A system and method for leveling and monitoring the printing quality of a 3D printing platform based on vision and laser fusion detection.

Abstract

This invention discloses a 3D printing platform leveling and printing quality monitoring system and method based on vision and laser fusion detection. The system includes a vision detection module, a laser ranging module, a fusion calculation module, an AI recognition and decision-making module, and a control feedback module. Both the vision detection module and the laser ranging module establish bidirectional communication connections with the fusion calculation module to match the timing of image and distance information acquisition. This application effectively solves the problems of insufficient accuracy and instability in traditional technologies. By constructing a collaborative detection structure through the vision detection module and the laser ranging module, a triangulation subsystem of laser and vision detection modules is formed, completely overcoming the inaccuracy problem of traditional probes on highly reflective surfaces or magnetic plates. Simultaneously, it avoids the interference of environmental temperature, humidity, and material properties on measurement accuracy, significantly improving the stability and accuracy of the detection.

Description

Technical Field

[0001] This invention relates to the field of additive manufacturing technology, specifically to a system and method for leveling and monitoring the printing quality of a 3D printing platform based on vision and laser fusion detection. Background Technology

[0002] In the field of additive manufacturing, FDM 3D printers have been widely used in production and R&D scenarios across multiple industries due to their advantages of controllable cost and ease of operation. Currently, platform leveling and print quality monitoring of FDM 3D printers mainly rely on a variety of traditional technologies, including mechanical contact detection, capacitive ranging, laser single-point ranging, traditional probe (such as BLTouch) detection, and purely visual solutions. Among these, purely visual solutions often achieve basic detection functions by recognizing the texture of the printing bed using a camera. These existing technologies provide support for the basic operation of 3D printing.

[0003] Existing technologies suffer from numerous prominent problems and struggle to meet the demands of complex scenarios. Manual leveling is time-consuming and labor-intensive, with extremely poor repeatability, severely impacting printing efficiency and consistency. Traditional probes are prone to positioning inaccuracies on highly reflective surfaces or magnetic plates, exhibiting insufficient stability. Mechanical contact and capacitive ranging methods suffer from slow response times, and their measurement accuracy is easily affected by environmental temperature, humidity, and the characteristics of the printing materials. Laser single-point ranging offers limited functionality and cannot achieve multi-dimensional detection. Purely visual solutions are highly dependent on lighting conditions and material reflectivity, exhibiting poor robustness and a tendency to misjudge. Furthermore, in multi-nozzle, multi-material, or multi-color printing scenarios, frequent first-layer switching makes it difficult for existing technologies to simultaneously address issues such as poor adhesion, abnormal extrusion volume, and uneven linewidth. They cannot monitor first-layer adhesion quality and nozzle output status in real time, nor do they possess early warning capabilities for material blockages or interruptions, only responding passively after a malfunction occurs. Currently, no publicly available technology deeply integrates laser ranging with visual AI recognition to form a composite detection solution, leaving a significant gap in high-precision, highly robust, and automated detection technologies adaptable to multiple scenarios.

[0004] Therefore, there is an urgent need for a composite detection system and method that integrates the core advantages of laser ranging and visual AI recognition. This technology should have functions such as high-precision non-contact measurement, real-time monitoring of the quality of the first layer of molten extruded material, and early warning of material blockage. At the same time, it should support multi-nozzle slide switching detection, automated self-calibration process and firmware macro linkage compensation, so as to completely solve the shortcomings of existing technologies in terms of accuracy, robustness, automation and scene adaptability, fill the technical gap of laser and visual AI fusion detection, and improve the success rate of 3D printing and product consistency. Summary of the Invention

[0005] The purpose of this invention is to provide a 3D printing platform leveling and printing quality monitoring system and method based on vision and laser fusion detection, so as to solve the problems existing in the prior art mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a 3D printing platform leveling and printing quality monitoring system based on vision and laser fusion detection, comprising a vision detection module, a laser ranging module, a fusion calculation module, an AI recognition and decision-making module, and a control feedback module;

[0007] Both the visual detection module and the laser ranging module establish a bidirectional communication connection with the fusion computing module to match the acquisition timing of image information and distance information.

[0008] The fusion computing module and the AI ​​recognition and decision-making module communicate bidirectionally.

[0009] The signal output terminal of the AI ​​recognition and decision-making module is connected to the signal input terminal of the control feedback module;

[0010] The control feedback module establishes a communication connection with the main control board of the 3D printer to obtain real-time nozzle position data and transmit adjustment commands to the main control board.

[0011] The vision detection module is used to acquire image information of the nozzle, printing platform and first layer of molten extruded material (bead); the laser ranging module is used to project a laser and measure the distance between the nozzle and the printing platform; the fusion calculation module is used to receive and process image information and distance information; the AI ​​recognition and decision-making module is used to analyze the processed data to generate control commands; and the control feedback module is used to execute control commands to achieve leveling and anomaly handling.

[0012] Preferably, the visual inspection module and the laser ranging module are arranged at a preset angle to form a collaborative inspection structure, constituting a triangulation subsystem of laser and visual inspection modules;

[0013] The laser and vision inspection module triangulation subsystem combines image processing from the vision inspection module with slide / nozzle position information to achieve high-precision measurement of the surface morphology of the printing bed (bed height map), quality inspection of the first layer of molten extruded material (bead), and nozzle / box blockage detection.

[0014] Before or during each print, the system automatically performs laser scanning or rapid point measurement through the triangulation subsystem of the laser and vision detection module, generates a height map in real time, and automatically compensates for Z deviation through firmware macros.

[0015] The system simultaneously monitors the first-layer extrusion line width and brightness through the collaborative work of the vision detection module and the laser ranging module, identifies abnormalities such as low adhesion, excessive flattening, or clogging, and can trigger alarms, pause, or switch to backup nozzles when material blockage or filament stoppage is detected.

[0016] Preferably, it also includes a self-calibration module, which is connected to the visual inspection module, the laser ranging module and the fusion calculation module respectively;

[0017] The self-calibration module uses the standard checkerboard calibration method to solve for the parameters and distortion coefficients within the visual detection module. The standard checkerboard calibration method specifically includes:

[0018] A1) Place the standard chessboard grid in different positions on the printing platform, and the vision detection module collects multiple sets of chessboard grid images;

[0019] A2) Perform corner detection and sub-pixel level precise extraction on each group of images, and establish the correspondence between the pixel coordinates and world coordinates of the checkerboard corner points;

[0020] A3) Based on Zhang Zhengyou's calibration algorithm, a calibration equation is constructed, and the parameters (including focal length and principal point coordinates) and distortion coefficients (including radial distortion coefficient and tangential distortion coefficient) of the visual detection module are solved by iterative optimization.

[0021] A4) Control the laser ranging module to project lasers to at least 3 sampling points with known Z heights, and the vision detection module to acquire laser line images of each sampling point and extract the laser line pixel coordinates on the image plane;

[0022] A5) The laser plane equation is obtained using the least squares fitting algorithm;

[0023] A6) Combining the principle of laser triangulation, the relative position calibration of the visual inspection module and the laser ranging module is completed by mapping the pixel position of the line on the image plane to the corresponding known Z height. The final calibration error does not exceed 0.05-0.1mm.

