One thing one code whole life cycle tracking system

Abstract

The application relates to the technical field of intelligent manufacturing, in particular to a one-product-one-code whole life cycle tracking system. The system comprises an image binding module, a commodity automatic warehousing module, a safe shelf identification module and a commodity whole life cycle tracking module; commodity production data are acquired; a commodity unique serial number is generated according to the commodity production data; image acquisition is carried out based on the commodity unique serial number, and a commodity serial number image is obtained; label wrinkle data are obtained through label wrinkle detection according to the commodity serial number image; label fracture data are obtained through label fracture evaluation based on the label wrinkle data, and label optimization data are obtained by optimizing a label production process based on the label fracture data; commodity automatic warehousing data are obtained through commodity automatic warehousing simulation based on the label optimization data; and inventory dynamic management data are obtained through inventory dynamic management according to the commodity automatic warehousing data. The application improves commodity management precision and warehouse in-out efficiency based on intelligent manufacturing technology.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, and in particular to a one-item-one-code full lifecycle tracking system. Background Technology

[0002] Traditional one-item-one-code full lifecycle tracking systems rely on a single method of product identification binding, using only two-dimensional visual encoding methods such as barcodes or QR codes. This fails to deeply bind the product's structure and manufacturing process parameters, making it difficult to establish stable and counterfeit-resistant identification at the source. Image acquisition typically uses fixed angles and single lighting conditions, failing to adapt to the dynamic environment of multi-angle and complex lighting in assembly line operations. This leads to decreased image resolution and recognition stability, affecting tracking accuracy. The warehousing process generally lacks a label status recognition mechanism, failing to effectively detect issues such as warping, wrinkles, or damage to labels during handling and stacking, resulting in tracking path interruptions. Label quality optimization lacks a closed-loop feedback mechanism, failing to dynamically adjust label materials, structures, or printing process parameters based on label failure data from actual use, making systematic label stability optimization difficult. Inventory management relies heavily on fixed sensors or manual inventory checks, lacking real-time location recognition and dynamic storage optimization capabilities linked to product status data. This is particularly problematic in dense storage scenarios, where there is insufficient identification of hidden security risks such as weld cracks and label obstruction. Summary of the Invention

[0003] Therefore, it is necessary for the present invention to provide a one-item-one-code full lifecycle tracking system to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, the one-item-one-code full lifecycle tracking system includes the following modules:

[0005] The image binding module is used to acquire product production data; generate a unique product serial number based on the product production data; and acquire an image based on the unique product serial number to obtain the product serial number image.

[0006] The automated product warehousing module is used to detect label wrinkles based on product serial number images to obtain label wrinkle data; to evaluate label breakage based on label wrinkle data to obtain label breakage data; to optimize label production processes based on label breakage data to obtain label optimization data; and to simulate automated product warehousing based on label optimization data to obtain automated product warehousing data.

[0007] The safety shelf identification module is used to perform dynamic inventory management based on automated goods receiving data, obtain dynamic inventory management data, calculate shelf utilization based on dynamic inventory management data, identify shelf welding cracks based on shelf utilization, obtain shelf crack data, and determine the safety shelf area based on shelf crack data.

[0008] The product lifecycle tracking module analyzes the product shipping path based on the safety shelf area to obtain product shipping path data; it simulates product delivery based on the product shipping path data to obtain product delivery data; and it verifies the serial number based on the product delivery data to obtain serial number verification data, which is then transmitted to the product management system to execute the product lifecycle tracking task.

[0009] This invention establishes a multi-dimensional, dynamically adaptive, one-item-one-code full lifecycle tracking system by constructing an image binding module, an automated product warehousing module, a safety shelf identification module, and a product lifecycle tracking module. This overcomes numerous technical bottlenecks in traditional systems regarding identifier binding, image acquisition, label quality recognition and optimization, inventory safety monitoring, shipping path analysis, and serial number verification. The system introduces a product production data-driven mechanism in the product identification binding stage, deeply binding manufacturing process parameters to the serial number during the serial number generation phase, ensuring the uniqueness and unforgeability of product identification. Compared to traditional two-dimensional encoding methods such as barcodes and QR codes, this method constructs an identity mapping model based on manufacturing process characteristics, improving the data reliability of the tracking starting point. During image acquisition, the system dynamically adjusts the image acquisition mechanism based on the generated unique product serial number, enabling the linkage of multi-angle, high-frame-rate image sensors. It automatically adapts to complex lighting and viewing angle changes in the production line environment, effectively improving the clarity and stability of the acquired images and enhancing the robust recognition capability of product serial number images. In the goods receiving stage, the system introduces a label wrinkle detection and breakage assessment process. This not only enables real-time monitoring of label status but also extracts physical deformation characteristics from failed label data collected during actual receiving, forming a closed-loop feedback path. Based on this feedback data, the system dynamically optimizes label process parameters such as material, size, adhesive strength, and printing parameters, thereby gradually improving overall label production quality and damage resistance, and reducing tracking breakpoints caused by label failure. Simultaneously, the receiving simulation stage fully considers real-world variables such as warehousing routes, transfer cycles, and stacking strength in its structural design. Through physical modeling, it dynamically verifies the goods receiving process, providing parameter support for actual deployment. In the inventory management stage, a goods status data linkage mechanism is introduced on top of traditional warehouse location identification. By dynamically analyzing goods receiving behavior and label status changes, the system improves the timeliness of inventory information. The system can calculate shelf utilization based on real-time warehouse location status. Combined with image recognition results, it automatically detects micro-cracks, deformation areas, or label obstruction at shelf welding points, achieving proactive detection of hidden structural damage. This assists the system in automatically delineating safe shelving areas, preventing goods from being stacked in locations with potential structural risks. In the goods dispatching process, the system breaks away from static path planning, integrating actual physical parameters of the dispatching path, such as shelf slope, aisle width, and ground vibration intensity. Combined with inventory data and safe shelving areas, it dynamically performs optimal path simulation analysis. This approach avoids the risk of secondary damage to fragile goods caused by high-vibration, high-slope aisles, improving the rationality of dispatching path selection and logistics stability.The serial number verification section introduces an image-based fault-tolerant recognition mechanism, which no longer relies solely on string comparison. Instead, it repairs and completes images and patterns by addressing visual anomalies such as incomplete, scratched, or tilted markings in the product shipping images. This ensures that serial number comparison and tracking loop maintenance can still be completed even when the printing is incomplete, significantly enhancing the system's fault tolerance and tracking continuity. Attached Figure Description

[0010] 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:

[0011] Figure 1 This is a schematic diagram of the modules of the one-item-one-code full lifecycle tracking system of the present invention;

[0012] Figure 2 This is a detailed functional flowchart of the image binding module of the present invention;

[0013] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0014] The technical method of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this invention.

[0015] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0016] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0017] To achieve the above objectives, please refer to Figures 1 to 2 This invention provides a one-item-one-code full lifecycle tracking system, which includes the following modules:

[0018] S1: Image binding module, used to acquire product production data; generate a unique product serial number based on the product production data; and acquire an image based on the unique product serial number to obtain the product serial number image;

[0019] In this embodiment, an industrial vision acquisition terminal is deployed at the end of the product production line. It employs a 2448×2048 resolution global shutter monochrome industrial camera (such as a Basler acA2440-75gm) equipped with a 5mm fixed-focus lens and uses a white LED array (wavelength approximately 400nm~700nm) for uniform vertical illumination. This vision terminal connects to an embedded control unit (NVIDIA Jetson AGX Xavier) via the GigE industrial camera acquisition interface to achieve high-speed image acquisition control and edge computing processing. Based on the product production batch information (including production date, shift number, production process parameters, etc.) provided by the upstream MES system, a unique serial number is generated using a built-in serial number generation rule. This rule consists of 12 digits, including a 2-digit product type code, a 4-digit production timestamp (format: YYMM), a 2-digit production line number, and a 4-digit serial number. For example: A11205230125. After generation, the serial number is printed onto a PVC label using a thermal transfer printer (Zebra ZT411) in bold, 12-point font with a 0.8mm letter spacing, and automatically bound to the product during packaging. Subsequently, an industrial camera simultaneously captures a high-resolution image of the labeled area, and the serial number image is extracted using image processing algorithms. The image extraction steps include image binarization (threshold set to 0.6 grayscale value), region contour recognition (area threshold of 400 pixels² or higher), character segmentation (with a standard spacing of 1.5mm), and region enhancement (CLAHE histogram equalization). After processing, the image file is named the serial number in PNG format and stored in a local cache, then written to the product master database to bind the unique serial number to the actual product image.

[0020] S2: Automated Goods Inbound Module, used to detect label wrinkles based on product serial number images to obtain label wrinkle data; to evaluate label breakage based on label wrinkle data to obtain label breakage data; to optimize label production process based on label breakage data to obtain label optimization data; and to simulate automated goods inbound based on label optimization data to obtain automated goods inbound data.

[0021] In this embodiment, panoramic images of the labels in the serial number image area are acquired by dual vision detection positions (one color industrial camera with a resolution of 1920×1200 on each side) set up in the warehousing channel. The images are transmitted to a GPU server (using an RTX A6000) to execute a label wrinkle detection algorithm. This algorithm extracts the label edges based on Canny edge detection (with upper and lower thresholds of 70 and 150, respectively), and then uses Hough line transform to analyze the number of broken line segments. If more than 5 continuous curve segments with an edge curvature change angle greater than 30° are detected, they are marked as wrinkled labels. For wrinkled labels, feature map orientation gradient analysis (using the Sobel operator, window size 3×3) is used to further determine whether there are label breaks. If the gray-level abrupt change area exceeds 15% of the label area and is distributed in the central axis region (center ±10mm), it is judged as a break. The breakage data is recorded by number and fed back to the label optimization system. The label optimization data consists of the breakage rate (%) and the breakage distribution density (numbers / mm). 2 The data, along with the label paper handling force (N), is input to the thermal transfer process control module. Process optimization is achieved by changing the label material to PET substrate, adjusting the label thickness from 0.08mm to 0.12mm, increasing the label temperature from 200℃ to 215℃, and reducing the printing speed from 120mm / s to 100mm / s. The optimized scheme is implemented in the next batch of labels, and a version mark is added to the system. After optimization, based on the existing inbound path simulation engine (using the AnyLogic simulation platform), the process of goods entering the storage location under different label states is simulated. Input parameters include label detection pass rate (90%), handling distance (1.5m), shelf height difference (0.6m), and relative humidity of the storage area (45%). The simulation yields inbound efficiency data and label integrity rate data, and outputs automated inbound data for use in the next module.

[0022] S3: Safety shelf identification module, used for dynamic inventory management based on automated goods receiving data, to obtain dynamic inventory management data; calculate shelf utilization rate based on dynamic inventory management data; identify shelf welding cracks based on shelf utilization rate, to obtain shelf crack data; and determine safety shelf areas based on shelf crack data.

