The current research focuses on the problem of ensuring quality and efficiency control of the automatic welding machine’s welding process. Taking into account the characteristics of analysis of welding experimental data, the orthogonal experiment technique is used for obtaining sample data, and the variance analysis is utilized for finding parameters that have the largest impact on the results of the experiments. Using the mentioned findings, the three-layer BP neural network structure was developed. In terms of genetic algorithms, real-number coding was chosen as a method of encoding the individuals, and the optimization problem became the extreme value problem of the objective function, thus developing the GA-based BP neural network. Depending on the optimal combinations of process parameters found using orthogonal experiments in the cases of flash butt welding and welding joint stretching, the number of individuals involved in the iterative process is equal to 80. Using the GA-based BP neural network, one can find optimal combinations of process parameters of the weld time-value transformation in the range between 1.0 and 9.0. With the use of such neural networks, welding quality is guaranteed, since its error prediction performance falls within the range of (-0.03; 0.02). The optimal combinations of welding process parameters are found automatically.
