Chauhan, JaydeepJaydeepChauhanGourishetti, SaichandSaichandGourishettiRohe, MaximilianMaximilianRoheSennewald, MartinMartinSennewaldHildebrand, JörgJörgHildebrandBergmann, Jean PierreJean PierreBergmann2023-11-092024-02-282023-11-092023-11-01https://publica.fraunhofer.de/handle/publica/45665610.22032/dbt.58732This study explores the potential of audible range airborne sound emissions from Gas Metal Arc Welding (GMAW) to create an automated classification system using neural networks (NN) for weld seam quality inspection. Irregularities in GMAW process (oil presence, insufficient shielding gas) may lead to porosity imperfections in weld seams. Using Directed Energy Deposition-Arc additive manufacturing, aluminum (Al) and steel wall structures were produced with varying shielding gas flows or applying oil. Acoustic emissions (AE) generated during the welding process were captured using audible to ultrasonic range microphones. Mel spectrograms were computed from the AE data to serve as input to NN during training. The proposed model achieved notable accuracies in classifying both Al weld seams (83% binary, 68% multi-class) and steel welds (82% binary, 58% multi-class). These results demonstrate that employing audible range AE and NN in GMAW monitoring offers a viable method for low-latency monitoring and valuable insights into improving welding quality.enFeinwerktechnikPräzisionsmessungIndustrie 4.0MechatronikSystemtechnikAnalyse IndustriegeräuscheDDC::600 Technik, Medizin, angewandte Wissenschaften::620 Ingenieurwissenschaftenconference paper