[0024] Preferably, it also includes a cloud learning module, which is communicatively connected to the AI ​​recognition and decision-making module. The cloud learning module is used to receive historical detection data and processing results uploaded by the AI ​​recognition and decision-making module, and then send the optimized AI model parameters back to the AI ​​recognition and decision-making module after training with the data.

[0025] Preferably, when adapting to a multi-nozzle FDM 3D printer, the vision detection module and the laser ranging module can acquire appearance images and measure distances to nozzles in non-working positions, and the AI ​​recognition and decision-making module can determine the nozzle status based on the acquired data and generate a list of available nozzles.

[0026] A method for leveling a 3D printing platform and monitoring printing quality in any of the aforementioned systems, characterized by comprising the following steps:

[0027] S1) Preparation before printing

[0028] The laser ranging module projects a monochromatic narrow laser line with a selectable wavelength of 650nm visible light, 405nm or 850nm infrared light to the vicinity of the print bed or nozzle. The vision inspection module captures the laser line at a fixed observation angle. Combining the known parameters within the vision inspection module with the laser plane equation, the Z height of the point is inferred from the pixel position of the line on the image plane, i.e., laser triangulation. Combined with the XY motion of the 3D printer, multi-point rapid scanning is completed on several sampling points to generate a print bed height map with a grid of 20×20 to 50×50. The Z deviation is automatically compensated by firmware macros.

[0029] S2) Platform Leveling

[0030] The AI ​​recognition and decision-making module analyzes the flatness deviation of the printing platform based on the height map, outputs the corresponding Z-axis compensation command, and the control feedback module communicates with the main control board to adjust the nozzle height step by step through the servo motor. After adjustment, the laser ranging module verifies the distance to ensure compensation accuracy.

[0031] S3) First Floor Printing Monitoring

[0032] During the actual printing of the first layer, the vision inspection module performs real-time image sampling of the area directly below or to the side of the nozzle, and combines the laser line to identify the outline of the extrusion line, measure the line width, outline height and brightness or color, and determine whether there is flattening, insufficient adhesion, overflow or material breakage.

[0033] S4) Abnormalities and Material Blockage Detection

[0034] By analyzing continuous frames around the nozzle, the system detects material blockage characteristics, such as no extruded filament pattern (no continuous material flow after filtering), sudden decrease in filament width, carbon buildup or increased dripping near the nozzle, or abnormal matching between the extrusion rate and the actual stepper motor axis steps. If an abnormality is detected in the first layer of printing, the extrusion rate or nozzle height is adjusted first. If a material blockage or filament stoppage is detected, a self-repair process of heating, back-pulling, and then extrusion is triggered first. If the self-repair fails, an alarm is triggered, printing is paused, or a backup nozzle is switched.

[0035] Preferably, step S1 further includes a data preprocessing sub-step:

[0036] S0-1) The fusion computing module performs noise reduction and background subtraction preprocessing on the images acquired by the visual inspection module;

[0037] S0-2) Combine the position information of the slide or nozzle to perform spatiotemporal alignment operation on the preprocessed image data and the distance data of the laser ranging module;

[0038] S0-3) After eliminating data delay bias, a standardized dataset is generated and transmitted to the AI ​​recognition and decision-making module.

[0039] Preferably, it also includes an additional step of monitoring the nozzle in a non-operating position:

[0040] SP1-1) The laser ranging module and the vision inspection module continuously collect images of the nozzle's appearance and distance data to the surrounding area in the non-working position;

[0041] SP1-2) The fusion calculation module extracts the characteristics of carbon deposits and dripping material at the nozzle end;

[0042] The SP1-3 AI recognition and decision-making module combines the aforementioned features to determine the risk of material blockage. If a risk exists, a preprocessing instruction is sent through the control feedback module.

[0043] Preferably, when adapting to a multi-nozzle FDM 3D printer, an additional nozzle switching step is also included:

[0044] SP2-1) Before the slide table switches the nozzle, the vision inspection module and the laser ranging module sample the image and distance of the target nozzle;

[0045] SP2-2) The fusion calculation module verifies the nozzle position and status;

[0046] After the SP2-3 AI recognition and decision-making module confirms that the nozzle is available, the control feedback module sends a slide switching command;

[0047] After the SP2-4 switch is completed, the detection benchmark is updated synchronously to ensure the accuracy of subsequent printing.

[0048] Compared with the prior art, the beneficial effects of the present invention are:

[0049] 1) This application effectively solves the problems of insufficient accuracy and lack of stability in traditional technologies. By constructing a collaborative detection structure through the vision detection module and the laser ranging module, a triangulation subsystem of laser and vision detection modules is formed. Combined with the XY motion of the printing platform, multi-point rapid scanning is achieved to generate a high-precision bed surface height map with a grid of 20×20 to 50×50. The internal parameters and distortion coefficients of the vision detection module are solved by using the standard checkerboard calibration method and Zhang Zhengyou calibration algorithm. The laser plane equation is obtained by least square fitting of at least three known Z height sampling points. The calibration error is controlled within the range of 0.05 to 0.1 mm. This completely overcomes the inaccuracy problem of traditional probes on highly reflective surfaces or magnetic plates. At the same time, it avoids the interference of environmental temperature and humidity and material properties on measurement accuracy, and significantly improves the stability and accuracy of detection.

[0050] 2) This application can be directly adapted to multi-nozzle FDM 3D printers. The visual inspection module and laser ranging module acquire appearance images and measure distances of non-working nozzles. The AI ​​recognition and decision-making module judges the nozzle status based on the acquired data and generates a list of available nozzles. Before switching nozzles, the system ensures smooth and accurate switching through processes such as image and distance sampling of the target nozzle, position and status verification, and availability confirmation. At the same time, the system monitors the line width, contour height, brightness, and color characteristics of the first layer of molten extruded material in real time, accurately identifies abnormalities such as poor adhesion, excessive flattening, and material blockage. For material blockage problems, it triggers a self-repair process of heating, back-pulling, and re-extrusion, which completely solves various printing failures caused by frequent multi-nozzle switching and meets the complex needs of multi-material and multi-color printing.

[0051] 3) This application adopts a composite solution that combines laser ranging and visual inspection, which avoids the excessive dependence of a purely visual solution on the surface texture and lighting conditions. The laser ranging module supports narrow laser line projection of multiple wavelengths such as 650nm visible light, 405nm or 850nm infrared light. Combined with the noise reduction, background subtraction preprocessing and data spatiotemporal alignment operations of the fusion calculation module, the influence of material reflection characteristics and data delay deviation is effectively eliminated, so that the system can maintain stable detection performance under different lighting and material scenarios, which greatly improves the applicability of the technology.

[0052] 4) This application completely replaces the time-consuming and laborious manual leveling by automating the pre-printing preparation, platform leveling, first-layer monitoring, and anomaly handling processes. The entire process requires no manual intervention, significantly improving printing efficiency. The integrated computing module and the main control board of the 3D printer communicate and link together, realizing automatic Z-deviation compensation by firmware macros, further ensuring printing accuracy. At the same time, the cloud learning module continuously receives historical detection data and anomaly handling results, optimizes AI model parameters through data training, and sends them back to the AI ​​recognition and decision-making module, which continuously improves the system's accuracy in recognizing various abnormal states, achieving long-term iterative optimization of the technology, and extending the system's service life and adaptability. Attached Figure Description

[0053] Figure 1 This is a flowchart of the standard chessboard calibration method for this application;

[0054] Figure 2 This is a flowchart of the 3D printing platform leveling and printing quality monitoring method of this application;

[0055] Figure 3 This is a flowchart of the data preprocessing sub-steps in this application;

[0056] Figure 4 This is a flowchart of the additional steps for nozzle non-operating position monitoring in this application;

[0057] Figure 5A flowchart of the additional nozzle switching steps in this application;

[0058] Figure 6 This is a schematic diagram of the structure of this application.