[0023] In this embodiment, the safety shelf identification module first retrieves automated product warehousing data, including specific storage location numbers, shelving times, product specifications, and label integrity. UWB positioning tags (positioning accuracy ±10cm) embedded in each shelf row are used to record the precise current location of the products. A matching RFID reader (such as Impinj Speedway R420) periodically scans the shelf RFID chips to record the product's storage status. Based on the storage location time series data, a time sliding window (sliding period of 3 hours) is used to count the actual number of products carried on each shelf layer, and this is divided by the shelf's maximum design capacity (provided by structural design data) to obtain the shelf utilization rate, rounded to three decimal places. If the shelf utilization rate exceeds 95%, and there are no significant product displacement records in the same area within the past 48 hours, the shelf is determined to be a high-voltage load area. To further assess the safety of this area, structural vibration sensors (sampling frequency 500Hz, sensitivity 50mV / g) deployed at key shelf nodes are used to record triaxial vibration waveforms after each manual shelving operation. Wavelet packet transform is used to analyze signal energy changes. If the proportion of high-frequency signal energy in a certain node area increases by more than 20% from the average, and the shelf utilization rate exceeds the set safety threshold of 90%, then ultrasonic crack detection is performed on the welded node of that shelf (using a 25MHz probe, with a maximum detection depth of 12mm). If the signal reflection amplitude in the detection result is greater than 80dB and the waveform repeats more than twice, then the node is marked as having a weld crack, its spatial coordinates are located, and it is marked as a danger zone.

[0024] S4: The product lifecycle tracking module analyzes the product shipping path based on the safety shelf area to obtain product shipping path data; simulates product delivery based on the product shipping path data to obtain product delivery data; verifies the serial number based on the product delivery data to obtain serial number verification data, and transmits it to the product management system to execute the product lifecycle tracking task.

[0025] In this embodiment, the product lifecycle tracking module calls the safety shelving area map and combines it with product outbound demand data for path analysis. The system uses Dijkstra's algorithm to construct a 3D shelving map structure, where each node represents a safe and feasible passage. Edge weights are set based on passage width (unit weight is 1 / w, where w is the width in meters), slope (increases weight by 1.5 for every 5% increase in slope), and ground friction coefficient (increases weight by 1 for values ​​below 0.5), comprehensively constructing a path cost function. The path search range is limited to within the safety shelving area, generating the shortest path solution and writing it into the outbound task list. Based on the path solution, the outbound simulation module is called to schedule a three-axis mobile AGV (load ≤ 60kg) to complete the outbound simulation task. The scheduling instructions are set to a maximum running speed of 0.7m / s, a turning angle not exceeding 30°, and an obstacle avoidance distance of at least 0.3m. For each shelving task point completed along the path, the completion time, path deviation angle, actual time consumed, and other data are automatically recorded, and the outbound simulation data is output. After the shipment simulation is completed, the serial number verification process begins, employing a dual mechanism: the first layer uses OCR to identify the serial number extracted from the shipment image (accuracy threshold must reach above 98%); the second layer uses a structural matching comparison algorithm, employing the Levenshtein distance algorithm to analyze and extract the differences between the serial number and the original number in the database, allowing a maximum edit distance of 1; exceeding this distance results in a verification failure. Verified product serial numbers, along with their paths, inbound / outbound status, label process version number, shelf number, and other data, are collectively compiled to form a complete tracking record. This record is then transmitted in real-time to the product master management system via the MQTT protocol and stored in the full lifecycle database, completing the closed-loop tracking process.

[0026] As an example of the present invention, reference is made to... Figure 2 As shown, Figure 1 A functional flowchart of the image binding module is shown in this embodiment. The functions of the image binding module include:

[0027] S11: Obtain product production data and extract production batch features to obtain production batch data;

[0028] In this embodiment, a Siemens S7-1500 series PLC control system is deployed on the production line, and a Siemens ET 200SPI / O module based on the Profinet industrial Ethernet communication protocol is installed to collect data. Each key production equipment has an independent data acquisition point. The types of data collected include raw material number (barcode scanner scan result), production timestamp (generated by the PLC's built-in RTC), operation procedure number (assigned based on the process flow diagram in Siemens TIA Portal), equipment number (encoded with a fixed production line ID and location number), equipment operating status (obtained using the Modbus protocol to acquire operating status bit information, 0 for shutdown, 1 for operation), and process environmental parameters (using thermocouples to detect station temperature, model K thermocouple, accuracy ±1℃, sampling period of 1 second). The collected data is uniformly packaged in JSON format and transmitted to the InfluxDB database, with the timestamp field set as the primary key. Every 10 minutes, a batch cycle is created as a production batch unit. Aggregation operations are performed based on the fields "material_code", "workstation_id", "start_time", and "end_time". The aggregation logic is as follows: if all products generated within a consecutive 600-second window share the same raw material number and equipment number, they are grouped into one batch. The batch number consists of the format "PRD + year / month / day + production line ID + time period number", such as "PRD20250528-L05-T06", and is written as the primary key index into the batch table "production_batch_list".

[0029] S12: Analyze product structure parameters based on production batch data;

[0030] In this embodiment, based on each batch number generated in S11, the fields "material_code" and "assembly_step_code" are retrieved from the database and used as indexes to query the local structural parameter mapping database. The database is built using PostgreSQL and named "product_structure_definitions". The database's built-in table fields include: product type number (e.g., ST3L is a three-layer stacked structure), number of layers (number of layers, integer value 1-5), connection method (e.g., ultrasonic welding, threaded fixing), key structural dimensions (e.g., length 60mm, width 40mm, thickness 5mm), number of connection points (defined according to CAD drawings, e.g., 8 point-to-point welding positions), tolerance range (written in ±0.03mm units), connection method standard number (e.g., USW-03 represents the third type of ultrasonic welding standard), and material code and structural component number binding information. The parsing process is completed by a Python script, which reads data through the psycopg2 database connection interface. The field comparison logic is as follows: if the "material_code" in the batch data is "A7075T6", it is mapped to the structure code "ST3L", and then mapped to the parameter set containing the above-mentioned structure hierarchy, connection method, and other fields through the structure code. The parsed structure parameters are stored in the form of a structure, including the contents of all structure fields and the batch number to which they belong, and a structure parameter description document is generated, named "BATCH_batch number_STRUCT.json".

[0031] S13: Obtain product production traceability data by binding production timestamps to product structure parameters;

[0032] In this embodiment, by calling the PLC's historical operation records and process logs, and relying on the PLC log buffer module (LogDataBlock), the key process node times for each product are extracted, including the initialization start time (field init_time), welding step completion time (field weld_finish_time), dimensional inspection completion time (field dim_check_time), and final packaging completion time (field final_pack_time). All times are represented in UTC format, accurate to milliseconds, for example, "2025-05-28T14:32:56.712Z". This time data is centrally collected through the Siemens WinCC SCADA system and synchronized with the database using the industrial middleware platform KepServerEX. The timestamp data is uniformly stored in a table named "product_time_log", with "product_id" as the unique foreign key. Subsequently, the fields "structure_code", "layer_number", and "joint_type" are extracted from the structural parameter document and combined with the above four time nodes and the equipment number "workstation_id" to form a complete traceability data record. Each record is organized in structured JSON format, with the field order fixed as: {product_id,structure_code,init_time,weld_finish_time,dim_check_time,final_pack_time,workstation_id}, and written to a table named "product_traceability" with "product_id" as the primary key. Each field must be fully populated and null values ​​are not allowed.

[0033] S14: Based on the product production traceability data, perform encrypted hashing of the product identifier to obtain the identifier code;

[0034] In this embodiment, in step S13, the fields "structure_code", "init_time", "final_pack_time", and "workstation_id" are extracted and concatenated in a fixed order. The specific concatenation rule is as follows: "structure_code + init_time + workstation_id + final_pack_time". An example concatenation is shown below: "ST3L20250528143256712WS19020250528143501421". This string is then irreversibly encrypted using the SHA-256 hash algorithm. The hash algorithm is implemented by calling the SHA256_Init(), SHA256_Update(), and SHA256_Final() functions in the openssl encryption library. Before processing, the original concatenated string is processed according to UTF-8 encoding format and converted into a byte stream input. The generated hash result is a 256-bit binary value, which is converted to a 64-bit hexadecimal string, such as "a7d1e1c74beaeed14c1bda4a6d03dbf40794fe8b1c132a7a0f9b1fc2e8384ef3". This hash result is the product identification code. The identification code is stored in the "product_hash_index" table, with the table structure: {product_id VARCHAR, hash_code CHAR(64), creation_time TIMESTAMP}, where "product_id" is a foreign key, consistent with the unique product number in S13, and the hash_code field has a unique index.

[0035] S15: Standardize the sequence format based on the identifier encoding to generate a unique serial number for the product;

[0036] In this embodiment, a unique product serial number is generated based on the hash code. The serial number format is set to "YYYYMMDD+LINEID+LAYER+HASH10", where YYYYMMDD comes from the date value extracted from the init_time field, LINEID comes from the first three digits of workstation_id (e.g., L19 for WS19), LAYER comes from structure_code (e.g., 3L for ST3L), and HASH10 is the first 10 characters of the identifier code. Example: If the identifier code is "a7d1e1c74beaeed1...", then HASH10 is "A7D1E1C74B", which indicates that the encoded serial number is specifically as follows: "20250528L193LA7D1E1C74B". The regular expression r'\d{8}L\d{2}\dL[A-F0-9]{10}' is used to verify the validity of the serial number format. All generated results are written to the "product_identity_list" table. The table structure fields include: {product_id, serial_number, hash_code, serial_time, serial_format_version}, where serial_format_version is the current version identifier, which is fixed as "V1.0". All field values ​​must be fully filled and there should be no empty values. The serial_number field is set with a unique index to prevent generation conflicts.

[0037] S16: Collect images based on the unique serial number of the product to obtain the product serial number image.

[0038] In this embodiment, a vision acquisition module consisting of a Basler acA2440-35um industrial camera is deployed at the end of the packaging production line. The camera has a resolution of 2448×2048 pixels and a shutter speed of 8500μs. Image acquisition is triggered by a PNP-type reflective photoelectric switch emitting a level signal to control the image acquisition time. The camera is equipped with a CCTV lens (25mm focal length, 50×50mm field of view), and a 470nm blue LED ring light is installed in front of the lens. A diffuser is also added to enhance the contrast of the QR code area. The unique serial number is printed using a Zebra ZT510 thermal transfer printer in 300dpi mode with a 24-point font size. The serial number is fixed at the upper left corner of the front of the product, with the edge no more than 5mm from the edge. After image acquisition, the `cv2.findContours()` function from the OpenCV library is called to identify the serial number text region. `cv2.getRectSubPix()` is used to crop the image to a standard 300×300 pixel region, and `cv2.resize()` is called to resize the image to a uniform 400×400 pixels before saving it as a PNG file. The image is named "SERIAL_SerialNumber_Timestamp.png", and the file storage path is " / data / product_images / YYYY / MM / DD / ". The file index is synchronously written to a table named "serial_image_index", with fields: {serial_number, image_path, image_capture_time}.

