A Quality Evaluation Method for Laser Welding of Al Alloys Through Neural Networks

Abstract

The authors propose an integrated methodology to evaluate the quality of Aluminium alloy butt joints welded by laser. The method, starting from the observation of the results of an experimental investigation, focuses on the definition of a omni-comprehensive quality index for welded joints. This index is obtained using the limits of imperfections for the quality level defined by the ISO 13919 standard. A neural network system has been developed to classify and evaluate the different welds. The experiments were performed on Al 6110 T61 alloy, welded using a 6 kW CO 2 laser beam on plane sheets with a continuous butt joint. Computerised image processing has been used to recognise and to quantify the imperfections in the weld cross section. The defects have been divided into groups, as required by the EN 26520 standard. Due to the huge number of measurements required to imperfections, the artificial neural network very greatly simplifies the relationship between the quality index and the main process parameters. The neural network was trained with a set of data containing very different welding parameter choices. Application of the system aids process parameter selection that has proved to be in good agreement with quality levels measured on experimental welds made under the same conditions.