**1. Introduction**

Nowadays, titanium alloys, especially Ti-6Al-4V (TC4), have acquired extensive applications in aerospace, aircraft, automotive, biomedical, and chemical industries [1–3]. This is primarily due to their superior performance characteristics, such as low density, good corrosion resistance, high heat resistance, and high specific stiffness and specific strength. In many fields of manufacturing and processing industry, gas tungsten arc welding (GTAW) [4,5] plays a critical role, which is viewed as one of the traditional and significant material processing techniques. The mechanical and physical properties of titanium alloys are greatly affected by the microstructure evolution during molten solidification [6]. In particular, Kou [7] proposed that the dramatic transformation of temperature and solute concentration directly gave rise to an unfavorable influence on the dendritic growth. Salimi [8] developed a 3D transient analytical solution to the heat conduction problem in different plates with a circular moving heat source. The analytical results were validated by the finite element (FE) method and experiments. During metal solidification, to simulate temperature field change and microstructure transition for a laser-based additive manufacturing processing, a successful cellular automaton-finite element (CA-FE) model was proposed by Zhang [9]. Specially, it should be noted that the direct observation and measurement of the dynamic solidification process of the welding molten pool is very difficult to achieve in experiments.

With the continuous advancement of computational technology, several approaches have been developed for modeling microstructure evolution in the weld molten pool. These approaches mainly include Monte Carlo (MC), phase field (PF), and cellular automata (CA). Anderson [10] uncovered the features of crystal growth and studied the relationship between the dendritic growth and undercooling by the MC method. Then, the temperature of the weld process was integrated with the simulation of grain growth in a computationally efficient manner [11]. However, due to lacking a physical basis, this MC method failed to perform quantitative analysis, which also greatly limited its application. Another employed a phase field (PF) method to quantitatively simulate the dendritic growth process. Qin et al. [12] simulated the solidification of a multicomponent and developed the multiphase systems based on the PF model. To simulate microstructure morphology and solute distributions of the Al-4 wt% Cu alloy in a welding molten pool under transient conditions, Wang et al. [13] developed a quantitative phase field model. However, this model needs a huge amount of computational time, resulting from requiring an extremely fine mesh. To this end, Rappaz et al. [14] proposed a CA method which characterizes the discrete temporal and spatial microstructure evolution using a network of regular cells. In the gas tungsten arc welding (GTAW) process, Zhan et al. [15,16] simulated the dendrite morphology based on the CA method. In addition, the CA method has two advantages: low computational cost and ease of coupling with the macroscopic thermal model considering the transfer of heat and mass in complex geometries. During the solidification of the TC4 alloy, Wang et al. [17] investigated the grain microstructure by means of a method where CA was coupled with FE in the molten pool. Chen et al. [18] proposed a 3D CA-FE model to permit the simulation of grain structure solidification during multiple passes of gas tungsten arc welding (GTAW) and gas metal arc welding (GMAW). Here, the CA-FE method was better used for simulating the temperature distribution and microstructure formulation in the molten pool. To date, there has been little investigation on predicting the temperature distribution and dendritic morphology of the columnar to equiaxed transition (CET) and obtaining experimental validation during Ti-6Al-4V alloy gas tungsten arc welding.

In this study, the temperature field and dendritic morphology were simulated by a CA-FD method. The FE software ABAQUS (version 6.13, Dassault SIMULIA, Providence, RI, USA) was used to compute the thermal field evolution under different welding current conditions. Moreover, the CA model was built to simulate the microstructure evolution of the CET process. The simulated results were verified by corresponding experiments.
