A method, system and device for detecting the quality of a fire barrier based on rotational motion

Abstract

A kind of combustion layer quality detection method, system and device based on rotary motion, by combining weighing sensor, 2D industrial camera and 3D line scanning sensor organically, the three-dimensional scanning of combustion layer can be completed by the present application, and weight detection of propellant grain, appearance collection and character recognition can be realized, in the process that rotating platform drives propellant grain to rotate, 3D line scanning sensor always remains stationary, by carrying out line-by-line scanning to propellant grain surface, three-dimensional data information of propellant grain surface is acquired;Raw multi-source data from data acquisition is received, and specific pretreatment for propellant grain combustion layer surface detection is executed.3D point cloud geometry information, 2D image texture features, OCR character recognition results and weighing data are comprehensively utilized, to construct comprehensive evaluation model of combustion layer quality, to realize defect detection, classification and severity quantification.The present application comprehensively utilizes mechanical automation and advanced visual detection algorithm, to realize efficient and accurate detection of combustion layer quality.

Description

Technical Field

[0001] This invention relates to the field of industrial vision inspection and control, and in particular to a quality inspection and control system for fire-limited layers based on rotational motion. Background Technology

[0002] like Figure 1 As shown: Agent 3' is installed inside support cylinder 1'; flame-retardant layers 2' are provided on both ends of agent 3'; axial through holes 4' are provided on agent 3', and the agent on the side of the explosive column and the side wall of the support cylinder are presented in a fence-like manner.

[0003] In industrial production, accurate testing of the quality of the flame-restriction layer of solid propellant grains is crucial for ensuring product safety and stability. Flame-restriction layer testing primarily focuses on the uniformity of its thickness and weight, as the thickness and uniformity of the flame-restriction layer significantly impact the combustion performance of the propellant grain; therefore, flame-restriction layer quality testing is of paramount importance.

[0004] Traditional visual inspection methods for detecting flammable tarpaulins typically involve reciprocating motion in the XY directions and vertical movement in the Z-axis to achieve full coverage. This approach has significant drawbacks: firstly, the complex multi-axis motion complicates the equipment structure, increasing cost and maintenance difficulty; secondly, the time-consuming multi-axis motion process leads to low inspection efficiency, failing to meet the inspection requirements of large-scale, high-efficiency production scenarios. Furthermore, potential mechanical errors during multi-axis motion can affect the accuracy of the results, failing to accurately capture subtle quality issues within the flammable tarpaulin. Current technologies lack a system capable of achieving full coverage inspection through a single rotational motion and possessing adaptive parameter adjustment capabilities. Therefore, an innovative inspection solution is urgently needed to overcome the shortcomings of traditional methods. Summary of the Invention

[0005] The purpose of this invention is to provide a method, system, and device for detecting the quality of fire-restricted layers based on rotational motion. This invention comprehensively utilizes mechanical automation and advanced visual inspection algorithms to achieve efficient and accurate detection of the quality of fire-restricted layers.

[0006] 1. Mechanical structure

[0007] The mechanical structure of this invention mainly consists of a robotic arm, a rotating platform, and a 3D line scan sensor. To meet the requirements of appearance image, character recognition, and weight detection for flammability restriction layer inspection, the system also integrates a 2D industrial camera, an OCR character recognition auxiliary light source assembly, and a weighing sensor module. The robotic arm precisely picks up the test cartridge from a specially designed loading rack and places it stably on the rotating platform. The robotic arm possesses high-precision positioning capabilities, ensuring that the test cartridge is accurately placed in the center of the rotating platform each time, laying the foundation for the accuracy of subsequent inspections.

[0008] After the propellant charge is grasped, the robotic arm first places it on a weighing sensor at the front of the system to quickly measure its total weight. The weighing module uses a high-precision weighing unit to determine whether the weight of the fuel-limiting layer is within the specified range, and transmits the measurement results to the software system in real time for recording and analysis. After weighing, the robotic arm then transfers the propellant charge to a position under a 2D industrial camera. The 2D camera and its accompanying light source form a vision acquisition component that can acquire images of the top or side of the propellant charge for inkjet character recognition, visual inspection, and 3D point cloud positioning assistance.

[0009] The rotating platform is one of the key components of the entire mechanical structure, designed to support the drug cartridge and achieve uniform rotation. During the testing process, once the drug cartridge is placed on the rotating platform, the platform begins to rotate at a preset speed. The rotation speed of the rotating platform can be flexibly adjusted according to actual testing needs to ensure that the 3D line scan sensor can acquire high-quality testing data. By using a single rotating axis to replace the traditional multi-axis reciprocating motion, the mechanical structure is greatly simplified, sources of motion error are reduced, and the consistency of posture after weighing and 2D image acquisition is ensured.

[0010] The 3D line scan sensor plays a crucial role in data acquisition within the entire detection system. It is fixedly mounted in a specific location, with the detection width starting from the origin on the top surface of the drug cartridge and gradually expanding radially until it covers the entire radius. As the rotating platform drives the drug cartridge to rotate, the 3D line scan sensor remains stationary, acquiring three-dimensional data information of the drug cartridge surface by scanning line by line. This design, with a fixed sensor and rotating workpiece, avoids vibration and positioning errors caused by the sensor's own movement, improving the stability and consistency of data acquisition.

[0011] By organically combining a weighing sensor, a 2D industrial camera, and a 3D line scan sensor, the mechanical structure of this invention can not only complete the three-dimensional scanning of the flame retardant layer, but also realize the detection of propellant weight, appearance acquisition, and character recognition, providing more comprehensive and richer detection data for subsequent software algorithms. This design approach, while maintaining structural simplicity, significantly improves the system's detection capabilities and information integrity, making the detection results more accurate and reliable.

[0012] 2. Software Algorithm Section

[0013] The software system adopts a layered architecture, but its module functions and data processing flow are specifically designed and work collaboratively around the specific task of detecting the three-dimensional morphology and geometric characteristics of the propellant charge's flame-limiting layer. The core functions and interactions of each layer are as follows:

[0014] Data Acquisition Layer: Utilizing multiple industrial communication interfaces such as EtherCAT, RS485, and USB 3.0, this layer enables real-time communication and control with hardware devices including robotic arms, rotary platforms, 3D line scan sensors, 2D industrial cameras, and weighing sensors. Its core task is not only data acquisition but also ensuring the provision of high-quality, synchronized, and consistent multi-source raw data for subsequent fuel-limited kerb quality analysis. This includes 3D point clouds, 2D images, character information, and propellant weight information, achieving precise coordination throughout the entire detection process. Specifically, this includes:

[0015] 1) Rotary platform control: Rotation speed control commands (adjustable from 5–30 r / min) are sent to the rotary platform via EtherCAT bus, and encoder feedback signals are received in real time to achieve precise and uniform rotation. The rotation speed can be dynamically adjusted according to the specifications of the drug cartridge and the detection accuracy requirements to ensure uniform point cloud density during scanning.

[0016] 2) 3D Point Cloud Acquisition: The system sends a synchronous trigger signal to the 3D line scan sensor, acquiring line scan data at a frequency of 500–2000 Hz. As the propellant grain rotates, the 3D line scan sensor scans the surface of the flame-retardant layer line by line, generating radial line scan point cloud data with the center of the top surface of the propellant grain as the origin. All point cloud data includes timestamps and angle information, facilitating the subsequent reconstruction of a complete 3D model.

