high-pressure fiberglass pipe winding process by triggering local corrections only when necessary.

[0021] In a preferred embodiment of the present invention, the physical data parsing module further includes a time-series noise reduction unit, which performs Kalman filtering on the time-series tension data to extract the true tension sequence, eliminate mechanical resonance noise interference, and fuse the true tension sequence into the standard time-series dataset.

[0022] This embodiment provides a noise reduction mechanism for tension timing; specifically, in the aforementioned high-pressure hydrogen storage fiberglass pipe winding scenario, the guide nozzle, tension roller and frame body may all generate mechanical vibration, especially when the spindle speed increases or the guide path is switched, the waveform collected by the tension sensor will be mixed with resonance components. If such raw tension data is directly input into spatial mapping and mechanical simulation, the vibration of the mechanical structure may be misjudged as the actual stress change of the fiber, thus causing incorrect residual tension estimation and falsely triggered local mechanical reconstruction. The timing noise reduction unit performs state-tracking smoothing on the tension signal, rather than simple averaging; The basic idea is that the actual tension of the fiber has a continuous evolution characteristic in physics. Unless there is a breakage, slippage or a sudden change in direction, it will not have discrete steps and high-frequency oscillations driven by non-process commands within a single sampling period. Mechanical resonance, on the other hand, often manifests as periodic oscillations, phase lag or short-term high-frequency ripples. Based on this difference, the time-series noise reduction unit comprehensively references the tension change trend of the previous moment and the current sensor reading in each sampling cycle, outputs a real tension sequence that is closer to the actual process, and then integrates it into the standard time-series dataset, so that the subsequent residual tension estimation is based on the actual force on the fiber rather than the sensor vibration reading. To facilitate understanding, the following example is provided in conjunction with a specific application scenario: Suppose that the original tension of three consecutive sampling points shows a pattern of stable-peak-stable, while the spindle speed, carriage speed, and landing position do not show any process changes that can explain the peak during the same period. In this case, the timing noise reduction unit will treat the middle peak as a mechanical disturbance and output a tension value that is more continuous with the preceding and following trends. Conversely, if the tension rise occurs simultaneously with the guide nozzle reversal, the fiber wrap angle increase, or the end cap curvature abruptly, the rise will be retained because this better reflects the actual stress changes in the process. As a fault-tolerant mechanism for abnormal situations, if the sensor drifts for a long time, causing the tension sequence after noise reduction to be smooth but deviating from the calibration range, the system needs to call the sensor zero-point calibration results or equipment maintenance records for secondary verification. If a fiber breakage or instantaneous loosening of a fiber bundle occurs during the current working condition, the sudden drop in tension is itself a critical anomaly that cannot be smoothly erased. Therefore, when the timing noise reduction unit detects a continuous drop accompanied by pose mismatch, it should retain the abnormal shape and output a fault marker to the subsequent modules. When the same high-pressure fiberglass pipe enters the head transition section, due to the trolley reversal and the change in the contact state of the guide wheel, a series of high-frequency ripples appear on the original curve of the tension sensor. If the original curve is used directly, the system will misjudge that the fiber bundle repeatedly experiences sudden increases and decreases in tension at multiple adjacent landing points, thus generating a series of false local defect areas; After introducing the noise reduction mechanism in this embodiment, the system extracts a continuous and stable real tension sequence and retains the tension downward trend only in the interval where the guide wire path actually undergoes a compression change. As a result, the subsequent residual tension calculation and local stiffness correction are more in line with reality. The purpose of this mechanism is to distinguish the mechanical resonance noise in the tension sensor from the actual process load changes, thereby achieving a more reliable force reconstruction input and avoiding false defect identification and unnecessary computational overhead caused by pseudo fluctuations.

[0023] In a preferred embodiment of the present invention, when the local perturbation correction unit executes the local stiffness matrix direct correction algorithm, it calls a preset local stiffness matrix direct correction formula based on the matrix inversion lemma, calculates the local defect dimension reduction inversion term based on the product of the pre-stored global stiffness matrix inverse and the low-rank matrix, and combines the local defect dimension reduction inversion term with the pre-stored global stiffness matrix inverse to solve for the updated global displacement field vector.

[0024] This embodiment provides a mechanism for direct correction of local stiffness matrix; specifically, based on the aforementioned dual-ring simulation architecture which is already able to identify abnormal regions, this embodiment further addresses the problem of how to complete mechanical updates within physical ticks after anomalies are identified. If the global re-inversion method of the entire pipe fitting is still used, even if the anomaly only occurs in the local finite element mesh of the head transition section, the entire large-scale finite element system will be recalculated, resulting in a significant increase in simulation delay. This embodiment uses a pre-set formula for direct correction of the local stiffness matrix based on the matrix inversion lemma. Its technical essence lies in utilizing the following physical law: when the anomaly only changes the stiffness of the local mesh element, most of the load-bearing skeleton of the global structure does not change in essence. Therefore, it is not necessary to overturn the previously obtained inverse global stiffness matrix. It is only necessary to add the incremental impact brought by the anomaly region in a dimensionality reduction manner. The inverse term of the local defect dimension reduction represents the additional influence channel of this small abnormal region on the overall structural response; after the system combines this additional influence with the original pre-stored global stiffness matrix inverse matrix, the updated global displacement field vector can be obtained. The following is an example illustrating a specific application scenario: Assume that most areas in the baseline model are in normal condition, with only local elements S1 and S2 experiencing stiffness changes due to fiber angle deviations; the traditional approach would recalculate the entire model from G1 to Gn. In this embodiment, only the low-maintenance positive terms related to S1 and S2 are extracted to form a local inverse object with a scale much smaller than the full model, and then its effect is reflected back to the pre-stored global displacement field vector. The specific local calculation and combination logic is as follows: Let the inverse of the pre-stored global stiffness matrix be... The product of the low-rank matrices is denoted as ,in Describes the first low-dimensional matrix. Describes the second low-dimensional matrix. This represents the transpose of the second low-dimensional matrix; the system generates the local defect dimensionality reduction inverse term by the local product of the target dimension determined by the affected degrees of freedom:

