Experimental design
Thin walls are manufactured and filmed using a near-infrared camera to monitor the area around the melt pool with a side view angle. The manufacturing technology is mature enough to make straight thin walls by controlling the thermic. Indeed, an idle time between layers is commonly used to control the thermal accumulation [24]. Two walls are manufactured: one with a short idle time of 2 seconds where a large heat accumulation can be observed and another with an idle time of 30 seconds where a little heat accumulation can be observed. Images are taken in-operando with the near-infrared camera. From these images, a thermal metric is extracted. After manufacturing, each wall is measured with a 3D scan to extract the width deviation in function of the layer and the longitudinal position.
The width and the thermal metric are compared at two scales. First, the thermal deviation is qualified at the wall scale. At this scale, the ends of the wall are not studied because their thermal conditions are particular. The data from the geometrical scan and the thermal metric are truncated, averaged and compared. Then, the thermal deviation is qualified at the bead scale. In this case, all the data of the layer are kept. The thermal particularity of the ends of the wall allows quantifying the sensitivity of the device.
Experimental set up
Manufacturing conditions
The WAAM machine is composed of a 6-axis robots Yaskawa MA1440 and a positioner with 2 axes and a welding station Fronius T.P.S. CMT 4000 Advanced. The material is an AlMg3Cr alloy. The wall is 150 mm long and it is composed of 42 layers. It is built on a 250 mm*250 mm*5 mm aluminum plate in AlMg3 at an initial temperature of 60°C.
To improve the initial thermal conditions, 2 beads are deposited on each side of the wall, at 20 mm, with a 30 cm/min robot travel speed (TS) and a 6.1 m/min synergic wire feed speed on the RCU (WFS_S). Then, the first 2 layers of the wall are deposited with a TS of 30 cm/min and a WFS_S of 5 m/min with a zigzag strategy. Finally, 40 layers are deposited with a zigzag strategy and the chosen idle time with the parameters presented in Table 1. Figure 2 illustrates the deposition strategy.
Table 1
Manufacturing parameters.
Material
|
Wire diameter
|
CMT Law
|
WFS_S
|
TS
|
Layers
|
Wall length
|
Delta z per layer
|
Idle time between layers
|
Deposition strategy
|
---|
AlMg3Cr
|
1.2 mm
|
875
|
4 m/min
|
40 cm/min
|
40
|
150 mm
|
2.3 mm
|
2s or 30s
|
Zig-Zag
|
In-Operando Measurement of the Melt pool
The CMOS camera is a Mako G-040B monochromatic with a Kowa LM35JC lens (focal of 35 mm). To measure the melt pool in-operando, the CMOS camera is fixed on the welding torch to maintain a constant distance between the camera and the robot driven tool center (Fig. 3). Figure 4 illustrates typical image obtained with the camera.
The welding process used for this experiment is based on a CMT law. In this case, the welding station produces a periodic electric arc with an approximate period of 10 to 20 ms. Images are taken when the electric arc is off. Indeed, that avoids images saturated by the electric arc or by the emission of the plasma of the shield gas masking the thermal emission of the melt pool.
To trig the camera, a zero-voltage detection is necessary which corresponds to the stop of the electric arc. Then the image is taken after a delay. If the delay is too short, a residual plasma around the melt pool deteriorates the image for analysis. If the delay is too long, there is not enough time to take the image with the correct exposure time (Fig. 5). Empirically, the delay is set to 1.2 ms and the exposure time to 0.5 ms considering that the short-circuit time is approximately 2ms (Fig. 5). Considering that a CMT cycle has an approximate duration of 10 to 20 ms, an idle time of 100 ms is set after a trigger to limit dataFigure 6. Thus, the acquisition frequency is around 8 Hz, this means approximately one image every 5 electrical arcs. The camera configuration is summed up in Table 2.
Table 2 : Camera configuration
Camera
|
Frame per second
|
Exposure time
|
Trigger Delay
|
Idle Time
|
Gain
|
Resolution
|
Bit depth
|
Objectif
|
Focal
|
---|
Mako G-040B
|
~ 8
|
0.5 ms
|
1.2 ms
|
100 ms
|
22
|
728 x 544
|
8
|
KOWA LM35JC
|
35 mm
|
Measurement of the geometrical deviation
When the wall is built, the shape of the wall is obtained using a structured-light 3D scan. The GO!SCAN SPARK™ has been used. Its accuracy is up to 0.05 mm, with a mesh resolution of 0.5 mm. Using the VXelements software, the point cloud is filtered and exported as an STL file.
Data processing
Image processing
The aim of the image processing is to automatically obtain a thermal metric from the captured image.
Images are taken during the process and a metric is extracted automatically. To exacerbate the light intensity, an image process is used [21]. A logarithm is applied to each image. Then a threshold at 45% of the bit depth of the image is applied. All values under this threshold are set to 0, and the others are linearly distributed. (Fig. 7)
Although there is a physical trigger, some images have too many perturbations due to projections or a shorter short circuit (Fig. 8). To sort out these images, the mean intensity of each image is calculated, and a histogram is plotted in function of the percentile of the mean image intensity of the sample (Fig. 9). The first five percent of the lower intensity images and last twenty percent of the higher intensity images are excluded because their characteristics are clearly different than the other images of the sample.
Last, a thermal metric of the process zone is extracted using the python library OpenCv [25]. All the images are monochromatic with a bit depth of 8. First, a binary threshold is applied with a value equal to 100. A morphological closing is applied to the output of the threshold with 4 × 4 one kernel. The biggest white spot (value of 1) is kept (Fig. 10). This spot is called the thermal processed area and its surface is called the thermal metric. Then, the images are labeled with the layer number.
The thermal metric obtained is different from the melt pool. Filters such as canny or adaptative threshold, usually used to get the boundary of the melt pool for alloys with higher solidus temperature, do not give expected result for aluminum alloys. However, as the contour made with the binary threshold is linked to the thermic, a correlation between the value of the thermal metric and the melt pool can be expected.
Scan process
The goal of the scan process is to obtain a map of the wall width as a function of the layer and the x-axis.
First, the STL file is imported using GOM Inspect 2018. A new basis is defined by the wall in the x direction and the vertical is on the z-axis (Fig. 11). The wall is sliced along the x-axis every millimeter. One hundred and fifty-one slices are created. These slices are exported as 151 2D (\(\overrightarrow{y},\overrightarrow{z})\) point clouds.
For each slice, only the left and right side of the wall are kept. Points from the plate or the top surface (2.3 mm under the highest point) are deleted. Moreover, the 4th first beads are not considered (4x2.3 mm = 9.2 mm). The least mean square line of the point cloud is calculated. It represents the middle plan of the wall in the current slice. For each point, the normal distance between the point and the line is calculated, corresponding to the half wall thickness (Fig. 11). That allows plotting the wall thickness as a function of the wall elevation. Considering the layer surface constant for each layer, the width and height of each layer as a function of the x-coordinate are determined.
Online Material
The data generated during each experiment is available with the creative commons license CC BY-NC 4.0. The raw images, the scan on STL format, the thermal metric shorted by layer and the geometry of the bead function of the layer and its x coordinate are shared. [26]