A VQ-based approach to thermal image analysis for printed circuit boards diagnosis

This paper proposes a novel method to analyze the thermal image of a printed circuit board (PCB) for fault detection. In this method, a gold thermal image is first generated from the thermal images of the PCB in normal operation, and then compressed into a codebook with a certain number of codewords. Each codeword represents a block of image size four by four. Each block in the thermal image for the board under test (BUT) is then encoded in the same way. The codewords in the codebook are arranged in ascending order with respect to their mean values. Any abnormal functional block in BUT can be identified by comparing the codeword index with that of the corresponding block in the gold thermal image. The memory size for storing the templates for comparison is, thus, significantly reduced without diagnosis performance degradation. Also, there is not a necessity for feature extraction such as the feature-based diagnostic methods. In addition, an adaptive threshold criterion is proposed to improve the detection sensitivity. From the experimental results, this proposed method is demonstrated to be very effective in abnormal functional block identification for PCBs based on the thermal image. Furthermore, this method is highly modularized for hardware implementation and parallel realization to speed up the processing time.