RESEARCH OF MECHATRONIC EXCAVATOR LOADING CONTROL SYSTEMS

Abstract

Excavation with a single-bucket excavator usually consists of three main operations: separation and capture of soil from the massif, its movement and subsequent placement. Calculations of productivity and setting the working dimensions of excavators during their design are directly related to the concept of excavator cut and the volume of excavated soil. To solve this problem, it was proposed to introduce a system for monitoring and alerting the condition of excavator teeth in real time. This system will allow employees to assess the condition of the teeth in a timely manner and take measures to replace them, which will prevent equipment downtime and save money and time. One of the options for implementing this system is to install Shovel Metrics equipment on excavators. This equipment allows you to receive information about the condition of excavator teeth using sensors mounted on the bucket of the excavator. Another option is to install a camera sensor on the excavator boom, develop a program code for reading from the image and a warning system. Implementation of a real-time monitoring and alert system for the condition of excavator teeth will solve the problem of unscheduled replacement of excavator teeth and their entry into the crushing plant. This will lead to significant economic benefits for the company, namely: reducing the number of equipment downtime; reducing the cost of unscheduled replacement of excavator teeth; reducing the likelihood of excavator teeth entering the crushing plant. This paper presents an artificial intelligence solution for excavator bucket control based on deep neural networks to obtain accurate and practically applicable data, which provides real-time status updates. The neural network acts as a pixel classifier and assigns a label to each pixel in the excavator bucket image. This classifier, combined with post-processing, provides a comprehensive inspection solution that can detect missing teeth, track wear, and detect fragmentation. This new deep neural network architecture replaces the previous algorithm, which used traditional computer vision techniques to extract information from incoming video footage while providing the specified product functions. Like all deep learning solutions, Shovel Metrics™ will only improve with large training data sets.