Abstract: 3D printing of variable-component composites is a cutting-edge direction in the development of additive manufacturing, which is an important technology to realise gradient material structure. Online regulation of process parameters to adapt to the material component changes during 3D printing is a difficult problem in manufacturing variable-component composites. Combining machine learning algorithms with 3D printing process, and on the basis of small-sample training in additive manufacturing, establishing an algorithmic model of the relationship between process parameters and extruded volume, so as to explore the effectiveness of machine learning algorithms to regulate the process parameters of variable-component composites 3D printing. The screw extrusion 3D printing equipment is used to collect experimental data for different material components. Five machine learning algorithms, SVR support vector regression, BP neural network, RF random forest, RBF neural network and Kriging model, were used to predict the extrusion volume, which is used to adjust the process parameters according to different material components. From the results, it can be seen that when extrusion volume prediction is performed, the training sample size should be more than 30 groups in order to ensure the prediction results. Moreover, SVR algorithm is the most suitable for the small sample size prediction situation among the five machine learning algorithms, and it has the highest extrusion volume prediction accuracy. 3D printing experiments of variable-component composites are carried out, and the process parameters are adjusted according to the material components during the printing process. The sample pieces are printed with good quality, which verifies the effectiveness of the SVR algorithm to regulate the process parameters.

Key words: additive manufacturing, variable component composites, machine learning, prediction of process parameters