REAL-TIME MOLTEN POOL AREA EXTRACTION FOR CONTROL OF GAS

TUNGSTEN ARC WELDING

Y. Z. Wang, Z. Jandric and R. Kovacevic*

ABSTRACT

This paper presents a real-time molten pool area extraction method for the control of gas tungsten arc welding (GTAW) with a wire filler in the hybrid rapid prototyping and tooling (RP&T) technique based on welding and CNC milling. The molten pool shape and size were strongly influenced by the two-dimensional geometry of the built part. A coaxial machine vision unit is used to acquire images of molten pool and its surrounding area through the torch. After image processing, the boundary of the molten pool is extracted. Because the molten pool is partially hidden by the tungsten electrode, its boundary is incomplete. The molten pool area is selected as the feedback to the closed-loop control of the GTAW process. In order to calculate the area of the molten pool with an incomplete boundary, a neural network is used. By changing welding parameters, the neural network is trained for different sizes of molten pools. The testing of the neural network is performed by another group of molten pool images.

INTRODUCTION

Nowadays, industry is under constant and growing pressure for shortening the time-to-market lead time. Rapid prototyping (RP) techniques have been developed for fulfilling this demand. For many years, rapid prototyping techniques have been used almost exclusively on materials such polymers, waxes, or paper (Ref. 1). The rapid prototyped parts had limited dimensions, and they had a tendency to distort due to shrinkage.

Three-dimensional welding has the ability to rapidly produce strong, fully dense metallic parts in the form of layers (Refs. 2, 3). To increase the surface quality and accuracy of the metallic parts made by the 3D welding, a promising hybrid rapid prototyping and tooling (RP&T) technique based on gas tungsten arc welding (GTAW) and CNC milling has been under development at the Research Center for Advanced Manufacturing (RCAM) at Southern Methodist University, Dallas, TX (Ref. 4). This hybrid process allows the fabrication of metallic parts with high dimensional accuracy and with complex external and internal geometrical features.

In the hybrid RP&T technique, the GTAW process with a wire filler provides depositing materials and energy to fuse the added material. For the needs of RP&T, the welding process is required to provide lower heat input and lower bead penetration that is exactly opposite to the conventional welding process. In order to generate uniform and fairly flat beads and layers, a metal deposition rate and molten pool size should be invariant of the alternations of the welding

Research Center for Advanced Manufacturing, Southern Methodist University

1500 International Pkwy, Suite 100, Richardson TX 75081

corresponding author.

Tel: 1-214-768-4865, Fax: 1-214-768-0812, Email: kovacevi@seas.smu.edu

conditions. The geometry of the molten pool can provide abundant, accurate and instantaneous information about the welding process. Hence, closed-loop control of the welding process, with the information of the molten pool size as a feedback, is a necessity.

Although several indirect approaches, such as monitoring the pool oscillation, the ultrasonic technique, infrared sensing, and radiography, have been proposed to sense the geometry of the molten pool, machine vision, being non-intrusive, seems to be the most promising technique for obtaining the clear geometry of a molten pool.

Machine vision systems have been extensively used for the molten pool sensing and seam tracking for more than two decades, mostly with the idea to simulate the eye sensing and process actuation of experienced human welders. Previous research works for sensing the molten pool in GTAW with machine vision was conducted without a wire filler.

In the GTAW process with a wire filler, it is necessary to ensure that the wire is always fed in front of the moving arc. In order to avoid blocking the optical path between the molten pool and the CCD camera by the wire filler, a coaxial machine vision unit, developed by Richardson et al. (Ref. 5), is used in this paper. The lens of a CCD camera is coaxially mounted with the electrode. The electrode tip is used to block the bright core of the arc light from the overpowering exposure on the CCD target. After the image of the molten pool is captured, the Laplacian operator is used to detect the edges. Then, the image was segmented into a binary one with an adaptive threshold. The boundary of the molten pool is extracted from the binary image.

The molten pool size and shape in the hybrid RP&T are strongly influenced by the twodimensional geometrical features of the part to be built. Fig. 1 demonstrates the molten pools at the flat surface (point 6), the edge (point 5), and the corners with different angles (points 1 to 4), while keeping the welding parameters constant. This paper is concentrated on the acquisition and processing of the shape of the molten pool at the flat surface (point 6).

Figure 1. The shapes of the molten pool as a function of 2D geometrical features of substrate

Fig. 2 illustrates the schematic presentation of the hybrid RP&T machine, consisting of a central controller, a CNC milling machine, a GTAW unit, and a coaxial machine vision unit. The central controller acquires the measured data and controls the welding current, wire feeding speed, and

wire feeder's position, while the workpiece motion is controlled by the CNC machine microprocessor. The filler metal is fed directly into the molten pool. The step motor 1 is used to ensure that the wire is always in front of the moving arc. The step motor 2 is used to control the filler metal feeding speed.