[0017] 3) Weighing data acquisition: Before placing the explosive charge onto the rotating platform, the robotic arm first places it on a weighing sensor to measure its total weight. The weighing data is transmitted to the host computer in real time via RS485, providing auxiliary data for judging the consistency of the thickness and density of the fire-limiting layer.

[0018] 4) Two-dimensional image acquisition and character recognition: Before formal testing, a 2D industrial camera acquires images of the top or side of the drug column. The system automatically calls the character recognition (OCR) algorithm module to extract features such as inkjet number and batch information for product traceability and test data binding, ensuring that the test results of each drug column are traceable.

[0019] 5) Data Synchronization and Unified Marking: All data streams from all sensors (3D point cloud, 2D image, OCR recognition, weighing information) are uniformly timestamped globally to achieve accurate synchronization of multi-source data. The system automatically sends a "start detection" signal before detection begins, completing the alignment of multiple module states and task triggering to ensure that the correspondence between data is matched one-to-one and the timing is consistent.

[0020] Data Processing Layer: This layer receives raw multi-source data from the data acquisition layer, including 3D line scan point clouds, 2D industrial camera images, OCR character recognition results, and weighing sensor data. It then performs specific preprocessing for the detection of the flammability restriction layer surface of the explosive charge. The task of this layer is to improve data quality, complete spatial alignment and multimodal fusion, and provide reliable input for feature extraction and quality assessment in the core algorithm layer.

[0021] 1) Noise Reduction and Data Cleaning:

[0022] 3D point cloud data: By using statistical filtering algorithms, discrete noise points caused by ambient light, dust or mechanical vibration are removed, improving the spatial consistency and continuity of point cloud data.

[0023] 2D image data: Median filtering and adaptive histogram equalization algorithms are used to eliminate uneven lighting and image noise, enhance character edge and surface texture details, and ensure the accuracy of subsequent character recognition and appearance analysis.

[0024] OCR character data: The recognized character results are filtered by confidence level to remove low-confidence recognition results and ensure the reliability of the data.

[0025] 2) Coordinate correction and data alignment:

[0026] Based on the high-precision angle feedback from the rotating platform, the discrete line scan point cloud data obtained from each scan is converted and reconstructed into a complete three-dimensional model of the top surface of the drug grain centered on the rotation axis.

[0027] Correct geometric errors caused by eccentric installation of the propellant grains (positioning deviation ≤ ±0.1 mm) or slight tilting to ensure the spatial accuracy of the model.

[0028] Spatial calibration of 2D images is performed, and the image coordinates are mapped to 3D coordinates in a unified manner to achieve the fusion display of surface texture and geometric shape.

[0029] Synchronously associate the OCR-recognized characters with the drug column detection number to complete the data index matching and ensure that the data of each drug column corresponds one-to-one.

[0030] 3) Establishing a datum and correlating it with quality data:

[0031] The ideal design plane of the flame-restricted layer is fitted from the reconstructed 3D model and used as the reference plane for calculating the geometric features such as the thickness and flatness of the coating layer.

[0032] The system calculates the deviation from the theoretical weight by combining the total weight data of the drug column measured by the weighing sensor. If the deviation exceeds the set threshold, the system automatically marks the drug column as "weight abnormal" and binds this information to geometric and texture features in a synchronous manner.

[0033] The final output is high-quality detection data after noise processing, coordinate correction, and multi-source fusion, which provides a basis for the comprehensive evaluation of the algorithm's core layer.

[0034] The core layer of the algorithm is the key to achieving accurate quality assessment of the fuel-limited pavement. It is responsible for extracting features from multi-source preprocessed data and performing intelligent pattern matching and comprehensive judgment. This layer comprehensively utilizes 3D point cloud geometric information, 2D image texture features, OCR character recognition results, and weighing data to construct a comprehensive assessment model for the quality of the fuel-limited pavement through a multi-dimensional feature fusion algorithm, thereby achieving defect detection, classification, and severity quantification.

[0035] 1) Feature extraction

[0036] Key features for quality assessment are extracted from the multi-source detection data output from the data processing layer, specifically including:

[0037] Geometric features (3D point cloud): Calculate the overall flatness of the flame-retardant layer surface and the height distribution relative to the reference design surface to evaluate the uniformity of the coating thickness; identify macroscopic geometric defects such as pits and protrusions, and extract their depth, height, area and other parameters.

[0038] Texture features (3D / 2D fusion): By combining point cloud normal vector changes and local curvature analysis with grayscale gradient features of 2D images, the micro-roughness and linear defects of the surface, such as cracks and scratches, are identified, and their length, direction, width and other indicators are calculated.

[0039] Weight characteristics (weighing data): Analyze the deviation between the measured weight of the propellant and the theoretical standard value, and use it as a reference indicator for abnormal coating thickness or material density.

[0040] Identification and traceability features (OCR data): Extract OCR character recognition results and bind them with detection numbers to achieve traceability of detection results and sample group statistics, providing index information for subsequent database updates and model self-learning.

[0041] 2) Pattern matching and quality assessment

[0042] Multi-parameter threshold comparison: The extracted geometric features (such as flatness and height standard deviation) and texture features (such as crack number and surface roughness), weight deviation and other parameters are compared with the preset standard thresholds in the database in a multi-dimensional way.

[0043] Defect identification and classification:

[0044] ① Determination of Defects: Determine whether the characteristics exceed the standard range to determine whether a quality defect exists;

[0045] ② Defect type determination: Pattern matching is performed through feature combination relationships. For example, a sudden change in local height accompanied by an annular texture abnormality is determined as "bulge", a continuous linear height depression with consistent texture direction is determined as "crack", and local high-frequency noise on the surface may be determined as "roughness abnormality".

[0046] ③ Defect severity classification: Based on the defect's geometric dimensions (area, length, depth) and the magnitude of its threshold deviation, defects are classified into three levels: "minor", "moderate", and "severe", providing a basis for subsequent repair suggestions and process feedback.

[0047] 3) Intelligent Judgment Mechanism and Model Optimization

[0048] Decision model mechanism: This system adopts a hybrid model of decision tree and rule engine based on multi-parameter features, which can automatically select matching parameter sets according to different specifications of propellant and material type (such as different thicknesses of flame-limiting layers) to achieve adaptive evaluation.

[0049] Self-learning and database updates: The system can automatically optimize feature thresholds based on historical detection results and manually reviewed labels, continuously improving detection accuracy.

[0050] Abnormal data linkage mechanism: When a certain drug column is judged to be abnormal in geometric or texture features, the algorithm will automatically compare its weight with the inkjet printing information to determine whether there is a batch or process abnormality, and generate linkage alarm information.

[0051] User Interaction Layer: This layer provides a dedicated visual operation and result display interface for the quality inspection of fuel-limited pavements, serving as a crucial bridge connecting the inspection algorithm and operators within the system. This layer not only handles functions such as setting inspection parameters, monitoring the inspection process, and visualizing results, but also integrates multi-source data display, traceability management, and report generation modules, achieving intuitive, intelligent, and closed-loop information management of the inspection process.

[0052] Operators can preset detection parameters and standard feature thresholds for different specifications of drug cartridges (50–150 mm in diameter) in the interface, including rotation speed, scanning resolution, allowable weight deviation range, and OCR recognition confidence threshold. After the detection task is started, the interface displays the detection status and equipment working progress in real time, including the robotic arm's movement status, rotary platform speed feedback, 3D line scan data sampling frequency, 2D camera image acquisition, and weighing sensor feedback values, ensuring that the operation process is transparent and controllable.