[0025] in, Let represent the identity matrix corresponding to the dimension of this local product. The dimension of the inverse term for this local defect dimension reduction depends only on the affected degrees of freedom specific to this local region, thus reducing the computational resource complexity of the inversion. The system generates a direct update operation term based on the matrix inversion lemma and this term:

[0026] The direct update operation item is then compared with the inverse of the pre-stored global stiffness matrix. The combined method performs subtraction compensation correction on the load vector, and directly calculates the updated global displacement field vector; This correction method fully preserves the mechanical information of non-abnormal areas because these areas are still supported by the pre-stored global stiffness matrix inverse matrix, with only a small amount of local compensation added. As a fault-tolerant mechanism for abnormal situations, if the spatial distribution of local abnormal regions is too discrete, or multiple consecutive defect bands are coupled with each other, causing the scale of the dimension reduction inversion term to increase rapidly, the system can first cluster and partition the abnormal regions, and then correct them separately according to the partitions. If low-dimensional features cannot be maintained after partitioning, it indicates that the current anomaly has evolved from local perturbation to mesoscale structural change. At this time, the system should switch to local sub-model reconstruction or even global reconstruction to ensure the credibility of the results, rather than mechanically adhering to the direct correction mode. In the same batch of pipe fittings, the system detected a local defect in a section of continuous tape due to tension decay and path deviation; the only actual affected areas were the lamination direction and compaction state of several adjacent units. After the local perturbation correction unit calls the preset local stiffness matrix direct correction formula, it does not need to resolve the entire high-pressure fiberglass pipe. Instead, based on the previously stored global stiffness matrix inverse, it only performs dimensionality reduction processing on the additional influence corresponding to the abnormal area of ​​the shoulder, and then outputs the updated global displacement field vector. In this way, the system can provide an updated structural response even before the physical entanglement has ended; The purpose of this mechanism is to limit the mechanical effects of local anomalies to a manageable, small-scale correction range, thereby achieving compatibility between high-fidelity finite element simulation and real-time synchronization requirements.

[0027] In a preferred embodiment of the present invention, the low-rank matrix product is formed by multiplying the transposes of the first low-dimensional matrix and the second low-dimensional matrix, wherein the first low-dimensional matrix represents the influence distribution of the local mesh degree of freedom association, and the second low-dimensional matrix represents the corresponding state feedback; the local perturbation correction unit maps the current fiber landing point spatial three-dimensional coordinates to the local mesh elements divided in the reference finite element model, and determines the number of columns of the first low-dimensional matrix and the second low-dimensional matrix based on the number of degrees of freedom of the local mesh elements.

[0028] This embodiment provides a local mesh mapping and low-dimensional construction mechanism. Specifically, in the previous embodiment, although it has been explained that direct correction can be performed on abnormal regions, if the landing point obtained by the sensor cannot be accurately mapped to the local mesh element in the finite element model, the low-rank correction will lose its effective position, causing the abnormal influence to be projected to the wrong region, thus destroying the credibility of the simulation. The local perturbation correction element first maps the spatial three-dimensional coordinates of the current fiber landing point to the surface of the reference finite element model and finds the corresponding local mesh element. The above mapping process has clear engineering significance: the sensor and kinematic module describe the landing point of the fiber bundle on the surface of the real mandrel, while the finite element model describes the discretized load-bearing structure. A point-to-element placement relationship must be established between the two. After finding the local unit, the system determines the number of columns of the first low-dimensional matrix and the second low-dimensional matrix based on the number of degrees of freedom of the local unit and its adjacent coupled units. The essential meaning is: if the anomaly only affects a small number of node degrees of freedom, the low-dimensional matrix only needs to cover these affected degrees of freedom; if the anomaly region crosses the interface between adjacent layers or the curvature change region, the coverage should be appropriately expanded to avoid only correcting the surface layer and missing the interlayer transmission effect. As a specific simplified example: Suppose that the landing point P is mapped to the mesh cell E5, which together with the adjacent cells E4 and E6 undertakes the layering transition at that point, then the system does not include all degrees of freedom of the entire model in the correction, but only constructs the first low-dimensional matrix and the second low-dimensional matrix around the local degrees of freedom related to E4 to E6. Specifically, the first low-dimensional matrix contains the influence distribution vector of the local mesh degrees of freedom association, and the second low-dimensional matrix contains the corresponding state feedback vector. The system constructs the aforementioned low-rank matrix product that can describe local stiffness changes by multiplying the first low-dimensional matrix with the transpose of the second low-dimensional matrix; if P is near the boundary of two elements, it can be assigned to both elements simultaneously according to proximity weights to avoid the problem of the landing point being located at the boundary and being incorrectly associated on one side. As a fault-tolerant mechanism for abnormal situations, if the landing point coordinates are mapped outside the model boundary, it usually indicates that there is an anomaly in the geometric parameters, sensor calibration, or time alignment. At this time, the system should mark the moment as a mapping failure and suspend the local stiffness correction of the landing point. If the landing point is located in the same curvature region at the tip of the end cap, a single surface unit is insufficient to represent the actual contact bandwidth of the fiber bundle. Therefore, the landing point can be expanded into a contact patch, and influence weights can be assigned to multiple local units to improve the stability of contact modeling. In the aforementioned high-pressure fiberglass pipe end cap shoulder, the fiber bundle falls at a position with rapid curvature transition due to the change in guide wire posture; the system maps the three-dimensional coordinates of this landing point to the local grid cells of the end cap transition zone, and identifies that the strip-shaped region involves several surface and subsurface cell degrees of freedom; Using these local degrees of freedom as the scope, construct the first low-dimensional matrix and the second low-dimensional matrix, then transpose and multiply them to generate a low-rank matrix product, so that subsequent direct corrections only act on the truly affected limited area, rather than spreading to the entire tube. The purpose of this mechanism is to establish a traceable correspondence between the actual landing point and the discrete grid, and to compress and correct the dimension accordingly, thereby achieving a balance between accurate local anomaly localization and controllable computational burden.