[0039] Preferably, the image acquisition based on the product's unique serial number in the image binding module includes:

[0040] Generate image acquisition instructions based on the product's unique serial number;

[0041] In this embodiment, image acquisition control instructions are constructed by calling the instruction generation module based on the product's unique serial number. The acquisition control instructions consist of five fields: serial number ID, acquisition angle identifier (set to four angles: 0°, 45°, 90°, 135°), exposure time parameter (unit: ms, default: 100ms), light source intensity value (integer value from 0 to 255, default value: 200), and shooting resolution specification (set to 1920×1080 pixels). The generation of image acquisition control instructions relies on a matching mechanism between the established unique serial number and the shooting template library. During matching, the product's category label is used as the main index field, and the corresponding standard shooting parameter combination is retrieved from the template library. The control instructions are encapsulated in hexadecimal encoding format to ensure compatibility with the image acquisition control module's parser standard during serial communication.

[0042] The image acquisition device is scheduled according to the image acquisition command, and the device is triggered to capture the image signal;

[0043] In this embodiment, image acquisition scheduling is implemented by a PLC controller, specifically a Siemens S7-1500 series PLC, which communicates synchronously with the industrial control host via PROFINET. The image acquisition device is a Basler acA1920-40uc industrial camera. After receiving the image acquisition command, the PLC parses the acquisition angle and resolution parameters according to the command fields. Subsequently, the PLC controls the servo motor to drive the rotating bracket to a specified angle position with an accuracy of ±0.2 degrees. An Omron E6B2-CWZ6C encoder is used to provide feedback on the angle position. Once the set position is reached, the PLC sends a 5V high-level shooting trigger signal to the camera through its output port. The signal duration is 20ms, used to control the camera shutter. After the shooting action is completed, the camera uploads the image data to the image processing server via a USB 3.0 interface.

[0044] The original image dataset is obtained by acquiring multi-angle product images based on the device's shooting signals.

[0045] In this embodiment, image acquisition employs a four-angle shooting method with angles of 0°, 45°, 90°, and 135°. One shooting operation is performed at each angle, ultimately obtaining four complementary product images. The exposure time for each shot is strictly set to 100ms, and a fixed-value exposure control method is used to disable automatic exposure. Acquired images are saved in RAW format, uncompressed and unpreprocessed, with each image file approximately 2MB in size. The file naming convention is "serial number_angle number_timestamp.raw". After image acquisition, the image acquisition workstation centrally schedules and archives the images according to their serial numbers into the primary product image data directory, forming the original image dataset. Each product number corresponds to a folder containing four images.

[0046] The original image dataset is subjected to resolution correction to obtain standardized image data.

[0047] In this embodiment, the `cv2.resize()` function from the OpenCV 4.5.1 library is used to standardize the image resolution, with the target size uniformly set to 1920×1080 pixels. Before correction, the image size is checked; if the size is incorrect, linear interpolation (INTER_LINEAR) scaling is performed. If the image aspect ratio is abnormal (aspect ratio error exceeds ±5%), edge padding is added without changing the pixel content. The fill color is the RGB average of the image edge 5 pixels. After image size standardization, the image color space is uniformly converted to BGR format, and grayscale processing is performed to generate a grayscale image. The grayscale conversion uses a weighted average method (Y = 0.299R + 0.587G + 0.114B), and histogram equalization is used to improve image contrast. Standardized image data is stored in PNG format, and each image is named according to the rule "serial number_angle number_std.png".

[0048] By binding a unique serial number to the product based on standardized image data, a product serial number image is obtained.

[0049] In this embodiment, the binding of image file naming and unique serial numbers is achieved through a database index. A table named "product_image_binding" is created based on MySQL 8.0, with fields including: serial number (CHAR(32)), image path (VARCHAR(255)), acquisition angle (INT), and shooting time (DATETIME). After each standardized image is entered into the database, the image processing thread extracts the image path and filename, binds them using the serial number instruction queue in memory, and writes them to the database. This database supports bidirectional queries, allowing image location by serial number and serial number lookup by image path. The final definition of the product serial number image is the image set corresponding to the product in this binding result set, supporting subsequent image recognition and anomaly detection.

[0050] Preferably, the label wrinkle detection based on the product serial number image in the automated product warehousing module includes:

[0051] Identify the outline of serial number characters from the product serial number image;

[0052] In this embodiment, character contour recognition employs a combination of morphological processing and contour extraction. The standardized image is first subjected to Canny edge detection, with a low threshold of 50 and a high threshold of 150. Then, a 3×3 rectangular structuring element is used for two erosion processes, followed by a dilation process to break up non-character boundary lines and highlight character boundaries. The OpenCV function `cv2.findContours()` is called to extract contours, with the contour retrieval mode set to `RETR_EXTERNAL`, retaining only the outermost contour. When filtering character contours, preliminary filtering is performed based on contour area (threshold set to 500 pixels) and contour aspect ratio (limited to 0.2–1.5), retaining suspected character regions and outputting the vertex coordinate set of the contour polygon for subsequent character line deformation analysis.

[0053] Calculate the character line deformation rate based on the character outline of the serial number;

[0054] In this embodiment, the character line deformation rate is calculated using linear regression and contour baseline comparison. For each character contour, the boundary point sets of its upper and lower edges are extracted. Least square linear fitting is performed on each set of boundary points to obtain the slopes of the fitted lines for the upper and lower edges (denoted as k_top and k_bottom). If the character is printed normally, the difference between k_top and k_bottom should be within ±0.05. The deformation rate is defined as |k_top-k_bottom|, in dimensionless floating-point numbers. When the deformation rate exceeds 0.1, the character is considered to have significant bending or stretching. The deformation rate data is mapped one-to-one with the character region positions to construct a character deformation rate matrix for subsequent detection.

[0055] Label warping data is obtained by detecting label warping based on the character line deformation rate.

[0056] In this embodiment, based on the deformation rate data of character contours, characters in the edge region (within 50 pixels of the image edge) are extracted. If the deformation rate is continuously greater than 0.1, it is determined to be a label edge lifting region. Using these region coordinates as the center, the detection area width is expanded to 20 pixels, and surrounding image blocks are extracted for edge gradient analysis. The Sobel operator is called to calculate the gradient direction and intensity of the image blocks. If the average gradient intensity is greater than 30 and the direction is concentrated within ±30°, lifting is determined to exist. The coordinate range and gradient vector direction of this region are extracted to form the label lifting data structure, which serves as the target area for subsequent laser contour acquisition.

[0057] Laser 3D contour acquisition is performed based on label warping data to obtain the label warping 3D contour.

[0058] In this embodiment, a Keyence LJ-V7000 series laser profilometer was used for 3D contour acquisition. The light source wavelength was set to 405nm, the scanning speed to 800Hz, and the scanning resolution to 5μm. Based on the coordinate range in the warped edge data, the moving platform was controlled to position the product within the scanning range. A PLC was used to control the laser head's start and stop. During scanning, X, Y, and Z 3D point data were automatically recorded, with the Z-axis representing the warped height and an accuracy of 0.01mm. The contour data was exported in CSV format, with fields representing the X, Y, and Z coordinates. The acquisition time for a single warped edge area did not exceed 0.5 seconds. After each acquisition task, coordinate normalization was performed to align all point cloud data to the product's center point.

[0059] Shear stress was calculated based on the three-dimensional contour of the label warped edge to obtain shear stress data;

[0060] In this embodiment, after acquiring the 3D contour of the label edge warping, shear stress needs to be calculated based on this 3D data. During the operation, the 3D contour data is imported into a 3D processing platform in point cloud format. The platform selected is PolyWorksInspector, and the point cloud format is loaded as .ply. The system divides the label edge warping area into partitions using a regular grid of 5 mm by 5 mm, extracts the coordinates of points within each grid, and fits a local surface. The fitted surface uses the least squares method to fit a quadratic surface. By extracting the changes in the tangent slope of the local surface in the X and Y directions, the deformation tensor of the point in different directions is obtained. Combined with the shear modulus value in the material parameters, which is set to 1.6 x 10^8 Pascals, the shear stress value corresponding to each grid is calculated according to the shear deformation in each direction. Finally, a shear stress distribution matrix is ​​established in the entire image. This matrix structure represents the shear stress value corresponding to each grid, in Pascals, and the matrix is ​​exported as a CSV file for subsequent steps.

[0061] Identify shear stress concentration regions based on shear stress data;

[0062] In this embodiment, based on the exported shear stress data, a heatmap rendering of the shear stress field is performed using a data visualization tool. The Matplotlib tool is used to process the shear stress matrix using the `imshow` function, mapping the shear stress values ​​to color depths to represent intensity distribution. A critical shear stress threshold of 8 MPa is defined; empirical analysis shows this value can be used to identify the stress range at which the label film enters the plastic deformation stage. Grid locations with stress values ​​greater than this threshold are marked, and the `findContours` function of OpenCV is used to extract all continuous stress regions satisfying this condition. Noise interference regions with excessively small areas are further excluded from the marked results; an area threshold of 3 square millimeters is set, retaining only connected regions with areas greater than this threshold as effective shear stress concentration regions. The minimum bounding rectangle boundary of each effective region is extracted using the `boundingRect` function of OpenCV, and the principal direction angle of each region is extracted using principal component analysis. This angle represents the principal direction of shear stress in that region. All results are saved in JSON format, containing the region's number, boundary coordinates, center position, average shear stress value, and principal direction angle information.

[0063] The wrinkle orientation is predicted based on the shear stress concentration area, and wrinkle orientation data is obtained.

[0064] In this embodiment, the principal orientation angle of the identified shear stress concentration region is used as the basic reference for predicting the wrinkle direction, and the stress direction information in each region is further processed. The principal orientation vector is extracted using the PCA method for the shear stress value distribution within each region. The orientation vectors of all stress points are averaged to obtain the global average orientation angle of the shear stress direction within each region. This angle is the direction of potential wrinkle formation. The predicted wrinkle orientation angles of each region are uniformly organized into a structured list, recording the region number, predicted wrinkle orientation angle, and orientation vector information, and an annotation layer is established by aligning it with the original labeled 3D point cloud map.

[0065] Label wrinkle detection is performed based on wrinkle direction data to obtain label wrinkle data.

[0066] In this embodiment, label wrinkle detection is performed based on the predicted wrinkle direction angle. First, the predicted direction angle is used as the reference direction for image directional filtering. A directional gradient map is constructed by extracting the Z-axis height change along the predicted direction in the 3D point cloud. During this process, the Sobel operator is used to perform first-order differentiation on the point cloud data along the predicted direction to identify regions of abrupt height changes on the label surface. A mutation identification threshold of 0.4 mm is defined, and the minimum continuous mutation edge length is set to 10 mm to ensure the exclusion of discontinuous interference edges. All mutation edges meeting the above conditions are connected using a continuous edge tracking algorithm and identified as suspected wrinkle paths. Simultaneously, the maximum curvature position and its direction are extracted on each path. All identified wrinkle paths are summarized by region number, and the recorded content includes path length, start and end point coordinates, direction angle, maximum curvature position coordinates, and curvature value. Finally, all detection data is summarized and output as an XML structure document, which is used for updating structured fields in the label status tracking database. After this wrinkle information is bound to the product's unique serial number, it can serve as one of the constraints in the product warehousing quality control process, realizing a closed-loop labeling of the quality status of a one-item-one-code full lifecycle tracking system.

