Open circuit and short circuit of chip can greatly affect the performance of the circuit board. Existing fault detection methods have some shortcomings, such as narrow scope of application and low accuracy. The infrared images series of circuit board contains multiple category information of faults. Using its time domain features can improve the accuracy of fault diagnosis. Therefore, we propose a chip open/short circuit defects inspection method based on infrared images series. Firstly, the mean temperature series of the critical area of chip during the response process of the power-on procedure is recorded and smoothed by Savitzky Golay convolution smoothing method. The time domain features are extracted，which are optimized by principal component analysis. Then support vector machine classification model is constructed, whose parameters are optimized by particle swarm algorithm to effectively distinguish different types of circuit board defects. In order to prove the validness of the method proposed before, a variety of solder ball open/short circuit experiments are carried out on a CPU chip of circuit board. The research results show that the cross-validation classification accuracy of the optimized SVM model in the test dataset is as high as 96.90%, which proves the validness of the method for detecting the chip open/short defects in this paper.