Intelligent glass fiber drum unloading method and system based on digital twinning

Abstract

The invention discloses an intelligent glass fiber drum unloading method and system based on digital twinning, and the method comprises the steps: collecting a yarn roll forming signal of a fiber forming machine, the positioning coordinates of a creel vehicle and the data of a weighing sensor in real time, and generating a multi-source synchronous data set; based on the multi-source synchronous data set, the three-dimensional coordinates of the center of the yarn roll and the lifting distance of the manipulator are solved, and a grabbing parameter set is output; the grabbing parameter set is issued to a servo control system, a manipulator is driven to complete yarn ball grabbing-weighing-placing operation, and a physical execution verification data packet is synchronously generated; and the motion trail of the manipulator and the stock state of the creel car are reconstructed in real time in the digital twin virtual model, and an abnormal early warning instruction is triggered and pushed to operation and maintenance personnel. By means of the embodiment of the invention, accurate positioning and real-time deviation correction in the yarn ball grabbing process can be achieved, and the reliability and production efficiency of drum unloading operation are improved.

Description

Technical Field

[0001] This invention belongs to the field of digital twin technology, and in particular to a method and system for intelligent unloading of glass fiber tubes based on digital twins. Background Technology

[0002] In the glass fiber production process, the unloading of yarn bundles has long relied on manual operation or semi-automated equipment, resulting in problems such as low positioning accuracy, insufficient efficiency, and difficulties in data traceability. Traditional robotic gripping systems, lacking real-time data feedback, are prone to gripping deviation or collision risks and cannot dynamically optimize the work process. Although digital twin technology has been applied in the field of intelligent manufacturing, a complete solution for multi-source data synchronous acquisition, virtual real-time verification, and physical execution closed-loop control has not yet been formed in the glass fiber unloading scenario. Existing methods struggle to coordinate the control of yarn bundle morphology changes and the dynamic positioning of the yarn carrier, leading to production interruptions or quality fluctuations. Summary of the Invention

[0003] The purpose of this invention is to provide a digital twin-based intelligent unloading method and system for glass fiber, which addresses the shortcomings of existing technologies and enables precise positioning and real-time correction during the yarn ball grasping process, thereby improving the reliability and production efficiency of unloading operations.

[0004] One embodiment of this application provides a digital twin-based intelligent unloading method for glass fiber, the method comprising: The fiber forming machine's yarn ball forming signal, yarn frame car positioning coordinates, and weighing sensor data are collected in real time by photoelectric sensors and PLC controllers to generate a multi-source synchronous dataset with timestamps. Based on the multi-source synchronous dataset, the three-dimensional coordinates of the yarn ball center and the lifting distance of the robotic arm are calculated by the core positioning algorithm, and the grasping parameter set is output. The grasping parameter set is sent to the servo control system to drive the robot arm to complete the yarn ball grasping-weighing-placement operation. Simultaneously, the equipment status, weight data, and position coordinates during the operation are uploaded to the digital twin virtual model to generate a physical execution verification data package. In the digital twin virtual model, the movement trajectory of the robotic arm and the inventory status of the yarn rack are reconstructed in real time. The physical execution verification data is compared with the preset execution constraints. If the physical execution verification data deviates from the preset execution constraints, an abnormal warning instruction is triggered and pushed to the operation and maintenance personnel.

[0005] Optionally, the step of collecting yarn ball forming signals, yarn frame car positioning coordinates, and weighing sensor data from the fiber forming machine in real time via photoelectric sensors and a PLC controller to generate a timestamped multi-source synchronous dataset includes: A laser interferometric grid positioning system is deployed on the yarn frame car track to emit phase-modulated laser beams in real time. The displacement of the yarn frame car is calculated through the reflected signals, and the original positioning coordinate stream is output. A high frame rate CMOS photoelectric sensor is installed at the outlet of the fiber forming machine to capture changes in the microstructure of the yarn ball surface, and a morphological edge detection algorithm is combined to generate a pulse signal of the yarn ball forming state. The weighing sensor data stream, the original positioning coordinate stream, and the forming pulse signal are input into the encrypted PLC gateway, and a hardware timestamp module driven by a cold atom clock is embedded to output encrypted data packets with nanosecond-level time tags. The encrypted data packets are reassembled in the time domain by a timestamp alignment engine to eliminate transmission jitter and generate a multi-source synchronized dataset containing four-dimensional data including timestamp, coordinates, weight, and forming status.

[0006] Optionally, based on the multi-source synchronous dataset, the three-dimensional coordinates of the yarn ball center and the lifting distance of the robotic arm are calculated using a core positioning algorithm, and a set of grasping parameters is output, including: Extract the yarn ball forming state pulse signal from the multi-source synchronous dataset, identify the yarn ball boundary feature points through the wavelet ridge detection algorithm, and output the boundary point cloud cluster. Input the boundary point cloud cluster into the vibration-resistant least squares fitter, combine the yarn frame car positioning coordinates to compensate for the mechanical vibration offset, and output the optimized yarn ball center fitting normal vector. Based on the geometric relationship between the normal vector and the coordinate system of the robot arm, the opening and closing angle of the gripper and the normal angle of the contact surface are calculated, and the quaternion of the gripping posture is output. Based on the dynamic prediction of the yarn ball centroid offset using weighing sensor data, the anti-sway lifting trajectory equation is calculated through a rigid body dynamics model, and the lifting distance-time series is output. By fusing the grasping posture quaternion and the lift distance-time series, a grasping parameter set containing spatial coordinates, posture angles, lift curves, and clamping force thresholds is generated.

[0007] Optionally, the step of sending the grasping parameter set to the servo control system to drive the robotic arm to complete the yarn ball grasping-weighing-placement operation, and simultaneously uploading the equipment status, weight data, and position coordinates during the operation to the digital twin virtual model, generating a physical execution verification data package, including: The captured parameter set is compiled into machine instruction code, digital signature is attached, and then sent to the servo control system to trigger the robot arm action chain and output the execution log hash value. The robot arm collects joint encoder data, six-dimensional force sensor readings, and visual positioning feedback in real time. It detects execution deviations through an action sequence verification algorithm and outputs action compliance flags. When the compliance flag is true, the original verification data block is generated by encapsulating the device status, weight data, location coordinates, and execution log hash value. The original verification data block is compressed using a Merkle tree and uploaded to the digital twin to generate a physical execution verification data package with a blockchain fingerprint.

[0008] Optionally, the real-time reconstruction of the robotic arm's motion trajectory and the yarn rack inventory status in the digital twin virtual model, comparing the physical execution verification data with preset execution constraints, and triggering an anomaly warning command and pushing it to maintenance personnel if the physical execution verification data deviates from the preset execution constraints, includes: Load the physical execution verification data package into the digital twin virtual model to reconstruct the set of robot arm motion trajectory points and the three-dimensional heat map of yarn rack inventory; The actual trajectory point sequence and the virtual preset trajectory are input into a 3D convolutional neural network, and the trajectory deviation score is output. If the trajectory deviation score exceeds the preset score threshold, or the yarn ball weight deviation exceeds the preset deviation threshold, a primary abnormal event label is generated; When a primary anomaly event label is triggered, the multi-source data correlation analysis engine is started to calculate the standard deviation multiple of the path execution time and the historical mean. If it exceeds 2 times the standard deviation, it is upgraded to a severe anomaly event label. The system automatically pushes early warning instructions to operations and maintenance personnel based on the level of abnormal event labels, and simultaneously marks the abnormal points in the twin model and generates suggested correction solutions.

[0009] Another embodiment of this application provides a digital twin-based intelligent fiberglass unloading system, the system comprising: The data acquisition module is used to acquire the yarn ball forming signal, yarn frame car positioning coordinates and weighing sensor data of the fiber forming machine in real time through photoelectric sensors and PLC controller, and generate a multi-source synchronous dataset with timestamps. The output module is used to calculate the three-dimensional coordinates of the yarn ball center and the lifting distance of the robotic arm based on the multi-source synchronous dataset and the core positioning algorithm, and output the grasping parameter set. The drive module is used to send the grasping parameter set to the servo control system, drive the robot to complete the yarn ball grasping-weighing-placement operation, and simultaneously upload the equipment status, weight data and position coordinates during the operation to the digital twin virtual model to generate a physical execution verification data package; The verification module is used to reconstruct the robot arm's motion trajectory and the yarn rack inventory status in real time in the digital twin virtual model. It compares the physical execution verification data with the preset execution constraints. If the physical execution verification data deviates from the preset execution constraints, it triggers an abnormal warning command and pushes it to the operation and maintenance personnel.

[0010] Another embodiment of this application provides a storage medium storing a computer program, wherein the computer program is configured to execute the method described in any of the preceding claims when running.

[0011] Another embodiment of this application provides an electronic device including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the method described in any of the preceding claims.

