**1. Introduction**

The double-pulsed gas metal arc welding (DP-GMAW) process is a mature arc welding operation technology that is prevalently employed in modern industrial manufacturing occasions. Though this process is designed based on the traditional pulsed-GMAW (P-GMAW), it has some significant merits when compared to the P-GMAW, such as the DP-GMAW process can reduce the porosity incidence [1] and improve the solidification cracking susceptibility [2]. Also, the process has better gap bridging ability [3], and better ability to control the mode of droplet transfer than those of the P-GMAW process [4]. This new technique is an effective variation of the traditional P-GMAW process, in which the pulsing current aiming to metal transfer control is overlapped by a thermal pulsation [5], which induces changes of temperature and stress of the welding pool [6]. Hence, it has been paid more and more attention in academic research and actual industrial production areas in recent years.

The difference between P-GMAW and DP-GMAW processes is the current waveform. During the DP-GMAW process, the current waveform is composed of a rhythmic thermal pulse phase (TPP) and a thermal base phase (TBP) [7], which have different frequencies and amplitudes, and the sum of durations of the two phases is equal to a thermal period (TP) [8]. Because of the existence of double pulses, it was also been named the twin-pulsed GMAW process in some published literature. To assure a stable welding process and obtain weld beads with high quality, researchers and scholars have put a lot of efforts to change the waveforms. The original waveform was a usual types of square waveforms where strong current and weak current were alternately appeared during the process. Under the circumstance, arc quenching may appear when actual currents switch between TPP and TBP, because the wire feeding equipment may not be able to catch up with the variation of the current pulse with adequate speed due to the mechanical inertia [8]. To alleviate the sudden changes of the current amplitude differences in these two phases, corresponding improvements have been revealed, such as trapezoid waveform, or sinusoidal waveform [8,9]. By means of the changes, the changes of the currents were replaced by gradual switches, which can improve the stability of the arc to achieve a stable droplet transfer [10]. Apart from arc quenching, splashes, short-circuit or open-circuit of the electrical system may frequently occur during the process if the parameters setting and matching are improper. All of these phenomena should be carefully considered during system design and operation in order to decrease the cost of production and improve the actual efficiency.

According to the principle of DP-GMAW, there are many parameters requiring proper setting during the process, such as the thermal period and corresponding frequency, twin pulse current change and duty cycles in two phases, and so on. Also, currently the welding robot has been employed more and more in manufacturing, and the robot traveling speed is also an important parameter. Hence, this is a typical multi-parameter system, and how to obtain an optimal operational parameters combination for achieving satisfactory performance is a challenge work for all users. In practical application, testing and justifying each parameter on the welding quality can cost so much and cannot be accepted in the majority of occasions.

No matter which type of welding technology is employed, quality estimation or evaluation is so important. For example, for resistance spot welding, the tensile-shear strength of the weld can be used for evaluating the welding quality [11]. For pulsed GMAW products, which is the weld bead, the quality involves more elements, such as crack, appearance, penetration, microstructure, and so on [12]. In general, these different elements should be properly combined to yield one reliable quality criterion. Nowadays, with developing computer technology, artificial intelligent (AI) technology has demonstrated many achievements. In welding research area, AI technology has been also employed, such as in quality estimation of resistance spot welding [11], or in optimal parameter prediction in double-wire-pulsed metal inert gas (MIG) arc welding [13]. As for the quality evaluation of arc welding products, many previous contributions have paid a lot attention to it. Casalino et al. [14] employed neural networks to establish a relation between process parameters and geometry of the molten zone of the welds, and then used a fuzzy C-means clustering algorithm to evaluate the quality. Wu et al. [15] used a Kohonen network to monitor the welding process and evaluate the quality in a GMAW process. The inputs were the probability density distribution (PDD) of the welding voltages and the class frequency distribution (CFD) of short circuiting times, and the network can recognize and classify the undisturbed and intentionally disturbed GMAW experiments. It can be noticed that the AI technology can be proper in estimating the quality of weld bead and exert remarkable effects.

This work aimed to explore how to obtain an optimal parameters matching in order to obtain the weld bead with satisfactory quality, while researching the influential levels of the key operational parameters on the different performances of the weld beads. The DP-GMAW process involves various input parameters, and the quality of weld bead also includes a lot of evaluation criteria. To achieve preliminary goals, two main contents have been included in this work. The first was

the optimal parameters machining. Among various operational parameters, few key operational parameters were selected to do experiments. To achieve the desired effects and decrease experimental complexity, orthogonal experimental design, which is an important experimental design method to explore the system effects typically involving multiple factors and multiple levels [16], was employed. This experimental design method can reduce the workload and involves corresponding methods of analyzing the experimental results, so to yield more reliable conclusions [17].

After employing orthogonal experimental design to obtain weld beads using different operational parameter combinations, an appropriate quality evaluation method can be used to estimate the experimental results. Because the quality of weld bead involves various elements, the quality evaluation method should also be a multi-input system. In addition, to clearly reflect the evaluation results, the output is better when quantitatively presented. In this work, considering the characteristics of weld beads and application of current AI technology, fuzzy comprehensive evaluation (FCE), which was an effective evaluation method based on the fuzzy sets and fuzzy mathematics, was chosen to conduct the quality evaluation for the weld beads. FCE was introduced in the 1960s, and has become an effective multi-factor decision-making tool for comprehensive evaluations so far. During the actual application, combining with the expert experiences, this method can make a full and comprehensive refection on the evaluation criteria and the influence factors of fuzziness, and produces evaluation results closer to the actual situation [18]. It has been used in a lot of different areas, such as in power policy making [19], teaching and education performance evaluation [20], motion performance evaluation of autonomous underwater vehicle [21], water resources carrying capacity [22], distinct heating system evaluation [23], real estate investment risk research [24], quality assessment for compressed remote sensing images [25], and other relative areas.

In this work, the DP-GMAW process based on an industrial robot operation was conducted, the objective was seeking optimal operational parameters combination in order to obtain weld bead with satisfactory quality, and obtaining the influential levels of different operational parameters on the performances of the weld bead. During the process, according to principle and operational characteristics of this process, some key operational parameters were selected to design orthogonal experiments, and then the FCE method was employed to do quality evaluation according to relative experimental results and obtained optimal operational parameters combinations. Advanced experimental designing methods and quantitative quality evaluation methods were effectively combined in this work, and the contribution can serve the current DP-GMAW process improvement and parameter optimization.
