Continuous fiber 3d printing process monitoring method based on artificial intelligence image recognition

Abstract

A continuous fiber 3D printing process detection method based on artificial intelligence image recognition selects a process monitoring technology based on computer vision and pattern recognition to monitor the continuous fiber reinforced composite material printing process, acquires the continuous fiber printing image through the camera, combines the neural network classification model to realize the image recognition classification function, and further realizes the automatic detection of the printing defects in the printing process; the present application can realize the automatic monitoring of defects in the continuous fiber 3D printing process, and can lay a foundation for subsequent real-time printing control or repair, and further realize the intelligent controllable printing quality, improve the qualified rate of continuous fiber 3D printing products, reduce material waste and shorten the manufacturing time.

Description

Technical Field

[0001] This invention belongs to the field of composite material 3D printing technology, specifically relating to a method for detecting the continuous fiber 3D printing process based on artificial intelligence image recognition. Background Technology

[0002] 3D printing technology, originally known as rapid prototyping or additive manufacturing, is a bottom-up manufacturing process that uses layer-by-layer material deposition to create solid objects. Unlike traditional subtractive manufacturing processes such as machining, it significantly reduces the requirements for production molds and the limitations on design complexity. Fiber-reinforced composite material structures exhibit significantly improved mechanical properties, possessing advantages such as high strength, high modulus, low specific gravity, corrosion resistance, and good thermal stability, leading to its widespread application in various fields. However, with the development of continuous fiber-reinforced 3D printing technology, accompanying quality issues have become increasingly prominent, especially in fields such as rehabilitation medicine and aerospace, where high demands are placed on the reliability and stability of composite material 3D printed products.

[0003] In continuous fiber reinforced 3D printing, the bonding between the fiber and the matrix has a critical impact on the quality and performance of the printed parts. A series of printing defects are prone to occur during continuous fiber 3D printing, such as fiber path misalignment, fiber breakage, fiber pull-out, and matrix fracture, leading to a decline in the quality and performance of the composite 3D printed parts. Therefore, process monitoring technology for composite 3D printing is particularly important to improve printing accuracy and ensure the overall performance of the printed parts.

[0004] Existing studies have detected printing defects mostly related to material extrusion amount (Zeqing Jin, Autonomous in-situ correction of fused deposition modeling printers using computer vision and deep learning, Manufacturing Letters, Volume 22, 2019). Internationally, there is currently no research on the detection of specific fiber-matrix bonding defects in continuous fiber 3D printing, and no research on multi-camera-guided additive manufacturing process monitoring. Existing studies mainly monitor images of each layer during the 3D printing process (He Ketai, 3D Printing Process Fault Diagnosis Method and Device: CN109968671B, 2020-07-10; Wang Di, A 3D Printing Process Monitoring Method and Device Based on Real-Time Camera Capture: CN106925784A, 2017-07-07). Ordinary single-camera-guided printing process detection is also used (William Jordan Wright, In-situ optimization of thermoset composite additive manufacturing via deep learning and computer vision, Additive Manufacturing, Volume). (58, 2022.) The image information near the nozzle is incomplete.

[0005] In summary, the current lack of process monitoring and defect identification methods for continuous fiber 3D printing technology leads to quality problems such as fiber misalignment in composite material printed parts, which limits the widespread application of composite material parts in industry. Summary of the Invention

[0006] In order to overcome the shortcomings of the prior art, the present invention aims to provide a continuous fiber 3D printing process detection method based on artificial intelligence image recognition. On the one hand, it can realize automatic monitoring of defects during the continuous fiber 3D printing process, and on the other hand, it can lay the foundation for subsequent real-time printing control or repair, thereby realizing intelligent control of printing quality, improving the pass rate of continuous fiber 3D printed products, reducing material waste, and shortening manufacturing time.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] A method for detecting continuous fiber 3D printing process based on artificial intelligence image recognition is proposed. This method uses computer vision and pattern recognition technology to monitor the printing process of continuous fiber reinforced composite materials. It acquires images of continuous fiber printed parts through a camera and combines them with a neural network classification model to achieve image recognition and classification, thereby realizing the automatic detection of printing defects in the printing process.

[0009] The continuous fiber reinforced composite material includes continuous fibers as reinforcing materials and resin as a base material. The continuous fibers include carbon fibers, glass fibers, aramid fibers, flax fibers, etc.; the resins include polylactic acid (PLA), TPU, ABS, PEEK, PPS, etc.

