Create Your First Project
Start adding your projects to your portfolio. Click on "Manage Projects" to get started
Computer Vision - Object Detection Model
Project type
Machine Learning
Date
August 2022 - September 2022
Location
San Jose
Throughout the manufacturing process of tape drives, there are countless areas where machines can malfunction or certain operations performed on the tape drives failed. This leads to wasted resources, time, and ultimately money lost for IBM.
One such process involves a wire bond machine making 80+ microscopic bonds between metal plates on the tape drive. The wire being used for bonding is half the diameter of a human hair, therefore requiring this bonding process to be extremely precise. Unfortunately, due to the fact that this is such a precise operation, there are a considerable amount of tape drives that go through the manufacturing process and experience defective bonds, and there is not a streamlined way to fix these. At the current moment, the technicians working in the manufacturing lines must put these drives under a 500x microscope and attempt to remove these defective bonds by hand, resulting in an incredibly low recovery rate, as doing this without further damaging the drive is difficult to do.
In order to help remedy this situation, the Tape Drive Department thought it would be best to develop an automated process to detect and remove faulty wire bonds. The first step in creating something that does this was to develop a object detection model that could identify defective bonds made by the machine, which was what I was tasked to do.
This was my first computer vision project, so this project went through several phases, many of them producing sub-optimal models. However, after exploring opencv in python and applying image pre-processing on the images used to train the machine learning model, I was able to develop a model through tensorflow that identified faulty bonds at a considerably high accuracy (95%).
The next step in the project would be to pipeline this machine learning model into an actual machine that would be used to take pictures of drives, identify the defective bonds, and somehow remove them. Considering this is not my expertise, this job has been given to somebody else.