Automation of the paper cup machine and detection method based on real-time image processing

The number of companies producing paper cups in our country is increasing day by day. Paper cup printing machines are produced in countries such as China, Korea, etc. There are also domestically produced machines in Turkey. These machines can print paper cups in sizes called 4, 7, 8, 12 oz. Machines are controlled by motors such as servo motors and asynchronous motors. With the help of sensors and heating resistances mounted on the machine, high quality cups with less waste loss are printed. In this study, a system that controls the previously produced cup press machine with servo motor, asynchronous motor and PLC, makes image processing based error detection and removes the faulty product is designed. For error detection, classification using the deep learning methods Yolo and Haarcascade algorithms through a real-time camera, and real-time object detection using the OpenCv library. The detected faulty cup is taken from the production line with a piston and thrown into a box. Commands are given to the servo motor driver with the commands from the sensors, the servo motor encoder and the user. Asynchronous drive parameters are set on the drive. All commands can be entered by the user on a screen and the number of products produced can be monitored over the system installed on the screen. In the artificial intelligence part of the study, the data obtained from the dataset we created beforehand for the detection of faulty cups were classified with the models of the YOLO algorithm in the Python language and Google Colab environment. After the models were trained, data from the camera were detected in real time using OpenCv libraries on Jupyter Lab (integrated development environment). The data trained using the Haarcascade algorithm were detected in real time using the OpenCv library on PyCharm. The effect of different deep learning methods on performance in the real-time system has been examined and successful results have been obtained. The most successful result was obtained with 90.8% accuracy in real-time application using Yolov5x architecture.