[0059] In the picture:

[0060] 1. 3D printer; 2. Printing platform; 3. Nozzle; 4. Vision inspection module; 5. Laser ranging module. Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] In the description of the invention, it should be noted that the terms "upper," "lower," "inner," "outer," "front end," "rear end," "both ends," "one end," and "the other end," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0063] In the description of the invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installed," "equipped with," "connected," etc., should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0064] In the description of the invention, it should be noted that the execution order of the steps is not limited by the sequence number. The possible changes in the order of some steps, the synchronous execution of steps, and the split execution of steps are all within the scope of protection of this application.

[0065] Please see Figure 1-6 The present invention provides a technical solution: a 3D printing platform leveling and printing quality monitoring system based on vision and laser fusion detection, including a vision detection module, a laser ranging module, a fusion calculation module, an AI recognition and decision-making module and a control feedback module;

[0066] Both the visual inspection module and the laser ranging module establish a bidirectional communication connection with the fusion computing module to match the acquisition timing of image information and distance information;

[0067] The fusion computing module and the AI ​​recognition and decision-making module communicate bidirectionally.

[0068] The signal output terminal of the AI ​​recognition and decision-making module is connected to the signal input terminal of the control feedback module;

[0069] The control feedback module establishes a communication connection with the main control board of the 3D printer to obtain real-time nozzle position data and transmit adjustment commands to the main control board;

[0070] The system includes a vision inspection module for acquiring image information of the nozzle, printing platform, and first layer of molten extruded material (bead); a laser ranging module for projecting a laser and measuring the distance between the nozzle and the printing platform; a fusion calculation module for receiving and processing image and distance information; an AI recognition and decision-making module for analyzing the processed data to generate control commands; and a control feedback module for executing control commands to achieve leveling and anomaly handling.

[0071] Specifically, this application fundamentally solves the problems of loosely connected modules, communication delays, and functional fragmentation in traditional detection systems. Through a bidirectional communication design between the visual inspection module and the laser ranging module, the timing of image and distance information acquisition is matched, laying the foundation for the accuracy of subsequent data fusion. The collaboration between the fusion computing module and the AI ​​recognition and decision-making module enables intelligent transformation from raw data to control commands, avoiding the inefficiency of relying on manual analysis in traditional systems. Direct communication between the control feedback module and the main control board ensures rapid execution of adjustment commands, solving the problem of decreased printing accuracy caused by delayed command transmission in traditional systems. The entire architecture design balances modularity and integration, facilitating individual debugging and maintenance of each module while achieving functional synergy through tight communication connections, providing a stable and reliable hardware and logical foundation for the implementation of all subsequent technical features.

[0072] Reference manual attached Figure 6 The 3D printer, printing platform, and nozzles are all existing technologies. Both the vision inspection module and the laser ranging module can be mounted on the print head and can move with the print head. The fusion computing module, AI recognition and decision-making module, and control feedback module can all be located externally to the 3D printer and are not shown in the attached diagram.

[0073] Example 1: Specific configuration and connection of system modules

[0074] The visual inspection module uses a global shutter CMOS camera, equipped with an M12 fixed-focus lens and an external IR filter, with a frame rate of 30-90fps, a resolution of 5MP (1920×1080), and a pixel size of ≥2.0μm to improve the signal-to-noise ratio in low-light environments;

[0075] The laser ranging module uses a narrow-line laser projector that can emit 650nm visible light, 405nm or 850nm infrared light, with a line width of <0.2mm (when the working distance is 200mm).

[0076] The fusion computing module uses a Raspberry Pi 4 / 8 as the core processing unit and is equipped with the OpenCV image processing library;

[0077] The AI ​​recognition and decision-making module integrates convolutional neural networks (CNN) and LSTM networks for image feature recognition and abnormal trend prediction.

[0078] The control feedback module communicates with the 3D printer's main control board (Duet / Klipper) via serial port, MQTT, or HTTP protocol, and connects to the servo motor to perform Z-axis height adjustment.

[0079] In this embodiment, the module connection method is as follows: the visual detection module and the laser ranging module establish bidirectional communication with the fusion computing module through the GPIO interface, and the communication delay is controlled within 10ms to ensure the timing synchronization of image acquisition and distance measurement; the fusion computing module transmits standardized data to the AI ​​recognition and decision-making module through WebSocket, and the control commands generated by the AI ​​recognition and decision-making module are output to the control feedback module in the form of digital signals; the control feedback module directly controls the laser start / stop and alarm device through the GPIO interface, and at the same time interacts with the main control board in real time with nozzle position data and adjustment commands.

[0080] The visual inspection module and the laser ranging module are arranged at a preset angle to form a collaborative inspection structure, constituting a triangulation subsystem of the laser and visual inspection modules;

[0081] The laser and vision inspection module triangulation subsystem combines image processing from the vision inspection module with slide / nozzle position information to achieve high-precision measurement of the surface morphology of the printing bed (bed height map), quality inspection of the first layer of molten extruded material (bead), and nozzle / box blockage detection.

[0082] Before or during each print, the system automatically performs laser scanning or rapid point measurement through the triangulation subsystem of the laser and vision inspection module, generates a height map in real time, and automatically compensates for Z deviation through firmware macros.

[0083] The system simultaneously monitors the line width and brightness of the first extrusion layer through the collaborative work of the vision inspection module and the laser ranging module, identifies abnormalities such as low adhesion, excessive flattening or clogging, and can trigger alarms, pause or switch to backup nozzles when material blockage or filament stoppage is detected.

[0084] Specifically, by constructing a triangulation subsystem for laser and vision inspection modules, combined with slide / nozzle position information and firmware macro linkage, the problems of insufficient accuracy, limited functionality, and delayed response to anomalies in traditional inspection technologies are solved. The angle design of the triangulation subsystem breaks through the field-of-view limitations of traditional laser single-point ranging, enabling comprehensive measurement of the print bed surface morphology. The generated high-precision height map provides accurate data support for Z-deviation compensation, completely solving the inaccuracy problem of traditional probes on highly reflective surfaces or magnetic plates. The first-layer quality inspection function, through dual monitoring of line width and brightness, achieves accurate identification of various anomalies such as low adhesion and excessive flattening, filling the technical gap of traditional systems' inability to monitor the first-layer status in real time. The rapid identification and automatic handling process for material blockage and filament stoppage avoids printing failures caused by the escalation of faults. Especially in multi-nozzle printers, the backup nozzle switching function significantly improves the continuity of multi-material and multi-color printing. This application integrates leveling, monitoring, and anomaly handling, achieving end-to-end assurance from pre-printing to printing, significantly improving printing accuracy and success rate.

