Based on the principal component analysis-support vector machine (PCA-SVM) model, one method is proposed to predict the joint morphology with the near infrared radiation signal. The correlation between the change laws of signals and the weld formation morphology is investigated and the optimization of process parameters is realized. Six kinds of characteristic parameters of signals in time domain are extracted and the principal component analysis is carried out to obtain the comprehensive evaluation index of joint morphology. Based on the input characteristics of signals， the classification prediction is done by using the support vector machine. The results show that, the near infrared radiation signals can reflect the change of weld state during the welding process, the characteristic changes of different defects have great difference, and the clear recognition exists. The proposed prediction model can accurately identify weld appearance with accuracy up to 96.6%.