Real-time printing defect detection method and system based on artificial intelligence

Abstract

The invention provides a real-time 3D printing defect detection method and system based on artificial intelligence, and aims to solve the problems of dependence on single visual information, lack of dynamic tracking and insufficient model generalization ability in the prior art. According to the method, the printing process is comprehensively perceived through multi-modal data fusion, and defects are analyzed and classified in real time by using a deep learning model. The model output comprises defect category, position, abnormal score and severity evaluation, and the defect development trend can be dynamically tracked and early warning can be carried out in advance. Through comparison with a printed digital model, the system can accurately identify geometric deviation, and early intervention is realized in combination with trend analysis. When serious defects are detected, the system automatically triggers response measures, such as printing pause, parameter adjustment or correction scheme provision, to form a complete quality control closed loop. According to the method, the detection precision and robustness are remarkably improved, and incremental learning is supported to adapt to different printing environments.

Description

Technical Field

[0001] The present invention belongs to the field of 3D printing, and in particular, relates to a real-time printing defect detection method and system based on artificial intelligence. Background Art

[0002] Traditional 3D printing defect monitoring relies primarily on manual observation and a small number of sensors, which presents numerous shortcomings. Traditionally, operators often observe the printing process through surveillance cameras or remotely view video feeds, but the human eye struggles to maintain stable monitoring over extended periods and detect subtle changes. Simple sensors built into consumer printers (such as the end-of-filament switch) can only detect obvious faults, but are unable to effectively identify the numerous subtle quality issues that accumulate layer by layer. Early automated detection methods based on image processing required manual feature extraction and were often limited to identifying defects of specific shapes. This was cumbersome and lacked generalizability.

[0003] To address these issues, existing patents and products have introduced artificial intelligence into 3D printing defect monitoring. Among them, Chinese patent CN202110337787A proposes a real-time online monitoring method based on a neural network. This method utilizes an industrial camera mounted above the printer workbench to capture a top-view image, converting it into a grayscale image. This image is then fed into a pre-trained convolutional neural network defect segmentation model for pixel-level classification. Each pixel in the image is classified as background, normal, or one of several preset defect categories (such as blistering, collapse, wrinkling, and missing threads). The defect area is then calculated, and an alarm is triggered if a preset threshold is exceeded. This method leverages deep learning to automatically extract features, overcoming the limitations of manual feature extraction and achieving a certain level of real-time monitoring effectiveness.

[0004] However, this intelligent monitoring solution lacks dynamic tracking of defect evolution. Currently, it primarily issues alerts based on defect area exceeding limits in a single frame of image, failing to provide early warning of gradual, cumulative development or subtle changes.

[0005] Visual information alone is not comprehensive. Although some faults may not be visually apparent in the early stages, they may be accompanied by abnormal sounds, vibrations, or temperature changes. Relying solely on image monitoring will reduce the robustness of detection.

[0006] The model's generalization ability is limited. The distribution of training data determines the model's adaptability. Changing the material color, printer model, or camera angle may reduce the model's detection accuracy. The model also lacks automatic calibration and online learning capabilities. Summary of the Invention

[0007] The technical problem to be solved by the present invention is: in response to the problems existing in the prior art, the present invention provides a real-time printing defect detection method and system based on artificial intelligence.

[0008] The technical solution of the present invention is: In a first aspect, a method for real-time printing defect detection based on artificial intelligence is provided, comprising the following steps: Step S1: Acquire a real-time image sequence of a 3D printing process, and simultaneously acquire sound, infrared, and vibration data of the 3D printing process, and combine the real-time image sequence, sound, infrared, and vibration data into multimodal data; Step S2: pre-processing the multimodal data collected in step S2; aligning and fusing the multimodal data collected from different sensors in time and space to form a multi-channel feature input synchronized with the printing process; Step S3: Inputting the preprocessed multimodal data into a pre-trained artificial intelligence defect detection model, the artificial intelligence defect detection model including at least one deep neural network for real-time analysis of the state during the printing process; the model output of the artificial intelligence defect detection model includes characterization information of the current printing state, including defect segmentation results and / or defect category determination of the printing area, and abnormality assessment indicators; based on the model output, determining whether a printing defect exists and the defect type; if a defect is detected, obtaining its position and area feature information in the current layer or current frame; if no defect is detected, continuing to cyclically monitor subsequent printing data; Step S4: For the defects detected in step S3, the severity of the defects is evaluated by integrating the multi-frame timing information and the printed digital model data; Step S5: Compare the severity of the defect evaluated in step S4 with a preset threshold. When the severity of the defect exceeds the safety threshold, immediately trigger a predetermined response measure. Step S6: After the printing job is completed, the defects detected during the entire printing process and the processing decision log are saved. If parameter adjustment corrective measures are taken during the printing process, the relevant data is fed back to the training module of the artificial intelligence defect detection model to perform incremental training or update the artificial intelligence defect detection model.