[0085] Example 2: Layout and Workflow of the Triangulation Subsystem

[0086] The vision inspection module and the laser ranging module are rigidly fixed to the upper part of the printing cavity by a bracket, forming a triangular angle of 25-35° (this angle range balances measurement accuracy and field of view coverage, avoiding depth insensitivity due to an excessively small angle or occlusion due to an excessively large angle). Before printing, the system controls the slide to move the nozzle along the XY direction via the main control board. The laser ranging module continuously projects a narrow laser line onto the printing bed surface, and the vision inspection module simultaneously acquires laser line images at various positions. Combining the real-time position information of the slide, the system calculates the Z height of each sampling point using the laser triangulation principle, generating a bed surface height map with a grid of 20×20 to 50×50. The fusion calculation module converts the height map data into a format recognizable by firmware macros, and the main control board executes macro instructions to automatically compensate for Z deviation, with a compensation step accuracy of 0.01mm.

[0087] During the first layer printing process, the vision inspection module focuses on the area directly below and to the side of the nozzle, acquiring extruded part images in real time. Combined with the contour features formed by laser line projection, it extracts parameters such as line width and brightness. The AI ​​recognition and decision-making module compares the acquired parameters with preset thresholds (line width threshold 0.2-0.8mm, brightness threshold adjusted according to material characteristics). When the line width is less than the lower limit or the brightness is lower than the threshold, it is determined to be a low adhesion or under-extrusion anomaly. When the line width exceeds the upper limit, it is determined to be an over-flattening or over-extrusion anomaly. If three consecutive frames show no extruded filament shape, the filament width suddenly decreases by more than 50%, or the carbon buildup / dripping characteristics near the nozzle continuously increase, it is determined to be a material blockage or filament stoppage anomaly. The system immediately triggers an alarm, and the control feedback module sends a pause command to the main control board. If it is a multi-nozzle printer, it automatically switches to the backup nozzle and performs a cleaning operation.

[0088] Comparative Example 1: Performance Comparison between Traditional Probe Detection and This System

[0089] Using the same FDM 3D printer, print bed leveling and first-layer quality monitoring tests were conducted using both a traditional BLTouch probe (Comparative Example 1) and this system (Example 2). The test materials were three common consumables: PLA, ABS, and PETG. Test bed surfaces included ordinary glass, highly reflective metal, and magnetic adsorption. Each scenario was printed 10 times. The test results are as follows:

[0090] Leveling accuracy: The leveling error of the comparative model on the high reflectivity metal bed and magnetic adsorption bed is 0.15-0.3mm, while the leveling error of this system is controlled within 0.05-0.1mm on all bed surfaces, which is more than 50% higher.

[0091] Anomaly recognition rate: The comparative model cannot recognize first-layer quality anomalies such as low adhesion and excessive flattening, and the material blockage recognition lag time is 3-5 seconds. This system has a recognition rate of over 98% for various anomalies in the first layer, and the material blockage recognition lag time is ≤0.5 seconds.

[0092] Printing success rate: The average success rate of the comparative model in multi-material printing is 72%, while the average success rate of this system is improved to over 95%.

[0093] According to another aspect of this application, it also includes a self-calibration module, which is connected to the visual inspection module, the laser ranging module and the fusion calculation module respectively;

[0094] The self-calibration module uses the standard checkerboard calibration method to solve for the intrinsic parameters and distortion coefficients of the visual inspection module. The standard checkerboard calibration method specifically includes:

[0095] A1) Place the standard chessboard grid in different positions on the printing platform, and the vision detection module collects multiple sets of chessboard grid images;

[0096] A2) Perform corner detection and sub-pixel level precise extraction on each group of images, and establish the correspondence between the pixel coordinates and world coordinates of the checkerboard corner points;

[0097] A3) Based on Zhang Zhengyou's calibration algorithm, a calibration equation is constructed, and the parameters (including focal length and principal point coordinates) and distortion coefficients (including radial distortion coefficient and tangential distortion coefficient) of the visual detection module are solved by iterative optimization.

[0098] A4) Control the laser ranging module to project lasers to at least 3 sampling points with known Z heights, and the vision detection module to acquire laser line images of each sampling point and extract the laser line pixel coordinates on the image plane;

[0099] A5) The laser plane equation is obtained using the least squares fitting algorithm;

[0100] A6) Combining the principle of laser triangulation, the relative position calibration of the visual inspection module and the laser ranging module is completed by mapping the pixel position of the line on the image plane to the corresponding known Z height. The final calibration error does not exceed 0.05-0.1mm.

[0101] Specifically, by adding a self-calibration module and defining a standardized calibration algorithm process, the problems of complex calibration, easy accuracy drift, and reliance on manual operation in traditional detection systems are solved. The standard checkerboard calibration method combined with Zhang Zhengyou's algorithm can accurately solve the intrinsic parameters and distortion coefficients of the visual inspection module, effectively eliminating the impact of lens distortion on image acquisition and providing a precise parameter basis for subsequent laser line pixel coordinate extraction and 3D reconstruction. Laser plane fitting with at least three known Z-height sampling points ensures the accuracy of the laser plane equation and avoids fitting errors caused by single sampling points. Relative position calibration between the visual inspection module and the laser ranging module ensures the measurement accuracy of the triangulation subsystem, enabling the system to maintain stable detection performance during long-term use. The entire self-calibration process requires no professional technical personnel; the system can automatically complete image acquisition, parameter solving, and calibration adjustment, reducing the system's maintenance difficulty and usage threshold. The calibration error is controlled within a high-precision range of 0.05-0.1mm, providing reliable accuracy assurance for functions such as print bed leveling and first-layer quality inspection.

[0102] Example 3: Specific calibration procedure for the self-calibration module

[0103] Step A1: Select a standard 12×9 checkerboard grid (square side length 10mm), fix it on the printing platform, and move the checkerboard grid in the XY plane using a slide table. At different Z heights (0mm, 1mm, 2mm, 3mm, 4mm) and different rotation angles (0°, 15°, 30°, -15°, -30°), the visual inspection module collects a total of 20 sets of checkerboard grid images to ensure that the images cover the entire camera field of view.

[0104] Step A2: Use OpenCV's findChessboardCorners function to detect corners in each group of images, and use the cornerSubPix function to extract them at the sub-pixel level with an extraction accuracy of 0.1 pixels. Establish the correspondence between the world coordinates (with the top left corner of the chessboard as the origin and the side length of the square as the unit length) and pixel coordinates of the chessboard corners, and remove mismatched corner data (corner points with a matching error of more than 0.5 pixels).

[0105] Step A3: Based on Zhang Zhengyou's calibration algorithm, construct a calibration equation containing intrinsic parameters, distortion coefficients, and extrinsic parameters. Solve the equation using the Levenberg-Marquardt iterative optimization algorithm to obtain the intrinsic parameters (focal length fx = 1000 ± 50 pixels, fy = 1000 ± 50 pixels, principal point coordinates cx = 960 ± 10 pixels, cy = 540 ± 10 pixels) and distortion coefficients (radial distortion coefficients k1 = -0.3 ± 0.05, k2 = 0.1 ± 0.03, tangential distortion coefficients p1 = 0.001 ± 0.0005, p2 = 0.0005 ± 0.0002).

[0106] Step A4: Control the printing platform to move precisely to three known height positions Z=0mm, Z=2mm, and Z=4mm via the main control board. The laser ranging module projects laser lines to each position, and the vision detection module acquires the corresponding laser line images and extracts the pixel coordinates (u,v) of the laser lines on the image plane.