The coaxial machine vision unit used for the molten pool image acquisition, the image processing, and the molten area extraction, is composed of a computer equipped with the frame grabber, a CCD camera, and the image processing software. The image resolution is 640x480 pixels with 256 gray-levels.

The experiments are carried out with the following parameters:

The workpiece traverse speed is 23 cm/min;

• The wire feeding speeds are 20 and 40

cm/min;

System
Controller

Wire Feeder Speed Control

Path
Generation

Positioning

Wire Feeder Speed Control

CNC
Central
Machine Controller

Timer V/O
Board for the PC

The welding currents are changed from 100A to 133A with a 3-A interval.

The welding voltages are changed from 9V to 11V.

• The filler metal is mild steel with a diameter of 0.89 mm.

Wire Feeder

DESCRIBING MOLTEN POOL GEOMETRY

In conventional GTAW without a wire filler, the shape of the front part of the molten pool always keeps semicircle with the half-maximum-width as the radius despite different welding conditions, but the sharpness of the rear of the molten pool is changed with the penetration. Also, the shape of the molten pool can be thought as symmetric along the welding direction (Ref. 6). The following geometrical parameters were used by different researchers to characterize the geometry of the molten pool: the maximum width, the maximum length, the half-length, the area, and the rear angle.

Ru

R,

R10

(a) Complete

R12

R

(b) Incomplete

Figure 5. The illustration of the selected characteristic parameters

The selection of the characteristic parameters is
crucial. Three criteria must be satisfied. First, all
parameters must have a common reference point.
Second, the area of the molten pool must be
calculated using the selected characteristic
parameters. Third, although more parameters could describe the
molten pool more accurately, the increase in the number of selected
characteristic parameters may complicate the calculation and
increase the calculation time. Thus, the number of parameters must
be optimized.

Intuitively, the center of the molten pool is a suitable reference point. But because of the difficulty of finding the molten pool center, a relative fixed point is another choice. The position of the electrode shadow, the dark spot in Fig. 4, is unchanged in all images. The top point L of the electrode shadow is always inside of the molten pool. Based on this position, the reference point R is determined as follows:

Figure 4. Definition of reference point

Direction of welding

Find the line segment AB that passes through point L and is perpendicular to the direction of welding as shown in Fig. 4. AB intersects with the boundary of the molten pool at points A and B, respectively;

Find the middle point R of AB. R is defined as the reference point for all selected characteristic parameters.

For a closed-contour of the molten pool, in which each radius vector from the center of the boundary intersects the boundary at only one point, a one-dimensional series of real numbers can be acquired by measuring the length of successive radius vectors that are angularly equispaced. This series is denoted as {R(a)}, where R(a) is the radius vector at the angle a.

In our case, the boundary of the molten pool is incomplete. A series of radius vectors is selected as the parameter to calculate the area of the partially hidden molten pool. Fig. 5(a) shows the simulation of a completed molten pool and a series of radius vectors composed of R1 R12. The angle between two adjacent radius vectors is 30°. From Fig. 4, it can be seen that the front part of the molten pool is hidden by the electrode shadow. It was experimentally determined that only part of the radius vectors, R~R7, can be extracted from every image. The rest, R R12, are not always available. Two radius vectors, denoted as Rs and R12 are used to connect the reference point R and the two end points. As shown in Fig. 5(b), the angles a1 and a2 may not be 30°. Thus, Ry to Rs and R12 plus a1 and a2 are the parameters used to calculate the area.

MOLTEN POOL BOUNDARY EXTRACTION

After the image of the molten pool is taken, it is smoothed using a median filter. A Laplacian operator is used to detect the edges. Then, the image is enhanced and segmented into a binary one with adaptive thresholding. All possible edge points are detected. The molten pool boundary is then extracted by making use of the continuity of the molten pool boundary with the following steps as shown in Fig. 6:

Draw a line that passes through the reference point R, defined in previous section, and is perpendicular to the direction of welding (Y direction) as shown in Fig. 6(a);

Find all possible edge points that the line intersects. For example, points 1-6 in Fig. 6(a); Starting from these points, trace the connected edge points to a negative Y direction to form edges, while counting the lengths of all edges;

Reserve the longest edge as the upper part of the molten pool boundary as shown in Fig. 6(b), and ignore the others;

Starting from points 3 and 5, along the positive Y direction, trace the possible edge points connected with the upper part of the boundary to find its lower part. The result of this operation is show in Fig. 6(c).

(a) original image

(b) possible edges (d) extracted boundary Figure 7. Molten pool boundary extraction

Based on extracted boundary, nine radius vectors and two angles are determined to calculate the area of the molten pool. Fig. 7(a) shows the original image acquired under the following welding conditions: the workpiece traverse speed is 23 cm/min; the wire feeding speed is 40 cm/min; the welding current is 115A; the welding voltage is 10.5V; and the filler metal is mild steel with a