[0009] Specifically, the evaluation of the severity of the defect in step S4 includes: calculating the growth rate or cumulative impact based on the development trend of the defect in the continuous layer, judging the functional importance of the physical part corresponding to the defect in the finished product based on the G-Code or digital model provided by the printer, and quantifying the defect severity score based on the defect area ratio and shape characteristics.

[0010] Specifically, the predetermined response measures in step S5 include: issuing an alarm to the operator to prompt the specific defect type and location, automatically pausing the printing job, and providing a correction plan or executing a pre-set parameter adjustment strategy based on the defect type. Otherwise, the alarm will not be triggered and the subsequent printing process will continue to be monitored. After the defect response measures are executed, the system returns to step S1 to continue monitoring until the printing job is completed.

[0011] Preferably, the artificial intelligence defect detection model is a multi-task neural network architecture, which simultaneously outputs multiple judgment results related to printing quality: its first output is the category and location of the printing defect, and the second output is the score or probability of whether there is an abnormality in the printing process; the predefined defect categories include at least drawing, warping, layer shift dislocation, under-extrusion, over-extrusion, and uneven surface roughness, and the model gives an abnormality score for unknown types of abnormal situations while identifying any of the above defects.

[0012] Preferably, the defect determination in step S3 includes introducing a reference comparison of the printed digital model, obtaining target cross-sectional contour data corresponding to the current printed layer, and comparing the reference contour with the contour in the actual printed image; if it is found that the actual printed shape deviates from the reference model, it is determined that a printing defect exists, and the geometric error is quantified by calculating the offset distance and overlap between the reference contour and the contour in the live image, and this is used as one of the bases for defect identification.

[0013] Preferably, the reference model comparison process includes: extracting edge features of the printed layer in the printed image, and matching them with the theoretical contour of the layer to detect displacement errors; when it is found that the actual printing height or position of the local area is lower than the theoretical model or higher than the theoretical model, and the actual boundary exceeds the theoretical range, it is determined as a corresponding defect and its position is recorded.

[0014] Preferably, the defect severity assessment in step S4 includes an analysis of the trend of defect development over printing time; for the defects identified in step S3, their area or characteristic changes are continuously monitored in each subsequent layer. When the area of ​​the defect at the same position gradually expands for n consecutive layers, or any defect indicator shows a continuous upward trend and the increase rate exceeds a preset threshold, even if the current layer does not exceed the absolute alarm threshold, it is determined in advance that the defect has a serious trend, thereby triggering an early warning.

[0015] Preferably, the response measures in step S5 include: automatically pausing the printer and locking the current state, sending an alarm message and recommended processing measures to the user terminal; when the defect type and cause are identifiable, the system provides a corresponding correction plan, which includes adjusting the extruder temperature, reducing the printing speed, increasing the cooling fan speed or modifying the printing path of the current layer. If the defect level is lower than the first set threshold and is correctable, the system performs a predetermined process parameter adjustment after pausing and then continues printing to try to repair the defect or prevent the defect from further deteriorating; if the defect severity is higher than the set threshold and cannot continue, the printing job is terminated.

[0016] Preferably, the target cross-sectional profile data is derived from printer G-Code instructions or original CAD model slice data.

[0017] In a second aspect, a real-time printing defect detection system based on artificial intelligence is provided, the system comprising: Multimodal data acquisition module, used to obtain real-time image sequences of the 3D printing process, and simultaneously obtain sound, infrared and vibration data of the 3D printing process to form multimodal data; A multimodal data preprocessing module is used to preprocess the multimodal data collected in step S1 and align and fuse the multimodal data collected from different sensors in time and space to form a multi-channel feature input synchronized with the printing process; A defect detection module is configured to input preprocessed multimodal data into a pretrained artificial intelligence defect detection model, the artificial intelligence defect detection model comprising at least one deep neural network for real-time analysis of the state of the printing process; the model output of the artificial intelligence defect detection model includes characterization information of the current printing state, including defect segmentation results and / or defect category determination of the printed area, and an abnormality assessment index; based on the model output, the presence and type of printing defects are determined; if a defect is detected, its position and area feature information in the current layer or current frame are obtained; if no defect is detected, subsequent printing data is continuously monitored in a loop; An evaluation module is used to evaluate the severity of defects detected by the defect detection module by integrating multi-frame timing information and printed digital model data; The measure response module is used to compare the severity of the defect assessed by the assessment module with the preset threshold. When the severity of the defect exceeds the safety threshold, the predetermined response measure is immediately triggered; The parameter optimization module is used to save the defects detected during the entire printing process and the processing decision log after the printing job is completed. If parameter adjustment corrective measures are taken during the printing process, the relevant data will be fed back to the training module of the artificial intelligence defect detection model to perform incremental training or update the artificial intelligence defect detection model.

[0018] Compared to existing technologies, this invention offers the following advantages: It transcends the traditional alarm system, which relies solely on alarms based on defect area exceeding limits in a single image frame. By utilizing continuous time-series data tracking, trend extrapolation, anomaly score accumulation, and multimodal data fusion, it can dynamically capture defects that accumulate layer by layer during the printing process, including subtle changes in the early stages. This mechanism not only allows the system to detect the growth trend of defects as they first appear, but also provides timely warnings to operators and recommends corrective measures to prevent further deterioration, significantly improving print quality and stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments or exemplary embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or exemplary descriptions. Obviously, the drawings described below are only some embodiments of the present application and should not be regarded as limiting the scope. For ordinary technicians in this field, other drawings can be obtained according to the drawings without paying creative work.