[0067] Preferably, the automated product warehousing module optimizes the label production process based on label breakage data, including:

[0068] Detecting the fiber orientation of materials based on label wrinkle data;

[0069] In this embodiment, after obtaining the label wrinkle data, to detect the material fiber arrangement direction, it is necessary to locate the wrinkle region in the three-dimensional contour data and perform micro-texture analysis on the wrinkle path along the vertical direction. First, using the wrinkle path data extracted in the previous step, a rectangular region with a length of 20 mm and a width of 5 mm is extracted with the center of each path as the reference, and point cloud profile data is extracted within this region at a step size of 0.1 mm. By extracting the height change in the Z-axis direction and the projection change in the X-axis or Y-axis direction, a surface texture gradient map of each region is constructed. A two-dimensional Fourier transform is used to perform frequency domain analysis on the texture gradient map to extract the dominant frequency direction angle, which is the material fiber arrangement direction angle in degrees. This angle is then bound to the region coordinates and output as a structured data table. Each region record field includes wrinkle number, dominant texture direction angle, texture direction intensity value (in dB), and coordinates of the region center point.

[0070] Of particular importance is the fiber orientation of the test material, which includes:

[0071] Based on the texture extraction processing of the label wrinkle data, a grayscale texture map of the wrinkle region is obtained;

[0072] In this embodiment, the collected label image data is first segmented locally on the label surface using an image cropping tool to extract the wrinkled areas caused by tension interference. The label image resolution is no less than 300 DPI, and the image size is maintained at 800×600 pixels or more to ensure clear image details. After the wrinkled areas are determined, a grayscale mapping algorithm is used to convert the RGB label image into a grayscale image. In the weighted average conversion formula used, the weights of the red, green, and blue channels are set to 0.299, 0.587, and 0.114, respectively. After the grayscale image is generated, a local variance filter is called to extract areas with significant grayscale differences. The filter kernel window size is 5×5 pixels, and the sliding step size is 1 pixel. Subsequently, a Gabor filter is applied to enhance the texture direction features. The Gabor kernel parameters are set to a frequency of 0.25, and the direction angle θ is set every 15° from 0° to 180°. The response amplitudes at all angles are superimposed and saved as a texture response map. The output image is the grayscale texture map of the wrinkled area.

[0073] Calculate the texture direction gradient based on the grayscale texture map of the folded region;

[0074] In this embodiment, the Sobel gradient operator is used to perform two-dimensional spatial derivative operations on the grayscale texture map of the wrinkled region. The Sobel operator performs filtering operations in the x and y directions, with convolution kernels of [-1,0,1],[-2,0,2],[-1,0,1] and [-1,-2,-1],[0,0,0],[1,2,1], respectively. After convolution, gradient maps in the Gx and Gy directions are obtained. Then, the texture direction angle is calculated for each pixel using the directional gradient calculation formula arctangent(Gy / Gx), and this angle is mapped to the range of 0–180°. To remove background noise interference, a pixel gradient magnitude threshold of 20 is set, and pixel angle data below this threshold are discarded. Next, the effective angle data are grouped into 5° sets for angle frequency statistics, and the statistical results form a direction histogram, which serves as the input data source for subsequent fitting.

[0075] Based on the texture direction gradient, the main direction of the folds is fitted to obtain linear fitting data of the main direction of the folds.

[0076] In this embodiment, based on the obtained orientation histogram data, a minimum mean square error fitting method is used to perform a linear regression fitting on the angular frequency distribution curve, aiming to extract the densest orientation region. First, the orientation histogram is smoothed using a weighted average method with a moving average window width of three angular intervals to smooth noise fluctuations. In the smoothing result, the location of the maximum frequency and its ±10° range are selected as the main fitting segment, and scatter plot data is constructed based on the angle values ​​and their frequencies within this segment. Then, the least squares formula is used to fit the angle-frequency scatter plot, obtaining the slope and intercept of the linear fitting equation. The maximum response angle corresponding to the fitting result is considered as the main fold direction angle. The final output includes linear fitting data of the main fold direction, containing parameters such as the fitting slope, fitting intercept, fitting confidence interval, and fitting angle, with angle precision retained to one decimal place.

[0077] The fiber arrangement direction of the material can be inferred from the linear fitting data of the main fold direction.

[0078] In this embodiment, the linear fitting angle of the main fold direction is used as the candidate main fiber direction, and direction mapping is performed according to the reference standard set by the material production process. Using the known label printing direction as the reference coordinate axis, the printing direction is defined as 0°. Directions consistent with it are defined as parallel directions, and directions perpendicular to it are defined as perpendicular directions. If the angle between the fitted angle and the reference direction is less than 15°, the fiber arrangement direction is determined to be "parallel"; if the angle between the fitted angle and the reference direction is between 75° and 105°, it is determined to be "perpendicular"; if neither of these conditions is met, it is marked as "tilted," and the specific angle is recorded. Finally, the material fiber arrangement direction data, including the arrangement direction labels (parallel, perpendicular, tilted) and corresponding angle values, is output and simultaneously uploaded to the material data label library of the "One Item One Code Full Lifecycle Tracking System" as a record item for fiber structure traceability.

[0079] Calculate the directional offset based on the fiber arrangement direction of the material;

[0080] In this embodiment, after obtaining the fiber arrangement direction of the material, a comparison with the reference fiber arrangement direction in the original design process is needed to calculate the directional offset. This design direction data comes from the standard fiber direction defined in the label manufacturing process database, set as a horizontal direction, i.e., an angle of 0 degrees. The difference between each actually measured material fiber arrangement angle and the standard angle is calculated to obtain the directional offset angle, with the result in degrees. To eliminate errors caused by directional symmetry, all angle differences are calculated using the smallest included angle method, for example, the offset angle is the smaller value between |θ_actual–θ_standard| and 360–|θ_actual–θ_standard|. Simultaneously, to quantify the offset intensity of each region, the offset index is calculated by combining the amplitude of the dominant frequency component in the Fourier frequency domain results with the offset angle: offset P = offset angle × dominant texture direction amplitude, with the unit being degrees × dB. All offset data is output as a CSV file, with columns including region number, offset angle, dominant texture direction intensity, and offset value.

[0081] The imbalance of material fiber arrangement is determined based on the directional offset, and the imbalance data of material fiber arrangement is obtained.

[0082] In this embodiment, to determine the imbalance state of material fiber arrangement, a threshold judgment standard for offset degree needs to be set. Based on actual measurement statistics, the critical value of offset angle is set to 15 degrees, and areas with an intensity of less than 25dB in the main texture direction are considered low signal-to-noise ratio areas. Areas with an offset angle greater than 15 degrees and an intensity of less than 25dB in the main texture direction are marked as "strong imbalance," areas with an offset angle between 5 and 15 degrees or an intensity of 25 to 35dB in the main texture direction are marked as "medium imbalance," and the rest are marked as "weak imbalance" or "normal." The above judgment rules are automatically executed by a Python script, which reads the CSV file line by line and generates new structure fields according to the conditions. The output data structure includes: area number, offset angle, intensity, offset degree, and imbalance level. The imbalance level field is of integer type, with 0 indicating normal, 1 indicating weak imbalance, 2 indicating medium imbalance, and 3 indicating strong imbalance. This result is used to establish a mapping model between label manufacturing data and fiber arrangement state, and is written into the data field of the corresponding label number in the database.

[0083] Predicting crack initiation points based on material fiber arrangement imbalance data;

[0084] In this embodiment, after extracting the fiber arrangement imbalance state, it is necessary to further predict crack initiation points. This process is based on "strong imbalance" and "moderate imbalance" regions, and stress concentration trend analysis is performed on the point cloud data within these regions. During the operation, the Z-axis height change rate, surface curvature, and three-dimensional angle jump variables are extracted within a circular area with a diameter of 10 mm centered on the imbalance region. The Z-axis height change rate is extracted using a linear fitting slope with a step size of 0.2 mm, the surface curvature is calculated using Gaussian curvature, and the angle jump variable is obtained from the included angle of the surface tangents. The following thresholds for initiation indicators are set: a Z-axis slope greater than 0.3, a curvature value greater than 0.002 mm^-2, and an angle jump greater than 10 degrees are considered to indicate a risk of crack initiation. If two of the three parameters meet the conditions, the point is marked as a crack initiation point, and its spatial coordinates are recorded. If multiple crack initiation points exist in each tag, the three points with the largest curvature values ​​are extracted as key prediction points, sorted by curvature value. The final output data on crack initiation points is in JSON format, with fields including label number, crack number, spatial coordinates (X, Y, Z), local maximum curvature, directional jump variable, and slope value.

[0085] Tag fracture assessment was performed based on crack initiation points to obtain tag fracture data;

[0086] In this embodiment, when performing label fracture assessment based on the aforementioned crack initiation point data, fracture path estimation is performed using the connectivity between initiation points and the stress propagation trend. Specifically, the Euclidean distance between initiation points is calculated. If the distance between two initiation points is less than 5 mm and they are aligned at similar orientation angles (less than 15 degrees), a potential fracture path is considered to exist between the two points. A crack connectivity graph is constructed in this manner. For each connectivity graph, the path length, average path curvature, and number of Z-axis jumps are calculated. If the path length is greater than 10 mm or the average curvature exceeds 0.0015 mm^-2, the path is marked as a potential fracture path. All paths are integrated and output as a fracture data structure, containing path number, start and end coordinates, total path length, average curvature, maximum height jump value, and crack grade (weighted by length and curvature, grade 1 for minor, grade 2 for moderate, and grade 3 for severe). The data is exported in XML format and written to the label database.

[0087] The die-cutting contour position was traced back based on the label breakage data;

[0088] In this embodiment, the starting position of the path for cracks of level 2 or 3 in the label fracture data is used to trace back the corresponding die-cutting contour position. Specifically, the crack starting position is aligned to the label CAD contour layer in the label plane coordinate system, and the layer is stored in SVG vector format. Through coordinate matching, the shortest distance from the crack starting point to the die-cutting edge is calculated. If this distance is less than 3 mm and the direction is perpendicular to the edge segment, the fracture origin is considered to be related to the die-cutting edge. The path number and position range corresponding to this die-cutting edge segment in the die-cutting design file are recorded. All crack starting points are bound to their path numbers and output to form a die-cutting contour crack mapping table. The data structure includes the label number, crack number, corresponding die-cutting contour number, edge segment start and end point coordinates, and the relative angle of crack origin.