[0053] After the detection is completed, the system will visualize and render the multidimensional detection results in the form of a virtual disk:

[0054] 3D point cloud defect presentation: The location, type and severity of defects on the surface of the fire-retardant layer are intuitively marked with different colors and highlighted areas (for example, cracks are marked with red lines and pits are marked with yellow areas).

[0055] 2D image association display: Users can click on any area on the disc to view the corresponding 2D camera image details and OCR recognition results, realizing the fusion and comparison of 3D structure and image information.

[0056] Weight and pass / fail indicator: The interface synchronously displays the comparison results between the measured weight of the drug column and the standard value. When the deviation exceeds the set threshold, it is automatically marked as "weight abnormal" and highlighted to prompt the operator to further verify.

[0057] Inspection record traceability: The system automatically retrieves historical inspection records by using the inkjet number or batch number identified by OCR, enabling quality traceability from a single drug cartridge to the batch level.

[0058] After the system completes the test, a complete test report will be automatically generated, including:

[0059] Geometric and texture parameters of the flame retardant layer surface (flatness error, maximum defect size, roughness index, etc.);

[0060] Quality grade and defect distribution statistics chart;

[0061] Analysis of propellant column weight deviation;

[0062] OCR batch identification and testing timestamp information.

[0063] The report can be exported as PDF, Excel, or database record format for subsequent production analysis and quality tracking. The interface supports one-click report generation, zoom-in viewing of defect details, and batch result comparison, providing production quality control personnel with an efficient and intuitive operating experience.

[0064] Interaction logic: The software implementation process for adaptively adjusting the detection strategy.

[0065] This system features an adaptive detection strategy, implemented in software using a multi-source data closed-loop control process. The system automatically adjusts detection parameters based on real-time data collected during the detection process, including 3D point clouds, 2D images, character recognition results, and changes in weighing data, achieving a dynamic balance between detection accuracy and efficiency. This closed-loop control mechanism comprises the following four stages:

[0066] 1) Real-time data quality monitoring

[0067] During the rotational scanning of the drug column, the core layer of the algorithm monitors and analyzes the multi-source detection data transmitted by the data acquisition layer in real time, including 3D point cloud, 2D image, character recognition results and weighing information.

[0068] The system calculates several key metrics during each detection cycle, such as point cloud density, signal-to-noise ratio, rate of change of local curvature, image brightness uniformity, OCR character recognition confidence, and weight fluctuation amplitude. These metrics are used to determine the reliability and data integrity of the current detection status.

[0069] If an increase in surface texture complexity, abnormal lighting, decreased character recognition confidence, or unstable weighing data is detected during the scanning process, the system will automatically record the state and enter the decision-making stage.

[0070] 2) Decision-making

[0071] The system compares real-time analysis results with various preset thresholds in the database, including "data quality threshold," "feature complexity threshold," "identification confidence threshold," and "weighing deviation tolerance." The system will trigger an adaptive parameter adjustment mechanism when it detects any of the following conditions:

[0072] The appearance of sudden changes in local curvature or a decrease in point cloud density may indicate a suspected defect area, which requires confirmation with higher precision data.

[0073] The surface texture features are exceptionally complex, making it difficult for the detection algorithm to converge stably.

[0074] Uneven brightness in 2D images or low confidence level in OCR recognition;

[0075] The drift of the weighing feedback signal exceeds the allowable range, affecting the judgment of the quality of the fire-restricted layer.

[0076] Through this stage of comparison and judgment, the system can intelligently identify the target area and specific reasons that require optimization of detection parameters.

[0077] 3) Parameter adjustment execution

[0078] Once the condition is triggered, the system sends optimization commands to each device module through the data acquisition layer, performing adaptive parameter adjustments. Specifically, this includes:

[0079] For areas with suspected defects or complex textures: the system automatically reduces the rotation speed of the rotating platform (e.g., from 15 r / min to 8 r / min) and simultaneously increases the sampling frequency of the 3D line scan sensor and the exposure accuracy of the 2D camera to acquire denser point cloud data and clearer image information per unit area.

[0080] To address excessive point cloud noise, the system will enable the sensor's built-in multiple average scan mode or superposition filtering algorithm to suppress random noise and improve the signal-to-noise ratio.

[0081] For OCR recognition anomalies: automatically adjust the lighting intensity or retake the current area to optimize the clarity of character edges and improve the recognition rate.

[0082] Regarding weighing drift: The system pauses the current detection cycle, performs zero-point calibration, or prompts the operator to reset the sensor. Through this automated adjustment mechanism, the system can dynamically correct the detection strategy without manual intervention.

[0083] 4) Effect verification and iterative optimization

[0084] After parameter adjustments are completed, the system immediately enters the effect verification phase, analyzing and comparing the re-collected data. If the data quality, OCR accuracy, and weighing stability all meet the set requirements, the system will resume the normal testing process and record the adjustment results. If the threshold conditions are still not met, the system will automatically return to the decision-making phase, recalculate the adjustment plan, and start the next round of parameter optimization. To avoid excessive iteration causing testing delays, the system sets a maximum number of adjustments and a time limit. If the desired effect is not achieved after exceeding these limits, the system will mark the drug column as "testing pending verification" and indicate this in the testing report.

[0085] This invention simplifies the testing process, improves testing efficiency and accuracy, and reduces equipment complexity and cost through innovative mechanical structure design and advanced software algorithms. This enables efficient and accurate testing of the quality of fire-restricted pavements, meeting the high quality requirements of modern industrial production. The beneficial effects of this invention are:

[0086] 1. Significantly Improved Detection Efficiency: By employing a rotating platform to drive the rotation of the propellant column and a fixed 3D line scan sensor, the complex movements in multiple directions required by traditional detection methods are avoided, greatly shortening the detection time and improving efficiency. Simultaneously, the integrated 2D industrial camera and weighing sensor provide additional auxiliary data, further reducing the need for extra processing steps. This allows the system to complete multi-dimensional detection simultaneously in a single process, meeting the high-efficiency requirements for quality inspection of fire-limited layers in large-scale production scenarios.

[0087] 2. Significantly Improved Detection Accuracy: The innovative mechanical structure design reduces potential errors during mechanical movement, while the adaptive adjustment of detection strategies in the software algorithm optimizes detection parameters in real time based on the actual conditions of the flame-restricted layer surface. By combining 3D point cloud data, 2D images, character recognition (OCR) data, and weighing data, the detection system can more accurately capture subtle defects and quality issues on the flame-restricted layer surface, improving the accuracy of detection results, especially when handling complex textures and minute defects.

[0088] 3. Reduced Equipment Complexity and Cost: Compared to traditional multi-axis motion detection equipment, the mechanical structure of this invention is simpler, reducing the number and complexity of mechanical parts, thus lowering manufacturing costs and maintenance difficulty. Simultaneously, the efficient algorithms and optimized architecture of the software system improve system operating efficiency, reduce hardware resource requirements, and further reduce overall costs. Furthermore, the integration of a 2D camera, OCR, and weighing sensor for multi-sensor collaborative operation further reduces the need for additional equipment and simplifies the overall system design.

[0089] 5. High adaptability: The adaptive adjustment function of the detection strategy in the software system enables the detection and control system of this invention to adapt to the quality detection needs of different types and specifications of fire-limited pavements. By adjusting the rotation speed, scanning accuracy, and data acquisition frequency according to different specifications of propellant, the system can cope with the challenges of different production conditions and detection tasks, and has a wide range of application scenarios and good versatility.