[0020] Figure 1 This is an overall flow chart of Example 1 of the present invention; Figure 2 This is the main flow chart of Example 1 of the present invention; Figure 3 This is a flowchart of step S6 in Example 1 of the present invention; Figure 4 This is a system block diagram of embodiment 2 of the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Generally, the components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.

[0022] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application for protection, but merely represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0023] Example 1: refer to Figure 1 and Figure 3 , a real-time printing defect detection method based on artificial intelligence, comprising the following steps: Step S1: Acquire a real-time image sequence of the 3D printing process, and simultaneously acquire sound, infrared, and vibration data of the 3D printing process, and combine the real-time image sequence, sound, infrared, and vibration data into multimodal data.

[0024] Before the printer begins printing, the operator loads the STL or G-Code file of the model to be printed into the monitoring system's software interface. The system preprocesses the model, extracting the contour information of each layer for reference. Once printing begins, a conventional camera and an infrared camera capture images of the print area at a rate of approximately two frames per second. Due to the slow printing speed (each layer takes approximately tens of seconds), 2 FPS ensures that at least multiple images are captured for each layer. Simultaneously, a microphone continuously records the sound stream, which the system divides into short audio segments corresponding to conventional image frames and infrared graphics frames. As printing progresses, the data acquisition module continuously provides image frames and sound segments, adding time stamps to align the three.

[0025] This step expands the scope of defect detection by increasing the collection and utilization of acoustic data. Many printing faults are accompanied by unique acoustic signatures, such as the rhythmic "clicking" sound produced by an extruder jam, and unusual noise and vibration caused by loose machine components. By integrating the audio captured by the microphone into the model through spectral analysis, the system can detect anomalies that are difficult to detect using images alone. Compared to approaches relying solely on vision, integrating acoustic signals significantly improves sensitivity to dynamic faults such as poor feeding and vibration. Once an abnormal acoustic pattern is detected, the system can issue an early warning before a defect forms on the surface, enhancing the timeliness and reliability of detection. This multimodal fusion approach fully utilizes microphone sensor information readily available from existing equipment and offers strong practical applicability. Infrared data also expands the utilization of infrared thermal imaging. Abnormal temperature fields during the printing process often indicate quality issues. For example, excessive cooling in a localized area can lead to warping and cracking, while excessive temperatures can cause overmelting, collapse, or scorching and discoloration. By incorporating infrared images, this method can monitor the temperature distribution of various parts of the printed part in real time and correlate thermal anomalies with geometric anomalies. For example, if an infrared image shows a temperature in a specific area that is significantly lower than the surrounding area and continues to drop, it can be inferred that warping may occur in that area (due to detachment from the heated bed). Alternatively, poor interlayer bonding can lead to localized overheating or undercooling spots. This information is difficult to obtain with visible light images alone. By fusing thermal imaging with visible light, the system can more comprehensively perceive defects, especially providing a basis for predicting early-stage warping and cracking, thereby improving monitoring accuracy. Furthermore, the recent decrease in the cost of infrared sensing hardware has made it affordable for small consumer printers, making this improvement practically feasible.

[0026] Step S2: pre-process the multimodal data collected in step S1, and align and fuse the multimodal data collected from different sensors in time and space to form a multi-channel feature input synchronized with the printing process.

[0027] For each new image, the system first performs distortion correction and denoising (using pre-calibrated camera parameters to eliminate lens distortion). The image is then scaled to the required input size for the neural network (e.g., 512 x 512 pixels). For sound data, the system calculates its power spectral density over the corresponding time window and extracts features such as the Mel-frequency cepstral coefficients to produce a feature vector representing the sound pattern at that moment. At this point, the system has normalized both the visual and sound data. Next, the system fuses these two features: specifically, the sound features are added as an additional channel, spliced ​​into a specific position in the image feature tensor. In other words, a multi-channel input is constructed, where the first three channels contain the RGB (or grayscale) image information, and the fourth channel contains a matrix mapped to the same size as the image based on the sound features (for example, by mapping the main frequency bands of sound energy distribution to uniform values ​​on the image plane, or by repeatedly padding the sound features to form a feature map of the same size as the image). This simple fusion method is efficient in implementation, allowing the subsequent neural network to "see" both image and sound information simultaneously. In this embodiment, the fourth channel is the infrared intensity, the fifth channel is the sound feature mode, and so on. The fused multi-channel tensor is then sent to the defect detection module.

[0028] Step S3: Input the preprocessed multimodal data into a pre-trained artificial intelligence defect detection model, which includes at least one deep neural network for real-time analysis of the status during the printing process; the model output result of the artificial intelligence defect detection model includes characterization information of the current printing status, and the characterization information includes the defect segmentation result and / or defect category determination of the printing area, as well as an abnormality degree assessment index; based on the model output result, determine whether there is a printing defect and the defect type; if a defect is detected, obtain its position and area feature information in the current layer or current frame; if no defect is detected, continue to cyclically monitor subsequent printing data.

