LSSVM Model for Penetration Depth Detection in Underwater Arc Welding Process

Abstract

For underwater arc welding, it is much more complexity and difficulty to detect penetration depth than land arc welding. Based on least squares support vector machines (LSSVM), welding current, arc voltage, travel speed, contact-tube-to-work distance, and weld pool width are extracted as input units. Penetration depth is predicted in underwater flux-cored arc welding (FCAW). For improvement prediction performance, the LSSVM parameters are adaptively optimized. The experimental results show that this model can achieve higher identification precision and is more suitable to detect the depth of underwater FCAW penetration than back propagation neural networks (BPNN).