[0012] Compared with existing technologies, this invention provides an intelligent unloading method for glass fiber based on digital twins. It collects real-time data from the yarn ball forming signal of the fiber forming machine, the positioning coordinates of the yarn carrier, and the weighing sensor data to generate a multi-source synchronous dataset. Based on this dataset, it calculates the three-dimensional coordinates of the yarn ball center and the lifting distance of the robotic arm, outputting a set of grasping parameters. This set is then sent to the servo control system to drive the robotic arm to complete the yarn ball grasping, weighing, and placement operations, simultaneously generating a physical execution verification data package. The robotic arm's motion trajectory and the yarn carrier's inventory status are reconstructed in real-time within the digital twin virtual model, triggering anomaly warning commands and pushing them to maintenance personnel. This enables precise positioning and real-time correction during the yarn ball grasping process, improving the reliability and production efficiency of the unloading operation. Attached Figure Description

[0013] Figure 1 A hardware structure block diagram of a computer terminal for an intelligent glass fiber unloading method based on digital twins provided in an embodiment of the present invention; Figure 2 A schematic flowchart of an intelligent glass fiber unloading method based on digital twin provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of a glass fiber intelligent unloading system based on digital twin, provided as an embodiment of the present invention. Detailed Implementation

[0014] The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0015] The present invention first provides a method for intelligent unloading of glass fiber based on digital twins. This method can be applied to electronic devices, such as computer terminals, specifically ordinary computers.

[0016] The following detailed explanation uses a computer terminal as an example. Figure 1 This is a hardware structure block diagram of a computer terminal for a glass fiber intelligent unloading method based on digital twins, provided as an embodiment of the present invention. (See diagram below.) Figure 1 As shown, the computer device includes a processor, memory, and network interface connected via a system bus, wherein the memory may include non-volatile storage media and internal memory.

[0017] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions that, when executed, cause the processor to perform any intelligent fiberglass unloading method based on digital twins.

[0018] The processor provides computing and control capabilities, supporting the operation of the entire computer device.

[0019] The internal memory provides an environment for the execution of computer programs in non-volatile storage media. When the computer program is executed by the processor, it enables the processor to execute any intelligent unloading method for glass fibers based on digital twins.

[0020] This network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 1 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0021] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0022] See Figure 2 The present invention provides an intelligent unloading method for glass fiber based on digital twins, which may include the following steps: S201 uses photoelectric sensors and a PLC controller to collect real-time data on yarn ball forming signals, yarn frame car positioning coordinates, and weighing sensor data from the fiber forming machine, generating a multi-source synchronous dataset with timestamps. Specifically, a laser interferometric grid positioning system can be deployed on the yarn frame car track to emit phase-modulated laser beams in real time, calculate the displacement of the yarn frame car through the reflected signals, and output the original positioning coordinate stream; The core components of a laser interferometric grid positioning system are installed at key locations on both sides of the yarn carrier's running track. This system consists of a high-precision laser emitter (LE), an optical reflector array (ORA), and a distributed photoelectric detector (PD). The laser emitter (LE) uses a semiconductor laser source with wavelength stability of ±0.1 picometers (pm) to continuously emit a phase-modulated laser beam (PMLB) upwards onto the track. The phase of this beam is sinusoidally modulated at a fixed frequency (e.g., 10 MHz) by a built-in electro-optic modulator (EOM). As the yarn carrier moves along the track, a retroreflective target (RRT) mounted on its bottom reflects the laser beam back to the detector (PD). Due to the Doppler effect, a phase difference (PDiff) is generated between the moving reflected light and the reference light. A photodetector (PD) converts the received optical signal into an electrical signal, which is then input into a high-speed phase demodulation chip (PDC). This chip calculates the phase offset (PO) between the emitted and reflected light in real time and, combined with the laser wavelength (λ), accurately calculates the instantaneous displacement (DIS) of the yarn carrier. For example, if a 90-degree phase offset (corresponding to a quarter wavelength) is detected, combined with the known laser wavelength of 632.8 nanometers (nm), the yarn carrier can be calculated to have moved 158.2 nanometers (nm). This process is performed continuously at a frequency of 1000 times per second (1kHz), forming a continuous raw position coordinate stream (RPCS) with a coordinate accuracy down to the sub-micrometer level (0.1 micrometers, μm).

[0023] To ensure positioning accuracy is unaffected by environmental interference, the system integrates multiple compensation mechanisms: Temperature Compensation Module (TCM): Platinum resistance temperature sensors (Pt100) are deployed along the track to monitor ambient temperature variations (TV) in real time. The module dynamically corrects for the impact of track length variations on the laser path based on the thermal expansion coefficient (TEC).

[0024] Vibration Filtering Algorithm (VFA): A MEMS accelerometer (MA) mounted on the track base acquires mechanical vibration data (VD) in real time. An adaptive Kalman filter (AKF) is run by a digital signal processor (DSP) to separate the true displacement (TD) of the yarn frame carriage and vibration noise (VN) from the raw displacement signal.

[0025] Air Refractive Index Correction (ARC): Barometers (BAR) and hygrometers (HYG) monitor air parameters and calculate the refractive index change (RIC) based on Edlen's Formula to compensate for its impact on laser wavelength. All compensation data is integrated into the coordinate calculation process in real time using hardware-level parallel processing to ensure the long-term stability of RPCS.

[0026] The raw positioning coordinate stream (RPCS) is temporarily stored using a ring buffer (RB), and each frame of data contains: Time Marker (TM): Local clock count value; Absolute Coordinate (AC): Three-dimensional spatial position (X, Y, Z); Displacement Vector (DV): The component of movement relative to the previous frame; Confidence Flag (CF): A data validity rating (0-100%) based on the signal-to-noise ratio (SNR).

[0027] Data is transmitted in real time to the Central Processing Unit (CPU) via Gigabit Industrial Ethernet (GIE) using User Datagram Protocol (UDP), providing basic positioning information for subsequent synchronization.

[0028] A high frame rate CMOS photoelectric sensor is installed at the outlet of the fiber forming machine to capture changes in the microstructure of the yarn ball surface, and a morphological edge detection algorithm is combined to generate a pulse signal of the yarn ball forming state. A high-frame-rate CMOS photoelectric sensor (HF-CMOS) is rigidly mounted 50 cm directly above the yarn exit of the fiber forming machine. This sensor is equipped with a global shutter (GS) imaging chip, a resolution of 1280×1024 pixels, and an adjustable frame rate (up to 2000 frames per second). The sensor lens employs a telecentric optical design (TOD) to eliminate perspective distortion and integrates a ring LED cold light source (RLCS) to provide shadowless illumination. As the glass fiber yarn (YarnCake, YC) is peeled from the rotating bobbin of the forming machine, its surface develops a unique microstructural texture (MT) due to differences in fiber winding density. The HF-CMOS continuously captures surface images (SI) of the yarn at a preset frame rate (e.g., 500 FPS). Each image is transmitted to a Field-Programmable Gate Array (FPGA) for preprocessing: first, non-uniformity correction (NUC) is performed to eliminate pixel response differences, and then flat-field correction (FCC) is performed to compensate for uneven illumination.

[0029] The preprocessed image is input into the Morphological Edge Detection Algorithm (MEDA) pipeline: Image Binarization (IB): Adaptive Threshold Segmentation (ATS) is used to dynamically calculate the threshold based on the local pixel grayscale distribution, converting the image into a black and white binary image (BI).

[0030] Morphological filtering (MF) first performs an opening operation (OO) – erosion (ER) followed by dilation (DL) – to eliminate minute noise; then it performs a closing operation (CO) – dilation followed by erosion – to fill small holes. The structuring element (SE) uses a 3×3 pixel circular kernel (CK).

[0031] Edge Extraction (EE): The Canny Operator (CO) is applied to the filtered binary image. Through a four-step process of Gaussian Filtering (GF), Gradient Calculation (GC), Non-Maximum Suppression (NMS), and Double Threshold Hysteresis (DTH), the sub-pixel precision boundary contour of the yarn clump is extracted.

[0032] Feature Quantization (FQ): Calculates geometric parameters such as perimeter (PERI), area (AREA), and circularity (CIRC).

[0033] Based on the contour feature changes of consecutive frames, the system generates a yarn formation state pulse signal (YFSPS): Peeling Start Signal (PSS): When the contour area (AREA) is detected to increase by more than 5% for 3 consecutive frames, a high-level pulse (HLP) is triggered.

[0034] Layer Completion Signal (LCS): A rising edge (RE) is generated when the circularity (CIRC) reaches 0.92 or higher and is maintained for 20 milliseconds (ms).

[0035] Abnormal Winding Signal (AWS): If the Perimeter Change Rate (PCR) exceeds the preset range (e.g., ±3% / frame), a Fault Code Pulse (FCP) is generated.

[0036] The pulse signal is converted into a 24V industrial level signal through an optocoupler isolator (OI), with an electromagnetic interference (EMI) immunity of up to 10V / m.

