Research On Internal Defect Detection Of 3D Printed Lattice Structure Based On Improved YOLOv4 Algorithm

Lattice structure is a kind of lightweight multifunctional material with high porosity.It has been widely used in aerospace,weapon equipment and other fields because of its high specific strength and specific stiffness,strong shock absorption and noise reduction ability.3D printing is an important means to prepare lattice structures,but structural defects will inevitably occur during the preparation process,affecting its mechanical properties and physical response.The internal defect detection methods of lattice structures are mostly scanned by industrial CT to obtain tomographic images and perform manual or algorithm detection.However,the fault image has the characteristics of gradual change and small defect scale,and the labor intensity of manual detection is easy to cause false detection and missed detection.The existing defect detection algorithms are not effective for small-scale defect detection.Therefore,this paper takes the titanium alloy lattice structure prepared by SLM technology as the research object,and carries out high-precision detection of internal defects of 3D printing lattice structure based on deep learning method.The main research contents are as follows :(1)According to the CT scan data of lattice structure,the internal defect data set of lattice structure is constructed.Firstly,the industrial CT is used to scan the titanium alloy lattice structure to obtain the tomographic image,and the noise is filtered to improve the image data quality.The internal defect types of lattice structure are clarified,the causes of defects are analyzed,and the image data is expanded according to the characteristics of defect category,quantity and morphology.Finally,the Labelimg platform is used to label its defect information,complete the construction of the data set,and provide high-quality and sufficient scale image data for the subsequent model.(2)A deep learning model is constructed to achieve high-precision detection of lattice structure defects.According to the characteristics of lattice structure fault images,such as gradual change,small defect scale and irregular shape,a Backbone backbone network with adaptive feature extraction function of defect scale and morphology is constructed based on deformable convolution method to enhance the feature extraction ability of the model for small-scale defects.In order to suppress the invalid feature information and enhance the difference between the image defect area and the background area,that is,the defect location should be assigned greater weight,a YOLOv4-NC4 defect detection model based on the Attention mechanism is proposed to enhance the model ’s ability to locate defects.Comparative experiments and ablation experiments are designed to verify the effectiveness of the model.(3)Lightweight defect detection model based on pruning strategy.In order to improve the inference speed of the defect detection model and reduce its parameter scale,a model lightweight method based on pruning strategy is proposed.By cutting the channels with less feature information in the model,the model is ’ slimmed down ’ and its inference speed is enhanced.(4)Development of online defect detection system for 3D printing lattice structure.Based on the above algorithm,a 3D printing lattice structure internal defect detection system is developed to realize the intelligent and high-precision detection of the internal defects of the lattice structure.And realize the functions of visualization of test results,saving statistics,model retraining and so on.