As the fourth industrial revolution is underway, scientists and engineers are focusing on integrating modern intelligent technology into traditional manufacturing and industrial practices, improving automation, self-monitoring, and analyzing and diagn...
As the fourth industrial revolution is underway, scientists and engineers are focusing on integrating modern intelligent technology into traditional manufacturing and industrial practices, improving automation, self-monitoring, and analyzing and diagnosing problems without manual intervention.
In the ever-developing consumer market and industry, people's requirements for computers are also changing. As a critical device responsible for data storage, hard disks have increasingly higher reading and storing speeds. That makes solid-state drives (SSD) quickly enter the user's field of vision. However, Damage to components on the internal printed circuit board (PCB) due to user misoperation or other reasons is a common cause of SSD failure. Although these faults can be repaired manually, there are usually a large number of components on the PCB, so it is very labor-intensive to detect component faults. Furthermore, some components such as capacitors and resistors only have slight color differences in the middle part in appearance, so in the case of brightness changes or electrothermal discoloration caused by long-term use, it is easy to cause detection errors.
The field of artificial intelligence is getting more and more attention. As a branch of artificial intelligence, deep learning has an increasing influence on object recognition and image separation. The rapid development of deep learning has made object recognition more and more efficient. In order to explore this, this thesis use the convolutional neural network to automatically locate and classify all the components on the PCB, guide the robotic arm to use the robotic arm to measure and find the faulty components automatically, and finally achieve the purpose of detecting the fault. In a further step, the faulty component is quickly locked. Through experiments, the total correct detection rate of all component categories is 99.28%, proving the method's effectiveness.