Integrating human expertise to optimize the fabrication of parts with complex geometries in WAAM

Wire arc additive manufacturing (WAAM) has emerged as a versatile solution for fabricating parts with complex geometries in recent years. However, the existing deposition parameter planning methods struggle to offer continuous and precise parameters when the part geometry varies dynamically due to the long-term dependence, strong coupling, and hysteresis properties of the WAAM process. To address this challenge, this research introduces an advanced algorithm for generating accurate and continuous deposition parameters by learning and utilizing the welding skills of proficient human welders. The first step involves capturing kinematic and welding parameter data from proficient human welders during practical welding processes. Following this, a human skill learning algorithm is developed based on a combination of the adaptive neuro-fuzzy inference system (ANFIS) architecture and particle swarm optimization (PSO) to analyse and model human motions and bead deposition results. Lastly, a practical backward model is established to generate continuous deposition parameters for weld beads with varying geometry. The effectiveness of the proposed algorithm is validated through the fabrication of two practical WAAM parts. The root mean square error (RMSE) values between the target geometry and the ground truth geometry of the parts are 0.1648 and 0.1805 respectively. The result demonstrates the algorithm's superior ability in optimal deposition parameters planning for fabricating parts with complex geometries.