[0029] The defect detection module runs a convolutional neural network model. During its development phase, the model was trained using a large amount of printing process data, including both normal printing and various defect scenarios. Some of this training data came from public datasets, while others were obtained by repeatedly printing "defective test pieces" on our experimental printer. These test pieces had their parameters intentionally adjusted to produce typical defects such as stringing, warping, and under-extrusion, providing the model with labeled learning samples. The model employs a U-Net architecture combined with a multi-head classifier: the encoder extracts deep features from the fused image, the decoder outputs defect segmentation results pixel by pixel, and a global classification head outputs defect class probabilities. The model also includes a parallel, small multi-layer perceptron to predict possible fault cause parameters based on the encoder features (corresponding to the implementation of claim 6). Upon receiving new multi-channel data, the model of this embodiment immediately performs forward inference and produces an output. During the first few layers of printing, the system detected no abnormalities, and the model output was "no defect," with the anomaly score remaining low.

[0030] For example, in the case of a stringing defect, printing continues until the model begins printing two vertical columns, requiring the filament to be pulled across the gap. Around the 50th layer, the camera feed shows subtle stringing between the columns. Our model successfully segments and marks these stringing areas in this frame, classifying the defect as "stringing." Specifically, the segmentation output identifies a non-zero defect class at the location of the stringing pixel between the columns, assigning the class label "stringing." The global classification head's probability for the "stringing" class increases to 0.85, while the probabilities for the other defect categories remain low. Simultaneously, the anomaly score increases slightly, but remains at a moderate level. The model's predicted cause output now indicates "possibly due to insufficient wire withdrawal distance or excessive nozzle temperature." The system therefore records the currently detected defect as 1 - Type: Stringing, Location: Between the two columns at the top of the model, and Cause: Possibly insufficient wire withdrawal.

[0031] For example, a warping defect occurs near layer 100. With the support structures at the four bottom corners complete and the main model rapidly accumulating height, the camera image shows signs of a corner of the model slightly lifting off the print bed. In the image of layer 105, a shadow and gap appear where the left front bottom corner of the model meets the print bed. This is detected as an abnormal area in the segmentation output. The model classifies this as a "warp" defect with a probability of approximately 0.9. The corresponding anomaly score also increases significantly, as warping often indicates serious quality issues. The system therefore records Defect 2: Type: Warp, Location: The left front bottom edge of the model, Area: Approximately 20 square millimeters (calculated by converting pixel count to actual size). At this point, the global anomaly score is approaching the warning level, as warping could cause misalignment in subsequent layers or even part detachment. The model's cause output indicates "possibly insufficient build plate temperature or low ambient temperature." This analysis is consistent with the actual situation: It is assumed that excessive cooling air flow in the printing environment caused the left front corner to overcool, leading to warping.

[0032] For under-extrusion defects, for example: when printing is nearing completion (about the 180th layer), the upper part of the camera image shows that the surface texture of a layer on the top of the model is abnormal: gaps appear between several adjacent lines, and the filling is incomplete. This phenomenon is identified by the model as an "under-extrusion" defect (i.e., lack of material) with a probability of 0.8. At the same time, the microphone recorded intermittent "clicking" sounds from the extruder when printing this area, indicating that the gears are slipping and the wire feeding is not smooth. The model that integrates the sound signal therefore increases the anomaly score and strengthens the confidence in the judgment of under-extrusion. The system records defect 3 - Type: local under-extrusion, Location: local area of ​​the top layer, Area: approximately 5 square millimeters. The cause analysis module prompts: "It may be due to slight blockage of the nozzle or slippage of the filament."

[0033] By designing a multi-task output, the model not only identifies specific defect categories but also provides an overall anomaly score. This allows for both classification of known defects and alerting for unknown anomalies. This design, while covering a wide range of typical defects (such as stringing, warping, layer shift, under-extrusion, over-extrusion, and surface defects) common in consumer 3D printing, ensures the system can monitor issues of general concern to users. Furthermore, the introduction of an anomaly scoring mechanism ensures sensitivity to uncommon defects (new defects the model has never seen) to avoid complete misses. This provides greater flexibility than the rigid, fixed-category judgments of existing technologies. Furthermore, pre-defining multiple defect output categories facilitates user explanation of the specific problem in alerts, enhancing user-friendly human-computer interaction.

[0034] Step S4: For the defects detected in step S3, the severity of the defects is evaluated by integrating the multi-frame timing information and the printed digital model data; The severity assessment strategy we set is to observe the growth trend of defects and the importance of their location.