[0029] In a preferred embodiment of the present invention, when performing spatial geometric transformation operations, the kinematic mapping fast loop module uses preset core mold radius parameters and kinematic mapping equations that map time, displacement, and rotation angles to spatial coordinates to map the standard time-series dataset collected based on the time dimension to a three-dimensional cylindrical coordinate system, generating a point cloud coordinate set containing spatial topological relationships.

[0030] This embodiment provides a geometric mapping mechanism from the time domain to the spatial domain; specifically, in the aforementioned scheme, the fast loop needs to convert the continuously acquired time series into a spatial trajectory that can be processed by the mechanical model; This embodiment further clarifies that the mapping is not a single-point coordinate conversion, but rather generates a point cloud coordinate set containing spatial topological relationships, which is used to describe the continuous direction of the fiber laying path, the connection relationship between adjacent points, and the local layer coverage range. The spindle rotation angle reflects the circumferential position of the fiber rotating around the mandrel, the carriage displacement reflects the position of the fiber advancing along the axis, and the mandrel radius constrains both of these quantities to the actual winding surface. After mapping the standard time series dataset to a three-dimensional cylindrical coordinate system, the system obtains not isolated spatial points, but a series of point cloud trajectories arranged in the order of sampling. The adjacent points in this point cloud naturally have a topological relationship with the previous and next landing points, so it can be used to identify whether the fiber path is continuous, whether there are jump points, and whether there is local congestion or sparseness in the end cap transition zone. This spatial topological information is crucial for understanding local defects because, for the same angular deviation, if it occurs in a straight cylindrical section, the stress distribution gradient changes less; but if it occurs in a high-curvature transition zone, it may cause uneven interlayer fiber overlap and abrupt changes in local thickness. As a specific simplified example: assuming that the principal axis angles and trolley displacements corresponding to three adjacent sampling points in the standard timing sequence increase continuously, then after mapping, points P1, P2, and P3 are formed, and P1 is adjacent to P2 and P2 is adjacent to P3, which indicates that the fiber path moves forward continuously. If two points that are adjacent in time are abnormally far apart after being mapped in space, it usually means that there is a sampling error, slippage, or geometric parameter mismatch, and further verification is required. As a fault tolerance mechanism for abnormal situations, if the core mold radius parameter does not match the current workpiece specification, the mapped point cloud will deviate from the real surface as a whole, causing the landing point coordinates to lose correspondence with the finite element model. Therefore, matching core mold geometric parameters must be loaded before each changeover production. If the current workpiece is not a pure cylinder, but contains end caps, conical sections, or transition surfaces, the three-dimensional cylindrical coordinate system can be used as an intermediate representation. Then, combined with regionalized surface parameters, a secondary projection is completed to prevent the landing point from drifting in complex curved surface areas. When the high-pressure fiberglass pipe is wound on the same production line, the point cloud trajectory of the cylindrical section is a regular spiral band, while the curvature of the point cloud is significantly enhanced after entering the head transition section. After mapping the time series into a spatial point cloud using the preset core mold radius and corresponding kinematic equations, the system can clearly identify the increased trajectory density in a certain strip-shaped area on the shoulder of the end cap, indicating that the fiber layup has localized aggregation at that location; this information is sent to the slow loop to explain the possible local stiffness enhancement or stress concentration in that area. The purpose of this mechanism is to transform the original process sampling sequence into fiber laying trajectories with actual geometric meaning, thereby achieving a unified data foundation for subsequent angle evaluation, stress reconstruction, and local mesh positioning.

