The issues are considered of evaluation of the quality of resistance spot welding in real time, using the regression models and artificial neural networks. The structure of the neural network was studied, the optimal for resistance spot weldingthree-layer neural-network model 9-3-1 was developed, the inputs of which receive the signals of welding current, voltage on the electrodes and welding duration, and the welding nugget diameter is predicted at the output. Experimental verification of thetrained neural network model when welding a car sheet from carbon steel of 0,8 to 2 mm thickness, showed that when the welding parameters are changed in a broad range (from metal sticking together up to splashes), a reliable prediction of the weldingnugget diameter is provided. Due to that, the trained neural network, the program of which takes up about 3 KBytes, can be successfully used in the regulators of the resitance machines for spot welding of the car parts.
展开▼