[0010] The computer vision and pattern recognition uses dual cameras to capture real-time images of the continuous fiber reinforced 3D printing process in the horizontal and vertical directions at the nozzle. The captured image data is input into an artificial intelligence algorithm that has been trained in the early stage to achieve real-time artificial intelligence image recognition, that is, to judge the defects that occur during the printing process. The neural network classifier outputs the defect monitoring for the continuous fiber 3D printing process.

[0011] The artificial intelligence algorithms include binary classification neural network models, multi-class convolutional neural networks, K-nearest neighbor algorithms, etc.

[0012] A method for detecting continuous fiber 3D printing process based on artificial intelligence image recognition includes the following steps:

[0013] 1) Based on the structure of the continuous fiber printing platform and the defects that are expected to be monitored during the target printing process, determine the structure of the printed part used to collect the dataset and the image acquisition method for judging the printing defects.

[0014] 2) Design and build a dual-camera-guided FDM continuous fiber 3D printing platform to acquire image data during the printing process in both horizontal and vertical directions;

[0015] 3) Import the 3D model file of the printed part structure used to collect the dataset in step 1) into the 3D printer, and repeat the printing process to collect image data;

[0016] 4) By collecting images near the nozzle in the early stage, the image data is classified according to the specific circumstances of the defect and a dataset is constructed;

[0017] 5) Use the image dataset from step 4) to train the initial neural network model to improve accuracy and achieve automatic diagnosis of printing defects;

[0018] 6) The trained neural network model outputs monitoring results for defects during the continuous fiber 3D printing process.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] (1) This invention combines artificial intelligence technology with composite material 3D printing process and designs a composite material 3D printing process monitoring method based on artificial intelligence image recognition technology. It can be used to judge printing defects (e.g., whether the continuous fiber path of the reinforcement is placed in the middle of the matrix material) during the printing process, thereby improving the quality problems of composite material printed parts caused by printing defects.

[0021] (2) This invention proposes an innovative dual-camera follow-up continuous fiber reinforced 3D printing process monitoring. Currently, there is no international research on defect detection for continuous fiber reinforced 3D printing. Existing studies have detected printing defects that are mostly related to material extrusion amount. There is no international research on the detection of special fiber-matrix bonding defects in the continuous fiber 3D printing process, and there is currently no research on additive manufacturing process monitoring with multi-camera follow-up. Existing studies have monitored images of each layer in the 3D printing process, rather than the detailed image information near the nozzle in the printing process of this invention. The camera follow-up of this invention can improve the problem of insufficient timely detection of printing defects. This invention is different from ordinary single-camera follow-up printing process detection. It simultaneously collects images from two directions, which can effectively improve the shortcomings of not being able to collect complete information of the image near the nozzle. Attached Figure Description

[0022] Figure 1 A flowchart illustrating the monitoring process of camera-guided continuous fiber 3D printing in an embodiment of the present invention.

[0023] Figure 2 This is a schematic diagram illustrating whether the fiber path shifts during continuous fiber 3D printing according to an embodiment of the present invention.

[0024] Figure 3 This is a schematic diagram illustrating how the neural network model of this invention determines whether the fiber path has deviated.

[0025] Figure 4 This is a schematic diagram illustrating the training effect of the neural network model in an embodiment of the present invention. Detailed Implementation

[0026] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0027] Reference Figure 1 A method for detecting continuous fiber 3D printing process based on artificial intelligence image recognition includes the following steps:

[0028] 1) Based on the structure of the continuous fiber printing platform and the defects that are expected to be monitored during the target printing process, determine the printed part structure used to collect the dataset and the image acquisition method corresponding to the printing defects to be judged; in this embodiment, we take whether the fiber path is located in the middle of the matrix material PLA and whether there is fiber-matrix debonding as an example.

[0029] 2) To achieve vision-based monitoring of the composite material 3D printing process, a dual-camera-guided FDM continuous fiber 3D printing platform was designed and built to capture real-time images of the printing process at the nozzles in both horizontal and vertical directions.

[0030] 3) Import the 3D model file of the printed part structure used to collect the dataset in step 1) into the 3D printer, and repeat the printing process to collect image data;

[0031] To achieve real-time monitoring of fiber path deviation during FDM continuous fiber 3D printing, this embodiment first requires building a continuous fiber reinforced 3D printing platform with camera tracking, equipped with two cameras and a macro lens to acquire images of the surface condition of the printed part at close range near the nozzle; using 3D modeling software to design and model the camera bracket structure, which is then manufactured using a 3D printer; and completing the structural design of the printed part required for the dataset capture and generating 3D printing code.

