Defect Recognition of Resistance Spot Welding Based on Artificial Neural Network

摘要

The nugget size forecast model was built based on BP algorithm of Artificial Neural Network. The input parameters of the model are two characteristic number extract from electrode displacement curve. The output parameter of the model is nugget size. The model has three layers and the hidden layer have five nodes. The transfer function of hidden layer is Sigmoid function and the transfer function of output layer is linear function. Measured value and forecast value was analyzed by comparative method and the difference value between them was calculated. The results showed that 83% of the difference value is less than 1 millimeter. Based on the nugget size forecasted by the model, a method to identify incomplete fusion defect of resistance spot welding was suggested. In this method, 7 millimeter nugget size was regarded as criterion of recognition incomplete fusion defect of resistance spot welding. The result showed that recognition accuracy rate up to 94.3%.