Data-driven modelling of the fume emission rate for gas metal arc welding

Abstract Gas metal arc welding (GMAW) is used across various industries due to its versatility. However, it carries significant health risks, particularly from the inhalation of fumes. While noise and radiation are also concerns, the focus on fume emission rates (FER) has been increasing in importance. Despite extensive research, the effects of various welding process variables on FER are not yet fully understood. Traditional methods for measuring these emissions are time-intensive and expensive, especially when considering the plethora of available materials and welding techniques. To address this topic, several FER models are being discussed in this paper, based on transient electrical and optical process data. For this task, an extensive database of FER measurements has been acquired with the respective electrical and optical time series, with 240 welds for model training and an additional 28 welds for validation. This database has been used to train different machine learning (ML) algorithms with varying transparency regarding the correlation between the transient process data and the FER. The investigation revealed that basic statistical modeling, using multiple linear regression, can achieve an average FER prediction accuracy of $${\pm 25}{\%}$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mrow> <mml:mrow> <mml:mo>±</mml:mo> <mml:mn>25</mml:mn> </mml:mrow> <mml:mo>%</mml:mo> </mml:mrow> </mml:math> across the FER operating range, measured by the average deviation of the predicted FER value from the measured value over the entire test data set. This applies to a broad range of wire feed rates, from $${5}{\text {m} \, \text {min}^{-1}}$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mrow> <mml:mn>5</mml:mn> <mml:mrow> <mml:mtext>m</mml:mtext> <mml:mspace/> <mml:msup> <mml:mtext>min</mml:mtext> <mml:mrow> <mml:mo>-</mml:mo> <mml:mn>1</mml:mn> </mml:mrow> </mml:msup> </mml:mrow> </mml:mrow> </mml:math> up to $${12}{\text {m} \, \text {min}^{-1}}$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mrow> <mml:mn>12</mml:mn> <mml:mrow> <mml:mtext>m</mml:mtext> <mml:mspace/> <mml:msup> <mml:mtext>min</mml:mtext> <mml:mrow> <mml:mo>-</mml:mo> <mml:mn>1</mml:mn> </mml:mrow> </mml:msup> </mml:mrow> </mml:mrow> </mml:math> . Deep learning algorithms can improve the average FER prediction accuracy even further, reaching up to $${\pm 12}{\%}$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mrow> <mml:mrow> <mml:mo>±</mml:mo> <mml:mn>12</mml:mn> </mml:mrow> <mml:mo>%</mml:mo> </mml:mrow> </mml:math> using neural networks. However, they may lack the correlation insights provided by other ML approaches. A correlation analysis of the data revealed that further investigations of different filler materials, etc., can be reduced to a few significant FER measurements. The implications of these findings are far-reaching. By incorporating these predictive models into modern welding equipment, we can create virtual FER sensors or observers on-site. This not only improves process efficiency but also reduces welders’ exposure to harmful fumes while enabling a more comprehensive analysis of FER within the broader manufacturing context.