Adversarial autoencoder for detecting anomalies in soldered joints on printed circuit boards

Abstract

The inspection of solder joints on printed circuit boards is a difficult task becausedefects inside the joints cannot be observed directly. In addition, because anomalous samples arerarely obtained in a general anomaly detection situation, many methods use only normal samplesin the learning phase. However, sometimes a small number of anomalous samples are availablefor learning. We propose a method to improve performance using a small number of anomaloussamples for training in such situations. Specifically, our proposal is an anomaly detection methodusing an adversarial autoencoder (AAE) and Hotelling’s T-squared distribution. First, the AAElearns features of the solder joint following the standard Gaussian distribution from a large num-ber of normal samples and a small number of anomalous samples. Then, the anomaly score of asolder joint is calculated by Hotelling’s T-squared method from the features learned by the AAE.Finally, anomaly detection is performed by thresholding using this anomaly score. In experi-ments, we show that our method performs anomaly detection with few false positives in suchsituations. Moreover, we confirmed that our method outperforms the conventional method usinghandcrafted features and a one-class support vector machine