[0090] 6. High degree of algorithm intelligence: Through adaptive adjustment of detection strategies and multi-feature fusion analysis (including 3D point cloud, 2D image, OCR data, and weighing data), the system possesses strong generalization ability and robustness. Whether inspecting relatively flat surfaces or surfaces with complex textures, the system can intelligently adjust parameters to ensure accuracy and efficiency, meeting the quality inspection needs of fire-limited pavements of different specifications and processes.

[0091] 7. Significantly different from existing technologies: Compared to traditional detection methods based on two-dimensional images or fixed-path scanning, this invention combines rotational motion with 3D line scanning technology to achieve full coverage, high efficiency, and high precision detection of the three-dimensional morphology of the fire-limiting layer. Furthermore, the newly added 2D camera, OCR character recognition, and weighing sensor modules enable the system not only to detect surface defects but also to verify the appearance information and quality consistency of the propellant grains, demonstrating significant technological advantages and innovation. Attached Figure Description

[0092] Figure 1 This is a schematic diagram of the structure of a medicine column;

[0093] Figure 2This is a flowchart of the algorithm of the present invention;

[0094] Figure 3 This is one of the schematic diagrams of the mechanical structure of the device of the present invention;

[0095] Figure 4 This is the second schematic diagram of the mechanical structure of the device of the present invention;

[0096] Figure 5 This is a flowchart illustrating the use of the device of the present invention. Detailed Implementation

[0097] Figure 2 Algorithm flow:

[0098] The system's software algorithm adopts a multi-source data fusion and hierarchical processing architecture to achieve high-precision detection and intelligent analysis of the surface quality of the fire-limited layer. The overall process includes data acquisition, data preprocessing, feature extraction, pattern matching and quality assessment, and result output. The modules are interconnected efficiently through a unified communication bus and data interface.

[0099] During system operation, a 3D line scan sensor acquires three-dimensional point cloud data of the propellant column surface, a 2D industrial camera acquires two-dimensional images of the surface, an OCR module recognizes the inkjet-printed character information, and a weighing sensor measures the total weight of the propellant column. After timestamp synchronization and cache management, the multi-source data is uniformly transmitted to the host computer software platform, providing basic data support for subsequent fusion analysis.

[0100] The data preprocessing module is responsible for cleaning and improving the quality of multidimensional data. The system uses statistical filtering and geometric correction algorithms to denoise and spatially align point cloud data, eliminating discrete errors caused by lighting, dust, or vibration. Two-dimensional image data undergoes brightness equalization and edge enhancement to effectively improve the clarity of surface texture features. OCR recognition results are subjected to confidence correction and character re-recognition to improve accuracy and stability. Weighing data undergoes drift compensation and fluctuation detection to ensure the reliability of weight measurement. All preprocessed multi-source data are uniformly converted to a standardized format to ensure feature dimension matching and spatial consistency during the fusion calculation stage.

[0101] The feature extraction module performs fusion analysis on the processed data to extract multi-dimensional quality assessment parameters. The system extracts geometric features from 3D point clouds, including flatness, height distribution, thickness uniformity, and geometric parameters of local defects; it extracts surface texture features from 2D images, such as roughness, crack orientation, scratch distribution, and grayscale variation characteristics; and it calculates material density deviation and the consistency index of the weight of the fire-retardant layer from the weighing data. The OCR module provides identification tags for the tested samples, ensuring a unique correspondence between all data and test results in the database, enabling full traceability of test records.

[0102] The pattern matching and quality assessment module is the core of the entire algorithm. The system compares the extracted geometric, texture, and weight features with pre-set standard flammability-limiting layer models in the database, calculating a multi-dimensional similarity index using a weighted fusion algorithm. The quality assessment results include defect presence determination, defect type identification (such as bulges, cracks, dents, and abnormal roughness), and severity classification. The system employs machine learning algorithms (support vector machines, convolutional neural networks, or ensemble decision models) to train and update parameters on historical defect samples, thereby improving the algorithm's ability to identify subtle defects.

[0103] The test results summary module integrates multi-source analysis results and presents them in a visual manner. The system outputs a superimposed view of a 3D point cloud reconstruction model and a 2D image, annotating defect areas and type information; displays propellant weight deviation, OCR recognition status, and comprehensive quality score; and provides statistical charts of geometric errors and texture distribution on the surface of the fire-limiting layer. The test results can automatically generate standardized report files, including all quantitative indicators, analysis results, and traceability information, supporting three output formats: PDF, Excel, and database import, facilitating subsequent quality management and process optimization.

[0104] Through the above algorithm design, the system achieves a fully automated closed-loop detection process, from multi-source data acquisition and fusion calculation to intelligent evaluation and visualization output. The combination of multi-dimensional feature fusion and machine learning algorithms enables the system to possess high-precision, self-learning detection capabilities, providing reliable technical support and intelligent assurance for the quality inspection of fuel-limited layers.

[0105] Code architecture:

[0106] The software system adopts a layered architecture design, mainly including a data acquisition layer, a data processing layer, an algorithm core layer, and a user interaction layer. Each layer has a clear function and works closely together to ensure the accuracy and efficiency of the quality detection of the fuel-limited layer.

[0107] The data acquisition layer communicates in real-time with the robotic arm, rotary platform, 3D line scan sensor, 2D industrial camera, and weighing sensor via EtherCAT, RS485, and USB 3.0 interfaces. This layer not only acquires raw data but also ensures the synchronization and integrity of multi-source data, achieving consistent correspondence between 3D point clouds, 2D images, character information, and drug cartridge weight data. The rotation speed control and encoder feedback of the rotary platform, the triggering and frequency setting of the 3D line scan sensor, the image acquisition and OCR character recognition of the 2D industrial camera, and the data acquisition from the weighing sensor are all uniformly managed by the data acquisition layer. All acquired data is timestamped to ensure synchronized recording of data from all sources, providing a reliable foundation for subsequent processing and analysis.

[0108] The data processing layer performs preliminary processing on the acquired raw data. The 3D point cloud data undergoes noise reduction, coordinate correction, and alignment, and is reconstructed into a complete 3D model of the drug column centered on the rotation axis. The 2D images undergo brightness equalization, edge enhancement, and character region optimization to provide clear images for character recognition. OCR character recognition results undergo confidence verification and optimization, while weighing data undergoes drift compensation and stability processing. The data processing layer standardizes and unifies multi-source data, providing high-quality, multi-dimensional analytical input for the core algorithm layer.

[0109] The core algorithm layer is the central module for assessing the quality of fire-restricted veneer layers. This layer extracts geometric and texture features from 3D models and 2D images, and combines OCR characters and weighing information for multi-parameter analysis. Pattern matching and machine learning algorithms (such as support vector machines or convolutional neural networks) are used to determine the presence of defects, classify defect types, and grade their severity. The adaptive detection strategy module adjusts the rotational speed of the rotating platform and the sensor acquisition frequency based on real-time data quality and surface complexity, achieving dynamic optimization detection of complex fire-restricted veneer surfaces and improving detection accuracy and efficiency.

[0110] The user interface provides an intuitive graphical interface, allowing operators to set detection parameters and standard feature thresholds, select scanning modes, and start the detection process. Detection results are visualized in the form of a virtual disk, displaying the location, type, and severity of defects. A detailed detection report is generated, including geometric errors, defect dimensions, surface roughness, weight deviation, and OCR information, enabling traceability and convenient management of the detection results.

[0111] The layered architecture design ensures efficient collaboration in data acquisition, processing, analysis, and interaction, while also enabling multi-source information fusion, real-time adaptive control, and result traceability, providing stable and reliable software support for the quality inspection of fuel-limited kerb layers.

