Overview

Since the emergence of deep learning in 2012, factories have adapted machine learning technologies in order to optimize processes such as sorting. However, these solutions are often outdated and have a high monetary and environmental burden. Defect Detect aims to demonstrate the viability of using the MAX78000 ultra-low power AI microcontroller to perform these tasks at 10% of the typical cost and 1% of the typical power usage.

Our project will perform anomaly detection on a product as it comes down the conveyer belt. We will do this by utilizing images from a camera with an overhead view of the conveyer belt as well as a CNN algorithm that runs on the MAX78000 ultra-low power AI microcontroller. The defective items will be removed using a powerful vacuum which will extract then place said items in a separate bin. There will be a functionality to control the speed of the conveyer belt as needed from the display. Our performance will be measured at the end by how many of the defects we successfully identify and remove from the conveyer belt.

Block Diagrams

Hardware

Software

External Behavior Specification

Our project has the role of binary classification with one of the classes being objects and the other class being (defective objects). When the input of objects and defects run through the conveyor belt, the output should be close to perfection of division between the objects and defects. 

1. The project will have a camera that monitors the items on the conveyor belt and use a CNN algorithm to identify the anomaly objects.

2. Once an anomaly object is detected, the vacuum extraction system will be activated to remove the anomaly object.

3. When the belt is moving the encoder will be used to measure the speed of the belt. This will be used to measure the distance of where the identified objects are located and when we should turn on the vacuum.

4. Once there are too many items on the conveyor belt or the belt speed is too fast for the camera to accurately identify abnormal objects, the belt speed control system can be used to adjust(use a touch screen that increases or decreases) the belt speed to the right range.