The invention relates to the technical field of friction stir, in particular to a
friction stir welding digital model construction method, which comprises the following steps of: establishing a difference value sequence by collecting
original data such as coaxial point position temperature,
axial thrust and transverse displacement, strengthening cross fusion expression of multi-
physical field indexes among nodes, and establishing a multi-
physical field model; by extracting the included angle difference value of the middle section nodes and screening the key section of
spatial direction change, the graph neural network is utilized to propagate the connection relation between the nodes and optimize the structural expression, the structural recognition capability of
welding path mapping is enhanced, and the
welding path mapping accuracy is improved. According to the method, the dynamic characteristics of the path can be accurately mapped according to the change degree of thermal power,
coupling coding and
information fusion are carried out on multi-dimensional physical quantities of thermal power, displacement and angle, whole-process mapping from original process data to
welding path structure evolution and control parameter extraction is achieved, and the
structural robustness of path prediction and the sensitivity of
parameter control are improved.