[0112] Interaction logic:

[0113] The software system of this invention features an adaptive adjustment function for detection strategies. During the detection process, the system analyzes the collected 3D point cloud, 2D image, OCR character information, and weighing data in real time, and automatically adjusts the detection parameters according to the actual conditions of the flame-restricted layer surface. For example, when the system detects areas with complex surface textures, suspected defects, or difficulties in character recognition, it automatically reduces the rotational platform speed and increases the sensor acquisition frequency to obtain denser data points; while for areas with uniform surfaces and clear character recognition, standard parameters are used to improve detection efficiency. This interactive logic of adaptive adjustment of the detection strategy greatly improves the adaptability and detection accuracy of the detection system for different types of flame-restricted layers.

[0114] (1) Data preprocessing module

[0115] The system performs targeted preprocessing on various types of raw data, laying a reliable foundation for subsequent analysis. For 3D point cloud data, adaptive median filtering effectively suppresses noise interference while preserving edge details to the greatest extent possible; geometric calibration and deviation correction of sensor pose are completed based on a preset calibration board, eliminating inherent system errors; and missing points are repaired through data interpolation and smoothing, significantly improving data integrity. For 2D image data, illumination equalization and distortion correction are performed to ensure stable image quality; automatic ROI localization is achieved, providing accurate regional support for subsequent character recognition and appearance defect analysis. In OCR character data processing, preprocessing operations are performed on 2D images before character recognition, including image binarization, noise reduction, and rotation correction, optimizing the prerequisites for character recognition. For weighing data, the system automatically removes the effects of sensor zero drift and external vibration, ensuring the accuracy of weight measurement results.

[0116] (2) Feature extraction module

[0117] The system employs a multi-feature fusion strategy to comprehensively extract key quality indicators from the surface of the flame-restricted layer. For texture feature extraction, core texture parameters such as contrast, energy, and entropy are calculated based on the gray-level co-occurrence matrix, while simultaneously fusing 3D point cloud normal vectors and 2D image texture features to further enhance the accuracy of defect identification. In terms of geometric features, key geometric parameters such as thickness, flatness, and roundness of the flame-restricted layer are accurately extracted through contour fitting and curvature analysis, reflecting its structural morphological characteristics. Defect feature extraction focuses on typical defects such as pits, cracks, and bubbles, identifying their morphology and distribution characteristics, while also referencing 2D image information to enhance the recognition effect of defect edges. Character features are used to verify the accuracy of batch and process number information of the propellant cartridges through OCR recognition, ensuring the reliability of product traceability. Weight features are used to determine whether the weight of the flame-restricted layer is within the standard range, assisting in the determination of whether there are any abnormalities or missing materials.

[0118] (3) Pattern matching and classification module

[0119] This module precisely matches the extracted feature vectors with the pre-stored standard fuel-limiting layer features in the database. In the similarity calculation stage, a combination of Euclidean distance and cosine similarity is used to comprehensively and objectively evaluate the degree of feature matching, ensuring the scientific validity of the matching results. The defect classification stage uses Support Vector Machine (SVM) or Convolutional Neural Network (CNN) models to accurately classify the identified defect types, while also integrating OCR character data and weight data for auxiliary judgment, improving the reliability of the classification results. Furthermore, the module outputs a confidence score for each defect and character recognition, providing a reference for subsequent manual review and facilitating the graded processing of defects.

[0120] (4) Adaptive detection strategy module

[0121] The system features real-time feedback and parameter self-adjustment capabilities, dynamically optimizing detection strategies based on actual testing scenarios. Regarding rotational speed adaptation, the system automatically matches the optimal rotational speed based on the propellant diameter and surface complexity, ensuring uniformity in 3D scanning and image acquisition. For accuracy and frequency adjustment, the system dynamically adjusts 3D scanning accuracy and data acquisition frequency according to the surface complexity of the flame-retardant layer. When encountering blurred characters or OCR recognition failures, it automatically increases the number of image acquisitions to improve the recognition success rate. In terms of detection modes, it supports switching between multiple modes such as "fast scanning," "fine inspection," and "defect re-inspection," meeting both the efficiency requirements of production cycles and ensuring strict quality inspection standards. Simultaneously, the adaptive strategy fully considers multi-source data such as 3D geometric data, 2D image texture, OCR character recognition rate, and weighing deviation, achieving comprehensive optimization of the detection process through multi-source data fusion to ensure the comprehensiveness and accuracy of the detection results.

[0122] 3. Process Description

[0123] The inspection process is initiated by a robotic arm. Following a pre-programmed sequence, the robotic arm precisely picks up the propellant cartridge from the loading rack and places it stably on the rotating platform. During this process, a weighing sensor simultaneously measures the total weight of the cartridge, and the result is transmitted to the software system in real time. The presence of a flame-retardant layer is determined based on the total weight of the cartridge and the weight data provided by the preceding process. The weight provided by the preceding process represents the weight of the cartridge after it has been filled with propellant but before the flame-retardant layer is installed. The difference between the total weight and the weight provided by the preceding process is the weight of the flame-retardant layer. A 2D industrial camera simultaneously captures images of the cartridge, providing foundational data for subsequent Optical Character Recognition (OCR) and visual defect analysis. A 3D line scan sensor, with pre-set detection parameters, extends its detection width from the origin on the top surface of the cartridge outwards in a radial direction, ensuring coverage of the entire cartridge surface and remaining in standby mode.

[0124] After the propellant grain is placed in position, the rotating platform starts rotating at a preset uniform speed, and the 3D line scan sensor simultaneously scans the surface of the propellant grain line by line, acquiring three-dimensional data information in real time. Once the rotating platform has completed one revolution, the full surface scanning of the propellant grain is finished, and the hardware data acquisition phase is officially completed.

[0125] During the 3D scanning process, the 2D industrial camera continuously monitors the surface of the drug cartridge to capture potential appearance defects. At the same time, the OCR module reads the inkjet information on the surface of the drug cartridge (including production date, batch number, etc.) to retain key data for product traceability analysis.

[0126] All collected multi-source data (including 3D point cloud data, 2D images, OCR character data, and weighing data) are uniformly transmitted to the software system for subsequent processing. The software system first performs preprocessing operations on the raw data to remove noise interference and data deviations; then, through feature extraction and pattern matching analysis, it completes the quality status determination of the fuel-limited layer, specifically including: analyzing the thickness distribution, flatness, and obvious defects of the fuel-limited layer based on 3D point cloud data; verifying the accuracy of the propellant marking information with the help of OCR recognition results, and simultaneously detecting missing or misidentified inkjet markings; and determining whether the overall quality of the propellant meets the specifications based on the weighing data.

[0127] After the detection and analysis are completed, the system generates a disk shape that matches the top surface of the propellant column by fitting the detection results to line segments. The upper control system then displays the detailed detection results on the monitor for operators to view and analyze. The detection results cover core information such as the geometric characteristics of the propellant column, appearance defects, weight anomalies, and inkjet printing information, and automatically generate a detailed detection report to provide a basis for subsequent quality control and traceability management.

[0128] The entire process leverages the close collaboration between mechanical structures and software algorithms, fully utilizing the advantages of multi-sensor linkage to achieve efficient and accurate detection of the quality of the fire-limited layer.