[0031] In a preferred embodiment of the present invention, the mechanical simulation slow loop module further includes a performance defect evaluation unit, which calculates the local stress and strain distribution based on the updated global displacement field vector and extracts the characteristic values ​​of the local fiber micro-buckling evolution state. When the characteristic value of the local fiber micro-buckling evolution state exceeds the preset boundary of the material failure envelope surface based on multiaxial stress state, the performance defect assessment unit outputs a defect warning signal containing the predicted value of the burst pressure reduction; when the characteristic value of the local fiber micro-buckling evolution state does not exceed the boundary of the material failure envelope surface, the performance defect assessment unit outputs a normal state signal.

[0032] This embodiment provides a performance defect assessment and early warning mechanism; specifically, after the aforementioned scheme can obtain the updated global displacement field vector, if it only stays at the level of angular deviation or local stiffness change, it is still insufficient to guide process review and product quality judgment. The key engineering assessment lies in further determining whether this local anomaly will evolve into a pressure-bearing performance defect, especially whether it will lead to a drop in the burst pressure of high-pressure FRP pipe fittings; The performance defect assessment unit obtains the local stress and strain distribution based on the updated global displacement field vector and extracts the characteristic values ​​of the local fiber micro-buckling evolution state. The aforementioned microbuckling is not an abstract indicator, but rather an early sign of instability in composite materials under the combined effects of compression, shear, and insufficient interlaminar constraint. For fiberglass structures that are still in the winding stage, if the outer layer tension is too high, the inner layer support is insufficient, or the fiber angle deviates, causing the main load direction to be misaligned, microscale bending and corrugation may form locally. The system compares the characteristic value of the local fiber micro buckling evolution state with the preset material failure envelope boundary based on multiaxial stress state. Its physical meaning is to determine whether the current multiaxial stress state is still within the material's tolerance range. Among them, the material failure envelope boundary based on multiaxial stress state is constructed based on the Cai-Wu tensor failure criterion or the Hasing composite material damage criterion to accurately characterize the stress critical state under multiple failure modes of fiber tension, fiber compression, matrix tension and matrix compression. When the characteristic value of the local fiber microbuckling evolution state exceeds the boundary, the performance defect assessment unit outputs a defect warning signal, which not only indicates the existence of the abnormality, but also includes the predicted value of the burst pressure reduction. The reason for outputting this predicted value is that, for composite material structures such as pressure vessels, the ultimate concern is not simply geometric errors, but whether their pressure-bearing limit has been weakened. To calculate the specific predicted value of the burst pressure reduction, the system does not rely on subjective human experience reduction coefficients, but executes a prediction process based on damage mechanics: the system converts the extracted difference amount that exceeds the boundary of the material failure envelope into a local stiffness reduction factor, and performs material stiffness softening treatment on the abnormal local area. The system then drives the slow loop to perform a burst limit simulation with increasing load steps until the updated model exhibits a singularity or the updated global displacement field vector diverges. The system then extracts the corresponding critical value for internal pressure failure at this point. The difference between the failure threshold and the theoretical reference burst pressure specified in the design of this batch of pipe fittings is the predicted burst pressure reduction value generated based on the objective deduction of the current micro buckling defects. If the characteristic value of the local fiber micro-buckling evolution state does not exceed the boundary, the output state is normal, indicating that although there are local process fluctuations, they have not yet caused significant damage to the final pressure bearing capacity. As a specific simplified example: Suppose that a local area A has fiber angle deviation, but its stress redistribution is still limited to the surface layer and the interlaminar shear is not significantly amplified, then the evaluation result can remain normal; If local region B simultaneously experiences insufficient residual tension and abrupt curvature change, leading to a deflection of stress direction and inducing coordinated instability of the surface and subsurface layers, the characteristic value of the local fiber micro-buckling evolution state will cross the boundary of the material failure envelope surface based on the multiaxial stress state. The system will output an early warning and mark B as a key area for review. As a fault-tolerant mechanism for abnormal situations, if the quality of the updated global displacement field vector is insufficient, for example, if there is a long period of missing upstream input or the confidence of local correction is too low, the performance defect assessment unit should output a state to be reviewed instead of directly giving a deterministic prediction of blasting pressure. If the material database lacks the corresponding fiber volume fraction or failure envelope parameters in the cured state, it can temporarily revert to the conservative judgment mode, prioritizing the output of risk level and abnormal location to avoid giving distorted quantitative conclusions. In the aforementioned abnormal area of ​​the head shoulder, after the slow loop updates the global displacement field vector, it is found that the interlayer shear and local compression coupling effects in this area are enhanced, and the characteristic value of the local fiber micro buckling evolution state exceeds the boundary of the material failure envelope surface based on the current multiaxial stress state. The performance defect assessment unit then outputs an early warning, indicating that the location may become a sensitive point for burst initiation, and provides a trend of decreasing pressure bearing capacity of the batch of pipe fittings relative to the design baseline; if another cylindrical section has a slight angular deviation, but the stress and strain distribution is still within the safe range, the performance defect assessment unit marks it as normal and only includes it in the offline review. The purpose of this mechanism is to ultimately translate process deviations into structural performance information that can be used for quality assessment, thereby achieving a closed-loop connection from process monitoring to pressure risk prediction.