[0089] The die-cutting path error rate is calculated based on the die-cutting contour position; the chamfering structure parameters are optimized based on the die-cutting path error rate to obtain the chamfering structure optimization parameters;

[0090] In this embodiment, the die-cutting path error rate is calculated based on the aforementioned die-cutting contour position. The actual die-cutting trajectory is reconstructed and compared with the designed CAD contour. First, the actual die-cutting image is extracted and edge recognition is performed using a vision system. The extracted edge path is a continuous vector sequence, denoted as P_actual. The designed die-cutting path is imported as a CAD contour point sequence, denoted as P_design. The error of each segment is calculated using the average point distance of every 10 mm path segment. The specific formula for calculating the error rate is as follows: |P_actual(i) – P_design(i)| / L_design(i), where L_design is the design segment length. An error threshold of 0.5 mm is set. If the error of a single segment exceeds this threshold, the segment is marked as an abnormal segment. The ratio of the length of all abnormal segments to the total length is the overall error rate. The output error rate data includes label number, path number, total error rate, number and location of abnormal segments. Based on the error rate, die-cutting chamfer structure parameter optimization is performed, focusing on analyzing the chamfer curvature and chamfer radius in the error rate concentration area. Extract the locations of abnormal segments, and their corresponding CAD chamfer design parameters, including radius of curvature R and angle A. Establish a table mapping error rate to chamfer parameters, and extract the corresponding radius of curvature for the point with the maximum error rate. If the error is greater than 1%, the radius of curvature needs to be increased by 10%; if it is less than 0.5%, the radius of curvature remains unchanged; in between, the chamfer radius is adjusted linearly proportionally to the error. After updating the parameters, output a chamfer structure optimization table, with fields including label number, path number, original chamfer parameters, optimized chamfer parameters, and optimization percentage, in millimeters and percentages.

[0091] Calculate the die-cutting trajectory abruptness rate based on the die-cutting contour position; determine the die-cutting contour smoothness based on the die-cutting trajectory abruptness rate.

[0092] In this embodiment, the die-cutting trajectory abrupt change rate is calculated based on the die-cutting contour position. The tangent direction angle of each path segment is continuously calculated in the CAD path. If the angle between adjacent path segments is greater than 25 degrees, it is recorded as one abrupt change. The abrupt change rate = number of abrupt changes / total number of path segments. After calculating the abrupt change rate for each path, a abrupt change statistics table is generated. If the abrupt change rate is greater than 5%, it is marked as a non-smooth path. All path abrupt change rate results are output in JSON format, with fields including path number, total number of segments, number of abrupt changes, abrupt change rate value, and smoothness level.

[0093] By integrating the chamfer structure optimization parameters and the smoothness of the die outline, we obtain optimized production process data for labels.

[0094] In this embodiment, the optimized chamfer parameters and their mutation rate levels for each path are combined to form optimization operation suggestions for each path, and a list of optimization operations for all paths is compiled by label number. The output process data structure is as follows: label number, path number, optimized chamfer radius, mutation rate level, whether re-molding is required, and expected path error lower limit. All optimized process data are uniformly output as a structured database and written into a table, serving as input parameters for the subsequent label production simulation module.

[0095] Label production simulation was performed based on the label optimization production process data to obtain label optimization data.

[0096] In this embodiment, during the label production simulation stage, based on optimized process data, the SimCAM CNC die-cutting simulation system is used to import optimized die parameters and simulate the label die-cutting process. Material parameters are set to match the die parameters, including a material thickness of 60 micrometers, an elastic modulus of 2.2 GPa, and a die linear velocity of 0.3 m / s. The simulation process records the pressure distribution at the die contact point, the die feed trajectory, a fracture trend prediction heatmap, and the simulated three-dimensional label structure. All simulation data is structured and output as optimized label data, with fields including simulation number, fracture trend index, die-cutting accuracy, and die stress uniformity index. This data is ultimately used in the label traceability database to replace the original process data, forming "optimized label data" to support high-precision traceability and improvement recording of the label production path in the "one item, one code, full lifecycle tracking system."

[0097] Preferably, the automated product warehousing simulation based on tag-optimized data in the automated product warehousing module includes:

[0098] Import the optimized label data into the database simulation system;

[0099] In this embodiment, data synchronization with the industrial database is performed via the PLC control interface. The aforementioned tag optimization data is imported into the MES (Manufacturing Execution System) integrated database simulation module using the RS485 communication protocol. The imported data is in CSV standard format, with fields including "Label ID", "Label Size (mm)", "Label Material Code", "Optimized Fiber Arrangement Balance (%)", "Die Contour Smoothness Coefficient (%)", and "Breakage Probability Index (0-1)". During the import process, the data verification submodule performs field format verification, using the CRC32 algorithm to check the integrity of each row of data, with the error rate controlled within 0.01%. After the tag optimization data is imported, a 60-second buffer is set in the cache area to ensure that the writing is complete before entering the simulation execution stage.

[0100] In the inbound simulation system, the coefficient of friction of the commodity surface is set to 0.1-0.6, the center of gravity offset distance is 30-100mm, the surface reflectivity is 10%-90%, and the packaging compression modulus is 50kPa-500kPa.

[0101] In this embodiment, the physical attribute parameters of the goods are set through the HMI (Human Machine Interface) touch terminal and written into the parameter configuration table of the warehousing simulation system. The specific settings are as follows: The surface friction coefficient was measured statically using a physical friction testing instrument (model: FPT-C2000), with industrial-grade rubber sheet as the friction comparison material. The set range was 0.1 to 0.6, and the set particle size was 0.05. The center of gravity offset distance was measured using a three-dimensional coordinate measuring machine (model: Hexagon GLOBAL S) to measure the spatial displacement of the object's center of gravity relative to its geometric center. The set distance range was 30 mm to 100 mm, and the accuracy was ±0.1 mm. The surface reflectivity was measured using an integrating sphere spectrophotometer (model: CAS140CT), with a set range of 10% to 90% and an accuracy of ±2%. The packaging compression modulus was obtained by applying a constant rate of compressive load using a materials mechanics testing machine (model: Instron 5982) to obtain the stress-strain curve. The compression modulus was calculated as Δσ / Δε, with a set range of 50 kPa to 500 kPa and a minimum interval of 25 kPa.

[0102] In the warehousing simulation system, the warehousing conveyor belt running speed is set to 0.1m / s-1.5m / s, and the sorting robot arm gripping frequency is 30 times / min-180 times / min;

[0103] In this embodiment, the conveyor belt speed is set via servo drive control parameters, and the motor output speed is adjusted using a frequency converter (model: ABB ACS580). The operating speed is monitored in real time via a speed measuring wheel, with a set range of 0.1 m / s to 1.5 m / s and a set interval of 0.1 m / s. The gripping frequency is configured by a six-axis robotic arm controller (model: KUKA KR C4), and the robotic arm's motion cycle is calculated based on the maximum joint acceleration and minimum rotation angle. The robotic arm gripping frequency is set to 30 times / min to 180 times / min, with a frequency adjustment step of 15 times / min, and the optimal gripping cycle is corrected by matching the workstation spacing. After each gripping operation, the visual recognition feedback module confirms the label scanning data and triggers a feedback signal. The robotic arm's gripping rhythm is adjusted using PID control based on the feedback signal to synchronize the recognition and gripping rhythm.

[0104] In the warehouse entry simulation system, the label recognition device is set to have a recognition distance of 100mm-800mm, a recognition angle of ±45°, and a label surface reflectivity of 20%-80%.

[0105] In this embodiment, the identification device uses an industrial-grade CCD scanning head (model: KEYENCE SR-2000W). The identification distance is set via built-in parameters in the laser ranging module, ranging from 100mm to 800mm, with a granularity of 50mm. The identification angle is ±45° from the reference axis, adjusted by a built-in servo motor in the scanning head and corrected in real-time by a gyroscope, with the angle error controlled within ±1°. The label surface reflectivity is obtained by the ratio of incident light intensity to reflected light intensity, ranging from 20% to 80%, and is calibrated using label test pieces (materials include PET, coated paper, and self-adhesive). A pre-scan test is performed before each reflectivity setting to ensure the stability of the identification boundary range, with a setting error not exceeding ±3%.

[0106] Run the automated goods entry module and output the automated goods entry data.

[0107] In this embodiment, after all parameters are set, the automated warehousing simulation module is activated through the warehousing simulation control terminal. During system operation, the main control PLC module (model: Siemens S7-1500) schedules the execution order of each unit's logic, controlling the conveyor belt's conveying rhythm, the identification module's barcode scanning feedback, the robotic arm's grasping execution, and the stacking path planning according to the warehousing cycle. The entire warehousing process is monitored in real time in the SCADA system, with monitoring indicators including label recognition success rate, grasping accuracy, warehousing error distance, and item stacking stability. The simulation process lasts for 1200 seconds, recording core indicators such as the total number of items warehousing, the number of successfully identified items, and the number of grasping errors. An automated warehousing data table is exported, with fields including "Label ID," "Warehousing Timestamp," "Successful Recognition Flag," "Grasping Error (mm)," and "Final Positioning Offset (mm)." The exported format is standard XML for subsequent system calls.

[0108] Preferably, the safety shelf identification module includes the following functions:

[0109] Real-time batch entry is identified based on automated product entry data.

[0110] In this embodiment, during the real-time batch identification process based on the automated product warehousing data, the system first calls upon the timestamp, product serial number, packaging identifier, and conveyor belt receiving location information output by the automated product warehousing module. An warehousing data structure is then constructed, with fields including: product ID, batch number, warehousing time (accurate to the second), packaging barcode number, and warehousing workstation number. The product data is then grouped according to adjacent 30-second time intervals to generate an warehousing batch number in the format "IN-YYYYMMDD-HHMMSS-Workstation Number". This batch number is recorded by the scheduling management terminal and written to the real-time database table "realtime_batch_table".

[0111] Calculate the current inventory capacity based on real-time inbound batches;

[0112] In this embodiment, when calculating the current inventory capacity based on real-time inbound batches, the current status field `stock_status` of all batches of goods in the inventory database is queried, and the entry with the status "in stock" is selected to read its corresponding packaging volume (m³). 3 The data fields are summed, and the inventory capacity is calculated as ∑ (unit packaging volume). If the inventory area is divided into multiple physical warehouses, the warehouse number field also needs to be read for grouping and summarizing to form the current occupied volume statistics for each warehouse. The inventory capacity data is uploaded to the inventory analysis engine in JSON format.

[0113] Based on the current inventory capacity, a storage location priority analysis is performed to obtain storage location priority data;

[0114] In this embodiment, during the storage location priority analysis based on the current inventory capacity, the "available volume," "distance from the conveyor line (mm)," "usage frequency," and "historical loss rate" fields of each storage location are read, and weights are assigned to the priority calculation parameters accordingly. The recommended weighting coefficients are: distance weight 0.3, volume utilization weight 0.3, loss rate reverse weight 0.2, and usage frequency positive weight 0.2. A linear combination formula is used to calculate the comprehensive priority score, with a value range limited to 0-100. Storage locations with a priority score greater than 70 are identified as "high priority" and recorded in the priority allocation table slot_priority_table.

[0115] Dynamic inventory management is performed based on storage location priority data to obtain dynamic inventory management data.

[0116] In this embodiment, during the dynamic inventory management process based on storage location priority data, high-priority storage location numbers are retrieved and bound one by one to the products in the current real-time batch to be received. Each binding is done on a single batch basis. Before binding, it is verified whether the remaining volume of the storage location is greater than the volume of a single product. If so, the receiving storage location number is recorded and the corresponding storage location volume field is updated to "original value - product volume". The latest status of the storage location is also updated to "pre-allocated". A product-slot binding table, item_slot_bind_table, is generated, and the dynamic inventory allocation log is uploaded to the "inventory_dynamics_log" table.

