Autonomous prediction and planning of welding process parameters for joints with grooves using cascaded ANFIS

Autonomous prediction and planning of optimal welding process parameters for joints remain inadequately addressed in existing research. Typically, these parameters are determined through trial-and-error approaches, requiring extensive and tedious manual experiments, which not only incurs substantial labor and material costs, but also restricts development potential of digital welding systems. This paper introduces an innovative welding process prediction and planning framework that entails manual input of the welding speed only once, which can directly and autonomously output weld pass layout parameters and welding parameters for joints with symmetrical and asymmetric V-grooves. A five-cascade ANFIS architecture is designed in this framework. The ANFIS at each cascade is optimized by using gradient descent (GD), least square (LS), recursive least square (RLS), genetic algorithm (GA), particle swarm optimization (PSO) and simulated annealing (SA) to explore the optimal network model for each welding process parameter. The RMSE values for the number of weld layers N , the welding current in each pass I i and the welding voltage in each pass U i predicted by RLS-ANFIS are 0.000007, 2.324454 A and 0.000037 V, respectively. The RMSE values for the number of weld passes in each layer M i and the wire feed rate in each pass v i f predicted by LS-ANFIS are 0.000088 and 0.079375 m/min, respectively. The autonomous welding results of six joints with different geometries demonstrate that the proposed framework significantly outperforms existing welding process parameter determination methods in terms of inference speed, adaptability to groove specifications, operational simplicity and system autonomy.