[0107] Step A5: Based on the camera intrinsic parameters and the laser line pixel coordinates, the least squares fitting algorithm is used to construct the laser plane equation Ax+By+Cz+D=0, and the fitting error is controlled within 0.01mm.

[0108] Step A6: Based on the principle of laser triangulation, the pixel coordinates of the laser line on the image plane are reversed to the three-dimensional coordinates in space. By mapping the coordinates with the known Z-height, the relative position parameters of the visual detection module and the laser ranging module are adjusted to achieve accurate calibration of the two. After calibration, the Z-height measurement error at any sampling point does not exceed 0.08mm.

[0109] According to another aspect of this application, it also includes a cloud learning module, which is communicatively connected to the AI ​​recognition and decision-making module. The cloud learning module is used to receive historical detection data and processing results uploaded by the AI ​​recognition and decision-making module, and after training and optimizing the AI ​​model parameters through the data, it is sent back to the AI ​​recognition and decision-making module.

[0110] Specifically, by adding a cloud-based learning module, the system's detection performance has been continuously iteratively optimized, solving the problem of traditional detection systems having fixed algorithms and being unable to adapt to diverse scenarios. The cloud-based learning module can accumulate a large amount of historical data under different materials, lighting conditions, and bed types. Through data training, it continuously optimizes the AI ​​model parameters, enabling the AI ​​recognition and decision-making module to continuously improve the accuracy of identifying various abnormal states. Especially for printing scenarios with new materials or complex lighting conditions, it can quickly adapt and maintain high detection accuracy. This closed-loop mechanism of "data accumulation - model optimization - performance improvement" gives the system self-learning capabilities, avoiding the tedious manual adjustment of algorithm parameters required in traditional systems. At the same time, the historical data stored in the cloud can provide data support for subsequent technical upgrades and functional expansions, extending the system's technical lifecycle. Furthermore, the communication connection between the cloud-based learning module and the AI ​​recognition and decision-making module uses encrypted transmission, ensuring data security and privacy, balancing practicality and security.

[0111] Example 4: Workflow of the cloud-based learning module

[0112] After each print job, the AI ​​recognition and decision-making module automatically packages and uploads data such as printing material type, lighting conditions, bed type, detected anomaly type, anomaly feature parameters (such as line width deviation and brightness change rate), processing results, and printing success rate to the cloud learning module. The cloud learning module uses the TensorFlow framework to build a training model, dividing the uploaded data into a training set (70%), a validation set (20%), and a test set (10%). With anomaly recognition accuracy and processing success rate as optimization objectives, the module iteratively updates the weight parameters of the CNN and LSTM networks using the gradient descent algorithm.

[0113] For example, when the system first used PETG material to print under strong light conditions, the accuracy rate for identifying low adhesion anomalies was 85%. With the accumulation of data from similar scenarios (over 1000 sets of uploaded data), the cloud-based learning module trained and optimized the model parameters, increasing the accuracy rate for identifying low adhesion anomalies in this scenario to over 97%. The optimized model parameters were then transmitted back to the AI ​​recognition and decision-making module via the network, which automatically updated its local algorithm parameters, achieving continuous improvement in recognition performance without manual intervention.

[0114] When adapting to a multi-nozzle FDM 3D printer, this application allows the vision inspection module and laser ranging module to acquire appearance images and measure distances to nozzles in non-working positions, while the AI ​​recognition and decision-making module can determine the nozzle status based on the acquired data and generate a list of available nozzles.

[0115] Specifically, by acquiring images and measuring distances of nozzles in non-working positions using a visual inspection module and a laser ranging module, nozzle blockages and damage can be identified in advance, preventing print failures due to nozzle malfunctions. The generation of an available nozzle list provides a fast and accurate basis for nozzle switching decisions, significantly reducing response time. In multi-material printing, frequent nozzle switching can lead to poor adhesion and extrusion anomalies. This technology ensures a smooth and accurate switching process through pre-switching status confirmation and post-switching baseline updates, effectively solving the problem of unstable print quality caused by frequent multi-nozzle switching. This technical feature allows the system to fully adapt to the diverse needs of 1-7 head FDM 3D printers, significantly expanding the system's application scenarios and providing reliable assurance for large-scale multi-material, multi-color printing applications.

[0116] Example 5: Specific Implementation of Multi-Nozzle Adaptation

[0117] A 7-head FDM 3D printer was selected, with each nozzle evenly arranged along the slide. Non-working nozzles were positioned at preset locations on the side of the printing chamber. During the system's power-on self-test, the vision inspection module and the laser ranging module sequentially inspected each non-working nozzle: the vision inspection module acquired images of the nozzle end face and surrounding area, while the laser ranging module measured the distance between the nozzle end face and the print bed surface; the fusion calculation module extracted features such as nozzle carbon buildup, material dripping, and nozzle end face flatness from the images, and combined them with the distance measurement data to determine whether the nozzles were clogged, damaged, or misaligned; the AI ​​recognition and decision-making module, based on the feature analysis results, marked nozzles in normal condition as usable, generated a list of usable nozzles containing information such as nozzle number, condition assessment, and compatible materials, and stored it on the main control board.

[0118] During multi-material printing, when the system detects that the current working nozzle is clogged and the self-repair fails, the AI ​​recognition and decision-making module selects a backup nozzle that is suitable for the current material from the list of available nozzles. The control feedback module sends a nozzle switching command to the main control board, and the slide moves quickly to the backup nozzle position. At the same time, the vision detection module and the laser ranging module perform a secondary status confirmation on the backup nozzle to ensure that the switched nozzle can work normally.

[0119] According to another aspect of this application, a method for leveling a 3D printing platform and monitoring printing quality based on the above system is also provided, comprising the following steps:

[0120] S1) Preparation before printing

[0121] The laser ranging module projects a monochromatic narrow laser line with a selectable wavelength of 650nm visible light, 405nm or 850nm infrared light to the vicinity of the print bed or nozzle. The vision inspection module captures the laser line at a fixed observation angle. Combining the known parameters within the vision inspection module with the laser plane equation, the Z height of the point is inferred from the pixel position of the line on the image plane, i.e., laser triangulation. Combined with the XY motion of the 3D printer, multi-point rapid scanning is completed on several sampling points to generate a print bed height map with a grid of 20×20 to 50×50. The Z deviation is automatically compensated by firmware macros.

[0122] S2) Platform Leveling

[0123] The AI ​​recognition and decision-making module analyzes the flatness deviation of the printing platform based on the height map, outputs the corresponding Z-axis compensation command, and the control feedback module communicates with the main control board to adjust the nozzle height step by step through the servo motor. After adjustment, the laser ranging module verifies the distance to ensure compensation accuracy.

[0124] S3) First Floor Printing Monitoring

[0125] During the actual printing of the first layer, the vision inspection module performs real-time image sampling of the area directly below or to the side of the nozzle, and combines the laser line to identify the outline of the extrusion line, measure the line width, outline height and brightness or color, and determine whether there is flattening, insufficient adhesion, overflow or material breakage.

