On-line predication of underwater welding penetration depth based on multi-sensor data fusion

Using least squares support vector machines (LS-SVM) technology, a new multi-sensor data fusion model for online predication of underwater flux-cored arc welding (FCAW) penetration depth is presented. In this model, welding speed, wire feed rate, arc voltage, contact-tube-to-work distance (CTWD), and weld pool width are used as inputs, while the depth of welding penetration as output. The radial basis function (RBF) is chosen to be the kernel function and a new method of self-adaptive determination for optimizing LS-SVM parameters is proposed, which enhances the generalization performance of this model. The experimental results show that this model can achieve higher identification precision with a reasonably small size of training sample sets and is more suitable to predict the depth of underwater welding penetration on-line than back propagation neural networks (BPNN).