Research On Quality Evaluation Of FSW Based On Self-organizing Incremental Learning Neural Network

Since the advent of friction stir welding,it has been widely used in aviation,aerospace,high-speed rail and other fields due to its high productivity and high joint strength.However,there are still certain quality problems in actual production because of the complicated welding process and many parameters.On the other hand,with the rise of artificial intelligence,the application of machine learning in the field of welding has gradually developed into a hot research direction.Under such a background,in order to improve the stability of aluminum alloy structural parts,and improve the digitization and informationization of the friction stir welding process,this paper designs a data acquisition system equipped with vario us sensors and establishs a quality evaluation system based on self-organizing incremental learning neural networkUsing the principle of two-color infrared temperature measurement,the paper obtains the temperature of the joint of the agitator head shoulder workpiece.The principle of laser distance measurement is used to obtain the pressure of the mixing head.The principle of the capacitor miniature pendulum is used to obtain the inclination angle of the stirring head.and The paper reads two parameters of the stirring head speed and the welding speed because the friction stir welding machine is authorized by the OPC system.Based on the above method,a friction stir welding parameter acquisition system was established by the paper,which provided a data source for subsequent machine learning.The paper studies the basic principle of self-organizing incremental learning neural network.The algorithm is implemented by MATLAB software programming and the corresponding model is established.In order to verify the performance of the algorithm,some artificial data sets were used to simulate and analyze the established machine learning model.Aiming at some shortcomings of self-organizing incremental learning neural network,an improved self-organizing incremental learning neural network algorithm is proposed by the paper.In order to further verify the reliability of the improved self-organizing incremental learning neural network algorithm,by studying the principle of the double support vector regression algorithm,the improved self-organized incremental learning neural network and dual support vector regression machine are established by using MATLAB programming software.Their respective models are used to compare the performance of the two algorithms.The experimental results show that the improved self-organized incremental learning neural network has similar simulation effects to the double support vector regression machine,but the training time of the self-organizing incremental learning neural network algorithm is much smaller than that of the dual support vector regression machine.Based on the accumulated simulation training results,the paper has established the friction stir welding quality evaluation system and realized the nonlinear mapping relationship from welding parameters to welding quality.The feasibil ity of the quality evaluation system was verified by using the actual data collected from the production.