[0126] S4) Abnormalities and Material Blockage Detection

[0127] By analyzing continuous frames around the nozzle, the system detects material blockage characteristics, such as no extruded filament pattern (no continuous material flow after filtering), sudden decrease in filament width, carbon buildup or increased dripping near the nozzle, or abnormal matching between the extrusion rate and the actual stepper motor axis steps. If an abnormality is detected in the first layer of printing, the extrusion rate or nozzle height is adjusted first. If a material blockage or filament stoppage is detected, a self-repair process of heating, back-pulling, and then extrusion is triggered first. If the self-repair fails, an alarm is triggered, printing is paused, or a backup nozzle is switched.

[0128] Specifically, in the pre-printing preparation step, laser triangulation and multi-point rapid scanning generate a high-precision bed height map and achieve automatic Z-deviation compensation, completely replacing time-consuming and laborious manual leveling, significantly improving leveling efficiency and accuracy. The step-by-step adjustment and verification mechanism of the platform leveling step ensures the reliability of compensation accuracy and avoids over-compensation or under-compensation problems caused by one-time adjustments. The first-layer printing monitoring step achieves comprehensive monitoring of the extruded body status through real-time image sampling and multi-parameter measurement, which can quickly identify various anomalies such as flattening and insufficient adhesion, and can be dealt with in a timely manner. The graded handling mechanism of the anomaly and material blockage detection step, from parameter adjustment to self-repair, and then to alarm / nozzle switching, takes targeted measures for different anomalies, which not only avoids printing interruptions caused by minor anomalies, but also prevents printing failures caused by serious malfunctions, significantly improving printing continuity and success rate.

[0129] Example 6: Complete Execution Flow of Method Steps

[0130] Taking the printing process of PLA material on a 3-head FDM 3D printer as an example, the execution of the method steps is explained in detail:

[0131] S1) Pre-printing preparation: Select a 650nm visible light narrow laser line, the laser ranging module projects the laser onto the print bed surface, and the vision inspection module captures the laser line at a frame rate of 30fps; combining the parameters within the vision inspection module calibrated according to claim 3 with the laser plane equation, the Z height is inferred from the image plane pixel coordinates; the slide moves in a serpentine motion along the XY direction at a speed of 50mm / s to complete a multi-point rapid scan of a 40×40 grid (a total of 1600 sampling points), generating a print bed height map; the fusion calculation module converts the height map data into firmware macro instructions, and the main control board executes the instructions to automatically compensate for the Z deviation, with a compensation range of -0.2mm to +0.2mm.

[0132] S2) Platform Leveling: The AI ​​recognition and decision-making module analyzes the height map and finds that the Z-axis height deviation in the center area of ​​the print bed is +0.15mm and the deviation in the edge area is -0.1mm. It outputs the corresponding Z-axis compensation command. The control feedback module communicates with the main control board and adjusts the nozzle height step by step through the servo motor, adjusting by 0.02mm each step. After adjustment, the laser ranging module checks each sampling point and confirms that the Z-axis height deviation of all points is controlled within ±0.05mm, thus completing the leveling.

[0133] S3) First Layer Printing Monitoring: The first layer printing speed is set to 30mm / s. The vision detection module acquires images of the extruded material directly below the nozzle in real time, with each frame acquired in 33ms. Combined with laser line recognition of the extrusion line contour, the measured line width is 0.4mm (preset threshold 0.3-0.5mm), the contour height is 0.2mm, and the brightness value is 180 (preset threshold 150-220), which is considered normal. When printing to the 10th layer, the line width is detected to drop to 0.25mm and the brightness value drops to 140, which is considered an abnormality due to insufficient adhesion.

[0134] S4) Anomaly and Material Blockage Detection: For the insufficient adhesion anomaly, the control feedback module adjusts the extrusion rate from 5mm³ / s to 6mm³ / s, reduces the nozzle height by 0.03mm, and after adjustment, the detection linewidth recovers to 0.4mm, the brightness value recovers to 175, and the anomaly is resolved; when printing continues to the 50th layer, the system detects three consecutive frames without extruded filaments, the filament width drops sharply from 0.4mm to 0, the carbon buildup near the nozzle is enhanced, and the matching error between the extrusion rate and the actual stepper motor axis exceeds 30%, which is determined to be a material blockage anomaly; the system immediately triggers the self-repair process, increases the nozzle temperature from 200℃ to 208℃, back-pulls 3mm, and then extrudes 8mm. After self-repair, if there are still no extruded filaments, the system performs a switch to a backup nozzle operation and sends an alarm message to the user terminal.

[0135] Step S1 of this application also includes a data preprocessing sub-step:

[0136] S0-1) The fusion computing module performs noise reduction and background subtraction preprocessing on the images acquired by the visual inspection module;

[0137] S0-2) Combine the position information of the slide or nozzle to perform spatiotemporal alignment operation on the preprocessed image data and the distance data of the laser ranging module;

[0138] S0-3) After eliminating data delay bias, a standardized dataset is generated and transmitted to the AI ​​recognition and decision-making module.

[0139] Specifically, by adding a data preprocessing sub-step to the pre-printing preparation process, the problem of decreased detection accuracy caused by noise interference, spatiotemporal asynchrony, and inconsistent formats in the original data is solved. Image denoising and background subtraction preprocessing effectively eliminate environmental noise and background interference, highlighting the core features of the laser line and the extruded body, providing a clear data foundation for subsequent pixel coordinate extraction and contour recognition. Spatiotemporal alignment operations, combined with slide / nozzle position information, solve the problem of acquisition delay between image data and distance data, ensuring the consistency of data at the same sampling point and avoiding Z-height calculation errors caused by data asynchrony. The generation of standardized datasets unifies and standardizes the data format, reducing the data analysis difficulty of the AI ​​recognition and decision-making modules and improving data processing efficiency. The entire preprocessing process significantly improves data quality, providing accurate and reliable data support for subsequent height map generation, anomaly recognition, and other steps, further ensuring the detection accuracy and stability of the system. Especially in complex lighting or high-noise environments, it can effectively improve the system's anti-interference ability, making the detection results more accurate.

[0140] Example 7: Specific Operations of Data Preprocessing

[0141] The S0-1) fusion computing module uses a 5×5 Gaussian filter to denoise the laser line image acquired by the visual inspection module. The Gaussian standard deviation is set to 1.5, which effectively eliminates random noise in the image. The frame difference method is used for background subtraction. The blank bed surface image acquired before printing is used as the background reference to remove background interference and highlight the contour features of the laser line and the extruded body, thereby improving the contrast of the laser line pixels by more than 30%.

[0142] S0-2) The real-time position data of the slide / nozzle is obtained through the main control board (the sampling frequency is the same as the image acquisition frequency, both are 30Hz), and the acquisition timestamp of each image frame and the corresponding slide / nozzle position coordinates are recorded; the distance data of the laser ranging module is matched with the image data according to the timestamp, and the data with delay (delay time exceeding 5ms) is interpolated and corrected to ensure that the image data and distance data of the same sampling point correspond one-to-one in time and space.

[0143] S0-3) Convert the spatiotemporally aligned image data into an 8-bit grayscale image, convert the distance data into floating-point values, and generate a standardized dataset in the format of "sampling point coordinates-image features-distance value". The dataset is transmitted to the AI ​​recognition and decision-making module through a data compression algorithm (compression ratio of 3:1), with the transmission delay controlled within 10ms to ensure that the AI ​​module can receive and analyze the data in real time.

