In-situ Monitoring And Real-time Feature Extraction Of Plasma Arc Additive Manufacturing Process

Plasma Arc Additive Manufacturing(PAM)is an advanced metal Additive Manufacturing(AM)technology that can significantly reduce production cycle and cost,and is suitable for manufacturing various complex metal parts in many fields.Nondestructive testing methods with high precision and real-time monitoring are required for the PAM process.In the PAM process,the molten pool and plasma arc are key objects that characterize the dynamic manufacturing process and can be used for process prediction and real-time feedback control.Therefore,achieving in-situ realtime monitoring of the PAM process is extremely important.The thesis focuses on the feature extraction and data analysis of the molten pool and plasma arc through in-situ monitoring based on image processing,as well as preliminary analysis of the quality of the workpiece.Firstly,the current status of in-situ real-time monitoring of metal AM processes based on image processing is analyzed in detail,which provides an excellent theoretical basis and international perspective for achieving in-situ real-time monitoring of the PAM process.Then,the software and hardware parts of the in-situ monitoring system for the PAM process were designed,and an experimental platform was built.The experimental platform used a high-speed camera with a frame rate of 30,000 frames per second to achieve real-time shooting of the processing images.A large number of workpiece processing experiments were carried out based on this platform,and molten pool and plasma arc processing images were collected under different current intensities and plasma scanning speeds,forming a raw data set for Artificial Intelligence(AI)algorithm training.Next,a Full Convolutional Network(FCN)is proposed for real-time feature extraction of the molten pool and plasma arc.The separable convolution technology is used to reduce network parameters,and the dilated spatial pyramid structure is used to extract multi-scale features.With an average processing time of only 84 ms,the accuracy can reach 95.1%,effectively ensuring both the accuracy of image processing and the high-speed processing of data features.The image segmentation performance of the proposed AI algorithm is compared with four traditional image processing algorithms and three AI methods.The experimental results show that the FCN method proposed in the thesis has better image processing performance than the seven algorithms mentioned above,and can quickly and accurately extract the morphology features of both the molten pool and plasma arc.Finally,the obtained data is quantitatively and qualitatively analyzed and compared in detail,and the relationship between the average capture area(molten pool and plasma arc)and PAM process parameters(current intensity and scanning speed)is obtained.Based on the measured surface roughness of the samples,the quality of the samples is preliminarily analyzed,and its relationship with the average capture area of the molten pool is clarified.This study provides an effective new idea for in-situ monitoring of the PAM process,which will help improve the repeatability and stability of the PAM process,improve the quality of metal parts,and provide better solutions for applications in fields such as aviation,automobiles,and ships.