Research On Data-driven Quality Control Of Porosity Defects In Metal Additive Manufacturing

Metal 3D printing is an advanced manufacturing technology with significant advantages,including high material utilization,short production cycle,customization and the ability to produce complex structures.It is widely used in aerospace,defense,automotive,electronics,and biomedical fields.However,due to its specific manufacturing process,the product still has many quality issues.Porosity defects are the most common defects,so it is urgent to develop quality control technologies to improve the quality of additive manufacturing.In this study,we conducted in-depth research on the quality control of porosity defects in metal 3D printing products from three perspective including process parameter optimization,quality inspection,and online anomaly detection.The specific research work is as follows:In the research on process parameter optimization,this study proposes a method based on Bayesian hierarchical model to establish the relationship between process parameters and porosity images.Unlike existing research that focuses on porosity rate,this study uses porosity images to characterize the defect status.We first use the two-point correlation function(TPCF)to accurately capture the morphological features of the porosity image and establish its relationship with the process parameters.Then,we use the Bayesian hierarchical model to describe the hierarchy of the data and the randomness of the porosity image.This model can predict the porosity and reconstruct the porosity image based on the simulated annealing algorithm,and optimize the process parameters to achieve the optimal printing parameters.In the research on quality inspection technology,a method based on the order statistical method is proposed to model the relationship between two-dimensional image and its threedimensional distribution.Traditional ultrasonic detection methods like Archimedes methods,X-ray,CT scanning,and other methods have problems such as low accuracy or inability to detect the shape,size,and distribution of three-dimensional pores.This study proposes a more practical method based on part of the larger pores to estimate all three-dimensional distribution.Based on the geometric relationship between two-dimensional and three-dimensional pores in space,this study establishes the joint probability density function of the first 7)larger twodimensional particle clusters and the three-dimensional size distribution,thereby avoiding the actual difficulties of correctly identifying and accurately measuring the radius of small clusters.This method has low cost and high efficiency,and is more practical and robust in practical applications.Based on the exploration of the above problems,higher requirements for the detection accuracy of some precision parts are proposed.Taking into account the data truncation,a more accurate estimation method is achieved.Since the order statistical method only considers partial observations,so it is not enough to achieve precise detection.In order to improve the accuracy of quality inspection,a method is proposed to estimate the three-dimensional distribution of pores in space based on all two-dimensional scanning electron microscope images.The likelihood function of truncated data is established based on the relationship between two-dimensional observation data and three-dimensional distribution,thereby deriving the accurate estimation of pore size distribution,number density,and spatial distribution,as well as the defect volume ratio in three-dimensional space.This method greatly improving the accuracy of quality inspection.A sparse Gaussian process-based anomaly detection method was proposed for online monitoring and anomaly detection in metal additive manufacturing.This study designed an anomaly detection algorithm based on monitoring the melt pool temperature to achieve early warning of abnormal states during the manufacturing process.Specifically,the method includes two stages.In the first stage,a quadratic regression and sparse Gaussian process mixture model was used to model the melt pool temperature under normal conditions.By decoupling the effects of power and spatiotemporal variables on melt pool temperature using the mixture model,the temperature variations were captured accurately.In the second stage,anomaly detection was performed based on the deviation degree of the cumulative residual mean,and the Six Sigma method was used for quality monitoring and anomaly detection.Ultimately,the proposed method achieved excellent performance with high detection rate,low false alarm rate,and low latency rate.This study focuses on the pore defects in metal 3D printed products and addresses the practical and challenging issues that arise during the entire manufacturing cycle,including process parameter optimization,quality inspection,and online anomaly monitoring.This research helps to improve the manufacturing efficiency,reduce costs,and enhance the product quality of the metal additive manufacturing industry,making it of significant theoretical value and practical significance for the pore defect quality control in metal 3D printing.