Abstract

Printed Circuit Boards (PCBs) are fundamental to the operation of a wide array of electronic devices, from consumer electronics to sophisticated industrial machinery. Given this pivotal role, quality control and fault detection are especially significant, as they are essential for ensuring the devices' long-term reliability and efficiency. To address this, the thesis explores advancements in fault detection and quality control methods for PCBs, with a focus on Machine Learning (ML) and Deep Learning (DL) techniques. The study begins with an in-depth review of traditional approaches like visual and X-ray inspections, then delves into modern, data-driven methods, such as automated anomaly detection in PCB manufacturing using tabular datasets. The core of the thesis is divided into three specific tasks: firstly, applying ML and DL models for anomaly detection in PCBs, particularly focusing on solder-pasting issues and the challenges posed by imbalanced datasets; secondly, predicting human inspection labels through specially designed tabular models like TabNet; and thirdly, implementing multi-classification methods to automate repair labeling on PCBs. The study is structured to offer a comprehensive view, beginning with background information, followed by the methodology and results of each task, and concluding with a summary and directions for future research. Through this systematic approach, the research not only provides new insights into the capabilities and limitations of existing fault detection techniques but also sets the stage for more intelligent and efficient systems in PCB manufacturing and quality control.