Spectral diagnosis and defects prediction based on ELM during the GTAW of Al alloys

Abstract

Quantitative analysis of arc spectrum makes great contribution to studying the interaction between arc radiation and welding quality. In the present work, the electron temperature Te calculated by Boltzmann plot method is found to have correlation with welding quality and its shape distribution is consistent with results obtained by the infrared video camera. To acquire smooth feature values of the Te curve, wave packet transform (WPT) is employed to eliminate the effect of power pulse interference on the spectral signal. The relationship between welding quality and reconstructed Te curve is investigated. Furthermore, the mechanism of Te variation is discussed. In consideration of the limit of the single temperature cure, six more spectral signals are discussed for describing the weld seam state accurately. The above spectral features are used to build the prediction model of welding quality based on extreme learning machine (ELM). The proposed methodologies are verified to be effective with high prediction accuracy.