[0117] Calculate shelf utilization based on dynamic inventory management data;

[0118] In this embodiment, during the calculation of shelf utilization based on dynamic inventory management data, the physical carrying capacity and the actual currently allocated object volume data of each shelf are obtained. The "current occupied volume" of all sub-storage locations under that shelf number is queried from the table slot_occupancy_table, and a ratio is calculated with the "maximum carrying capacity" in the shelf definition. Shelf utilization rate = current occupied volume / maximum carrying capacity × 100%. The calculation results are saved in "shelf_utilization_table" for subsequent structural analysis.

[0119] Based on the shelf utilization rate, identify the welding cracks in the shelf and obtain shelf crack data;

[0120] In this embodiment, during the process of identifying welding cracks in the shelving based on shelving utilization, multi-channel vibration acceleration sensors and ultrasonic micro-crack detectors installed at the four corners of the shelving and welding nodes are used. Through a continuous data acquisition cycle of 2 hours, the vibration spectrum of the shelving structure is compared to see if there is any abnormal frequency band drift. Crack identification is based on the occurrence of more than 3 vibration band shifts with a range greater than 0.05Hz, and the identification of echo notch reflection peak intensity greater than 5dB in the ultrasonic echo as the detection standard. The coordinates of the crack nodes are recorded, and a shelving crack data file shelf_crack_data.csv is generated and uploaded to the equipment health diagnosis system.

[0121] Assess the structural load-bearing capacity of the shelving based on shelving crack data;

[0122] In this embodiment, when assessing the load-bearing capacity of the shelving structure based on shelving crack data, the correction factor for the impact of welding defects on load-bearing capacity in the national standard GB50017-2017 "Code for Design of Steel Structures" is used. The crack ratio is calculated as the weld crack length / total weld length, and a safety correction factor is introduced. If the crack length of a weld node exceeds 10% of the original weld length, the strength reduction factor of the corresponding component is set to 0.8; if it exceeds 20%, the reduction factor is 0.6. The load limit value is reassessed in conjunction with the shelving geometric model, the load-bearing capacity index is recalculated, and a structural assessment report, shelf_strength_report.json, is output.

[0123] Identify safe shelving areas based on the load-bearing capacity of the shelving structure.

[0124] In this embodiment, during the process of identifying safe shelving areas based on the shelving structure's load-bearing capacity, the load-bearing capacity value is compared with the current load. If the load-bearing safety factor = assessed load-bearing limit / current load, shelving areas with a value greater than 1.5 are marked as "safe areas," those with a value between 1.0 and 1.5 are "monitored areas," and those with a value less than 1.0 are "dangerous areas." The identification results are stored in the shelf_safety_zone_table, and the system scheduling module automatically avoids allocating goods to non-safe areas in the next inbound task, thereby realizing an inventory distribution management strategy based on structural safety.

[0125] Preferably, the safety shelving identification module identifies shelving welding cracks based on shelving utilization, including:

[0126] Identify high-utilization shelving areas based on shelving utilization rates;

[0127] In this embodiment, the shelving management system retrieves inbound and outbound data records. Each data record contains a unique identifier ("one item, one code"), shelving number, inbound time, outbound time, and item weight. The system summarizes the cumulative storage time and total weight of all items on designated shelves and compares this data with the shelf's maximum load-bearing capacity and standard working cycle. The usage of each shelf is statistically analyzed daily for a 7-day period. Using the rule engine configured in the warehouse management platform, areas with an average shelf utilization rate exceeding 80% are designated as high-utilization areas. This threshold is derived from the medium-to-high risk workload boundary defined in the structural fatigue standard for long-term full-load monitoring of shelves. After summarizing historical data, the system marks the high-utilization shelf numbers on the warehouse map and exports them as a "High-Utilization Shelf List" as the target area for noise detection.

[0128] Acoustic sensor data was collected based on high-utilization shelving areas;

[0129] In this embodiment, acoustic sensors are deployed at welded joints within the high-utilization shelving area. Each shelving unit has four sensor points, using industrial-grade piezoelectric ceramic acoustic sensors. Installation locations include the base beam, the intersection of vertical columns, the top support beam, and the weld seam of the middle support. Each sensor is adhered to the metal surface using specialized industrial adhesive, with the bonding area controlled within 2 square centimeters. A pressure clamp is used to maintain stability for 12 hours to ensure coupling performance. The signal cable is a metal-shielded cable, no longer than 5 meters, connected to the centralized acquisition controller. The acoustic acquisition controller is a TDSP-308, set to a continuous sampling time of 2 minutes and a sampling frequency of 1 million times per second. Data records are saved in raw waveform binary format on the local hard drive. After acquisition, the data is automatically archived to a designated file directory in the data center, and the data files are identified by shelving number and sensor number.

[0130] The acoustic emission sensor spectrum is obtained by performing a fast Fourier transform on the acoustic emission sensor data.

[0131] In this embodiment, the acoustic waveform data collected by each set of sensors is imported into a dedicated data processing module in the edge server. The signal is preprocessed, including high-frequency interference filtering and signal normalization. The processing module calls an FFT transform plugin, sets the data segmentation window to 1024 points, and outputs the processing result as an acoustic spectrum in the form of a frequency distribution, covering a frequency range from 10kHz to 500kHz. This processed data is saved in CSV format, with each row representing the sound intensity value at a frequency point. A spectrum file is generated for each channel and stored in a file named "shelf_number_channel_number_spectrum.csv", serving as the data source for subsequent abrupt change segment identification and high-frequency energy analysis.

[0132] Identifying wideband abrupt changes based on acoustic emission sensor spectrum; Identifying high-frequency bands based on acoustic emission sensor spectrum;

[0133] In this embodiment, the spectrum file is segmented, with the frequency divided into several intervals of 100kHz. Within each interval, the sound intensity changes are read, identifying segments where the sound intensity value fluctuates drastically between adjacent frequency points. If the fluctuation within a frequency interval exceeds twice the historical background noise fluctuation standard, it is marked as a sudden change segment. The background noise standard is calculated from the acoustic emission spectrum collected under no-load conditions. The system outputs a sudden change record file according to the start and end frequencies and location of the sudden change segment, and marks the sudden change segment as a candidate segment for structural anomaly signal, for cross-validation in the next step of energy density analysis. The frequency range in the spectrum is limited to 200kHz to 500kHz as the high-frequency band. All sound intensity data within this range is read and compared with the average sound intensity of the complete spectrum. If the overall sound intensity level of the high-frequency band is more than 15% higher than the set threshold of the average sound intensity of the entire band, this frequency band is marked as a high-frequency cluster segment. This standard threshold is derived from the statistical characteristics of a large number of normal shelf weld samples. The processing results generate a high-frequency aggregation record file, which contains information such as the start and end frequencies of the aggregation segment, average sound intensity, maximum sound intensity, and sound intensity standard deviation, and is used for high-frequency energy density assessment.

[0134] Calculate high-frequency energy density based on the high-frequency band;

[0135] In this embodiment, the spectral data marked as high-frequency clusters are analyzed by extracting sound intensity values ​​at each frequency point, converting them to linear energy units, and then performing aggregated statistics. The sum of the sound intensities at all frequency points is divided by the frequency range length to obtain the high-frequency energy density. Each acoustic emission channel is processed independently, and the results are output to a unified energy density file with the shelf number, sensor number, high-frequency range, and corresponding energy density value. Subsequently, the data from multiple sensor channels on the same shelf are integrated, and a distance-weighted average is used to improve the sensitivity to identify local weld anomalies. The final output is the high-frequency energy density index corresponding to each shelf number.

[0136] Based on the wideband abrupt change range and high-frequency energy density, the welding cracks of the rack were determined, and the rack crack data were obtained.

[0137] In this embodiment, the system retrieves broadband abrupt change bands and high-frequency energy density indicators, matching the two types of data across frequency ranges. If the abrupt change band corresponding to a certain sensor is entirely within the high-frequency convergence band, and the energy density corresponding to that band exceeds a preset safety threshold, then a crack signal is determined to exist in the weld area. This safety threshold is derived from the acoustic energy boundary value recorded in the welded structure load-bearing experiment, and its value is 90% of the minimum abnormal value in the reference sample. Simultaneously, by combining the arrival time difference of acoustic signals between multiple sensors, spatial inversion is performed through the geometric position of the sensor layout to calculate the three-dimensional coordinates of the sound source point, with the error range controlled within 15 mm. All identification results are written into the shelf crack data file, including shelf number, crack location, crack signal frequency range, energy density, identification time, and sound source location coordinates.

[0138] Preferably, the product lifecycle tracking module includes the following functions:

[0139] Based on the safety shelf area, aisle structure features are extracted to obtain aisle structure data;

[0140] In this embodiment, image modeling of the warehouse aisle structure is performed around the shelving clusters already marked as "safe shelving areas." A 3D laser scanner (FARO Focus S150) is used to collect 3D point cloud data at a distance of 50 meters and a resolution of 2 million points per square meter. The scanner is deployed at both ends of the shelving aisles and in the central transverse aisle area, with at least six measurement points. After scanning, the point cloud data is converted into a 3D structural model using a dedicated laser mapping system. The model includes annotations for wall locations, shelving boundaries, aisle edges, and ground undulations. Subsequently, a BIM modeling system is used to extract aisle structural parameters, including aisle width (mm), minimum turning angle (degrees), total aisle length (m), and ground height difference (mm). Finally, all parameters are summarized and output as an aisle structural data document in JSON format, with each aisle segment labeled with its corresponding shelving number.

[0141] Acquire outbound gate data; match outbound task nodes based on outbound gate data and channel structure data to obtain target outbound node data for goods;

[0142] In this embodiment, the data of the outbound passage is read in real time through an intelligent gate control system. The system is installed at each gate of the warehouse outbound passage, reading the gate number, current open status, bandwidth, passable time period, and geographical distance to the shelving area. The status table for each outbound passage is updated in real time using an RFID reader. The data format is CSV, containing fields including passage number, bandwidth (mm), door height (mm), availability flag (1 or 0), and shortest distance to each shelving unit (in meters). Matching rules are executed based on the following parameters: the shelving number associated with each product is compared with the aisle structure data, filtering out the outbound passages closest to the product, with a door width greater than 600mm, a door height greater than 1500mm, and currently in an "available" state; the channel number is used to determine whether the door is directly connected to the aisle where the shelving unit is located. Outbound doors that meet the above conditions are marked as target outbound nodes, and the output is "Product Target Outbound Node Data," which includes a unique product ID, target outbound door number, corresponding channel number, distance data, and outbound time period.