[0037] The weighing sensor data stream, the original positioning coordinate stream, and the forming pulse signal are input into the encrypted PLC gateway, and a hardware timestamp module driven by a cold atom clock is embedded to output encrypted data packets with nanosecond-level time tags. Data streams from three key data sources converge into the Encrypted PLC Gateway (EPG): Load Cell Data Stream (LCDS): The strain gauge load cell (SGLC) installed at the bottom of the yarn frame cart outputs the weight (WT) of the yarn spool at a sampling rate of 100 Hz, with a range of 0-50 kg and an accuracy of ±0.05% of full scale (FS).

[0038] Original Positioning Coordinate Stream (RPCS): A sequence of three-dimensional coordinates from the laser interferometer system.

[0039] Forming Pulse Signal (YFSPS): A digital pulse that characterizes the yarn bundle forming event.

[0040] EPG is equipped with three independent high-speed analog-to-digital converters (ADCs) and digital input channels (DICs) to acquire analog weight signals and digital pulse signals, respectively.

[0041] The core of data synchronization is a cold atom clock-driven hardware timestamp module (CAC-HTM): Cold Atom Clock (CAC): Based on the hyperfine transition (HT) of rubidium-87 atoms, with a frequency stability of 1×10⁻⁶. -13 / second (s). The atomic temperature was reduced to 1 micro Kelvin (μK) by laser cooling (LC) and magneto-optical trap (MOT), reducing the Doppler shift (DS).

[0042] Timestamp Embedding (TE): When each data sampling point arrives at the gateway, a dedicated hardware counter (HC) reads the 1 GHz clock signal from the atomic clock and generates a nanosecond timestamp (NT) in the format of a 64-bit integer (INT64) in International Atomic Time (TAI).

[0043] Anti-Delay Design (ADD): A field-programmable gate array (FPGA) is used to implement a nanosecond-level hardware interrupt response (HIR) to ensure that the timestamp embedding latency (TEL) is less than 10 nanoseconds (ns).

[0044] Time-stamped data enters the encryption pipeline: Packet Encapsulation (PE): Encapsulates the weight value, coordinate value, pulse status and timestamp of a single sample into a fixed-length data frame (FLDF), with a length of 128 bytes.

[0045] Hardware Encryption (HE): A dedicated encryption chip executes the AES-256-GCM algorithm (Advanced Encryption Standard with 256-bit key in Galois / Counter Mode) to generate a 128-bit authentication tag (AT) to prevent data tampering (TP).

[0046] Output Encrypted Data Packet (EDP): The encapsulated data is appended with a Frame Header (FH) and a Cyclic Redundancy Check (CRC) code, and output via a dual-redundant CAN-FD bus (Controller Area Network with Flexible Data-Rate) at a baud rate (BR) of 5 Mbps. Each EDP contains: Encrypted Payload (EP), Timestamp (NT), Authentication Tag (AT), and Data Source Identifier (DSI).

[0047] The encrypted data packets are reassembled in the time domain by a timestamp alignment engine to eliminate transmission jitter and generate a multi-source synchronized dataset containing four-dimensional data including timestamp, coordinates, weight, and forming status.

[0048] The Timestamp Alignment Engine (TAE) is deployed on the Central Industrial Server (CIS) and receives encrypted data packets (EDP) from multiple gateways. The engine first performs a decryption process: Hardware Decryption Card (HDC): Unencrypts AES-256-GCM encryption, verifies the Authenticator Tag (AT), and discards invalid packets (IPD). Data Deframing (DD): Extracting raw data such as timestamp (NT), coordinates (AC), weight (WT), and formation state flag (FSF); Buffer Classification (BC): Data is stored in independent First-In-First-Out (FIFO) queues based on the Data Source Identifier (DSI), with each queue having a depth of 1000 frames.

[0049] Key technologies for eliminating transmission jitter (TJ): Global Time Base (GTB): The server synchronizes with the Cold Atom Clock (CAC) via the Precision Time Protocol (PTPv2), with a clock offset (CO) of less than 100 nanoseconds (ns). Jitter Calculation (JC): For each data queue, calculate the difference between the Theoretical Arrival Time (TAT) and the actual timestamp (NT), which is the jitter value (JV). Dynamic Window Filter (DWF): Employs a sliding time window (STW) algorithm, where the window width (WW) is dynamically adjusted based on the historical jitter standard deviation (SD) (e.g., 3 times SD). Only data frames with jitter values ​​less than 1 millisecond (ms) within the window are retained.

[0050] Temporal Reorganization (TR) generates the final dataset: Timeline Matching (TM): Using 1 millisecond (ms) as the basic time unit (TimeUnit, TU), the filtered data is aligned to the Unified Timeline (UT) according to the timestamp (NT). Missing Data Handling (MDH): Interpolation Compensation (IC): The coordinate data is interpolated using cubic spline interpolation (CSI). State Hold (SH): The molding state flag retains the previous valid value; Weight Filling (WF): Weight data is marked as invalid (IW). 4D Dataset Construction (4DDC): Generates Structured Records (SRs), each containing: Timestamp (NT): Absolute time with nanosecond precision; Coordinates (AC): Three-dimensional spatial position (X, Y, Z); Weight (WT): mass of a yarn bundle in kilograms; Forming State (FSF): 8-digit code (e.g., 00000001 = peeling start, 00000010 = layering complete).

[0051] The dataset is stored in a Time Series Database (TSDB) and supports millisecond-level queries (MLQ).

[0052] S202, Based on the multi-source synchronous dataset, the three-dimensional coordinates of the yarn ball center and the lifting distance of the robotic arm are calculated by the core positioning algorithm, and the grasping parameter set is output; Specifically, the yarn ball forming state pulse signal can be extracted from the multi-source synchronous dataset, and the boundary feature points of the yarn ball can be identified by the wavelet ridge detection algorithm to output the boundary point cloud cluster. The system first filters the yarn formation state pulse signal from a multi-source synchronous dataset (containing four-dimensional data including timestamps, coordinates, weight, and forming state). This signal is generated by a high frame rate CMOS photoelectric sensor installed at the fiber forming machine outlet. Its working principle involves capturing hundreds of frames per second of yarn surface images (e.g., a frame rate of 500 FPS) and combining this with a morphological edge detection algorithm to analyze the microscopic changes in the yarn outline in real time. When the yarn formation reaches a preset diameter threshold (e.g., 80 cm), the sensor outputs a high-level pulse signal. The system extracts the time-series data of this pulse signal, precisely marking the start and end times of yarn formation (e.g., timestamp accuracy up to 10 milliseconds).

[0053] The algorithm employs a pulse signal input wavelet ridge detection algorithm. This algorithm uses the Morlet wavelet basis function to perform multi-scale decomposition of the signal and locates abrupt change points at the yarn bunch boundary by calculating the modulus maximum of the wavelet coefficients. For example, continuous scanning is performed within a scale parameter (SP) range of 5 to 20 to identify the frequency characteristics corresponding to the yarn bunch edge (e.g., high-frequency components corresponding to sharp edges). The algorithm automatically removes noise interference points caused by fiber movement (noise amplitude below a preset threshold of 0.05 volts), and finally outputs a set of spatial coordinates—boundary feature points. The three-dimensional coordinates of each point are jointly determined by the yarn frame positioning coordinates (from the laser interferometric grid positioning system) at the pulse signal trigger time and a fixed height value (determined by the exit height of the forming machine).

[0054] All boundary feature points constitute a boundary point cloud cluster. This point cloud cluster is optimized using a density-based clustering algorithm, merging neighboring points with a spatial distance less than a set threshold (e.g., 5 mm) and removing isolated points (clusters with fewer than 10 points are considered invalid). The final output point cloud cluster contains hundreds of discrete spatial coordinate points, accurately outlining the geometric outline of the yarn bundle and providing input for subsequent center point calculations. The point cloud data format is an N×3 matrix (N is the number of points, and 3 corresponds to X, Y, and Z coordinates), with timestamps included for synchronizing subsequent calculations.

[0055] Input the boundary point cloud cluster into the vibration-resistant least squares fitter, combine the yarn frame car positioning coordinates to compensate for the mechanical vibration offset, and output the optimized yarn ball center fitting normal vector. The boundary point cloud cluster is input into a vibration-resistant least squares fitter. The core of this fitter is the weighted least squares (WLS) method, which aims to fit the discrete point cloud into a spatial geometry (usually a cylinder, as the yarn bundle is tubular). A vibration compensation mechanism is introduced during the fitting process: first, high-frequency vibration components (frequency range 1–50 Hz) are extracted from the yarn frame positioning coordinate stream. The vibration spectrum is then analyzed using a Fast Fourier Transform (FFT) to identify the main vibration sources (such as 5 Hz low-frequency vibration caused by track unevenness or 20 Hz impact caused by motor start-stop). Based on the vibration amplitude (AMP, in millimeters) and phase (PHS), the weight (WT) of each point in the point cloud cluster is dynamically adjusted. For example, when the vibration amplitude is greater than 0.3 mm, the weight of the point collected at the corresponding time is reduced to 0.5 to suppress vibration interference.