[0144] According to another aspect of this application, an additional step of monitoring the nozzle in a non-operating position is also included:

[0145] SP1-1) The laser ranging module and the vision inspection module continuously collect images of the nozzle's appearance and distance data to the surrounding area in the non-working position;

[0146] SP1-2) The fusion calculation module extracts the characteristics of carbon deposits and dripping material at the nozzle end;

[0147] The SP1-3 AI recognition and decision-making module combines features to determine the risk of material blockage. If a risk exists, a preprocessing instruction is sent through the control feedback module.

[0148] Specifically, by adding an additional step of nozzle non-working position monitoring, preventative maintenance of non-working nozzles is achieved, solving the problem of traditional systems focusing only on working nozzles and neglecting the status of non-working nozzles, which can lead to potential malfunctions. Continuous appearance image acquisition and distance measurement can capture abnormal features such as carbon buildup, dripping, and positional deviation of non-working nozzles in real time, identifying the risk of material blockage in advance and avoiding printing failures caused by hidden faults after nozzle switching. The feature extraction algorithm of the fusion computing module can accurately separate abnormal areas, providing a reliable basis for risk assessment. The risk judgment and preprocessing command sending of the AI ​​recognition and decision module enables early intervention in faults. Residual material in the nozzle can be removed through simple heating, back-pulling, and re-extrusion operations, reducing the probability of material blockage without disassembling the nozzle for manual cleaning, greatly improving maintenance efficiency. This step expands the system's monitoring range from the working position to all nozzles, forming a full-nozzle coverage monitoring system, further improving the printing stability and continuity of multi-nozzle printers, and reducing the user's maintenance costs and operational difficulty.

[0149] Example 8: Specific Implementation of Nozzle Non-Operating Position Monitoring

[0150] SP1-1) The sampling frequency for non-working nozzle monitoring is set to 1Hz (data is collected once every 1 second). The laser ranging module measures the distance between the nozzle end face and the preset reference surface with a measurement accuracy of ±0.01mm. The vision inspection module collects appearance images of the nozzle end face and the surrounding 5mm range with an image resolution of 1920×1080 to ensure that fine features such as carbon deposits and dripping are clearly captured.

[0151] The SP1-2 fusion calculation module performs grayscale processing on the acquired images, uses the Canny edge detection algorithm to extract the nozzle end face contour, and uses a threshold segmentation algorithm (the threshold is dynamically adjusted according to the material properties) to separate the carbon deposit and dripping areas from the nozzle body, and calculates the area and proportion of the carbon deposit and dripping areas; at the same time, it combines distance data to determine whether there is a positional offset of the nozzle.

[0152] The SP1-3 AI recognition and decision-making module sets the carbon deposit area ratio threshold to 5%, the droplet diameter threshold to 0.3mm, and the position offset threshold to 0.1mm. When the carbon deposit area ratio of a non-working nozzle reaches 6% and the droplet diameter is 0.4mm, it is judged as a high risk of material blockage. The AI ​​recognition and decision-making module sends a preprocessing instruction to the main control board through the control feedback module. The main control board controls the nozzle to heat up to 200℃ (the temperature suitable for PLA material), back-pull 2mm, and then extrude 3mm to remove the residual molten material in the nozzle and reduce the probability of material blockage in subsequent use.

[0153] When adapting to multi-nozzle FDM 3D printers, this application also includes an additional nozzle switching step:

[0154] SP2-1) Before the slide table switches the nozzle, the vision inspection module and the laser ranging module sample the image and distance of the target nozzle;

[0155] SP2-2) The fusion calculation module verifies the nozzle position and status;

[0156] After the SP2-3 AI recognition and decision-making module confirms that the nozzle is available, the control feedback module sends a slide switching command;

[0157] After the SP2-4 switch is completed, the detection benchmark is updated synchronously to ensure the accuracy of subsequent printing.

[0158] Specifically, this application provides a standardized and high-precision switching process for nozzle switching scenarios in multi-nozzle FDM 3D printers, solving the problems of inaccurate nozzle positioning, unknown status, and inconsistent references that lead to decreased print quality in traditional nozzle switching. Target nozzle sampling and status verification before switching ensure accurate nozzle position and normal status after switching, avoiding print interruptions caused by switching to a faulty nozzle. The availability confirmation of the AI ​​recognition and decision-making module provides reliable assurance for sending switching commands, reducing the risk of erroneous switching. Post-switching detection reference updates enable the system to quickly adapt to the characteristics and working status of the new nozzle, solving the problem of mismatched detection references caused by individual differences between different nozzles, ensuring the adhesion quality and extrusion accuracy of the first layer printed after switching. The entire switching process is fully automated from sampling, verification, switching to reference updates, with a switching time controlled within 2 seconds. This significantly improves the efficiency and continuity of multi-nozzle printing, significantly ensuring the quality stability of multi-material and multi-color printing, and further expanding the system's application scenarios and practical value.

[0159] Example 9: Specific process for switching between multiple nozzles

[0160] Taking the switching of a 3-head FDM 3D printer from nozzle 1 (compatible with PLA material) to nozzle 2 (compatible with ABS material) as an example:

[0161] SP2-1) The slide moves to the non-working position of nozzle 2. The vision inspection module collects the end face image of nozzle 2 and the surrounding area image. The laser ranging module measures the distance between the end face of nozzle 2 and the printing bed surface and the relative position with the slide reference point. The sampling data includes nozzle end face flatness, carbon deposit area, dripping state, distance value, position coordinates, etc.

[0162] The SP2-2 fusion calculation module compares the collected distance value with the preset standard position (the calibration position of nozzle 2). The deviation is 0.03mm (threshold 0.1mm), and the position is judged to be normal. The image features are analyzed, and the carbon deposit area accounts for 2% (threshold 5%). There is no obvious dripping, and the status is judged to be normal.

[0163] The SP2-3 AI recognition and decision-making module, combined with the verification results of the fusion computing module, confirms that nozzle 2 is available and generates a switching permission command; the control feedback module sends a slide switching command to the main control board, which includes parameters such as the target nozzle number, switching speed, and positioning accuracy.

[0164] The SP2-4 slide moves to the working position of nozzle 2 at a speed of 100mm / s, with a positioning accuracy of ±0.02mm. After the switch is completed, the vision inspection module and the laser ranging module re-acquire the image and distance data of the working position of nozzle 2. The fusion calculation module updates the detection benchmark (including Z-axis benchmark, extrusion feature benchmark, etc.), and the AI ​​recognition and decision module updates the anomaly judgment threshold (the line width threshold for ABS material is 0.35-0.55mm, and the brightness threshold is 160-230) to ensure the accuracy of subsequent printing inspection.

[0165] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A 3D printing platform leveling and printing quality monitoring system based on vision and laser fusion detection, characterized in that, It includes a visual inspection module, a laser ranging module, a fusion computing module, an AI recognition and decision-making module, and a control feedback module; Both the visual detection module and the laser ranging module establish a bidirectional communication connection with the fusion computing module to match the acquisition timing of image information and distance information. The fusion computing module and the AI ​​recognition and decision-making module communicate bidirectionally. The signal output terminal of the AI ​​recognition and decision-making module is connected to the signal input terminal of the control feedback module; The control feedback module establishes a communication connection with the main control board of the 3D printer to obtain real-time nozzle position data and transmit adjustment commands to the main control board. The vision detection module is used to acquire image information of the nozzle, printing platform and first layer of molten extruded material; the laser ranging module is used to project laser and measure the distance between the nozzle and the printing platform; the fusion calculation module is used to receive and process image information and distance information; the AI ​​recognition and decision-making module is used to analyze the processed data to generate control commands; and the control feedback module is used to execute control commands to achieve leveling and anomaly handling.

