top of page
Welding specialist

Results

With our quality metric, we were able to differentiate between the expert, beginner, and intermediate welders in our limited data set.  A more standardized approach for data collection, would result in a better metric and thus more clear differentiation.

Owen_Weld.jpg

Figure 10: Beginner welder test trimmed DFT

01

Beginner Welder Q=.38

Most of the frequencies of movement are in the 1.5 to 3hz range, but no singular spikes are evident.  This implies that the welding motion was more erratic and no consistent movement occurred.

02

Intermediate Welder Q=.36

There is a clear spike around .7hz as expected, but there is also a lot of higher frequency noise which corresponds to less fluid and more inconsistent movements.

Zeno_Weld.jpg

Figure 11: Intermediate welder test trimmed DFT

Nate_Weld.jpg

03

Expert Welder Q=.19

There is a clear spike at .7hz and virtually no other significant frequencies.  This signifies a very consistent movement with a frequency of .7hz

Figure 12: Expert welder test trimmed DFT

Errors and Next Steps

1

The largest source of error with the system as it stands now is data collection. Because every welder holds the MIG gun slightly differently, and the phone can't always be placed in the exact same location, the axes are not always oriented the same. This means acceleration values are not significant, just their frequency.

The algorithm is also limited by the fact that it 'ranks' a welder by skill and gives generic advice as to how they could improve. Without aligning the axes with the direction of motion, or rotating the results to orient them, it is not possible to get enough data to give more specific feedback.

The most significant improvement that could be made is a mount to align the sensor with the axes of motion. A 3D printed mount for an accelerometer that attaches to the MIG gun would increase the reliability and accuracy of the data

4

With improved data collection there is the opportunity to give more specific feedback. As described in the motion model, the algorithm could analyze both the frequency of the welder's motion in the X and Y axis and the stability of their motion in the Z axis.

2

3

bottom of page