Title:

Automating Defect Detection with Machine Learning

General Electric (GE) Aviation manufactures bladed disks (blisks) that must meet high quality standards. Currently during the first quality inspection stage, visual inspection is done manually, and human error can result in defective parts being missed. Our project aimed to increase manufacturing efficiency by automating the visual inspection process to find defective parts more effectively. To accomplish this a machine learning algorithm was trained on various blisk images that were collected from both defective and non-defective parts. For automated inspection, a structure was built that includes a motor-controlled turntable, and a rigid mount for a digital microscope. The blisk will sit upon the turntable and the digital microscope will capture an image at a common defect location. As the system operates, the microscope will take a single image and feed it to the machine learning algorithm, which will determine if a defect is present. If no defect is present, the motor will turn the blisk and another image will be captured and analyzed. This process will continue until the entire circumference has been scanned, and the blisk is marked as passing or failing the inspection. The initial machine learning algorithm was consistently able to distinguish between defective and non-defective parts, and a final dataset of images will be collected to optimize the accuracy of the model.

Name

Comment*

Comments are viewable only by submitter

Captcha*