[0129] A device for detecting the quality of fire-limited kerb based on rotational motion includes a mechanical structure and a software algorithm. The mechanical structure consists of a robotic arm, a rotating platform, a 3D line scan sensor, a 2D industrial camera, and a weighing sensor. The software algorithm has functions such as data preprocessing, feature extraction, pattern matching, and adaptive adjustment of detection strategies. The mechanical structure and software algorithm work together to achieve multi-source detection of the fire-limited kerb quality. The robotic arm has a repeatability accuracy of ≤±0.02mm, a load capacity of ≥7kg, and is equipped with customized rubber grippers. It can pick up explosive charges weighing ≤5kg and with a diameter of 50-150mm from the loading rack and place them at the center of the rotating platform with a positioning deviation of ≤±0.1mm. The system drives the explosive charges to rotate at a uniform speed through the rotating platform. The fixedly installed 3D line scan sensor and 2D industrial camera perform scanning and image acquisition, while the weighing sensor records the weight of the explosive charges. This achieves full-coverage detection and multi-source data acquisition under a single rotational motion, replacing the traditional multi-axis reciprocating motion method, simplifying the mechanical structure and improving detection efficiency.

[0130] The rotating platform is an electric rotating platform with a table diameter ≥200mm, a maximum load capacity ≥20kg, and a rotation accuracy ≤±0.005°. It is equipped with a servo motor and encoder to achieve real-time speed feedback and closed-loop control, and the speed is adjustable in the range of 5-30r / min.

[0131] The 3D line scan sensor has a scanning accuracy of ≤±0.01mm, a scanning width of 50-200mm, and a data acquisition frequency of 500-2000Hz. It has a built-in laser calibration function and is fixedly installed 300-500mm directly above the rotating platform. The scanning starting point is aligned with the center of the top surface of the drug column, and the scanning extends radially to cover the top surface area of ​​drug columns with a diameter of 150mm or less. The 2D industrial camera is used to capture images of the top and side surfaces of the drug columns for appearance defect analysis and character recognition (OCR). The camera resolution is ≥5MP and the frame rate is ≥30fps. The weighing sensor is installed on the rotating platform or at the loading position to collect the total weight of the drug columns in real time with an accuracy of ≤±0.1g.

[0132] The computer hardware environment required for the software algorithm to run is as follows: CPU Intel Core i7-12700K or higher, memory ≥32GB DDR4, hard drive ≥1TB SSD, graphics card supporting CUDA acceleration, and operating system Windows 10 Professional (64-bit). The software adopts a layered architecture, including a data acquisition layer, a data processing layer, an algorithm core layer, and a user interaction layer. The data acquisition layer communicates with the robotic arm, rotary platform, 3D line scan sensor, 2D industrial camera, and weighing sensor through EtherCAT, RS485, and USB3.0 interfaces to achieve multi-source synchronous data acquisition.

[0133] The software algorithm processing flow is as follows: First, the raw data collected by the 3D line scan sensor is preprocessed to remove noise and correct deviations. Simultaneously, the 2D image is enhanced and geometrically corrected, and OCR character recognition is performed. Then, key features such as surface texture, geometry, two-dimensional appearance, character information, and weight of the flame-restricted layer are extracted. Subsequently, the extracted features are compared with the standard flame-restricted layer features in the database using a pattern matching algorithm to determine the quality problem, type, and severity. Finally, the results are summarized to generate a report, which is displayed on the monitor in a disc format through the upper control system. The pattern matching algorithm uses a machine learning classification model to achieve automatic defect identification and classification through multi-source feature fusion.

[0134] The adaptive detection strategy function in the software algorithm can dynamically adjust the rotational platform speed, 3D line scan sensor scanning accuracy and data acquisition frequency, 2D camera shooting parameters, and OCR recognition strategy based on the actual surface conditions of the fire-limited layer and multi-source data to optimize detection accuracy and efficiency. The rotational platform speed is automatically adjusted according to the propellant diameter (15 r / min for 50-100 mm, 10 r / min for 100-150 mm); the 3D line scan sensor scanning accuracy is adjustable between ±0.01 mm and ±0.05 mm, and the acquisition frequency is between 500-2000 Hz; the 2D camera exposure and frame rate can be automatically adjusted according to surface texture and lighting conditions; the OCR recognition strategy can dynamically increase the number of shots or adjust preprocessing parameters.

[0135] During testing, the robotic arm picks up the drug cartridge and places it on the rotating platform. The platform rotates at a constant speed, and the 3D line scan sensor and 2D camera simultaneously collect data. The weighing sensor records the weight of the drug cartridge. One rotation completes the hardware data acquisition. The software system processes the multi-source data and displays the results in a disc format. During mechanical structure installation, the concrete thickness of the robotic arm base is ≥300mm, the levelness deviation of the rotating platform surface is ≤0.02mm / m, and the data error of the 3D line scan sensor is ≤±0.01mm when calibrated with a calibration block. The system achieves full-coverage scanning of the top surface of the drug cartridge and synchronous acquisition of multi-source data through a single rotational motion, eliminating the need for multi-axis motion, reducing the risk of mechanical error accumulation, and improving the consistency and reliability of testing.

[0136] Equipment setup (such as) Figure 3 , 4 Select a robotic arm 4 with high-precision positioning capabilities, whose load capacity should meet the requirement of stably grasping the tested propellant charge. The rotating platform 1 must have high-precision rotation control capabilities to ensure stability and uniformity during rotation. The 3D line scan sensor 2 should be selected according to the accuracy and range requirements of the fire hazard layer detection, ensuring accurate acquisition of three-dimensional data information of the propellant charge surface. Simultaneously, install a 2D industrial camera 3 and its matching light source assembly to acquire images of the propellant charge appearance and perform character recognition (OCR); install a weighing sensor 5 to obtain the total weight of the propellant charge, providing auxiliary data for fire hazard layer quality assessment. Install and debug the robotic arm, rotating platform, 3D line scan sensor, 2D camera, and weighing sensor according to design requirements, ensuring secure connections between components, smooth movement, normal sensor operation, and synchronized data acquisition from multiple sensors. A display is mounted on the rotating arm 6.

[0137] 2. Software System Installation and Configuration: Install the developed software system on a computer with the appropriate computing capabilities. During installation, ensure that the communication interfaces between the software system and each hardware device (robotic arm, rotary platform, 3D line scan sensor, 2D camera, load cell) are correctly configured to enable smooth data transmission. Based on actual testing requirements, set initial testing parameters in the software system, such as the rotation speed of the rotary platform, the scanning accuracy of the 3D line scan sensor, the 2D camera acquisition parameters, OCR recognition settings, load cell calibration values, and data acquisition frequency.

[0138] 3. Testing process (e.g.) Figure 5 After the detection system is started, the robotic arm picks up the drug cartridges from the loading rack according to the preset program and performs the following operations in sequence:

[0139] The medicine column is placed on the weighing sensor, the total weight of the medicine column is collected and transmitted to the software system in real time.

[0140] The drug cartridge is moved to the position of the 2D industrial camera to capture an image of the drug cartridge surface and perform OCR character recognition.

[0141] The drug column is placed on a rotating platform, which then begins to rotate at a constant speed. Simultaneously, a 3D line scan sensor scans the surface of the drug column, collecting three-dimensional point cloud data.

[0142] After data acquisition, the software system automatically preprocesses, extracts features, and performs pattern matching analysis on all raw data, including: geometric and texture feature analysis of 3D point clouds, processing of 2D image defects and character recognition results, and comparison of weighing data. The system comprehensively analyzes various data to determine the quality of the fire-restricted layer and displays the test results in a disc format on the monitor, indicating the location, type, and severity of defects. Operators can assess the quality of the fire-restricted layer based on the test results displayed on the monitor and analyze and process the detailed information in the test report.