[0035] For example, the system calculated the length and number of wires at the 50th layer and continued to monitor the same area with each subsequent layer. It found that during the printing of layers 51 and 52, thin wires still appeared at this location, with similar lengths and no significant increase in size. Based on the model's digital reference, there was no design entity between the columns, so these wires did not affect the design geometry. The severity assessment module comprehensively determined that while wires existed, their size was within an acceptable range and had not increased. The impact on the final model was primarily aesthetic, with no functional impact. Therefore, the severity was classified as "minor." This level is below the warning threshold, so the system does not trigger a pause; instead, the wire defect is recorded and marked for subsequent processing. The system highlights the wired area between the columns in the 3D model view in the user interface and indicates that this is a minor defect and does not require interruption. Simultaneously, the model's cause analysis recommends checking the wire retraction settings. The operator is prompted to adjust the parameters (for example, increasing the wire retraction distance by 1mm) to reduce wires in subsequent prints. However, this example assumes that the operator does not intervene, and printing continues.

[0036] For warping defects, for example: the system initiates a strict evaluation strategy for warping defects. First, it compares with the reference model: it is found that the corner should be flat against the platform, but the actual height deviates by about 2 mm. Secondly, the trend is checked: from the 100th layer to the 105th layer, the image shows that the warping area is gradually expanding, and the gap grows from almost invisible to clearly visible, indicating that the warping is rapidly worsening. If no intervention is made, subsequent layers may be dislocated due to overhang, or the nozzle may hit the warped part. Based on these factors, the severity is rated as "serious", far exceeding the threshold and requiring immediate attention.

[0037] For under-extrusion defects, for example, for this under-extrusion defect of the top layer, the system evaluates its impact. Since the printing is almost completed, only a small amount of voids appear in the last few layers, which has little effect on the overall structural strength, but the surface quality is damaged. The model reference comparison shows that the area should be solidly filled, but there are actually about 20% voids. In terms of trend, the under-extrusion has just appeared in 1-2 layers and has not deteriorated further. However, judging from the sound of the extruder, the blockage may worsen. If more parts are printed, this blockage may completely interrupt the discharge. Taking all factors into consideration, the system rates the severity as "medium". Although it will not result in scrapping, measures need to be taken to ensure that subsequent printing tasks do not have more serious problems.

[0038] The present invention uses digital models as "standard answers" to accurately point out the location and extent of printing errors. For example, if the actual printed outline of a certain layer is one piece less than the theoretical model, it immediately indicates that a material shortage or layer drop has occurred; if the actual outline is offset outward, over-extruded burrs or filaments may appear. Through this layer-by-layer comparison, not only can the geometric deviations of the model be detected, but the type of deviation can also be determined. This is more deterministic than relying solely on image feature classification and reduces false positives and missed reports. For example, in the case of complex surface textures on parts, it may be difficult to determine whether details are missing by directly looking at the image model, but with the help of a digital model reference, "parts that should not exist or should not be missing" can be clearly found. Therefore, this improvement significantly improves the accuracy and range of defect detection, and is particularly important for monitoring dimensional accuracy and shape integrity, and expands the application scenarios of the system (such as quality assurance for the printing of functional components with high precision requirements).

[0039] Step S5: Compare the severity of the defect evaluated in step S4 with a preset threshold. When the severity of the defect exceeds the safety threshold, a predetermined response measure is immediately triggered.

[0040] Step S6: After the printing job is completed, the defects detected during the entire printing process and the processing decision log are saved. If parameter adjustment corrective measures are taken during the printing process, the relevant data is fed back to the training module of the artificial intelligence defect detection model to perform incremental training or update the artificial intelligence defect detection model.

[0041] The system stores the monitoring data for this print (including images, sensor data, three detected defects, and treatment measures) in a log database. Researchers can choose to use this new data for model retraining. For example, the detected "wiredrawing" image and its final processing mark (minor, no need to stop) can be used to enhance the model's ability to discern the severity of wiredrawing; the model can be trained with "warping"-related data to capture signs of warping earlier; and the model can be supervised using sound and image data from "under-extrusion" to more accurately correlate sound features with feeding anomalies. As data from more printing tasks accumulates, the model will continue to optimize, and its adaptability to various types of equipment and materials will also increase. This adaptive optimization mechanism reflects the continuous improvement capabilities of the present invention, enabling it to gradually learn to identify unknown new printing faults, significantly surpassing traditional solutions that require manual rule adjustments.

[0042] Through this approach, the present invention transcends the traditional reliance on single-frame imagery for alarms based on excessive defect area. By employing a combination of techniques, including continuous time-series data tracking, trend extrapolation, anomaly score accumulation, and multimodal data fusion, the system can dynamically capture defects that accumulate layer by layer during the printing process, including subtle changes in their early stages. This mechanism not only allows the system to detect the growth trend of defects as they first appear, but also provides timely warnings to operators and recommends corrective measures to prevent further deterioration, significantly improving print quality and stability.

[0043] Specifically, the evaluation of the severity of the defect in step S4 includes: calculating the growth rate or cumulative impact based on the development trend of the defect in the continuous layer, judging the functional importance of the physical part corresponding to the defect in the finished product based on the G-Code or digital model provided by the printer, and quantifying the defect severity score based on the defect area ratio and shape characteristics.

