**1. Introduction**

VPPA (Variable Polarity Plasma Arc) keyhole welding is an ideal method to achieve joints made of middle thick aluminum with high quality and efficiency [1]. There are two polarities in one current cycle: an EN (Electrode Negative) phase and an EP (Electrode Positive) phase. The cycle can achieve both deep penetration and remove the oxide layer on the surface of the base metal [2,3]. The keyhole digging process is very important, because the welding easily fails without proper keyhole generation [4]. However, the evolution of the keyhole and weld pool has not been clearly understood, especially in the digging process of VPPA welding. Therefore, it is necessary to explore the physical phenomenon and the mechanism of the digging process for achieving stable welding.

As a unique characteristic of plasma arc welding, keyhole behavior determines the process stability of plasma arc welding. Keyhole detection research has been carried out in view of DC-PAW (Direct Current Plasma Arc Welding) for steel. Liu et al. [5–7] directly captured the keyhole exit image using a CCD (Charge Coupled Device) camera. The keyhole exit deviation distance and keyhole size parameters were put forward to evaluate the thermal state, which provides the feasibility for accurately controlling the keyhole stability. Zhang et al. [8] developed a charge sensor to monitor the plasma cloud dynamics, corresponding to the keyhole status. Metcalfe's investigation shows that it is possible to indicate whether the keyhole is open or not yet formed by monitoring the efflux plasma. Its fluctuations can reflect the keyhole stability [9]. Regarding to VPPA welding of aluminum, Zheng et al. [10] focused on the front image sensing of the keyhole. An algorithm for extracting the keyhole's geometrical size was proposed to establish the linkages with weld formation. Wu et al. [11] developed a vision sensor system to acquire the keyhole images. A novel hybrid approach was used to recognize the keyhole status. The prediction is realized through the detection, which is a step forward for accurate control of the keyhole. In addition to directly observing the keyhole with a CCD camera, the unique signal characteristics of VPPA welding were used to measure the keyhole state. Saad et al. [12] achieved identification between three models of VPPA keyhole welding (no-keyhole, keyhole and cutting) using acoustic signal measurement. Wu et al. [13] also investigated the relationship between the keyhole geometry and acoustic signatures with a dual-sensor system. An extreme learning machine model was built for predicting keyhole geometry. However, these methods are mainly used to indirectly obtain the keyhole information. The keyhole boundary inside the weld pool in real time is difficult to measure with the above methods. An X-ray transmission observation system was successfully adopted to investigate the keyhole and weld pool dynamics [14–16], which highly contributed to the understanding of keyhole evolution. Anh et al. [17] adopted stereo synchronous imaging of tracer particles with two sets of X-ray transmission systems to clarify the weld pool formation process in DC-PAW.

The above research mainly focused on the keyhole weld pool evolution in a welding quasi-steady state rather than the keyhole digging process. Zheng et al. [18] pointed out that a smooth transition from start-up segment to main body segment was very important for welding stability. By optimizing the waveforms of current and plasma gas flow rate, a relative proper keyhole generation segment was obtained. A synchronous increase of gas flow rate and current contributes to the smooth transition. Chen et al. [19] added a pre-cleaning segment before the parameters increased for easier penetration. The numerical simulation was carried out to reveal the mechanism of the keyhole digging process of VPPA welding. The keyhole weld pool is still prone to be unstable when the welding condition (such as welding position) changes [20–22]. Due to the thermal-physical properties of aluminum alloys, welding defects, such as porosity and cutting, are easily generated if the welding process is unstable [23]. Therefore, it becomes very important to understand the plasma arc pressure output and analyze the instability mechanism of the keyhole weld pool. Han et al. [24] measured the plasma arc pressure and analyzed the effect of the arc shape on arc pressure in VPPA. It found that the arc pressure in the EP phase is smaller than in the EN phase when the arc current of different polarities are the same. The existence of a double arc in the EP phase makes the plasma arc pressure reduce further. Jiang et al. [25] measured the VPPA pressure using both pressure transducer and U-tube barometer methods, while the effects of welding parameters were analyzed. The influence of EP on the pressure output is minimal because its time ratio is much less than that of the EN phase. The increase of plasma gas flow rate cools the arc further, resulting in greater constraint of the arc, thus the arc pressure obviously increases. At present, the arc pressure change due to the polarity transaction process is not well understood. Also, the influence of pressure on keyhole weld pool evolution in VPPA welding digging process has not been studied.

Here, we have observed the fluctuation of molten pool surface in the digging process by high speed camera with a high frequency pulsed diode laser light source system. The keyhole boundary in real time was also obtained by an imaging system of an X-ray transmission. In order to analyze the factors influencing the keyhole stability, the plasma arc pressure is measured by the pressure transducer. Combining the energy and momentum balance between electrodes and arc, the physical mechanism of plasma arc pressure was obtained, based on which we optimized the pressure output. Finally, the optimized parameters were verified by the weld formation and weld pool free surface fluctuation situation.