[0056] The fitter performs spatial cylinder fitting. The cylinder model parameters are defined as the axis direction vector (ADV) and the center point coordinates (CPC). The algorithm solves for the optimal parameters by minimizing the sum of squared distances (DSS) from all points to the cylinder surface. The specific iterative process is as follows: Initial estimation: The initial axis orientation is determined using the eigenvectors of the covariance matrix (CM) of the point cloud cluster; Iterative optimization: The Levenberg-Marquardt algorithm (LMA) is used to adjust the axis direction and radius, and the residual (RES) is recalculated according to the weights after each iteration. After 10–15 iterations, stable cylindrical model parameters are output, including the three-dimensional coordinates (CPC_X, CPC_Y, CPC_Z) of the yarn ball's central axis and the axis direction vector (ADV_X, ADV_Y, ADV_Z).

[0057] The center axis direction vector is further transformed into an optimized yrn center normal vector. This normal vector is perpendicular to the end face of the yarn bundle (i.e., the bottom surface of the cylinder) and is generated by rotating the axis direction vector 90 degrees around the vertical axis. The directional accuracy of the normal vector reaches 0.1 degrees, which is used to guide the vertical gripping posture of the robotic arm's gripper. The final output parameters include: Center point coordinates: for example (1200.5 mm, 350.2 mm, 150.0 mm); Normal vector components: For example, (0.02, 0.99, 0.01) indicates a near vertical downward direction; All parameters are accompanied by a confidence score (CS, ranging from 0 to 1), calculated based on the fitting residuals and vibration compensation effect.

[0058] Based on the geometric relationship between the normal vector and the coordinate system of the robot arm, the opening and closing angle of the gripper and the normal angle of the contact surface are calculated, and the quaternion of the gripping posture is output. The system establishes a Robot Base Coordinate System (RBCS), with its origin located at the center of the robot arm mounting base and the Z-axis pointing vertically upwards. The normal vector fitted to the center of the yarn ball (denoted as vector N) is transformed from the world coordinate system (aligned with the laser positioning system) to the RBCS coordinate system. The transformation is achieved using a homogeneous transformation matrix (HTM), which is determined by the robot arm mounting position calibration parameters (e.g., base offset X=200 mm, Y=300 mm, Z=0 mm) and rotation angle (e.g., rotation around the Z-axis -5 degrees). The transformed normal vector is denoted as N_RBCS = (Nx_R, Ny_R, Nz_R).

[0059] The gripper opening angle (GOA) and contact surface normal angle (CSNA) are calculated based on N_RBCS: Gripper opening angle: Based on the yarn ball diameter (calculated from the cylinder radius R fitted by the point cloud cluster, for example, R=400 mm) and gripper mechanical parameters (minimum opening 100 mm, maximum opening 800 mm), the theoretical angle is calculated using the formula GOA=2×arcsin(R / L) (L is the gripper arm length, for example, L=500 mm), and a safety margin (for example, 5%) is added. The final GOA output is 60 degrees. Contact surface normal angle: refers to the angle between the gripper plane and the normal vector N_RBCS of the yarn end face. It is calculated by vector dot product (DotProduct, DP): CSNA = arccos(N_RBCS · Z_RBCS), where Z_RBCS=(0,0,1) is the unit vector of the Z-axis of the base coordinate system. If CSNA > 5 degrees, the robot arm posture needs to be adjusted so that the gripper plane is parallel to the yarn end face.

[0060] Grasping Pose Quaternion (GPQ) describes the spatial orientation of the gripper. The quaternion format is (Qw, Qx, Qy, Qz), and it is generated through the following steps: Calculate the rotation axis (RA) and rotation angle (RA) based on the normal vector N_RBCS and the base coordinate system Z-axis: RA = Z_RBCS × N_RBCS (cross product); RA = arccos(Z_RBCS · N_RBCS); Convert the rotation axis and angle to quaternions: Qw = cos(RA / 2); (Qx, Qy, Qz) = sin(RA / 2) ×RA_normalized.

[0061] For example, the output GPQ=(0.707, 0, 0.707, 0) represents a 90-degree rotation around the Y-axis. The quaternion data is bound to the center point coordinates to form a complete description of the gripping position and attitude.

[0062] Based on the dynamic prediction of the yarn ball centroid offset using weighing sensor data, the anti-sway lifting trajectory equation is calculated through a rigid body dynamics model, and the lifting distance-time series is output. Centroid Offset Prediction (COP): The system reads weight sensor data (WSD) from a multi-source synchronous dataset. This data is updated at a frequency of 100 Hz and records the real-time weight of the yarn bundle (e.g., 20.5 kg). The centroid offset (CO), caused by uneven weight distribution, is estimated through the following steps: Weight variation gradient analysis: Calculate the standard deviation (SD) of the weight data in the last 10 seconds. If SD > 0.3 kg, it is determined that there is a significant bias. Offset direction inference: Combine historical data of yarn ball forming pulse signals to locate the angular region where the weight change occurs (e.g., the pulse signal shows that the yarn ball is abnormally formed at a 300-degree azimuth angle). The final output is the centroid offset vector CO = (ΔX, ΔY, ΔZ), for example (15 mm, -10 mm, 5 mm), which means that the centroid is offset by 15 mm in the positive X direction.

[0063] The anti-swing lifting trajectory equation (ASLTE) is generated by the rigid body dynamics model (RBDM). The model simplifies the yarn ball into a mass-penulum system suspended at the end of the manipulator. The input parameters include: yarn ball mass M (20.5 kg), center of mass offset CO, wire rope length L (1.2 m), and target lifting height H (0.8 m).

[0064] The model solves for the optimal acceleration curve to suppress oscillation. The core steps include: State-space modeling: Establish a 4-dimensional state vector containing position, velocity, swing angle, and angular velocity; Optimal control solution: A controller is designed using a linear quadratic regulator (LQR) to minimize the swing energy (weighting coefficient Q=0.8) and energy consumption (weighting coefficient R=0.2). The output is an acceleration-time function a(t), for example, the acceleration is 0.3g in the first 0.5 seconds and decreases to 0.1g in the next 0.5 seconds.

[0065] The Lifting Distance-Time Sequence (LDTS) is generated by a quadratic integration of the acceleration a(t). The discretized time step is Δt = 0.05 seconds, and the calculation formulas are: velocity v(t) = v(t-1) + a(t) × Δt; distance s(t) = s(t-1) + v(t) × Δt.

[0066] The total duration of the sequence is dynamically adjusted based on the lifting height (e.g., 800 mm lifting in 1.0 second), and is subject to a maximum swing angle constraint (<2 degrees).

[0067] By fusing the grasping posture quaternion and the lift distance-time series, a grasping parameter set containing spatial coordinates, posture angles, lift curves, and clamping force thresholds is generated.

[0068] The Data Fusion Engine (DFE) receives four inputs: yarn ball center coordinates (CPC_X, CPC_Y, CPC_Z), grasping posture quaternions (Qw, Qx, Qy, Qz), lifting distance-time series (LDTS), and real-time weighing data (WSD).

[0069] Engine performs spatiotemporal alignment: mapping the center coordinates and quaternions to the robot arm's base coordinate system, and synchronizing the LDTS with the robot arm's motion clock via timestamps (error < 1 millisecond).

[0070] The clamping force threshold (GFT) is set based on the weight and material of the yarn bundle: Base pressure: F_base = WSD × safety factor (1.5) / gripper contact area (e.g., 0.02 square meters) → e.g., 1537.5 Newtons; Dynamic compensation: The displacement of the center of mass causes torque imbalance, and the additional pressure increment ΔF = |CO| × WSD / L (L is the gripper lever arm of 0.2 meters) → for example, ΔF = 102.5 Newtons; The final GFT = F_base + ΔF = 1640 Newtons, and is partitioned and distributed to different hydraulic units of the gripper (e.g., the pressure on the offset side increases by 10%).

[0071] The Grasping Parameter Set (GPS) is output as a structured data packet containing the following fields: Spatial Coordinates (SC): 3D coordinates of the target grab point (1200.5, 350.2, 150.0), in millimeters; Pose Angles (PA): Euler angles format converted from quaternions (Roll=0°, Pitch=0°, Yaw=45°); Lifting Curve (LC): LDTS sequence containing 100 time-distance data pairs; Gripping Force Threshold (GFT): 1640 Newtons, configured with three pressure zones; Verification Code (VC): A CRC32 checksum generated from all parameters (e.g., 0x58D4C3D9) to prevent transmission errors.

[0072] Data packets are sent to the servo control system via industrial Ethernet (EtherCAT protocol) to trigger the robot arm to perform operations.

