Weld Feature Prediction for Laser Beam Welding Using Neural Network Algorithms

Laser beam welding (LBW) is a prime candidate for the production of metallic aerospace components for high-performance environments. Laser processes provide a high degree of control over both the spatial and temporal heating profiles during welding. These unique qualities enable precise control of important weld properties such as penetration, width, microstructure, and defect formation. In many situations, the ability to non-destructively inspect these welds to insure quality and conformance to specifications may be difficult or even impossible with certain alloys and design morphologies. The presented research effort addresses the industrial need to improve weld feature prediction capabilities to aid in weld parameter selection and improve the efficiency of procedure development for laser beam welding. Weld geometry analysis was performed on transverse and longitudinal cross sections of partial penetration laser beam welds on stainless steels and nickel base alloys, including commercial available 304L SS and alloy 690. Weld geometry and solidification microstructure characteristics were related to input parameters (travel speed and laser power) to cover different weld modes, i.e. conduction, transition and keyhole mode. The obtained data was used to develop and optimize a neural network algorithms that can predict weld geometry features, most importantly weld penetration.