Arc spectral processing technique with its application to wire feed monitoring in Al–Mg alloy pulsed gas tungsten arc welding

Abstract The principal component analysis (PCA) is applied for three purposes: spectral line identification, redundancy removal and spectral characteristic signals extraction. The spectral information is classified as the first and the second principal components, associated with Ar I lines and metal lines, respectively. With the mean value method, pulse interference resulted from the pulse current is eliminated from the spectral signals. The relationships among these extracted signals and the defects resulted from wire feed are discussed and the results show that the second principal component is closely related to these defects while the first principal component has relationship with the arc states. To test validity of the extracted signals, a back-propagation neural network is designed and appropriately trained with “Early Stopping” technique to detect these defects automatically.