[0033] In a preferred embodiment of the present invention, the reference finite element model is constructed based on the constitutive equation of a three-dimensional orthogonal anisotropic material; the constitutive equation of a three-dimensional orthogonal anisotropic material establishes the mathematical mapping relationship between macroscopic mechanical performance parameters and microscopic fiber volume fraction and design winding angle, and pre-calculates and generates a pre-stored global stiffness matrix inverse matrix.

[0034] This embodiment provides a benchmark finite element model construction mechanism. Specifically, in the aforementioned scheme, the slow loop can perform regular refresh and local correction, but the premise is that the benchmark finite element model must be close enough to the actual material behavior of high-pressure fiberglass pipe fittings. If such composite materials are simplified to isotropic materials, even if the spatial landing point and tension estimation are correct, the final stiffness distribution and pressure prediction will deviate from reality. The baseline finite element model is constructed using the constitutive equation of a three-dimensional orthogonal anisotropic material. This is because the mechanical properties of wound fiberglass pipes differ significantly in the fiber direction, transverse direction, and thickness direction, and this difference is directly affected by the fiber volume fraction and the design winding angle. The fiber volume fraction determines the density of the load-bearing skeleton, and the design winding angle determines the matching relationship between the main force direction and the compressive load transmission direction. By establishing the mapping relationship between macroscopic mechanical performance parameters and these microscopic structural factors, the benchmark model can express the initial stiffness characteristics of different ply regions before startup and generate the inverse global stiffness matrix in advance for quick use during the production process. The above pre-calculations are not divorced from actual manufacturing conditions, but rather are the initial mechanical basis completed based on the established design layup scheme, material system and geometric boundaries of this type of pipe fitting; During the subsequent online process, the system does not build the entire structure from scratch, but rather superimposes local disturbances caused by the actual process on this chassis. This is the only way to move the large-scale computation to the design preparation stage. As a fault-tolerant mechanism for abnormal situations, if the production batch changes the fiber grade, resin system or target design angle, but continues to use the old benchmark model, all subsequent online simulations will introduce systematic errors. Therefore, whenever there is a substantial change in the process formula, the benchmark model should be rebuilt and the pre-stored inverse matrix should be updated. If the current workpiece is in the stage where the resin is not fully cured and it is necessary to consider time-varying material properties, a staged material parameter table can be introduced into the benchmark model to distinguish the different response basis between the initial winding state and the post-curing state. In the preparation stage of this 70MPa high-pressure fiberglass pipe fitting, the process engineer first established the orthogonal anisotropic material properties of each region based on the design angles of the circumferential layer, spiral layer and end cap transition layer of the cylindrical section, combined with the target fiber volume fraction, and generated a benchmark finite element model. In this way, when an angular deviation is detected in the end cap shoulder in subsequent online tests, the system corrects the local stiffness based on the real laminated structure, rather than making approximations on an oversimplified homogeneous model. The purpose of this mechanism is to provide an initial mechanical foundation for online digital twins that is consistent with the actual load-bearing mechanism of composite materials, so as to enable subsequent local correction results to have engineering interpretability and performance prediction credibility.

[0035] In a preferred embodiment of the present invention, the system is deployed in a hierarchical network architecture that includes edge computing nodes and cloud servers; the physical data parsing module and the kinematic mapping fast loop module reside on the edge computing nodes to perform low-latency data acquisition and spatial coordinate mapping; The slow-loop mechanical simulation module resides on the cloud server and performs asynchronous mechanical reconstruction based on the current three-dimensional coordinates of the fiber landing point, the actual winding angle, and the residual tension value output by the fast-loop kinematic mapping module.

[0036] This embodiment provides a hierarchical network deployment mechanism; specifically, based on the division of labor between fast and slow rings in the aforementioned scheme, this embodiment further illustrates the deployment method of the system at the engineering site; If all modules are concentrated on the local industrial control computer on the production line, although the data transmission path is short, large-scale finite element reconstruction will crowd out real-time acquisition and mapping resources; conversely, if all raw sensor data are directly uploaded to the remote end for unified processing, network jitter will affect the timely expression of the landing point and angle by the fast loop. This embodiment adopts a layered architecture that coordinates edge computing nodes and cloud servers; the physical data parsing module and the kinematic mapping fast loop module reside on the edge computing node because these two tasks directly face the encoder and tension sensor, requiring data access, anomaly removal, spatial landing point calculation and angle estimation to be completed within the device cycle, with low latency and high continuity as the priority objectives. The slow loop module for mechanical simulation resides on a cloud server because local perturbation correction, displacement field update, and subsequent performance evaluation require strong computing resources and a larger model cache. The edge end asynchronously sends the three types of key results that have been sorted out, namely the three-dimensional coordinates of the current fiber landing point, the actual winding angle, and the residual tension value, to the cloud. The cloud performs mechanical reconstruction based on these results without having to receive all the original high-frequency waveforms. As a specific simplified example: edge nodes can organize continuous sampling into records R1, R2, R3, with each record retaining only key features such as landing point position, angle, and residual tension; When the cloud receives R2, it can determine whether a local correction has been triggered without waiting for all the original electrical signals from R1 to Rn to be transmitted. This cloud-based approach reduces network load and avoids transmitting noise and redundant data along with the signal. As a fault-tolerant mechanism for abnormal situations, if a short-term disconnection occurs between the edge node and the cloud, the edge node can continue to perform local data collection and spatial mapping, and temporarily store key features in the cache, and retransmit them in batches after the network is restored. During the chain break, the temporal continuity of the cloud-based mechanical results will be interrupted. At this time, the system should mark the corresponding time period as pending calculation. If insufficient cloud computing resources cause slow loop queuing, the system can schedule according to the abnormal priority, prioritizing the processing of the head transition section, high curvature area and tension abnormal area, while delaying the processing of the normal cylindrical section. During the continuous night shift production of this batch of high-pressure fiberglass pipe fittings, edge nodes were deployed next to the winding machine to collect spindle angle, trolley displacement and tension information in real time, and complete spatial mapping in milliseconds; the cloud server was located in the enterprise data center to perform parallel mechanical reconstruction of abnormal areas from multiple winding machines. When the aforementioned abnormality occurs at the shoulder of the end cap, the edge end immediately uploads the spatial coordinates, actual winding angle, and residual tension at that location. The cloud then initiates local correction and returns the performance risk label at that location for quality engineers to view. The purpose of this mechanism is to reasonably allocate latency and computing power requirements according to different tasks, so as to achieve both the stability of on-site data acquisition and the scalability of complex simulations.