[0143] 4. System Maintenance and Optimization: Regularly inspect and maintain the mechanical structure to ensure the good mechanical performance of each component and that the motion accuracy meets requirements. Update and optimize the software system, continuously improving the algorithm based on problems encountered during actual testing and new requirements to enhance the system's detection performance and stability. The 2D camera, OCR module, and weighing sensor also require regular calibration and inspection to ensure clear image acquisition, accurate character recognition, and reliable weight measurement. Simultaneously, regularly perform overall calibration of the entire multi-sensor system to ensure the accuracy and reliability of the detection results.

Claims

1. A method for detecting the quality of fire-limiting pavements based on rotational motion, characterized in that: Includes the following steps Step 1) Data Collection Weighing data acquisition: First, place the medicine column on the weighing sensor to measure the actual weight of the medicine column and obtain the weighing information; Two-dimensional image acquisition and character recognition: The vision acquisition component acquires images of the top or side of the medicine column to obtain 2D image data. The character recognition algorithm module is called to extract the OCR character data of the medicine column based on the character recognition results. The OCR character data includes the inkjet number and batch information features. A 2D camera and its accompanying light source together form a vision acquisition component; 3D point cloud acquisition: During the rotation of the propellant column placed on the rotating platform, the 3D line scan sensor scans the surface of the flame retardant layer line by line, generating radial line scan point cloud data with the center of the top surface of the propellant column as the origin; Data synchronization and unified tagging: Weighing information from weighing data collection, 2D image data from 2D image collection, OCR recognition information, and 3D point cloud data are all uniformly tagged with a global timestamp; Step 2) Data Processing: Receive raw multi-source data from data acquisition, including 3D point cloud data, 2D image data, OCR character data, and total weight data of the propellant charge. Perform specific preprocessing for the detection of the propellant charge's flame-retardant layer surface to obtain multi-source detection data. The specific preprocessing includes: 1) Noise removal and data cleaning: 3D point cloud data uses statistical filtering algorithms to remove discrete noise points caused by ambient light, dust, or mechanical vibration; The 2D image data is processed using median filtering and adaptive histogram equalization algorithms to eliminate uneven lighting and image noise, and enhance character edge and surface texture details. OCR character data: Confidence filtering is performed on the recognized character results to remove low-confidence recognition results; 2) Coordinate correction and data alignment: Based on the high-precision angle feedback of the rotating platform, the discrete radial line scanning point cloud data obtained from each scan is converted and reconstructed into a complete three-dimensional model of the top surface of the drug grain centered on the rotation axis. Establish a reference coordinate system, with the engine housing axis as the theoretical center axis, and define the radial and axial measurement reference surfaces to obtain the eccentricity value of the propellant grain installation, and correct the geometric error caused by the eccentricity or slight tilt of the propellant grain installation. Spatial calibration is performed on 2D image data to uniformly map image coordinates to 3D model coordinates, thereby achieving the fusion display of surface texture and geometric shape; Synchronously associate the OCR-recognized characters with the drug column detection number to complete the data index matching and ensure that the data of each drug column corresponds one-to-one. 3) Establishment of datum planes and correlation of quality data: The ideal design plane of the flame-restricted layer is fitted from the reconstructed 3D model and used as the reference plane for calculating the thickness and flatness geometric characteristics of the flame-restricted layer. Based on the weighing information, the deviation between the weight of the fire-limiting layer measured by the weighing sensor and the theoretical weight is calculated; if the deviation exceeds the set threshold, the explosive charge is automatically marked as "weight abnormal" and this information is synchronously bound to geometric and texture features; Step 3) Quality assessment of the fire-restricted layer 1) Feature extraction From the multi-source detection data output by data processing, key features for fire-limited pavement quality assessment are extracted. These key features include geometric features, texture features, weight features, and identification and traceability features. 2) Pattern matching and quality assessment Multi-parameter threshold comparison: The extracted geometric features, texture features, and weight features are compared with the preset standard thresholds in the database in multiple dimensions; Defect identification and classification: Based on the comparison results, determine whether the features exceed the standard range, thereby determining whether there is a quality defect; for drug columns with quality defects, determine the defect type by pattern matching through feature combination relationships; Defect severity classification: Based on the defect's geometric dimensions and the magnitude of its threshold deviation, defects are classified into three levels: "minor", "moderate", and "severe". Step 1) During data acquisition, the detection strategy is dynamically adjusted based on the quality changes of 3D point cloud data, 2D image data, character recognition results, and weighing data collected in real time during the detection process. The dynamic correction detection strategy is as follows: 1) Real-time data quality monitoring During the rotational scanning of the drug column, the multi-source detection data transmitted by the data acquisition system is monitored and analyzed in real time. Multiple key indicators are calculated in each detection cycle. These key indicators include point cloud density, signal-to-noise ratio, local curvature change rate, image brightness uniformity, OCR character recognition confidence, and weighing fluctuation amplitude. These key indicators are used to determine the reliability and data integrity of the current detection status. If an increase in surface texture complexity, abnormal lighting, decreased character recognition confidence, or unstable weighing data is detected during the scanning process, the state will be automatically recorded and the process will proceed to the decision-making stage. 2) Decision-making The real-time analysis results are compared with multiple preset thresholds in the database, including "data quality threshold," "feature complexity threshold," "identification confidence threshold," and "weighing deviation tolerance." The system will trigger an adaptive parameter adjustment mechanism when it detects any of the following conditions: The appearance of sudden changes in local curvature or a decrease in point cloud density may indicate a suspected defect area, which requires confirmation with higher precision data. The surface texture features are exceptionally complex, making it difficult for the detection algorithm to converge stably. Uneven brightness in 2D images or low confidence level in OCR recognition; The drift of the weighing feedback signal exceeds the allowable range, affecting the judgment of the quality of the fire-restricted layer. Through this stage of comparison and judgment, the target area and specific reasons that need to optimize the detection parameters are identified; 3) Parameter adjustment execution Once the judgment condition is triggered, optimization instructions are sent to each device module to perform adaptive parameter adjustments. Each device module includes a vision acquisition component, a 3D line scan sensor, a weighing sensor, and a rotating platform. The 2D camera and its matching light source together form the vision acquisition component. Parameter adaptive adjustment includes: For areas with suspected defects or complex textures: automatically reduce the rotation speed of the rotating platform, and simultaneously increase the sampling frequency of the 3D line scan sensor and the exposure accuracy of the 2D camera, so as to obtain denser point cloud data and clearer image information per unit area; To address excessive point cloud noise: enable the multi-average scan mode or superposition filtering algorithm built into the 3D line scan sensor to suppress random noise and improve the signal-to-noise ratio; For OCR recognition anomalies: automatically adjust the lighting intensity or retake the current area to optimize the clarity of character edges and improve the recognition rate; For weighing drift: Pause the current detection cycle, perform zero-point calibration, or prompt the operator to reset the sensor; 4) Effect verification and iterative optimization After parameter adjustment is completed, the effect verification stage is entered, and the re-collected data is analyzed and compared. If the data quality, OCR accuracy and weighing stability all meet the set requirements, the normal detection process will be restored and the adjustment results will be recorded. If the threshold conditions are still not met, the system will automatically return to the decision-making stage, recalculate the adjustment plan and start the next round of parameter optimization. If the number of parameter optimizations exceeds the maximum number of adjustments and the time limit, and the desired effect is still not achieved, the drug column will be marked as "testing pending verification" and noted in the test report.