[0073] S203, the grasping parameter set is sent to the servo control system to drive the robot arm to complete the yarn ball grasping-weighing-placement operation, and the equipment status, weight data and position coordinates during the operation are uploaded to the digital twin virtual model to generate a physical execution verification data package; Specifically, the captured parameter set can be compiled into machine instruction code, a digital signature can be attached, and then the code can be sent to the servo control system to trigger the robot arm action chain and output the execution log hash value. Machine instruction code compilation process: The captured parameter set includes spatial coordinates (X, Y, Z values, in millimeters), attitude angles (pitch, yaw, roll, in degrees), lift curves (lift height versus time sequence), and clamping force thresholds (in Newtons). First, the Command Compiler Engine (CCE) converts these parameters into Machine Instruction Code (MIC) that the underlying hardware can recognize. For example, spatial coordinates (X=1200.5mm, Y=750.3mm, Z=200.0mm) are converted into servo motor pulse counts (PC), based on the rule that 1 degree of servo motor rotation corresponds to 1000 pulses (PC / Degree=1000). The time series in the lifting curve (e.g., height H=0mm at t=0s, H=150mm at t=2s) generates a continuous trajectory function using a cubic spline interpolation algorithm, which is then decomposed into position loop (PL) and velocity loop (VL) control parameters for the servo motor. The clamping force threshold (e.g., 50 Newtons) is mapped to the current threshold (CT) of the gripper servo motor, achieving precise conversion through the force-current calibration curve (FCCC).

[0074] Digital signature and instruction issuance mechanism: The compiled machine instruction code (MIC) must be appended with a digital signature (DSig) to ensure instruction integrity and source credibility. The signing process employs the Elliptic Curve Digital Signature Algorithm (ECDSA), using a 256-bit private key (PriK) to encrypt the instruction code hash value (HV) to generate the signature. The signed MIC is then sent to the Servo Control System (SCS) via a real-time industrial Ethernet protocol (such as EtherCAT). The SCS comprises a Multi-axis Motion Controller (MMC) and Servo Drivers (SDs). Upon receiving the instruction, the MMC decomposes it into independent motion instructions for each joint (e.g., the J1-J6 axes of a robotic arm) and sends them to each Servo Driver via Cyclic Synchronous Communication (CSC) at 1-millisecond intervals.

[0075] Action chain triggering and log hash generation: The servo control system triggers the robot action chain (RAC), which consists of three core stages: Grasping phase: The grippers move to the target coordinates (X,Y,Z) as instructed, contact the yarn ball with the calculated attitude angle (Pitch,Yaw,Roll), and apply a preset gripping force (e.g., 50±2 Newtons).

[0076] Weighing stage: After gripping, the robotic arm moves the yarn ball to the embedded weighing platform (WP), and the load cell (LC) reads the weight data (WD) with a sampling period of 10 milliseconds.

[0077] Placement stage: Based on the inventory coordinates of the yarn rack cart (e.g., A-3-5 position), the robotic arm executes an anti-sway lifting trajectory (e.g., an S-shaped acceleration and deceleration curve) to complete the placement.

[0078] All operational details (timestamp, joint angle, motor current, actual position) are recorded as an execution log (EL). The log is used to generate a unique hash value of the execution log (HEL) using the secure hash algorithm SHA-256 (Secure Hash Algorithm 256-bit), which is used for subsequent data integrity verification.

[0079] The robot arm collects joint encoder data, six-dimensional force sensor readings, and visual positioning feedback in real time. It detects execution deviations through an action sequence verification algorithm and outputs action compliance flags. Multi-source data acquisition system: Joint Encoder Data (JED): Each robotic arm joint (J1-J6) is equipped with an absolute encoder (AE) that provides real-time feedback of joint rotation angles (in degrees) with a resolution of 0.1 milliseconds. For example, the angle θ of axis J1 is 35.7° at t=1.5s.

[0080] Six-axis Force Sensor Reading (6DFSR): A six-axis force sensor mounted on the robotic wrist (measuring XYZ axis forces and torques around the XYZ axes), with a sampling frequency of 1 kHz. For example, at the moment of grasping, a Z-axis force Fz = 48.5 Newtons and an X-axis torque Tx = 2.1 Nm are detected.

[0081] Visual Positioning Feedback (VPF): An industrial camera (2048×1536 resolution, 120fps) deployed in the work area, combined with a sub-pixel corner detection (SCD) algorithm, outputs in real time the pixel-level offset between the edge of the yarn ball and the gripper (e.g., Δx=0.5 pixels, Δy=-0.3 pixels), which is converted into spatial offset (e.g., ΔX=0.1mm, ΔY=-0.06mm) through a calibration matrix (CM).

[0082] Action sequence verification algorithm: The core process of the Action Sequence Verification Algorithm (ASVA), which collects raw data and inputs it into the algorithm, is as follows: Data synchronization alignment: JED, 6DFSR, and VPF are aligned with nanosecond-level timestamps to construct a spatiotemporally unified data frame (DF).

[0083] Dynamic trajectory comparison: Input the actual joint angle sequence (such as the curve of the J1 axis angle changing with time) and the virtual preset trajectory (from the capture parameter set) into the Dynamic Time Warping (DTW) algorithm to calculate the trajectory similarity score (SS, range 0-100).

[0084] Force control compliance testing: The six-dimensional force data is compared with the preset clamping force threshold (50±2 Newtons) and anti-sway torque threshold (such as torque Tz around the Z-axis <1.0 N·m) to generate the Force Deviation Index (FDI).

[0085] Visual positioning verification: If the offset of the yarn ball position (ΔX, ΔY) in the visual feedback exceeds the tolerance (e.g., ±0.5mm), it is marked as a positioning anomaly (Position Anomaly Flag, PAF).

[0086] Compliance flag output: The algorithm integrates three indicators: trajectory similarity SS ≥ 95 points; force control deviation index FDI ≤ 0.1; and visual positioning without anomalies (PAF = False).

[0087] When all three conditions are met, the Compliance Flag Bit (CFB) is set to "True" (True / 1); if any condition is not met, "False" (False / 0) is output. For example, if an actual clamping force Fz = 53 Newtons (exceeding the threshold) is detected, the CFB is set to 0 and real-time correction is triggered.

[0088] When the compliance flag is true, the original verification data block is generated by encapsulating the device status, weight data, location coordinates, and execution log hash value. Data filtering and packaging conditions: The system initiates the data encapsulation process only when the compliance flag (CFB) is "true". This design ensures that only operational data conforming to process standards is used for digital twin model updates, preventing invalid or erroneous data from contaminating the twin. Equipment status includes: Temperature of motors at each joint of the robotic arm (e.g., J2 axis motor temperature T_motor=42.3℃); Servo drive load rate (e.g., SD3 load rate Load_rate=78%). Pneumatic gripper pressure (PGP = 0.55 MPa).

[0089] Multi-dimensional data integration: The Raw Verification Data Block (RVDB) contains four core types of information: Equipment Status (ES): Stored in key-value pair (KVP) format, for example, {"MotorTemp_J3": 41.2, "ServoLoad_Axis4": 82}.

[0090] Weight Data (WD): The stable value output by the weighing platform (e.g., net weight of yarn spool 12.53 kg) and the maximum fluctuation value during the weighing process (±0.05 kg).

[0091] Position Coordinates (PC): Includes three sets of coordinates: Pickup XYZ, Weighing XYZ, and Placement XYZ, with an accuracy of ±0.1mm.

[0092] Execution Log Hash (HEL): A 64-bit hexadecimal string (e.g., "e3b0c44298fc1c...") used to associate with the complete execution log.

[0093] Data structures and metadata additions: RVDB is encapsulated using a self-describing binary format (SBF): Header: Contains the data version number (DV=2.0), timestamp (TS=2025-04-11T14:23:45.123Z), and block length (BL=1024 bytes).

[0094] Body: Stores binary data streams of ES, WD, PC, and HEL according to a predefined structure.

[0095] Trailer: An additional Cyclic Redundancy Check (CRC-32) is added for transmission error detection (e.g., CRC=0xA3B5D7F1).

[0096] The original verification data block is compressed using a Merkle tree and uploaded to the digital twin to generate a physical execution verification data package with a blockchain fingerprint.

[0097] Merkle tree compression process: A Merkle Tree (MT) is a tree-like data structure used for efficiently verifying the integrity of large datasets. The original Validation Data Block (RVDB) is divided into fixed-size data chunks (DCs, e.g., 256 bytes each). A SHA-256 hash (called a leaf node hash, LNH) is calculated for each DC. Adjacent LNHs are concatenated and hashed again to generate the parent node hash (PNH). This process is recursively repeated until a unique root hash (RH) is generated, such as "6d4a7b...f9c1". This structure ensures that any tampering with a DC will result in a change to the RH.