2. The 3D printing platform leveling and printing quality monitoring system based on vision and laser fusion detection according to claim 1, characterized in that, The visual inspection module and the laser ranging module are arranged at a preset angle to form a collaborative detection structure, constituting a triangulation subsystem of laser and visual inspection modules; The laser and vision inspection module triangulation subsystem combines image processing from the vision inspection module with slide / nozzle position information to achieve high-precision measurement of the surface morphology of the printing bed, quality inspection of the first layer of molten extruded material, and nozzle / box blockage detection. Before or during each print, the system automatically performs laser scanning or rapid point measurement through the triangulation subsystem of the laser and vision detection module, generates a height map in real time, and automatically compensates for Z deviation through firmware macros. The system simultaneously monitors the line width and brightness of the first extrusion layer through the collaborative work of the vision detection module and the laser ranging module, identifies abnormalities such as low adhesion, excessive flattening, or clogging, and triggers an alarm, pauses, or switches to a backup nozzle when material blockage or filament stoppage is detected.

3. The 3D printing platform leveling and printing quality monitoring system based on vision and laser fusion detection according to claim 1, characterized in that, It also includes a self-calibration module, which is connected to the visual inspection module, the laser ranging module and the fusion calculation module respectively; The self-calibration module uses the standard checkerboard calibration method to solve for the parameters and distortion coefficients within the visual detection module. The standard checkerboard calibration method specifically includes: A1) Place the standard chessboard grid in different positions on the printing platform, and the vision detection module collects multiple sets of chessboard grid images; A2) Perform corner detection and sub-pixel level precise extraction on each group of images, and establish the correspondence between the pixel coordinates and world coordinates of the checkerboard corner points; A3) Based on Zhang Zhengyou's calibration algorithm, a calibration equation is constructed, and the parameters and distortion coefficients of the visual detection module are solved by iterative optimization. A4) Control the laser ranging module to project lasers to at least 3 sampling points with known Z heights, and the vision detection module to acquire laser line images of each sampling point and extract the laser line pixel coordinates on the image plane; A5) The laser plane equation is obtained using the least squares fitting algorithm; A6) Combining the principle of laser triangulation, the relative position calibration of the visual inspection module and the laser ranging module is completed by mapping the pixel position of the line on the image plane to the corresponding known Z height.

4. The 3D printing platform leveling and printing quality monitoring system based on vision and laser fusion detection according to claim 1, characterized in that, It also includes a cloud learning module, which is connected to the AI ​​recognition and decision-making module. The cloud learning module receives historical detection data and processing results uploaded by the AI ​​recognition and decision-making module, optimizes the AI ​​model parameters through data training, and then sends them back to the AI ​​recognition and decision-making module.

5. The 3D printing platform leveling and printing quality monitoring system based on vision and laser fusion detection according to claim 1, characterized in that, When adapting to a multi-nozzle FDM 3D printer, the vision inspection module and the laser ranging module can acquire appearance images and measure distances to nozzles in non-working positions. The AI ​​recognition and decision-making module determines the nozzle status based on the acquired data and generates a list of available nozzles.

6. A method for leveling a 3D printing platform and monitoring printing quality based on the system described in any one of claims 1-5, characterized in that, Includes the following steps: S1) Preparation before printing The laser ranging module projects a monochromatic narrow laser line onto the print bed surface or near the nozzle. The vision inspection module takes a picture of the laser line at a fixed observation angle. Combining the known parameters within the vision inspection module with the laser plane equation, the Z height of the point is inferred from the pixel position of the line on the image plane, i.e., laser triangulation. In conjunction with the XY motion of the 3D printer, multi-point rapid scanning is completed on several sampling points to generate a print bed height map with a grid of 20×20 to 50×50. The Z deviation is automatically compensated by firmware macro. S2) Platform Leveling The AI ​​recognition and decision-making module analyzes the flatness deviation of the printing platform based on the height map, outputs the corresponding Z-axis compensation command, and the control feedback module communicates with the main control board to adjust the nozzle height step by step through the servo motor. After adjustment, the laser ranging module verifies the distance to ensure compensation accuracy. S3) First Floor Printing Monitoring During the actual printing of the first layer, the vision inspection module performs real-time image sampling of the area directly below or to the side of the nozzle, and combines the laser line to identify the outline of the extrusion line, measure the line width, outline height and brightness or color, and determine whether there is flattening, insufficient adhesion, overflow or material breakage. S4) Abnormalities and Material Blockage Detection By analyzing continuous frames around the nozzle, the system detects material blockage characteristics, such as no extruded filament pattern (no continuous material flow after filtering), sudden decrease in filament width, carbon buildup or increased dripping near the nozzle, or abnormal matching between the extrusion rate and the actual stepper motor axis steps. If an abnormality is detected in the first layer of printing, the extrusion rate or nozzle height is adjusted first. If a material blockage or filament stoppage is detected, a self-repair process of heating, back-pulling, and then extrusion is triggered first. If the self-repair fails, an alarm is triggered, printing is paused, or a backup nozzle is switched.

7. The method for leveling and monitoring the printing quality of a 3D printing platform based on vision and laser fusion detection according to claim 6, characterized in that, Step S1 also includes a data preprocessing sub-step: S0-1) The fusion computing module performs noise reduction and background subtraction preprocessing on the images acquired by the visual inspection module; S0-2) Combine the position information of the slide or nozzle to perform spatiotemporal alignment operation on the preprocessed image data and the distance data of the laser ranging module; S0-3) After eliminating data delay bias, a standardized dataset is generated and transmitted to the AI ​​recognition and decision-making module.

8. The method for leveling and monitoring the printing quality of a 3D printing platform based on vision and laser fusion detection according to claim 6, characterized in that, It also includes an additional step for monitoring nozzles in non-operating positions: SP1-1) The laser ranging module and the vision inspection module continuously collect images of the nozzle's appearance and distance data to the surrounding area in the non-working position; SP1-2) The fusion calculation module extracts the characteristics of carbon deposits and dripping material at the nozzle end; The SP1-3 AI recognition and decision-making module combines the aforementioned features to determine the risk of material blockage. If a risk exists, a preprocessing instruction is sent through the control feedback module.

9. The method for leveling and monitoring the printing quality of a 3D printing platform based on vision and laser fusion detection according to claim 6, characterized in that, When adapting to multi-nozzle FDM 3D printers, an additional nozzle switching step is also included: SP2-1) Before the slide table switches the nozzle, the vision inspection module and the laser ranging module sample the image and distance of the target nozzle; SP2-2) The fusion calculation module verifies the nozzle position and status; After the SP2-3 AI recognition and decision-making module confirms that the nozzle is available, the control feedback module sends a slide switching command; After the SP2-4 switch is completed, the detection benchmark is updated synchronously to ensure the accuracy of subsequent printing.

Images