[0044] Specifically, the predetermined response measures in step S5 include: issuing an alarm to the operator to prompt the specific defect type and location, automatically pausing the printing job, and providing a correction plan or executing a pre-set parameter adjustment strategy based on the defect type. Otherwise, the alarm will not be triggered and the subsequent printing process will continue to be monitored. After the defect response measures are executed, the system returns to step S1 to continue monitoring until the printing job is completed.

[0045] Preferably, the artificial intelligence defect detection model is a multi-task neural network architecture, which simultaneously outputs multiple judgment results related to printing quality; its first output is the category and location of the printing defect, and the second output is the score or probability of whether there is an abnormality in the printing process; the predefined defect categories include at least drawing, warping, layer shift dislocation, under-extrusion, over-extrusion, and uneven surface roughness, and the model gives an abnormality score for unknown types of abnormal situations while identifying any of the above defects.

[0046] Preferably, the defect determination in step S3 includes introducing a reference comparison of the printed digital model, obtaining target cross-sectional contour data corresponding to the current printed layer, and comparing the reference contour with the contour in the actual printed image; if it is found that the actual printed shape deviates from the reference model, it is determined that a printing defect exists, and the geometric error is quantified by calculating the offset distance and overlap between the reference contour and the contour in the live image, and this is used as one of the bases for defect identification.

[0047] By extracting the deviation between image edges and theoretical contours, this invention can accurately measure layer-to-layer positional errors (misalignment) and height errors (missing layers or protrusions). For example, if the height of an area in the actual print is detected to be lower than the expected value, it can be determined that a collapse or layer omission has occurred; if the height exceeds the expected value, it indicates a local protrusion or excessive material ejection. Similarly, if the actual boundary exceeds the designed boundary, it indicates problems such as tailing or flying wire adhesion. These specific rules make the system's defect judgment more clear and reliable, effectively defining identification criteria for each geometric deviation. This multi-rule fusion detection method has greater interpretability and accuracy than a pure end-to-end network, helping to reduce potential misjudgments in deep learning models. Furthermore, the method of using G-Code templates to locate deviations is universal and applicable to parts of different shapes. Because the comparison benchmark is based on each part's own digital design, it improves the system's adaptability to diverse printing tasks.

[0048] Preferably, the reference model comparison process includes: extracting edge features of the printed layer in the printed image, and matching them with the theoretical contour of the layer to detect displacement errors; when it is found that the actual printing height or position of the local area is lower than the theoretical model or higher than the theoretical model, and the actual boundary exceeds the theoretical range, it is determined as a corresponding defect and its position is recorded.

[0049] Preferably, the defect severity assessment in step S4 includes an analysis of the trend of defect development over printing time; for the defects identified in step S3, their area or characteristic changes are continuously monitored in each subsequent layer. When the area of ​​the defect at the same position gradually expands for n consecutive layers, or any defect indicator shows a continuous upward trend and the increase rate exceeds a preset threshold, even if the current layer does not exceed the absolute alarm threshold, it is determined in advance that the defect has a serious trend, thereby triggering an early warning.

[0050] By recording the area changes of the same defect in consecutive layers and the changes in the abnormal scores output by the model, the present invention can identify potential problems that are gradually worsening. For example, if a slight warping continues to expand by 10% per layer, the system can judge that it will eventually develop into a serious defect based on the growth rate before the absolute alarm threshold is reached, thereby intervening in advance. This trend extrapolation method effectively compensates for the lag of fixed threshold judgment and prevents small defects from accumulating into big problems before they are discovered. Compared with the original technology that only alarms when the defect exceeds the threshold once, this improvement provides a more sensitive "early warning" function, which can significantly reduce scrap and material waste. At the same time, combined with trend information, it can also filter out some false positive alarms caused by instantaneous noise - if a small defect is detected in a frame but does not expand subsequently, the system can determine that its impact is limited and will not trigger a stop. This time series-based analysis reflects the application of systematic thinking to quality monitoring, making monitoring decisions more robust and intelligent.

[0051] Preferably, the response measures in step S5 include: automatically pausing the printer and locking the current state, sending an alarm message and recommended processing measures to the user terminal; when the defect type and cause are identifiable, the system provides a corresponding correction plan, which includes adjusting the extruder temperature, reducing the printing speed, increasing the cooling fan speed or modifying the printing path of the current layer. If the defect level is lower than the first set threshold and is correctable, the system performs a predetermined process parameter adjustment after pausing and then continues printing to try to repair the defect or prevent the defect from further deteriorating; if the defect severity is higher than the set threshold and cannot continue, the printing job is terminated.

[0052] Preferably, the target cross-sectional profile data is derived from printer G-Code instructions or original CAD model slice data.