[0037] In a preferred embodiment of the present invention, the timing status data includes spindle rotation angle data and trolley displacement data; the encoder connected to the physical data parsing module is an absolute encoder, the physical data parsing module obtains the spindle rotation angle data and trolley displacement data through the absolute encoder, and performs timestamp alignment processing on the spindle rotation angle data and trolley displacement data to eliminate acquisition delay error.

[0038] This embodiment provides a state acquisition and time alignment mechanism based on an absolute encoder; specifically, in the aforementioned spatiotemporal mapping scheme, if the spindle rotation angle and the trolley displacement cannot be interpreted under the same time reference, even if each measurement is accurate, the final actual winding angle and landing position will still deviate. Especially in the transition section of the end cap, the coordination between the main shaft and the trolley is more sensitive, and even millisecond-level time delay errors can be amplified into spatial trajectory deviation; The spindle rotation angle data reflects the fiber unfolding progress in the circumferential direction, and the carriage displacement data reflects the fiber advancing progress in the axial direction. In this embodiment, an absolute encoder is used to obtain these two types of data. Compared with incremental methods, absolute encoders can directly recover the current position after power failure and restart or short-term communication interruption, making them more suitable for continuous winding, which is sensitive to the cumulative error of the trajectory. The physical data parsing module performs timestamp alignment processing on the two types of data, so that the rotation angle and displacement at the same moment can correspond to the same real process instant; its engineering significance is to prevent the spindle information from being one step ahead and the carriage information from being one step behind, thereby misjudging a fiber path that was originally regular as a skewed path. As a specific simplified example: Suppose that the main axis angle at a certain moment belongs to time T1, while the displacement of the trolley actually comes from a later time T2 due to the delay in the acquisition link. If these two pieces of data are directly combined, a false landing point P that does not exist will be formed. After timestamp alignment, the system only combines angles and displacements from the same time base to restore the true landing point; if one side of the data is temporarily missing, it can wait for the other side to fill in the missing data within a short window, and then process it as a missing sample after the window expires. As a fault-tolerant mechanism for abnormal situations, if the absolute encoder experiences continuous drift or communication abnormalities, the system should perform cross-verification through the device zero-return reference bit, mechanical limit bit, or backup measurement channel. If valid data for both the main shaft and the trolley cannot be obtained simultaneously within the alignment window, the sampling period should not be forcibly used for spatial mapping to avoid introducing systematic pseudo-trajectories. For high-speed winding conditions, insufficient time alignment accuracy may also cause the winding angle to frequently jump across the threshold. Therefore, the system can add short-term trajectory continuity verification after alignment to filter out combination samples that obviously violate the smoothness of mechanical motion. When the high-pressure fiberglass pipe enters the head turning section, the main shaft continues to rotate while the trolley begins to decelerate and change direction. If the main shaft angle data arrives first and the displacement data arrives later, the unaligned system may misjudge that the fiber has deviated from the design path in advance and erroneously trigger local stiffness correction. After aligning the absolute encoder and timestamp in this embodiment, the system can accurately reconstruct the cooperative motion state at that moment, and trigger subsequent mechanical analysis only when the actual trajectory deviation occurs. The purpose of this mechanism is to ensure that the core motion quantities constituting the spatial trajectory have a unified time reference, thereby enabling reliable calculation of the actual winding angle and landing point coordinates.

[0039] In a preferred embodiment of the present invention, the operating frequency of the kinematic mapping fast loop module is synchronized with the data sampling frequency of the physical actuator; when the mechanical simulation slow loop module does not receive the trigger command of the local perturbation correction unit, it writes the actual winding angle and residual tension value output by the kinematic mapping fast loop module into the timing feature cache pool to generate feature data for offline process review.

