Laboratorní rotační model pro účely rozpoznávání předmětů

Abstract

This thesis focuses on the comprehensive design and implementation of a laboratory workstation for automated classification of machine components using deep learning methods. The theoretical part analyses the principles of convolutional neural networks and provides a detailed description of the AlexNet and GoogLeNet architectures. The practical output is a mechatronic rotational platform model driven by a DC motor with an L298N H-bridge. Control and data acquisition are managed by a Raspberry Pi 5 unit in conjunction with MATLAB, featuring a custom application developed in App Designer for real-time monitoring and control. The experimental section focuses on creating an original dataset of nine component types and subsequently comparing classification accuracy and computational performance among AlexNet, GoogLeNet, and MobileNetV2 networks. The result is a functional system achieving high recognition accuracy in real-time.