Abstract: Welding pass overlap is the essence of wire and arc additive manufacturing (WAAM) technology. Appropriate process parameter selection is of great significance to control the welding pass geometry and improve the dimensional accuracy of the molded parts. A prediction model of deep beilef network (DBN) optimized by adaptive cuckoo search (ACS) algorithm is constructed. The welding width and residual height of the weld pass are predicted based on the four technological parameters of the given nozzle height, welding current, welding speed and wire feeding speed. The optimal number of hidden layers and hidden elements are determined based on the experimental method, and the prediction model of WAAM weld pass size based on ASC-DBN is established. Simulation experiments are used to verify the performance of ASC-DBN prediction model. By comparing with the traditional models, the results show that the ACS-DBN model can effectively map the complex non-linear relationship between the weld pass size and welding process parameters of WAAM, and the prediction error of the weld pass size under the ACS-DBN prediction model is less than 6%, which has higher accuracy and stability compared with other prediction models.

Key words: wire and arc additive manufacturing(WAAM), bead size, prediction model, adaptive cuckoo search (ACS) algorithm, deep belief network(DBN)