[0098] Blockchain fingerprint generation: A blockchain fingerprint (BFP) is a unique identifier for a data packet in a distributed ledger. The generation process is as follows: Concatenate the Merkle root hash (RH), timestamp (TS), and device ID (e.g., Robot_ID=RS-205) into a string.

[0099] The string is signed with the private key of an asymmetric encryption algorithm (AEA) to generate a digital fingerprint signature (DFS).

[0100] Associate DFS with the original data packet and write it to a lightweight blockchain node (such as an edge blockchain based on Hyperledger Fabric, EBC).

[0101] Data packet upload and twin reconstruction: The compressed data packet is uploaded to the Digital Twin Platform (DTP) via a Zero-Trust Security Tunnel (ZTST). The platform then performs the following operations: Data decompression and verification: Merkle root hash (RH) is used to verify the integrity of the data blocks. If the verification passes, the original data (ES, WD, PC, HEL) is extracted.

[0102] Twins are updated in real time: The robot's movement trajectory is reconstructed into a red dynamic path line in virtual space based on the position coordinates (PC).

[0103] The inventory status of the yarn rack cart is updated in a 3D heatmap based on the placement coordinates, with areas of inventory growth displayed as gradient blue.

[0104] Equipment status data drives the health model of the virtual robot (e.g., the model turns yellow when the motor temperature exceeds the limit).

[0105] Generate a Physical Execution Verification Data Packet (PEVDP): containing raw data, a Merkle tree structure, and a blockchain fingerprint (BFP), which is stored in a twin database as trusted evidence of execution.

[0106] S204 reconstructs the robot arm's motion trajectory and the yarn rack inventory status in real time within the digital twin virtual model, compares the physical execution verification data with the preset execution constraints, and if the physical execution verification data deviates from the preset execution constraints, it triggers an abnormal warning command and pushes it to the operation and maintenance personnel.

[0107] Specifically, a physical execution verification data package can be loaded into the digital twin virtual model to reconstruct the set of robot arm motion trajectory points and the three-dimensional heat map of yarn rack inventory; After the Physical Execution Verification Data Packet (PEVDP) is transmitted to the Digital Twin Virtual Model (DTVM), the system first performs packet loading and parsing. This packet contains a Merkle Tree Compressed Data Block (MTCDB), whose core fields include: Robot Joint Encoder Data (RJED): Records the real-time angles (in degrees) of each joint. Six-Axis Force Sensor Readings (6DFSR): Includes X / Y / Z axial force (unit: Newton N) and torque (unit: Newton-meter Nm); Visual Positioning Feedback (VPF): Three-dimensional coordinates (in millimeters) of the robotic end effector captured by industrial cameras deployed in the workshop. Creel Car Inventory Status (CCIS): Quantity, weight, and occupied marker position of yarn spools in each compartment.

[0108] The system ensures data integrity through Blockchain Fingerprint Verification (BFV), and then calls the Spatial Mapping Engine (SME) to convert the raw data into parameters in a virtual coordinate system. For example, the robot arm base is set as the origin of the world coordinate system (0,0,0), the yarn carrier track extends along the X-axis, and the Z-axis points vertically upward. The RJED data is converted into a three-dimensional trajectory point sequence (TPS) of the end effector (gripper) using the Forward Kinematics Algorithm (FKA), with each point containing position coordinates (X,Y,Z) and a timestamp (nanosecond precision).

[0109] The core of Motion Trajectory Reconstruction (MTR) lies in point cloud interpolation and smoothing. The original TPS data may contain gaps due to sampling frequency limitations (e.g., 100Hz). The system employs a Cubic B-Spline Interpolation (CBSI) algorithm to insert smooth path points between adjacent trajectory points, generating a Continuous Trajectory Curve (CTC). Simultaneously, the 6DFSR data is mapped as an additional attribute layer for the trajectory curve: when the Z-axis force reading exceeds a preset threshold (e.g., 50N), an "Overload Flag" (OLF) is marked on the corresponding trajectory point; when torque fluctuation exceeds ±10%, a "Torque Vibration Flag" (TVF) is marked. These markers will serve as key weighting factors in subsequent deviation analysis. The final output is a Robot Trajectory Point Set (RTPS), containing 10,000 to 50,000 attributed 3D spatial points.

[0110] The reconstruction of the 3D heatmap (CC3DH) of the yarn car inventory relies on the fusion of multi-source data: Position occupancy status: Based on the position flag in CCIS (0 / 1 indicates empty / occupied), position cubes are rendered on the virtual yarn rack model. Empty positions are displayed in green, and occupied positions are displayed in red.

[0111] Yarn weight distribution: Weighing sensor data (unit: kilograms) is converted into gravimetric thermodynamic values ​​using the Gaussian Kernel Density Estimation (GKDE) algorithm. For example, a 20kg yarn ball generates a thermal field with a radius of 0.5 meters and a peak intensity of 100 in its storage compartment, which decreases with distance in adjacent compartments.

[0112] Inventory dynamic changes: When the robotic arm places a new yarn bundle, the system updates the warehouse status in real time and diffuses the thermal effect through a fluid dynamics simulation (FDS) algorithm. For example, the moment a new yarn bundle enters the warehouse, it triggers a "heat wave" (HW) that diffuses to the surrounding warehouses, with the wave speed set at 2 m / s and an attenuation coefficient of 0.7.

[0113] The final generated CC3DH is a three-dimensional thermal field model with a transparency channel. Operation and maintenance personnel can observe the inventory density distribution in 360° through VR devices, and key areas (such as overweight clusters and empty warehouse areas) will automatically be highlighted and flashed.

[0114] The actual trajectory point sequence and the virtual preset trajectory are input into a 3D convolutional neural network, and the trajectory deviation score is output. A Virtual Preset Trajectory (VPT) is an ideal path baseline in a digital twin model, predefined by process experts. Its data structure is a sequence of 5,000 keyframes, each containing: Target Position (TP): X / Y / Z coordinates (unit: millimeters); Allowable Deviation Range (ADR): Radius of the spherical tolerance zone (e.g., ±5mm); Time Window (TW): The earliest / latest timestamp (in nanoseconds) to reach that point.

[0115] The actual trajectory point sequence (RTPS) needs to be spatiotemporally aligned with the VPT. The system calls the Dynamic Time Warping (DTW) algorithm to compress both to the same length (e.g., 10,000 points) and uses a Spatiotemporal Normalization Layer (SNL) to eliminate the impact of execution speed differences.

[0116] The input to a 3D Convolutional Neural Network (3D-CNN) is a two-channel 3D voxel grid. Channel 1: Actual Trajectory Voxelization (ATV) Convert RTPS to a 200×200×200 voxel mesh (voxel side length 5mm). If the trajectory passes through a voxel, its value is 1; otherwise, it is 0.

[0117] Channel 2: Preset Trajectory Voxelization (PTV) VPT is converted according to the same rules, but with the addition of tolerance domain attributes: core trajectory voxel value = 1, tolerance domain voxel value = 0.5.

[0118] The network structure includes: Convolutional layer 1: 64 7×7×7 convolutional kernels, stride 2, extracting macroscopic path features; Pooling layer 1: Max pooling, kernel size 3×3×3, compressing spatial dimensions; Convolutional layer 2: 128 5×5×5 convolutional kernels to capture local jitter features; Feature fusion layer: stitches the dual-channel feature maps into a 256-dimensional tensor; Fully connected layer: Outputs a 1-dimensional deviation score (DS), ranging from 0 to 1 (0 = complete overlap, 1 = severe deviation).

[0119] The deviation score generation mechanism includes multi-factor weighting: Spatial Deviation Weight (SDW): The percentage of voxels whose actual trajectory exceeds the VPT tolerance range (e.g., number of out-of-tolerance voxels / total number of voxels × 0.6). Dynamic Behavior Weight (DBW): Frequency of OLF / TVF markers converted from 6DFSR data (e.g., deduct 0.05 points per marker). Timeliness Deviation Weight (TDW): The difference between the actual arrival time and the median TW (0.001 points are deducted for each millisecond of error).

[0120] The final scoring formula is: DS = min(1, 0.7×SDW + 0.2×DBW + 0.1×TDW).

[0121] For example: In a certain crawl, SDW=0.4 (40% of points are out of tolerance), DBW=0.15 (3 overloads detected), TDW=0.08 (average latency 80ms), then DS=0.7×0.4+0.2×0.15+0.1×0.08=0.328. This score will serve as the core basis for anomaly detection.

[0122] If the trajectory deviation score exceeds the preset score threshold, or the yarn ball weight deviation exceeds the preset deviation threshold, a primary abnormal event label is generated; The preset score threshold (PST) is dynamically adjusted according to process requirements: Standard operating condition: PST=0.25; High-speed mode: PST=0.35 (allows for greater path deviation); Precision operation: PST=0.15 (e.g., grabbing fragile yarn balls).

