The intelligent detection method study of PQFP solder joint defects based on improved neural network

Abstract

The quality of solder joint has a direct effect on the reliability of products, while the detection of solder joint defects plays an important part in the inspection of solder joint quality. So scientific methods should be selected to achieve the intelligent detection of solder joint defects. In this paper, a forward neural network with BP learning algorithm is introduced according to the relationship between geometric features vector of PQFP solder joint and solder joint defects. During training neural network with standard BP algorithm, there are some problems such as slow convergence and easy to trap into local minimum of the error function, etc. So genetic algorithm is brought in the neural network training, the specific approach is that firstly uses genetic algorithm to optimize the connection weights and thresholds of neural network, then trains the network with BP algorithm again. The results show that the improved training method can accelerate the convergence process and reduce the training error to better the network performance. Therefore, it is helpful to improve the classification of solder joint defects by applying this method in the detection of complex solder joint defects.