Improved Particle Filter-Based Modeling With Robotic GMAW for Weld Forming Prediction

Abstract

To solve the knowledge modelling problem of strong nonlinear characteristics such as multivariable, high coupling and stochastic interference in the GMAW process of hull structures, a clustering similarity particle filter (CSPF) method based on the consistency principle of the state trajectory is proposed to establish a knowledge model for welding dynamics. The modelling mechanism of the stochastic process is selected to construct a Hammerstein model with uncertain noise in the relationship between the welding parameters and geometric characteristics of the weld pool, and the Hammerstein model is identified using a step response test of constant specification and two parameter identification algorithms: the recursive least squares method and the final prediction error criterion. Thus, the state space equation of the accurate identification structure is provided for knowledge process modelling. Relying on the particle filter algorithm as the core of the modelling framework and theoretical analysis, SIS filtering and GPF prediction methods are adopted to obtain the combined trajectory formed by the current (original) and future multistage (modified) spatial state information, and the state similarity between the combined trajectory and the actual system trajectory is measured by the cluster analysis method. A new proposed distribution is generated under the guidance of similarity measurement to improve the particle degradation phenomenon, update the firstorder Markov process with new observation information to compensate and modify the importance weight calculation, replace the resampling strategy to eliminate the particle depletion problem, and then establish the welding knowledge model of state tracking and forming prediction. Through simulation and experimentation of hull structure GMAW, it is concluded that both the training effect of state tracking and the prediction accuracy of weld formation can meet the process requirements of ship welding by particle filter and its improved method. Meanwhile, integrated with the convergence theorem of the CSPF algorithm, compared with the standard particle filter and auxiliary particle filter, the training effect of state tracking is better in the application of the CSPF method in the process knowledge modelling of hull structure GMAW, and the prediction results of weld forming have the advantages of higher accuracy, stronger robustness, and better timeliness.