Desarrollo de sistema de banda transportadora para la detección de defectos en conservas enlatadas por medio de visión artificial

Abstract

The objective of this work is the design and implementation of a classification system for canned goods, through the use of artificial vision convolutional neural networks in Python 3. The system focuses on the use of a conveyor belt, which carries the cans to the chamber, for their respective analysis and classification based on the physical conditions of the can. For the implementation of this proposal, an artificial vision recognition system is used through the use of TensorFlow, for the respective analysis and correct operation of the system. The AI is trained using a dataset, which provides all kinds of defects and viable conditions of the cans. Once the system evaluates the element, it will take an action to push the can through the use of a piston, if it is not in good condition, otherwise the can will continue its corresponding path along the conveyor belt. Finally, for the design and training of the AI, the Python 3 environment was used, due to the stability and performance it offers for the creation of different types of AI, which is executed on a Raspberry Pi 4b in order to reduce economic costs and optimize resources. For the creation of the belt, mathematical calculations and simulations were carried out using Matlab that allow determining the correct operation and viability.