2. The method for detecting the quality of fire-limiting layers based on rotational motion according to claim 1, characterized in that: The method for extracting key features as described in step three): Geometric features: Calculate the overall flatness and height distribution of the flame-restricted layer surface based on 3D point cloud data to evaluate the uniformity of the flame-restricted layer thickness; identify macroscopic geometric defects such as pits and protrusions, and extract the depth, height, and area parameters of defects in the flame-restricted layer; Texture features: By combining point cloud normal vector changes and local curvature analysis with gray-level gradient features of 2D images, the micro-roughness and linear defects of the surface, such as cracks and scratches, are identified, and the length, direction, and width of the linear defects are calculated. Weight characteristics: The weight deviation between the measured weight of the propellant charge and the theoretical standard value is analyzed and used as a reference indicator for abnormal thickness or density of the flame-restricted layer. Identification and traceability features: Extract the OCR character recognition results and bind them with the detection number to achieve traceability of the detection results and sample group statistics.

3. The method for detecting the quality of fire-limiting pavements based on rotational motion according to claim 1, characterized in that: Step 3) Determination of the type of Chinese medicine column: Pattern matching is performed by combining features. Local height abrupt change accompanied by annular texture abnormality is determined as "bulge", continuous linear height depression with consistent texture direction is determined as "crack", and local high-frequency noise on the surface is determined as "roughness abnormality".

4. The method for detecting the quality of fire-limiting layers based on rotational motion according to claim 1, characterized in that: The test results are generated and displayed on the monitor in the form of a disc, indicating the location, type and severity of defects. Based on the displayed test results, the operator assesses the quality of the fire-restricted layer and analyzes and processes the information in the test report.

5. The method for detecting the quality of fire-limiting layers based on rotational motion according to claim 1, characterized in that: Step three) of the fire-limiting layer quality assessment also includes: 3) Intelligent Judgment Mechanism and Model Optimization Decision model mechanism: A hybrid model of decision tree and rule engine based on multi-parameter features is adopted to automatically select matching parameter set according to different specifications of pharmacopoeia and material type to achieve adaptive evaluation; Self-learning and database updates: Automatically optimize feature thresholds based on historical detection results and manually reviewed labels to continuously improve detection accuracy; Abnormal data linkage mechanism: When a certain drug column is judged to be abnormal in geometric or texture features, the algorithm will automatically compare its weight with the inkjet printing information to determine whether there is a batch or process abnormality, and generate linkage alarm information.

6. A quality detection system for fire-limited pavement based on rotational motion, characterized in that: include 1) Data Acquisition Layer Weighing data acquisition: First, place the medicine column on the weighing sensor to measure the actual weight of the medicine column and obtain the weighing information; Two-dimensional image acquisition and character recognition: The vision acquisition component acquires images of the top or side of the medicine column to obtain 2D image data, and calls the character recognition algorithm module to extract the OCR character data of the medicine column. The OCR character data includes the inkjet number and batch information features. A 2D camera and its accompanying light source together form a vision acquisition component; 3D point cloud acquisition: During the rotation of the propellant column placed on the rotating platform, the 3D line scan sensor scans the surface of the flame retardant layer line by line, generating radial line scan point cloud data with the center of the top surface of the propellant column as the origin; Data synchronization and unified tagging: Weighing information from weighing data collection, 2D image data from 2D image collection, OCR recognition information, and 3D point cloud data are all uniformly tagged with a global timestamp; (ii) Data processing layer: Receives raw multi-source data from data acquisition, including 3D point cloud data, 2D image data, OCR character data and total weight data of the propellant charge, and performs specific preprocessing for the detection of the surface of the propellant charge's flame-limiting layer to obtain multi-source detection data. III) Quality Assessment Layer for Flammability Restricted Layers 1) Feature extraction From the multi-source detection data output by data processing, key features for fire-limited pavement quality assessment are extracted. These key features include geometric features, texture features, weight features, and identification and traceability features. 2) Pattern matching and quality assessment Multi-parameter threshold comparison: The extracted geometric features, texture features, and weight features are compared with the preset standard thresholds in the database in multiple dimensions; Defect identification and classification: Based on the comparison results, determine whether the features exceed the standard range, thereby determining whether there is a quality defect; for drug columns with quality defects, determine the defect type by pattern matching through feature combination relationships; Defect severity classification: Based on the defect's geometric dimensions and the magnitude of its threshold deviation, defects are classified into three levels: "minor", "moderate", and "severe". Step 1) In the data acquisition layer, the detection strategy is dynamically adjusted based on the quality changes of 3D point cloud data, 2D image data, character recognition results and weighing data collected in real time during the detection process. The dynamic correction detection strategy is as follows: 1) Real-time data quality monitoring During the rotational scanning of the drug column, the multi-source detection data transmitted by the data acquisition system is monitored and analyzed in real time. Multiple key indicators are calculated in each detection cycle. These key indicators include point cloud density, signal-to-noise ratio, local curvature change rate, image brightness uniformity, OCR character recognition confidence, and weighing fluctuation amplitude. These key indicators are used to determine the reliability and data integrity of the current detection status. If an increase in surface texture complexity, abnormal lighting, decreased character recognition confidence, or unstable weighing data is detected during the scanning process, the state will be automatically recorded and the process will proceed to the decision-making stage. 2) Decision-making The real-time analysis results are compared with multiple preset thresholds in the database, including "data quality threshold," "feature complexity threshold," "identification confidence threshold," and "weighing deviation tolerance." The system will trigger an adaptive parameter adjustment mechanism when it detects any of the following conditions: The appearance of sudden changes in local curvature or a decrease in point cloud density may indicate a suspected defect area, which requires confirmation with higher precision data. The surface texture features are exceptionally complex, making it difficult for the detection algorithm to converge stably. Uneven brightness in 2D images or low confidence level in OCR recognition; The drift of the weighing feedback signal exceeds the allowable range, affecting the judgment of the quality of the fire-restricted layer. Through this stage of comparison and judgment, the target area and specific reasons that need to optimize the detection parameters are identified; 3) Parameter adjustment execution Once the judgment condition is triggered, optimization instructions are sent to each device module to perform adaptive parameter adjustments; each device module includes a vision acquisition component, a 3D line scan sensor, a weighing sensor, and a rotating platform; A 2D camera and its accompanying light source together form a vision acquisition component; Parameter adaptive adjustment includes: For areas with suspected defects or complex textures: automatically reduce the rotation speed of the rotating platform, and simultaneously increase the sampling frequency of the 3D line scan sensor and the exposure accuracy of the 2D camera, so as to obtain denser point cloud data and clearer image information per unit area; To address excessive point cloud noise: enable the multi-average scan mode or superposition filtering algorithm built into the 3D line scan sensor to suppress random noise and improve the signal-to-noise ratio; For OCR recognition anomalies: automatically adjust the lighting intensity or retake the current area to optimize the clarity of character edges and improve the recognition rate; For weighing drift: Pause the current detection cycle, perform zero-point calibration, or prompt the operator to reset the sensor; 4) Effect verification and iterative optimization After parameter adjustment is completed, the effect verification stage is entered, and the re-collected data is analyzed and compared. If the data quality, OCR accuracy and weighing stability all meet the set requirements, the normal detection process will be restored and the adjustment results will be recorded. If the threshold conditions are still not met, the system will automatically return to the decision-making stage, recalculate the adjustment plan and start the next round of parameter optimization. If the number of parameter optimizations exceeds the maximum number of adjustments and the time limit, and the desired effect is still not achieved, the drug column will be marked as "testing pending verification" and noted in the test report.

Images