[0123] When the real-time calculated DS > PST, a Trajectory Deviation Anomaly (TDA) is triggered. For example, if DS = 0.328 > PST = 0.25, the TDA level is marked as follows: TDA_Level = ceil((DS - PST) × 10) → In this example, (0.328-0.25)×10=0.78, which is rounded up to level 1.

[0124] Yarn Weight Deviation Detection (YWDD) runs independently: Extract the actual weight (AW) from the physical execution data packet. Retrieve the standard weight (SW) of this type of yarn from the process database. Calculate the absolute deviation (AD) = |AW - SW|; Calculate the relative deviation rate (RDR) = AD / SW × 100%.

[0125] The Preset Deviation Threshold (PDT) uses a two-condition trigger: Condition 1: AD > 0.5 kg (hard threshold); Condition 2: RDR > 3% (flexible threshold); For example: SW=20kg, AW=20.7kg → AD=0.7kg>0.5kg, RDR=3.5%>3%. If both conditions are met, a weight anomaly (WA) will be triggered.

[0126] Rules for generating Primary Anomaly Event Tags (PAETs): Single anomaly scenario: When only TDA or WA occurs, generate Level 1 labels; Composite anomaly scenario: TDA and WA occur simultaneously, generating Level 2 labels.

[0127] The tag data structure example includes: { "Event ID": "ANOM-20240411-0032", / / Year / Month / Day-Serial Number "Exception type": ["TDA_Level1", "WA"], "Key parameters": { "DS": 0.328, "PST": 0.25, "AD": 0.7, "SW": 20.0 }, "Spatiotemporal Positioning": "Robot #3@(X12.5,Y7.8,Z1.2), Yarn Cart B Zone #5 Warehouse" }

[0128] The label is written to the Distributed Anomaly Log (DAL) and simultaneously triggers visual calibration of the digital twin model: the anomaly location flashes red in the virtual scene and a floating label of the anomaly parameters is projected.

[0129] When a primary anomaly event label is triggered, the multi-source data correlation analysis engine is started to calculate the standard deviation multiple of the path execution time and the historical mean. If it exceeds 2 times the standard deviation, it is upgraded to a severe anomaly event label. The Multi-source Data Correlation Engine (MDCE) starts immediately after PAET is generated and performs a three-stage analysis: Phase 1: Time Series Association Extract the Path Execution Time (PET): the total time taken from the start of fetching to the end of placement (in seconds); Retrieve historical PET data (Historical PET Dataset, HPETD) from the last 30 similar operations. Calculate the mean value μ and standard deviation σ of HPETD. Calculate the Standard Deviation Multiple (SDM) = (this PET - μ) / σ.

[0130] Phase 2: Spatial-Mechanical Connections The robot arm trajectory is divided into functional sections: acceleration section, gripping section, lifting section, translation section, and placement section; Analyze the differences between the following parameters and historical benchmarks segment by segment: Maximum acceleration (MA, unit: m / s²) 2 ); Average force (AF, unit: N); Path Curvature Radius (PCR, unit: m); When the MA of any segment exceeds 150% of the historical mean, or the AF is lower than 70% of the mean, or the PCR is lower than 50% of the mean, it is marked as a dynamics anomaly (DA).

[0131] Phase 3: Severity Escalation Decision PAET is upgraded to a Critical Anomaly Event Tag (CAET) if any of the following conditions are met: Time dimension: SDM>2.0 (i.e., execution time exceeds μ+2σ); For example: μ=120 seconds, σ=5 seconds, PET=133 seconds → SDM=(133-120) / 5=2.6>2.0; Composite anomaly: PAET Level 2 and DA count ≥ 2 segments; Continued deterioration: The same device triggered a third PAET within 24 hours.

[0132] The upgraded CAET data structure adds the following fields: { ... / / Inheriting PAET fields Upgrade basis: "SDM=2.6 (>2.0)", Urgency level: "Urgent" "Auto Response": "Pause Robot Arm #3 Task Queue" }

[0133] The system automatically pushes early warning instructions to operations and maintenance personnel based on the level of abnormal event labels, and simultaneously marks the abnormal points in the twin model and generates suggested correction solutions.

[0134] Alert Instruction Push (AIP) employs a tiered strategy: Primary Exception (PAET): Push method: WeChat text message + email notification.

[0135] Content template: [Basic Warning] Robotic arm #3 has deviated from its trajectory at position (X12.5, Y7.8) (DS=0.328) and is overloaded (20.7kg / 20kg). The equipment is still running; it is recommended to check within 30 minutes.

[0136] Critical Abnormalities (CAET): Push notification methods: APP pop-up + SMS message + control room audible and visual alarm.

[0137] Content template: [Emergency Stop] Robotic arm #3 has been paused due to an abnormal path time (SDM=2.6)! Coordinates (X12.5, Y7.8, Z1.2). Please take immediate action! The push system integrates a Personnel Location Module (PLM) to prioritize notifying the maintenance personnel closest to the faulty equipment (such as those wearing UWB positioning tags within 50 meters).

[0138] Steps for implementing Twin Model Anomaly Marking (TMAM): Spatial anchoring: In the virtual scene, anomalies (such as the position of the robotic arm end effector) are bound to a pulse light source (PLS) with a pulse frequency of 2Hz. Red light represents CAET and yellow light represents PAET.

[0139] Data Perspective: Clicking on an anomaly activates the Holographic Data Lens (HDL), which floats in the view. Real-time sensor readings (six-dimensional force / visual coordinates); Similar historical anomalies (3 most recent records); Status of related equipment (motor temperature, gearbox vibration spectrum).

[0140] Impact diffusion simulation: Based on the Finite Element Stress Cloud (FESC) algorithm, the stress distribution of the mechanical structure is generated with anomaly points as the center, and the red highlighted areas indicate the locations of potential fatigue damage risks.

[0141] Correction Plan Suggestion (CPS) process: Root cause matching: Input the abnormal parameters into the Case-Based Reasoning Engine (CBRE) to match historical cases in the knowledge base with a similarity of >85%. For example, it matches case ID: CASE-20231205-142 (vibration caused location displacement).

[0142] Dynamic decision tree: Example of generating suggestions along the following branches: A [Trajectory deviation?] --> |Yes| B [SDM>2?]; A -->|No| C [Weight deviation?]; B -->|Yes| D[Check servo motor encoder]; B -->|No| E [Calibrate the visual positioning system]; C -->|Yes| F[Clean Weighing Sensor]; C -->|No| G[Check yarn ball specifications].

[0143] Example of output structured scheme: { "Serial Number": "FIX-0032", "Applicable Exception ID": "ANOM-20240411-0032", "Operation steps": [ 1. Use a laser calibrator to calibrate the base coordinate system of robot #3 (refer to document DOC-003) 2. Clean the weighing sensor in compartment #5 of section B of the yarn rack cart (anhydrous ethanol is required) 3. Execute the positioning accuracy test script TEST-002 and upload the results. ], Estimated time: 45 minutes Spare parts requirements: Calibration target #7, Cleaning kit #3 }

[0144] The solution automatically synchronizes to the mobile terminals of maintenance personnel and is linked to the equipment maintenance work order system (WOS).

[0145] As can be seen, the yarn ball forming signal of the fiber forming machine, the positioning coordinates of the yarn frame trolley, and the data from the weighing sensor are collected in real time to generate a multi-source synchronous dataset. Based on the multi-source synchronous dataset, the three-dimensional coordinates of the yarn ball center and the lifting distance of the robot arm are calculated, and a grasping parameter set is output. The grasping parameter set is sent to the servo control system to drive the robot arm to complete the yarn ball grasping-weighing-placement operation, and a physical execution verification data package is generated simultaneously. The robot arm's motion trajectory and the inventory status of the yarn frame trolley are reconstructed in real time in the digital twin virtual model, triggering anomaly warning commands and pushing them to maintenance personnel. This enables precise positioning and real-time correction of the yarn ball grasping process, improving the reliability and production efficiency of the unloading operation.

[0146] Another embodiment of the present invention provides a glass fiber intelligent unloading system based on digital twins, see [link to relevant documentation]. Figure 3 The system may include: The acquisition module 301 is used to acquire the yarn ball forming signal, yarn frame car positioning coordinates and weighing sensor data of the fiber forming machine in real time through photoelectric sensors and PLC controller, and generate a multi-source synchronous dataset with timestamps. Output module 302 is used to calculate the three-dimensional coordinates of the yarn ball center and the lifting distance of the robot arm based on the multi-source synchronous dataset and the core positioning algorithm, and output the grasping parameter set. The drive module 303 is used to send the grasping parameter set to the servo control system, drive the robot to complete the yarn ball grasping-weighing-placement operation, and simultaneously upload the equipment status, weight data and position coordinates during the operation to the digital twin virtual model to generate a physical execution verification data package; The verification module 304 is used to reconstruct the movement trajectory of the robotic arm and the inventory status of the yarn rack in real time in the digital twin virtual model, compare the physical execution verification data with the preset execution constraints, and if the physical execution verification data deviates from the preset execution constraints, an abnormal warning instruction is triggered and pushed to the operation and maintenance personnel.