[0040] This embodiment provides a synchronous sampling and offline replay caching mechanism; specifically, in the aforementioned scheme, the fast loop is responsible for high-frequency mapping, and the slow loop is responsible for on-demand reconstruction; However, during a large number of normal winding periods, the slow loop will not be continuously triggered; if the output of the fast loop is directly discarded at this time, the continuous record of the entire process history will be lost, which is not conducive to the subsequent analysis of the evolution process before and after the formation of a certain defect. The operating frequency of the kinematic mapping fast loop is synchronized with the data sampling frequency of the physical actuator, so that each sampled data can be interpreted in position, angle and tension within the same beat; This avoids the timing misalignment where the device has acquired a new state, but the fast loop is still processing the old state; for periods when local perturbation correction is not triggered, the slow loop is not completely idle, but writes the actual winding angle and residual tension value output by the fast loop into the timing feature buffer pool. This cache pool stores process features that provide important diagnostic basis for subsequent analysis, rather than all original waveforms. Its purpose is to support offline process review, such as tracing the tension decay trend, angle drift accumulation process, and warning intervals before anomalies occur in a sample that failed a burst test during the winding process. As a specific simplified example: Suppose that in a continuous sampling period, records C1, C2, and C3 do not exceed the angle distortion threshold, then the system will not initiate local mechanical correction, but will write the corresponding actual winding angle and residual tension into the buffer pool in sequence. If C4 triggers an anomaly in subsequent records, engineers can not only view C4 during offline playback, but also trace whether C1 to C3 have shown a slow offset trend; in this way, the system saves online computing resources and preserves the complete process causal chain. As a fault tolerance mechanism for abnormal situations, if the cache pool capacity is close to the upper limit, the system can perform rolling archiving by time fragmentation, layer number fragmentation, or workpiece number fragmentation, and prioritize the retention of critical windows before and after the occurrence of the abnormality. If sampling synchronization is disrupted, for example, if the fast loop period is lower than the sensor sampling frequency, the clock synchronization should be restored first before writing to the buffer; otherwise, the temporal continuity of the buffered data will be compromised. During network outages or peak cloud usage periods, edge nodes can cache the feature data locally first, and then upload and archive it after communication is restored. During the winding of the cylindrical section of the high-pressure fiberglass pipe, the actual winding angle of most sampling points was stable and the residual tension was normal. Therefore, the slow loop did not frequently trigger local corrections. However, the system still writes these features into the cache pool at a synchronous beat; when an anomaly occurs in the end cap shoulder and triggers an alert, the quality engineer can review the feature evolution of the previous tens of seconds and find that the residual tension has actually been slowly decaying, but has not yet exceeded the trigger threshold; thus, it can be determined that the anomaly is not an instantaneous occurrence, but is caused by the gradual change in the friction state of the guide path. The purpose of this mechanism is to preserve the historical characteristics of the process while ensuring online real-time performance, thereby achieving a data closed loop for anomaly tracing, process optimization, and sample quality review.

[0041] To verify the effectiveness of this simulation system, a comparative verification test was conducted on the continuous layup winding process of 70MPa high-pressure FRP pipe fittings. Test data shows that when the actual winding angle deviation exceeds the threshold and triggers the local perturbation correction unit, the average time for single mechanical reconstruction using the local stiffness matrix direct correction algorithm is reduced from 1250ms in the traditional global recalculation model to approximately 45ms, meeting the low-delay synchronous simulation requirements of high-frequency production line cycles. Simultaneously, a water pressure burst destructive test was conducted on the sample that outputs the system's performance defect warning signal. The predicted burst pressure reduction value given by the system is assumed to be... The actual explosion pressure value measured in the experiment was Its relative error According to the formula Calculations were performed, and statistical analysis of multiple sets of continuous tests showed the relative error. All deviations remained within 3.2%. Furthermore, compared to traditional global finite element analysis, the local stress field calculation deviation caused by the dimensionality reduction and inversion term update of the global displacement field in this system does not exceed 1.5%, proving that this system can accurately translate process deviations into pressure risk prediction indicators without sacrificing calculation fidelity.

[0042] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A fully automated layup and winding simulation system for high-pressure fiberglass pipe fittings based on digital twins, characterized in that, The system includes: a physical data parsing module, a kinematic mapping fast loop module, and a mechanical simulation slow loop module; The physical data parsing module connects to the encoder and tension sensor of the physical actuator, which includes a main shaft and a carriage. It collects the time-series state data of the rotation of the main shaft and the translation of the carriage, as well as the time-series tension data, and performs a data cleaning operation to remove outliers, generating a standard time-series dataset. The kinematic mapping fast loop module performs spatial geometric transformation operations on the standard time-series dataset based on the established three-dimensional cylindrical coordinate system to calculate the three-dimensional spatial coordinates of the current fiber landing point and the actual winding angle; and calculates the residual tension value at the current fiber landing point based on the time-series tension data and the tension transmission attenuation model including the friction coefficient and the contact wrap angle. The mechanical simulation slow loop module includes a reference finite element model and a local perturbation correction unit. The reference finite element model is pre-initialized and generates a pre-stored global stiffness matrix inverse matrix, load vector, and global displacement field vector. The mechanical simulation slow loop module calculates the absolute value of the angular deviation between the actual winding angle and the design winding angle preset based on the pipe fitting design parameters. When the absolute value of the angle deviation is greater than the preset angle distortion judgment threshold, the local perturbation correction unit is triggered. Based on the current three-dimensional coordinates of the fiber landing point and the residual tension value, the local stiffness change matrix caused by the fiber deviation is calculated, the local stiffness change matrix is ​​decomposed into a low-rank matrix product, and the global displacement field vector of the reference finite element model is updated based on the local stiffness matrix direct correction algorithm containing the matrix inversion lemma. When the absolute value of the angle deviation is not greater than the angle distortion judgment threshold, the state of the reference finite element model remains unchanged, and the pre-stored global stiffness matrix inverse matrix is ​​multiplied with the load vector to perform conventional displacement field calculation to update the global displacement field vector.