[0143] Based on the target shipment node data of the goods, a path feasibility analysis was conducted to obtain preliminary reachable path data;

[0144] In this embodiment, a geographic information modeling system (such as Revit or Autodesk Navisworks) is used to load channel structure data and shipping node data. Based on the spatial path between the node's starting point (current shelf location of the goods) and ending point (target shipping door), a depth-first graph search algorithm is used to construct all paths in the channel structure map. The path construction process is limited by the following parameters: the path turning angle must not be less than 45 degrees, the turning radius must not be less than 500 mm, and the effective passage width of the channel must not be less than 400 mm. For each path, the number of path segments, the length of each segment, the number of turning points, and the cumulative distance are calculated segment by segment. The path information is output as "preliminary reachable path data," including the number of each path segment, the coordinates of the starting and ending points, the segment length, the channel number, the corresponding channel width, the turning point information, and the ground elevation.

[0145] Calculate ground flatness based on preliminary reachable path data;

[0146] In this embodiment, based on the preliminary reachable path data, ground point cloud data of the corresponding segments of the path are read, and surface fitting is performed using a 3D ground fitting algorithm. Each path segment is divided into several grids, with a grid size of 100mm x 100mm. The maximum and minimum height difference of each grid is extracted, and the height fluctuation value is recorded. If the height difference exceeds 5mm, it is recorded as a "micro-step"; if the width of the undulating area is less than 400mm and the height difference is greater than 20mm, it is marked as a "step obstacle". The overall standard deviation of each path segment is calculated, and a ground smoothness factor is generated. The output is recorded as a "ground smoothness table", which includes the segment number, maximum height difference, average height change rate, obstacle segment number, and ground condition score.

[0147] Based on the flatness of the ground, the preliminary accessible path data is filtered to obtain the product shipping path data;

[0148] In this embodiment, all path segments with any of the following problems are removed from the preliminary reachable path data: First, there are stages with height changes greater than 20mm; second, the average ground undulation value (i.e., the average height standard deviation) is greater than 10mm; third, the width of the middle section of the path is less than 400mm; fourth, the maximum slope of any section of the path is greater than 12 degrees. The filtered path set is sorted in ascending order according to the total path length, and the shortest path is selected as the product shipping path. The output structure is "Product Shipping Path Data", which includes the starting point, ending point, path segment sequence, length of each segment, cumulative total length, ground features of each segment, and channel number.

[0149] Based on the product shipping path data, a product shipping simulation was conducted, in which the transportation channel width was set to 400mm-600mm, the maximum allowable slope ≤12° and the maximum step height difference ≤20mm, and the product shipping data was obtained.

[0150] In this embodiment, a simulation engine (such as FlexSim or AnyLogic) is invoked, and the shipping path data is set as the shipping trajectory in the imported 3D model of the warehouse aisle. The physical dimensions of the transport equipment are defined in the simulation system (length 800mm, width 350mm, height 450mm), wheel diameter 100mm, maximum climbable slope 12°, and maximum obstacle-crossing capability of 20mm steps. During the simulation, ground elevation and aisle boundary data are read point-by-point along the path to simulate the transport equipment's passage status in each section of the aisle, including equipment tilt angle, turning radius, whether it contacts obstacles, and whether it jams or stops. Each simulation lasts for no less than 10 runs, recording the success of each shipment. Finally, the product shipment data is output, including whether the path was successfully traversed, simulation time, maximum slope, maximum number of collisions, and movement speed records.

[0151] The serial number is verified based on the product shipment data to obtain serial number verification data, which is then transmitted to the product management system to perform product lifecycle tracking tasks.

[0152] In this embodiment, after the simulated and confirmed shipping route is completed, the "one item, one code" data corresponding to the unique identifier of the product is retrieved, and the full lifecycle tracking code of the product is obtained from the shipping task generation list. The RFID reader at the end of the route identifies the product tag information, including the product serial number, shipping time, route number, shelf outbound number, and shipping port number. The serial number read on-site is compared one by one with the corresponding shipping serial number recorded in the database. If they match completely, the verification is marked as successful. All verification results are written to the "Serial Number Verification Log" in real time, in CSV format, with fields including product ID, task number, actual shipping serial number, database matching serial number, verification time, and verification status code (1 for success, 0 for failure). The log data is pushed in real time via the MQTT protocol to the product management system host interface address "192.168.1.99:2083" for lifecycle tracking recording and related calls.

[0153] Preferably, the serial number verification based on product shipping data in the product lifecycle tracking module includes:

[0154] Read the product serial number based on the product shipping data;

[0155] In this embodiment, an industrial barcode scanning and identification device, specifically a KEYENCE SR-2000W series fixed barcode reader, is installed at the end of the loading and unloading channel. The device is mounted 200mm above the conveyor belt, with a 90° vertical overhead shooting angle and a reading area of ​​100mm x 100mm. Each item in the shipping data includes a path endpoint timestamp. Based on this timestamp, the barcode scanner's timed exposure function is triggered, with an exposure time set to 1.2ms and an aperture of F4. The content read includes the QR code and the plaintext characters beneath it, and the recognition format is Code128 or DataMatrix type barcode. After scanning, the original barcode string is used as the initial serial number value, and the timestamp, camera number, image frame number, and item ID are recorded to form a preliminary serial number record data table.

[0156] Collect serial number images based on product serial numbers;

[0157] In this embodiment, after the barcode scanning is completed, the image acquisition system is triggered to acquire a high-resolution image. The device used is a Basler acA1920-40gc industrial camera with a resolution of 1920×1200, BMP image format, and a frame rate of 20fps. The camera is mounted next to the barcode scanner, with a white LED backlight module as the light source, an illumination of 3000 lux, and a fixed exposure time of 2.0ms. During shooting, the image capture center area is aligned with the barcode area, and the ROI parameter is set to a range of 640×480 pixels to ensure complete acquisition of the barcode and surrounding characters. The acquired image is named using the "product ID + timestamp" rule and stored in the " / barcode_images / " directory on a shared server within the local area network. This image serves as the raw input for subsequent scratch detection, code overlap analysis, and other processing steps.

[0158] Barcode scratch detection is performed based on the serial number image to obtain barcode scratch data;

[0159] In this embodiment, after loading the above image, an edge detection algorithm is used to identify scratched areas. The Sobel operator is used to perform horizontal and vertical gradient detection on the image to identify the continuity of the black and white boundaries of the barcode. When the length of the missing boundary segment is greater than 3 pixels and the width of the missing segment is not less than 5 pixels, it is determined to be a scratched area. The starting coordinates, width, height, and grayscale mean of each scratched area are recorded as scratched area information. Based on the image resolution, each 100 pixels is equivalent to a real size of 5mm. Therefore, the barcode scratch judgment threshold is set as a valid scratch point if the continuous black and white break area is greater than 30 pixels. The output includes the image number, the total number of scratched areas, the position coordinates, width, height, and grayscale standard deviation of each area, forming a unified "barcode scratch data table".

[0160] Of particular importance is the barcode scratch detection feature in the product lifecycle tracking module, which includes:

[0161] The sequence number image is converted to grayscale to obtain a grayscale sequence number image.

[0162] In this embodiment, image grayscale processing is performed by linearly weighting the pixel values ​​of the red, green, and blue channels. The weighting factors are R: 0.299, G: 0.587, and B: 0.114. The acquired original image of the sequence number is read pixel by pixel, and the grayscale value of each pixel is calculated according to the above proportions, replacing the original RGB color information to generate a new single-channel image matrix. This operation is implemented using the image matrix channel processing function in OpenCV. All images use 8-bit grayscale precision, and the image resolution is uniformly adjusted to 640×480 pixels to ensure consistency in subsequent processing. The generated grayscale image is saved as a two-dimensional array structure, with each position representing a grayscale value ranging from 0 to 255, for subsequent boundary analysis.

[0163] Based on the grayscale sequence number image, the continuity of black and white boundaries is analyzed to obtain boundary fracture detection data;

[0164] In this embodiment, Sobel edge detection is performed on a grayscale image to obtain gradient information in the horizontal and vertical directions. The Sobel operator's convolution kernel size is set to 3×3, processing the horizontal gradient (Gx) and vertical gradient (Gy) respectively, and calculating the gradient magnitude G = √(Gx² + Gy²). The gradient map is binarized, and an edge response threshold of 80 is set (an empirical value determined by obtaining the average minimum boundary break response magnitude from the sampled training data). Pixels greater than the threshold are considered boundary points. The continuity of boundary lines is detected by row-wise scan. Boundary lines with a break length exceeding 10 pixels or a break frequency exceeding 3 times per 100 pixels are marked as "break boundaries," and their start and end positions, pixel coordinates, and gradient magnitude are recorded as boundary break detection data.

[0165] Based on the boundary fracture detection data, the coordinates of continuous fracture areas are extracted and marked as suspected scratch areas;

[0166] In this embodiment, regional connectivity analysis is performed based on the clustering characteristics of fracture points in the boundary fracture detection data. An 8-neighborhood connected region detection algorithm is used to mark the connected components to which all fractured pixels belong. A connection threshold is set at a minimum number of connected pixels ≥ 15 and a minimum width span ≥ 4 pixels; connected components meeting these criteria are extracted as "continuous fracture regions." The boundary coordinates (top left corner, bottom right corner), total number of pixels, and center point coordinates of each connected region are extracted and marked as suspected scratched area data. This data is stored as a structure array, including the fields: "bounding_box" (rectangular boundary), "area" (number of pixels), "center_coord" (centroid), and "fracture intensity value" (mean gradient of the fractured area).

[0167] Based on the suspected scratched areas, filter areas with an area greater than 30 pixels and a boundary span of more than 5 pixels, and mark the valid scratched areas;

[0168] In this embodiment, conditional filtering is performed on all suspected scratched areas extracted in the previous step. The filtering parameters are set as follows: the area threshold is set to 30 pixels, and the boundary span threshold is set to 5 pixels, that is, the width or height of the area boundary is ≥ 5 pixels. For each area, it is determined whether its "area" field is ≥ 30, and whether its "width" or "height" field in "bounding_box" is ≥ 5. If both conditions are met, the area is marked as a valid scratched area. The area is highlighted in the image, and the field "is_valid = True" is added to the data structure. All valid scratched area data is output in JSON format, and each record contains five fields: image number, sequence number, area coordinates, area, and span value.

[0169] Barcode texture breakage detection is performed based on the effective scratched area to obtain barcode scratch data.

[0170] In this embodiment, local texture directionality analysis is performed within each effective scratched area. First, the gray-level co-occurrence matrix (GLCM) of the image is extracted within the area, with a calculation window set to 21×21 pixels and offset directions set to 0°, 45°, 90°, and 135°. The contrast and second-order angular moment (correlation) in the four directions are calculated from the GLCM. If the coefficient of variation (CV) of the contrast in different directions within the area is >0.3, and the difference in directional correlation is >0.25, then the texture direction is considered misaligned or broken. Areas meeting the above criteria are defined as the final scratched areas. The final barcode scratch data includes: image number, serial number, scratched area coordinates, area texture contrast (CV), and correlation difference (Δcorrelation). The output is a standard structured data table in .csv or .json format for subsequent inkjet overlap analysis.