[0053] Through the above implementation examples, the workflow and effects of the present invention have been fully verified. Without changing the existing printer structure, the present invention successfully monitors and handles a variety of defects in the printing process by using simple additional equipment such as cameras and microphones and intelligent algorithms. From slight wire drawing defects to severe warping deformation, the system has made timely and accurate identification and response. Compared with manual visual observation or single sensor alarms, the intelligent monitoring system significantly improves the comprehensiveness and timeliness of defect discovery; and the automatic pause and parameter adjustment functions automate the steps that used to rely on manual intervention, reducing waste and labor costs. This real-time printing defect detection method based on artificial intelligence has the advantages of a wide detection range, fast response speed, low deployment cost, and sustainable learning optimization. It is particularly suitable for large-scale promotion and application in the field of consumer-grade 3D printing. Through the present invention, ordinary users' 3D printers can also have the ability to "consciously" monitor and correct errors, thereby greatly improving the printing success rate and product quality, and providing strong support for the popularization of additive manufacturing technology.

[0054] This invention truly forms a complete closed-loop quality control loop. When a serious defect is detected, printing is automatically paused, preventing further defective products, material waste, and even equipment damage (for example, a print head striking a lifted component). Specific defect type and location information is provided to the user, facilitating quick identification of the source of the problem. While existing technologies require operators to manually adjust parameters such as temperature based on alarm checks, this solution pre-maps defect types with remedial measures, enabling automatic or semi-automatic adjustments. For example, if an under-extrusion defect is detected, the print speed can be automatically reduced or the extrusion temperature increased to compensate. If severe stringing is detected, the wire withdrawal distance can be increased or the nozzle temperature lowered. This allows problems to be corrected in many cases without manual intervention, reversing failed prints. Even when manual intervention is required (such as cleaning a clogged nozzle), timely shutdown reduces the difficulty of subsequent cleanup and reduces material waste. Overall, this improvement upgrades a traditional monitoring system to an intelligent control system, maximizing the value of defect detection information and making it particularly useful for unattended or large-scale printing scenarios.

[0055] Implementation Example 2: refer to Figure 4 , provides a real-time printing defect detection system based on artificial intelligence, the system comprising: A multimodal data acquisition module is used to acquire a real-time image sequence of the 3D printing process, and simultaneously acquire sound, infrared, and vibration data during the 3D printing process, and combine the real-time image sequence, sound, infrared, and vibration data into multimodal data; A multimodal data preprocessing module is used to preprocess the multimodal data collected in step S1 and align and fuse the multimodal data collected from different sensors in time and space to form a multi-channel feature input synchronized with the printing process; A defect detection module is configured to input preprocessed multimodal data into a pretrained artificial intelligence defect detection model, the artificial intelligence defect detection model comprising at least one deep neural network for real-time analysis of the state of the printing process; the model output of the artificial intelligence defect detection model includes characterization information of the current printing state, including defect segmentation results and / or defect category determination of the printed area, and an abnormality assessment index; based on the model output, the presence and type of printing defects are determined; if a defect is detected, its position and area feature information in the current layer or current frame are obtained; if no defect is detected, subsequent printing data is continuously monitored in a loop; An evaluation module is used to evaluate the severity of defects detected by the defect detection module by integrating multi-frame timing information and printed digital model data; The measure response module is used to compare the severity of the defect assessed by the assessment module with the preset threshold. When the severity of the defect exceeds the safety threshold, the predetermined response measure is immediately triggered; The parameter optimization module is used to save the defects detected during the entire printing process and the processing decision log after the printing job is completed. If parameter adjustment corrective measures are taken during the printing process, the relevant data will be fed back to the training module of the artificial intelligence defect detection model to perform incremental training or update the artificial intelligence defect detection model.

[0056] The above-described embodiments merely represent preferred embodiments of the present invention. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous modifications, improvements, and substitutions without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention.

Claims

1. A real-time printing defect detection method based on artificial intelligence, characterized in that: The following steps are involved: Step S1: Acquire a real-time image sequence of a 3D printing process, and simultaneously acquire sound, infrared, and vibration data of the 3D printing process, and combine the real-time image sequence, sound, infrared, and vibration data into multimodal data; Step S2: preprocessing the multimodal data collected in step S1, and aligning and fusing the multimodal data collected from different sensors in time and space to form a multi-channel feature input synchronized with the printing process; Step S3: Inputting the preprocessed multimodal data into a pre-trained artificial intelligence defect detection model, the artificial intelligence defect detection model including at least one deep neural network for real-time analysis of the printing process status; the model output of the artificial intelligence defect detection model includes characterization information of the current printing status, including defect segmentation results and / or defect classification determination of the printed area, and abnormality degree assessment indicators; Based on the model output, determine whether there are printing defects and the type of defects. If a defect is detected, obtain its position and area feature information in the current layer or current frame. If no defects are detected, the subsequent printing data will continue to be monitored in a loop; Step S4: For the defects detected in step S3, the severity of the defects is evaluated by integrating the multi-frame timing information and the printed digital model data; Step S5: Compare the severity of the defect evaluated in step S4 with a preset threshold. When the severity of the defect exceeds the safety threshold, immediately trigger a predetermined response measure. Step S6: After the printing job is completed, the defects detected during the entire printing process and the processing decision log are saved. If parameter adjustment corrective measures are taken during the printing process, the relevant data is fed back to the training module of the artificial intelligence defect detection model to perform incremental training or update the artificial intelligence defect detection model.