2. The fully automated layup and winding simulation system for high-pressure fiberglass pipe fittings based on digital twins as described in claim 1, characterized in that, The physical data parsing module also includes a time-series noise reduction unit, which performs Kalman filtering on the time-series tension data to extract the real tension sequence, eliminate mechanical resonance noise interference, and fuse the real tension sequence into the standard time-series dataset.

3. The fully automated layup and winding simulation system for high-pressure fiberglass pipe fittings based on digital twins as described in claim 1, characterized in that, When executing the local stiffness matrix direct correction algorithm, the local perturbation correction unit calls a preset local stiffness matrix direct correction formula based on the matrix inversion lemma. Based on the product of the pre-stored global stiffness matrix inverse and the low-rank matrix, it calculates the local defect dimension reduction inverse term and combines the local defect dimension reduction inverse term with the pre-stored global stiffness matrix inverse to solve for the updated global displacement field vector.

4. The fully automated layup and winding simulation system for high-pressure fiberglass pipe fittings based on digital twins as described in claim 1, characterized in that, The low-rank matrix product is formed by multiplying the transpose of the first low-dimensional matrix and the second low-dimensional matrix, wherein the first low-dimensional matrix represents the influence distribution of the local mesh degree of freedom association, and the second low-dimensional matrix represents the corresponding state feedback. The local perturbation correction unit maps the current fiber landing point spatial three-dimensional coordinates to the local mesh elements divided in the reference finite element model, and determines the number of columns of the first low-dimensional matrix and the second low-dimensional matrix based on the number of degrees of freedom of the local mesh elements.

5. The fully automated layup and winding simulation system for high-pressure fiberglass pipe fittings based on digital twins as described in claim 1, characterized in that, When performing the spatial geometric transformation operation, the kinematic mapping fast loop module uses preset core mold radius parameters and kinematic mapping equations that map time, displacement, and rotation angles to spatial coordinates to map the standard time-series dataset collected based on the time dimension to a three-dimensional cylindrical coordinate system, generating a point cloud coordinate set containing spatial topological relationships.

6. The fully automated layup and winding simulation system for high-pressure fiberglass pipe fittings based on digital twins according to claim 1, characterized in that, The mechanical simulation slow loop module also includes a performance defect evaluation unit, which calculates the local stress and strain distribution based on the updated global displacement field vector and extracts the characteristic values ​​of the local fiber micro-buckling evolution state. When the characteristic value of the local fiber microbuckling evolution state exceeds the preset material failure envelope boundary based on multiaxial stress state, the performance defect evaluation unit outputs a defect warning signal containing the predicted value of burst pressure reduction. When the characteristic value of the local fiber microbuckling evolution state does not exceed the boundary of the material failure envelope, the performance defect evaluation unit outputs a normal state signal.

7. The fully automated layup and winding simulation system for high-pressure fiberglass pipe fittings based on digital twins as described in claim 1, characterized in that, The reference finite element model is constructed based on the constitutive equation of a three-dimensional orthogonal anisotropic material; The constitutive equation of the three-dimensional orthogonal anisotropic material establishes a mathematical mapping relationship between macroscopic mechanical performance parameters, microscopic fiber volume fraction, and the designed winding angle, and pre-calculates and generates the inverse of the pre-stored global stiffness matrix.

8. The fully automated layup and winding simulation system for high-pressure fiberglass pipe fittings based on digital twins according to claim 1, characterized in that, The system is deployed in a layered network architecture that includes edge computing nodes and cloud servers; The physical data parsing module and the kinematic mapping fast loop module reside on the edge computing node, performing low-latency data acquisition and spatial coordinate mapping. The mechanical simulation slow loop module resides on the cloud server and performs asynchronous mechanical reconstruction based on the current fiber landing point spatial three-dimensional coordinates, the actual winding angle, and the residual tension value output by the kinematic mapping fast loop module.

9. The fully automated layup and winding simulation system for high-pressure fiberglass pipe fittings based on digital twins as described in claim 1, characterized in that, The timing status data includes spindle rotation angle data and trolley displacement data; The encoder connected to the physical data parsing module is an absolute encoder. The physical data parsing module obtains the spindle rotation angle data and the trolley displacement data through the absolute encoder, and performs timestamp alignment processing on the spindle rotation angle data and the trolley displacement data to eliminate acquisition delay errors.

10. The fully automated layup and winding simulation system for high-pressure fiberglass pipe fittings based on digital twins according to claim 1, characterized in that, The operating frequency of the kinematic mapping fast loop module is synchronized with the data sampling frequency of the physical actuator. When the mechanical simulation slow loop module does not receive the trigger command from the local perturbation correction unit, it writes the actual winding angle and the residual tension value output by the kinematic mapping fast loop module into the time-series feature cache pool to generate feature data for offline process review.

Images