[0147] This invention also provides a storage medium storing a computer program, wherein the computer program is configured to execute the steps in any of the above method embodiments when running.

[0148] This invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the steps in any of the above method embodiments.

[0149] Specifically, the aforementioned electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the aforementioned processor, and the input / output device is connected to the aforementioned processor.

[0150] The above description, based on the embodiments shown in the figures, details the structure, features, and effects of the present invention. The above description is only a preferred embodiment of the present invention, but the present invention is not limited to the scope of implementation shown in the figures. Any changes made in accordance with the concept of the present invention, or equivalent embodiments modified to have equivalent changes, that do not exceed the spirit covered by the specification and figures, should be within the protection scope of the present invention.

Claims

1. A digital-twin-based intelligent glass fiber uncoiling method, characterized in that, The method comprises: Real-time acquisition of yarn ball forming signals, yarn rack car positioning coordinates and weighing sensor data of the fiber forming machine through photoelectric sensors and PLC controllers to generate a multi-source synchronous data set with time stamps; Based on the multi-source synchronous data set, the three-dimensional coordinates of the yarn ball center and the lifting distance of the manipulator are calculated by a core positioning algorithm, and a set of grabbing parameters is output; The set of grabbing parameters is sent to the servo control system to drive the manipulator to complete the yarn ball grabbing-weighing-placing operation, and the equipment state, weight data and position coordinates during the operation are uploaded to the digital twin virtual model to generate physical execution verification data packets; In the digital twin virtual model, the manipulator motion trajectory and the yarn rack car inventory state are reconstructed in real time, and the physical execution verification data is compared with the preset execution constraints. If the physical execution verification data deviates from the preset execution constraints, an abnormal early warning instruction is triggered and pushed to the operation and maintenance personnel.

2. The method of claim 1, wherein, The method comprises: Deploying a laser interference grid positioning system on the yarn rack car track to emit phase-modulated laser beams in real time, calculating the displacement of the yarn rack car through the reflected signals, and outputting the original positioning coordinate stream; A high-frame-rate CMOS photoelectric sensor is installed at the outlet of the fiber forming machine to capture the surface microstructure changes of the yarn ball, and a morphological edge detection algorithm is used to generate yarn ball forming state pulse signals; The weighing sensor data stream, original positioning coordinate stream and forming pulse signal are input into an encrypted PLC gateway, a hardware timestamp module driven by a cold atomic clock is embedded, and an encrypted data packet with nanosecond-level time labels is output; The encrypted data packet is reorganized in the time domain by a timestamp alignment engine to eliminate transmission jitter, and a multi-source synchronous data set containing four-dimensional data of timestamps, coordinates, weights and forming states is generated.

3. The method of claim 2, wherein, The method comprises: Extracting the yarn ball forming state pulse signal from the multi-source synchronous data set, identifying the boundary feature points of the yarn ball through a wavelet ridge line detection algorithm, and outputting the boundary point cloud cluster; The boundary point cloud cluster is input into an anti-vibration least squares fitter, the mechanical vibration offset is compensated combined with the yarn rack car positioning coordinates, and the optimized yarn ball center fitting normal vector is output; According to the geometric relationship between the normal vector and the base coordinate system of the manipulator, the opening angle of the gripper and the normal angle of the contact surface are calculated, and the grabbing attitude quaternion is output; Based on the weighing sensor data, the yarn mass center offset is dynamically predicted, the anti-swing lifting trajectory equation is calculated through a rigid body dynamics model, and the lifting distance-time sequence is output; The grabbing attitude quaternion and the lifting distance-time sequence are fused to generate a set of grabbing parameters containing spatial coordinates, attitude angles, lifting curves and clamping force thresholds.

4. The method of claim 3, wherein, The method comprises: The set of grabbing parameters is compiled into machine instruction code, and after a digital signature is attached, it is sent to the servo control system to trigger the mechanical hand action chain and output the execution log hash value; Real-time collection of mechanical hand joint encoder data, six-axis force sensor readings, and visual positioning feedback, detection of execution deviation through action sequence verification algorithm, and output of action compliance flag; When the compliance flag is true, encapsulate the device state, weight data, position coordinates, and execution log hash value to generate the original verification data block; Compress the original verification data block through the Merkle tree and upload it to the digital twin to generate a physical execution verification data package with a blockchain fingerprint.

5. The method of claim 4, wherein, The system includes: The acquisition module is configured to: Deploy a laser interference grid positioning system on the track of the creel car, emit phase-modulated laser beams in real time, calculate the displacement of the creel car through the reflected signals, and output the original positioning coordinate stream; Install a high-frame-rate CMOS photoelectric sensor at the outlet of the fiber forming machine to capture the surface microstructure changes of the yarn mass, and generate a yarn mass forming state pulse signal by combining a morphological edge detection algorithm; The output module is configured to: Based on the multi-source synchronous data set, calculate the three-dimensional coordinates of the yarn mass center and the lifting distance of the mechanical hand through the core positioning algorithm, and output the set of grabbing parameters; 6. A digital-twin-based intelligent glass fiber uncoiler system, characterized in that, The driving module is configured to: Send the set of grabbing parameters to the servo control system to drive the mechanical hand to complete the yarn mass grabbing-weighing-placing operation, and simultaneously upload the device state, weight data, and position coordinates during the operation to the digital twin virtual model to generate the physical execution verification data package; The verification module is configured to: Reconstruct the mechanical hand motion trajectory and the creel car inventory state in the digital twin virtual model, compare the physical execution verification data with the preset execution constraints, and if the physical execution verification data deviates from the preset execution constraints, trigger an abnormal early warning instruction and push it to the operation and maintenance personnel. The acquisition module is configured to:

7. The system of claim 6, wherein, Deploy a laser interference grid positioning system on the track of the creel car, emit phase-modulated laser beams in real time, calculate the displacement of the creel car through the reflected signals, and output the original positioning coordinate stream; Install a high-frame-rate CMOS photoelectric sensor at the outlet of the fiber forming machine to capture the surface microstructure changes of the yarn mass, and generate a yarn mass forming state pulse signal by combining a morphological edge detection algorithm; The output module is configured to: Based on the multi-source synchronous data set, calculate the three-dimensional coordinates of the yarn mass center and the lifting distance of the mechanical hand through the core positioning algorithm, and output the set of grabbing parameters; The driving module is configured to: Send the set of grabbing parameters to the servo control system to drive the mechanical hand to complete the yarn mass grabbing-weighing-placing operation, and simultaneously upload the device state, weight data, and position coordinates during the operation to the digital twin virtual model to generate the physical execution verification data package; The verification module is configured to: Reconstruct the mechanical hand motion trajectory and the creel car inventory state in the digital twin virtual model, compare the physical execution verification data with the preset execution constraints, and if the physical execution verification data deviates from the preset execution constraints, trigger an abnormal early warning instruction and push it to the operation and maintenance personnel. The weighing sensor data stream, the original positioning coordinate stream and the forming pulse signal are input into the encrypted PLC gateway, a hardware timestamp module driven by a cold atomic clock is embedded, and an encrypted data packet with a nanosecond-level time label is output; The encrypted data packet is recombined in the time domain through a timestamp alignment engine, transmission jitter is eliminated, and a multi-source synchronous data set containing four-dimensional data of timestamps, coordinates, weights and forming states is generated.

8. The system of claim 7, wherein, The output module is specifically used for: extracting the yarn ball forming state pulse signal from the multi-source synchronous data set, identifying the yarn ball boundary feature points through a wavelet ridge line detection algorithm, and outputting the boundary point cloud cluster; inputting the boundary point cloud cluster into an anti-vibration least square fitter, combining the positioning coordinate of the creel trolley to compensate the mechanical vibration offset, and outputting the optimized yarn ball center fitting normal vector; according to the geometric relationship between the normal vector and the base coordinate system of the manipulator, solving the opening and closing angle of the gripper and the normal angle of the contact surface, and outputting the grasping pose quaternion; based on the dynamic prediction of the yarn ball centroid offset by the weighing sensor data, calculating the anti-swing lifting trajectory equation through a rigid body dynamics model, and outputting the lifting distance-time sequence; fusing the grasping pose quaternion and the lifting distance-time sequence, and generating a grasping parameter set containing spatial coordinates, attitude angles, lifting curves and clamping force thresholds.

9. A storage medium, characterized by The storage medium has a computer program stored therein, wherein the computer program is configured to execute the method of any one of claims 1-5 when running.

10. An electronic device comprising a memory and a processor, characterized in that, The memory has a computer program stored therein, and the processor is configured to execute the computer program to execute the method of any one of claims 1-5.

Images