2. The real-time printing defect detection method based on artificial intelligence according to claim 1, characterized in that: The evaluation of the severity of the defect in step S4 includes: The growth rate or cumulative impact is calculated based on the development trend of the defect in consecutive layers. The functional importance of the physical part corresponding to the defect in the finished product is determined based on the G-Code or digital model provided by the printer. The defect severity score is quantified based on the defect area ratio and shape characteristics.

3. The real-time printing defect detection method based on artificial intelligence according to claim 1, characterized in that: The predetermined response measures in step S5 include: An alarm is issued to the operator to prompt the specific defect type and location, the printing job is automatically suspended, and a correction plan is provided or a pre-set parameter adjustment strategy is executed based on the defect type. Otherwise, the alarm is not triggered and the subsequent printing process continues to be monitored. After the defect response measures are executed, the system returns to step S1 to continue monitoring until the printing job is completed.

4. The method for real-time printing defect detection based on artificial intelligence according to claim 1, characterized in that: The artificial intelligence defect detection model is a multi-task neural network architecture that simultaneously outputs multiple judgment results related to print quality; Its first output is the category and location of the printing defect, and the second output is the score or probability of whether there is an abnormality in the printing process; the predefined defect categories include at least drawing, warping, layer shift dislocation, under-extrusion, over-extrusion, and uneven surface roughness. When the model identifies any of the above defects, it also gives an abnormality score for unknown types of abnormal situations.

5. The real-time printing defect detection method based on artificial intelligence according to claim 4 is characterized in that: The defect determination in step S3 includes introducing a reference comparison of the printed digital model, obtaining target cross-sectional contour data corresponding to the current printed layer, and comparing the reference contour with the contour in the actual printed image; If the actual printed shape is found to deviate from the reference model, a printing defect is determined to exist. The geometric error is quantified by calculating the offset distance and overlap between the reference contour and the contour in the live image, and this is used as one of the bases for defect identification.

6. The method for real-time printing defect detection based on artificial intelligence according to claim 5, characterized in that: The reference model comparison process includes: Extract the edge features of the printed layer in the print image and match them with the theoretical contour of the layer to detect displacement errors; When it is found that the actual printing height or position of a local area is lower than or higher than the theoretical model, or the actual boundary exceeds the theoretical range, it is determined as a corresponding defect and its position is recorded.

7. The method for real-time printing defect detection based on artificial intelligence according to claim 5, characterized in that: The defect severity assessment in step S4 includes: Trend analysis of defects over printing time; For the defects identified in step S3, continuously monitor the changes in their area or characteristics in each subsequent layer; When the area of ​​defects at the same location gradually expands for n consecutive layers, or any defect indicator shows a continuous upward trend and the increase rate exceeds the preset threshold, even if the current layer does not exceed the absolute alarm threshold, the defect is judged in advance to have a serious trend, thereby triggering an early warning.

8. The method for real-time printing defect detection based on artificial intelligence according to claim 5, characterized in that: The response measures in step S5 include: Automatically pause the printer and lock the current status, send an alert message and recommended treatment measures to the user terminal; When the defect type and cause can be identified, the system provides a corresponding correction plan, which includes adjusting the extruder temperature, reducing the printing speed, increasing the cooling fan speed, or modifying the printing path of the current layer. If the defect level is lower than a first set threshold and is correctable, the system performs a predetermined process parameter adjustment after a pause and then continues printing to attempt to repair the defect or prevent it from further deteriorating. If the severity of the defect exceeds the set threshold and cannot continue, the print job is terminated. A multimodal data acquisition module is used to acquire a real-time image sequence of the 3D printing process, and simultaneously acquire sound, infrared, and vibration data during the 3D printing process, and combine the real-time image sequence, sound, infrared, and vibration data into multimodal data; A multimodal data preprocessing module is used to preprocess the multimodal data collected in step S1 and align and fuse the multimodal data collected from different sensors in time and space to form a multi-channel feature input synchronized with the printing process; A defect detection module is configured to input preprocessed multimodal data into a pretrained artificial intelligence defect detection model, the artificial intelligence defect detection model comprising at least one deep neural network for real-time analysis of the state of the printing process; the model output of the artificial intelligence defect detection model includes characterization information of the current printing state, including defect segmentation results and / or defect category determination of the printed area, and an abnormality assessment index; based on the model output, the presence and type of printing defects are determined; if a defect is detected, its position and area feature information in the current layer or current frame are obtained; if no defect is detected, subsequent printing data is continuously monitored in a loop; An evaluation module is used to evaluate the severity of defects detected by the defect detection module by integrating multi-frame timing information and printed digital model data; The measure response module is used to compare the severity of the defect assessed by the assessment module with the preset threshold. When the severity of the defect exceeds the safety threshold, the predetermined response measure is immediately triggered; The parameter optimization module is used to save the defects detected during the entire printing process and the processing decision log after the printing job is completed. If parameter adjustment corrective measures are taken during the printing process, the relevant data will be fed back to the training module of the artificial intelligence defect detection model to perform incremental training or update the artificial intelligence defect detection model.

Images