Detecting Degradation using Structure-Borne Sound in Flame Torch Cutting

Flame cutting is a widely used metal processing technique whose performance depends critically on the condition of the cutting nozzle. Nozzle degradation can impair flame quality, compromise safety, and reduce process efficiency. This work investigates the feasibility of using structure-borne sound sensing for automated detection of nozzle wear. In a controlled laboratory setting, we compare alternative sensor placements, acquire an annotated dataset, and extract spectral features for classification. Support Vector Classifier, Symbolic Classification, and Multilayer Perceptron models are evaluated to assess diagnostic potential. Additionally, symbolic regression is employed to validate the findings. Within the constraints of a limited dataset, results indicate that intact and degraded nozzles can be distinguished with high accuracy under stable conditions. While these findings are not yet generalizable to full-scale industrial deployment, they provide a reproducible methodology, practical sensor placement insights, and a foundation for future studies incorporating greater data diversity and operational variability.