[0171] Based on barcode scratch data, inkjet printing overlap analysis is performed to obtain inkjet printing overlap data;

[0172] In this embodiment, based on the scratch detection results, pixel-level binarization is performed on the original image. A threshold value of 128 grayscale is used, with all pixels higher than 128 considered background and those lower than or equal to 128 considered foreground. Subsequently, connected component analysis is used to extract the contours of the characters. If the horizontal distance between two groups of characters is less than 10 pixels, and the contour boundaries are nested or share more than 20% of pixels, they are marked as overlapping areas. This analysis uses the `findContours()` function in OpenCV to perform contour finding, using area ratio and boundary overlap rate as standards. Areas with low contrast and edge intersection areas greater than 20 square pixels are considered overlapping areas. The output data includes the overlapping area number, top-left coordinates, number of overlapping characters, overlapping contour area, and minimum boundary distance between characters, forming a structured "inkjet overlapping data".

[0173] Based on the inkjet overlap data, character positions are aligned to obtain character alignment data;

[0174] In this embodiment, character segmentation and alignment are performed based on the recognition results of overlapping regions. Using the vertical projection histogram method, the pixel distribution of characters in the vertical direction is statistically analyzed to locate the center point of each character. Character separation is performed on the overlapping regions using the minimum bounding rectangle envelope method, forcibly dividing the overlapping blocks into two groups of characters. Each group must have a width greater than 20 pixels and a height greater than 40 pixels; all recognition regions smaller than this threshold are rejected. Subsequently, all characters are reordered according to their horizontal position, with the difference in the left boundary position of each character set not to exceed 15% of the average character width. After character sorting, a character number table is created, recording the character index number, original coordinates, new coordinates, and displacement deviation value. Finally, a "character alignment data table" is formed, ensuring that characters are arranged from left to right, and the image pixel coordinates corresponding to each character position are clearly defined.

[0175] The serial number is reconstructed based on the character alignment data to obtain the reconstructed serial number data.

[0176] In this embodiment, based on the aligned character regions, template matching is used to perform character recognition on each character patch. The template library stores a standard character image set, including A, Z, and all characters from 0 to 9. Each character template is 28×28 pixels in size, and the matching method uses the Normalized Cross-Correlation Coefficient (NCC) algorithm. Characters with a maximum correlation coefficient greater than 0.75 in the matching results are confirmed, while characters with a correlation coefficient less than this threshold are determined to be unrecognizable and marked with "#". All successfully recognized characters are concatenated into a string according to their sorted positions to construct the reconstructed serial number. The reconstructed serial number is output in JSON format, with fields including product ID, reconstructed serial number string, coordinate block of each character, and recognition confidence, forming a "Reconstructed Serial Number Data Table".

[0177] The serial number is verified based on the reconstructed serial number data to obtain the serial number verification data, which is then transmitted to the product management system to perform the product lifecycle tracking task.

[0178] In this embodiment, the reconstructed serial number data is compared character by character with the original serial number registered in the product database. The comparison logic follows a one-to-one correspondence based on the index; a perfect character match scores 1, while the presence of "#" or inconsistent characters scores 0. If the total score ratio is higher than 90%, it is considered "verifiable". The comparison result is marked as a verification status bit (1 for success, 0 for failure), and the comparison time, reconstructed serial number, original serial number, index of the difference character, and difference content are recorded. Finally, a "Serial Number Verification Log" is generated in CSV format, with fields including product ID, original serial number, reconstructed serial number, verification result, difference list, and timestamp. The log is sent via LAN UDP protocol to the main node of the product management system with IP address "10.10.0.20" and port 18550. After receiving the log, the system writes it to the database and performs the binding of the tracking chain node.

[0179] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0180] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A one-item-one-code full lifecycle tracking system, characterized in that, Includes the following modules: The image binding module is used to acquire product production data; generate a unique product serial number based on the product production data; and acquire an image based on the unique product serial number to obtain the product serial number image. The automated product warehousing module is used to detect label wrinkles based on product serial number images, obtaining label wrinkle data, including: Identify the outline of serial number characters from the product serial number image; Calculate the character line deformation rate based on the character outline of the serial number; Label warping data is obtained by detecting label warping based on the character line deformation rate. Laser 3D contour acquisition is performed based on label warping data to obtain the label warping 3D contour. Shear stress was calculated based on the three-dimensional contour of the label warped edge to obtain shear stress data; Identify shear stress concentration regions based on shear stress data; The wrinkle orientation is predicted based on the shear stress concentration area, and wrinkle orientation data is obtained. Label wrinkle detection is performed based on wrinkle direction data to obtain label wrinkle data; Label breakage assessment is performed based on label wrinkle data to obtain label breakage data. Then, label production processes are optimized based on this breakage data to obtain label optimization data, including: Detecting the fiber orientation of materials based on label wrinkle data; Calculate the directional offset based on the fiber arrangement direction of the material; The imbalance of material fiber arrangement is determined based on the directional offset, and the imbalance data of material fiber arrangement is obtained. Predicting crack initiation points based on material fiber arrangement imbalance data; Tag fracture assessment was performed based on crack initiation points to obtain tag fracture data; The die-cutting contour position was traced back based on the label breakage data; The die-cutting path error rate is calculated based on the die-cutting contour position; the chamfering structure parameters are optimized based on the die-cutting path error rate to obtain the chamfering structure optimization parameters; Calculate the die-cutting trajectory abruptness rate based on the die-cutting contour position; determine the die-cutting contour smoothness based on the die-cutting trajectory abruptness rate. By integrating the chamfer structure optimization parameters and the smoothness of the die outline, we obtain optimized production process data for labels. Label production simulation was performed based on the label optimization production process data to obtain label optimization data; Based on the tag optimization data, simulate the automated warehousing of goods to obtain automated warehousing data; The safety shelf identification module is used to perform dynamic inventory management based on automated goods receiving data, obtain dynamic inventory management data, calculate shelf utilization based on dynamic inventory management data, identify shelf welding cracks based on shelf utilization, obtain shelf crack data, and determine the safety shelf area based on shelf crack data. The product lifecycle tracking module analyzes the product shipping path based on the safety shelf area to obtain product shipping path data; it simulates product delivery based on the product shipping path data to obtain product delivery data; and it verifies the serial number based on the product delivery data to obtain serial number verification data, which is then transmitted to the product management system to execute the product lifecycle tracking task.

2. The one-item-one-code full lifecycle tracking system according to claim 1, characterized in that, The image binding module includes the following functions: Obtain product production data and extract production batch features to obtain production batch data; Analyze product structure parameters based on production batch data; Based on the product structure parameters and the production timestamp, product production traceability data is obtained; Based on the product production traceability data, the product identifier is encrypted and hashed to obtain the identifier code; The sequence format is standardized based on the identifier encoding to generate a unique serial number for the product. Image acquisition is performed based on the product's unique serial number to obtain the product serial number image.

3. The one-item-one-code full lifecycle tracking system according to claim 2, characterized in that, The image binding module includes image acquisition based on the product's unique serial number, which includes: Generate image acquisition instructions based on the product's unique serial number; The image acquisition device is scheduled according to the image acquisition command, and the device is triggered to capture the image signal; The original image dataset is obtained by acquiring multi-angle product images based on the device's shooting signals. The original image dataset is subjected to resolution correction to obtain standardized image data. By binding a unique serial number to the product based on standardized image data, a product serial number image is obtained.

4. The one-item-one-code full lifecycle tracking system according to claim 1, characterized in that, The automated product receiving module simulates automated product receiving based on tag-optimized data, including: Import the optimized label data into the database simulation system; In the inbound simulation system, the coefficient of friction of the product surface is set to 0.1-0.6, the center of gravity offset distance is 30-100mm, the surface reflectivity is 10%-90%, and the packaging compression modulus is 50kPa-500kPa. In the warehousing simulation system, the warehousing conveyor belt running speed is set to 0.1m / s-1.5m / s, and the sorting robot arm gripping frequency is 30 times / min-180 times / min; In the warehouse entry simulation system, the label recognition device is set to have a recognition distance of 100mm-800mm, a recognition angle of ±45°, and a label surface reflectivity of 20%-80%. Run the automated goods entry module and output the automated goods entry data.

5. The one-item-one-code full lifecycle tracking system according to claim 1, characterized in that, The safety shelf identification module includes the following functions: Real-time batch entry is identified based on automated product entry data. Calculate the current inventory capacity based on real-time inbound batches; Based on the current inventory capacity, a storage location priority analysis is performed to obtain storage location priority data; Dynamic inventory management is performed based on storage location priority data to obtain dynamic inventory management data. Calculate shelf utilization based on dynamic inventory management data; Based on the shelf utilization rate, identify the welding cracks in the shelf and obtain shelf crack data; Assess the structural load-bearing capacity of the shelving based on shelving crack data; Identify safe shelving areas based on the load-bearing capacity of the shelving structure.

6. The one-item-one-code full lifecycle tracking system according to claim 5, characterized in that, The safety shelving identification module identifies welding cracks in the shelving based on shelving utilization, including: Identify high-utilization shelving areas based on shelving utilization rates; Acoustic sensor data was collected based on high-utilization shelving areas; The acoustic emission sensor spectrum is obtained by performing a fast Fourier transform on the acoustic emission sensor data. Identifying wideband abrupt changes based on acoustic emission sensor spectrum; Identifying high-frequency bands based on acoustic emission sensor spectrum; Calculate high-frequency energy density based on the high-frequency band; Based on the wideband abrupt change range and high-frequency energy density, the welding cracks of the rack were determined, and the rack crack data were obtained.

7. The one-item-one-code full lifecycle tracking system according to claim 1, characterized in that, The product lifecycle tracking module includes the following functions: Based on the safety shelf area, aisle structure features are extracted to obtain aisle structure data; Acquire outbound gate data; match outbound task nodes based on outbound gate data and channel structure data to obtain target outbound node data for goods; Based on the target shipment node data of the goods, a path feasibility analysis was conducted to obtain preliminary reachable path data; Calculate ground flatness based on preliminary reachable path data; Based on the flatness of the ground, the preliminary accessible path data is filtered to obtain the product shipping path data; Based on the product shipping path data, a product shipping simulation was conducted, in which the transportation channel width was set to 400mm-600mm, the maximum allowable slope ≤12° and the maximum step height difference ≤20mm, and the product shipping data was obtained. The serial number is verified based on the product shipment data to obtain serial number verification data, which is then transmitted to the product management system to perform product lifecycle tracking tasks.

8. The one-item-one-code full lifecycle tracking system according to claim 7, characterized in that, The product lifecycle tracking module includes serial number verification based on product shipping data, which includes: Read the product serial number based on the product shipping data; Collect serial number images based on product serial numbers; Barcode scratch detection is performed based on the serial number image to obtain barcode scratch data; Based on barcode scratch data, inkjet printing overlap analysis is performed to obtain inkjet printing overlap data; The character positions are aligned based on the inkjet overlap data to obtain the character alignment data; The serial number is reconstructed based on the character alignment data to obtain the reconstructed serial number data. The serial number is verified based on the reconstructed serial number data to obtain the serial number verification data, which is then transmitted to the product management system to perform the product lifecycle tracking task.

Images