[0032] The dual-camera tracking design enables the acquisition of images along the X and Y directions near the nozzle. Since it is necessary to train a neural network image classifier with artificial intelligence image recognition capabilities, it is necessary to collect image information during the continuous fiber printing process as training data in the early stage, and then train the neural network model through machine learning.

[0033] 4) By collecting images near the nozzle in the early stage, the image data is manually classified according to whether the fiber path has been deviated, and a dataset is constructed.

[0034] 5) Use the image dataset from step 4) to train the initial neural network model to improve accuracy and achieve printing defect diagnosis based on whether the fiber path has deviated.

[0035] In this embodiment, image data of the continuous fiber reinforced 3D printing process was initially acquired using Action2 with dual cameras, and the initial video data was preprocessed. The acquired images were manually judged and classified to determine whether fiber path offset occurred, referring to... Figure 2The labeled images can be divided into two categories: "GOOD" (no fiber path deviation) and "BAD" (fiber path deviation). The labeled images are then randomly divided into training and test sets in a 9:1 ratio to complete the dataset creation. The initial neural network model for image classification is trained using the training set, and the model accuracy is verified on the test set. In this embodiment, the training set uses approximately 600 photos.

[0036] 6) The trained neural network model outputs monitoring results on whether the fiber path deviates during the continuous fiber 3D printing process, providing the possibility for real-time status monitoring or subsequent defect repair functions in the composite material 3D printing process.

[0037] Reference Figure 3 In this embodiment, an 18-layer ResNet is used as the initial neural network model for training. Through machine learning, the ResNet18 model learns feature points in various types of photos, thereby enabling the classification of the printed parts as "GOOD" or "BAD" during continuous fiber 3D printing by picking features from unlabeled photos captured by the servo camera. Figure 4 As shown, the initial neural network model was trained using the dataset classified by the aforementioned fiber path offset. During the complete traversal of the dataset 10 times, the loss function showed a decreasing trend, while the accuracy showed an increasing trend, demonstrating the effectiveness of the above-mentioned training of the image recognition model for fiber path offset. After training, an average classification accuracy of 92% was achieved, proving that the trained neural network model has a good classification effect for this feature.

[0038] This invention combines artificial intelligence image recognition theory with composite material 3D printing technology to achieve vision-based identification and judgment of printing defects during continuous fiber reinforced 3D printing. This lays the foundation for real-time control or repair of printing defects in the later stages. At the same time, it provides the possibility for continuous fiber 3D printing process detection in unmanned environments. It has potential application value in fields such as aerospace and biomedicine where there are requirements for high-precision, high-performance printed parts or unsupervised printing.

Claims

1. A continuous fiber 3D printing process detection method based on artificial intelligence image recognition, characterized in that: The process monitoring technology based on computer vision and pattern recognition is selected to monitor the continuous fiber reinforced composite printing process, the continuous fiber printed part image is collected by a camera, the image recognition classification function is realized by combining a neural network classification model, and then the automatic detection of printing defects in the printing process is realized; The computer vision and pattern recognition uses a double-camera follow-up to collect the horizontal and vertical direction real-time pictures of the continuous fiber reinforced 3D printing process at the nozzle, inputs the collected image data into the pre-trained artificial intelligence algorithm, realizes real-time artificial intelligence image recognition, that is, judges the defects occurring in the printing process; The neural network classifier realizes the output of the defect monitoring in the continuous fiber 3D printing process; The method comprises the following steps: 1) According to the continuous fiber printing platform structure and the defects expected to be monitored in the target printing process, the printed part structure for collecting the data set and the image collection mode corresponding to the expected judgment of the printing defects are determined; 2) A double-camera follow-up FDM continuous fiber 3D printing platform is designed and built to realize the collection of image data in the horizontal and vertical directions during the printing process; 3) The three-dimensional model file of the printed part structure for collecting the data set in step 1) is imported into the 3D printer, and the printing process is repeated to collect image data; 4) According to the specific conditions of the defects, the image data is classified and the data set is constructed by collecting the images near the nozzle in advance; 5) The initial neural network model is trained using the image data set in step 4) to improve the accuracy and realize the automatic diagnosis of the printing defects; 6) The trained neural network model outputs the monitoring results of the defects in the continuous fiber 3D printing process.

2. The method of claim 1, wherein: The continuous fiber reinforced composite material comprises continuous fibers as reinforcing materials and resins as base materials, the continuous fibers include carbon fibers, glass fibers, aramid fibers and flax fibers; the resins include polylactic acid (PLA), TPU, ABS, PEEK and PPS.

3. The method of claim 1, wherein: The artificial intelligence algorithm includes a binary classification neural network model, a multi-classification convolutional neural network and a K-value proximity algorithm.

Images