Abstract
The SMT Machine in PCB Assembly has been improving rapidly in the past decades. However, there are still defects occurring during the working process. In general, the defects can be classified into four categories, such as component missing, misalignment, counter polarity, and solder defect. This paper focuses on chip resistor and chip capacity inspection. We apply the algorithms in the technical literatures, such as Gray statistics method, Pattern matching method, Extreme point method and two algorithms that we proposed to inspect the on-line SMD samples. Due to the fact that single algorithm cannot differentiate completely to determine whether the SMD samples are qualified and it is difficult to classify the defects of SMD samples; therefore, we integrate the advantages of these five algorithms by Neural Network. Then we estimate the inspection results in four indexes, which are False alarm rate, Fault missing rate, Incorrect flaw classification and inspection time. The goal is to design a complete inspection procedure in order to detect the defects more correctly and conveniently. After being verified by numerical simulations and experiments, there are five inputs of Neural Network that have better inspection result. With these five inputs, the False alarm rate and Fault missing rate are both equal to zero. Therefore, Neutral Network is better than single algorithm in inspection and classification of the defects of SMD sample.