Development of a Recognition System (ML) and Intelligent Control of Manual Operations in Industrial Production

This work is dedicated to the implementation of tracking algorithms in an intelligent system for monitoring manual operations in industrial production. The goal of the work is to enhance the quality of an application that monitors the actions of an assembly operator in real-time and promptly notifies them of errors. The research stand used for testing the system is an automated workstation, which includes a computer, monitor, projector, and Basler industrial cameras. The stand allows for monitoring the correctness of operations, identifying typical errors and violations such as incorrect sequence of actions, mistakes in component selection, and breaches of safety regulations. In this work, object tracking algorithms such as MOSSE, KCF, DeepSRDCF, AutoTrack, SiamRPN, SiamRPN++, and SiamBAN were studied and integrated with the YOLOv8 detection model. The tracking algorithms were implemented in the developed control scenarios, which include the transfer of screws to the assembly area, correction of classification errors, monitoring the assembly of chassis and motors, and correction of false detections. The results obtained showed high accuracy and speed of the KCF, AutoTrack, SiamRPN, and SiamBAN algorithms when applied to the task of intelligent manual operations control. The improved version of the application was successfully tested on the stand and is ready for use in real industrial environments.